Distributed Reduplication (Linguistic Inquiry Monographs)

Linguistic Inquiry Monograph Fifty-Two

Distributed Reduplication

John Frampton

Distributed Reduplication

Linguistic Inquiry Monographs Samuel Jay Keyser, general editor

A complete list of books published in the Linguistic Inquiry Monographs series appears at the back of this book.

Distributed Reduplication

John Frampton

The MIT Press Cambridge, Massachusetts London, England

© 2009 Massachusetts Institute of Technology All rights reserved. No part of this book may be reproduced in any form by any electronic or mechanical means (including photocopying, recording, or information storage and retrieval) without permission in writing from the publisher. MIT Press books may be purchased at special quantity discounts for business or sales promotional use. For more information, please e-mail [email protected] or write to Special Sales Department, The MIT Press, 55 Hayward Street, Cambridge, MA 02142. This book was set in Times Roman by the author and was printed and bound in the United States of America. Library of Congress Cataloging-in-Publication Data Frampton, John. 1937– Distributed reduplication / John Frampton. p. cm. — (Linguistic inquiry monograph) Includes bibliographical references and index. ISBN 978-0-262-01326-0 (hardcover : alk. paper) — ISBN 978-0262-51353-1 (paperback : alk. paper) 1. Grammar, Comparative and general — Reduplication. I. Title P245.F73 2009 415—dc22 2009000726 10 9 8 7 6 5 4 3 2 1

Contents

Series Foreword Preface

ix

xi

Acknowledgments

xiii

Chapter 1 Introduction and Overview 1 Chapter 2 Transcription and Rule Application in Crossed Structures 15

2.1 The No Crossing Constraint 16

2.2 Rule Application in Structures with Long-Distance Geminates 21 2.3 The Retraction Condition 31 Chapter 3 The Morphophonology of Reduplication 33

3.1 Readjustment 35

3.2 Juncture Insertion is a Variety of Readjustment 37 3.3 Klamath Distributive and Intensive Reduplication 40 3.4 Erromangan: Evidence for Readjustment 43

vi

Contents

Chapter 4 Truncated Reduplication 51

4.1 Truncation at the Leading Edge (Reduplicant Truncation) 52 4.2 Trailing Truncation 57 4.3 Embedded Duplicants 60 4.4 Transposition, Metathesis, and Inxation 62

Chapter 5 Sources of Variation 67

5.1 Possible Juncture Insertion Rules 67

5.2 Possible Domains of Reduplicative Affixes 71 5.3 Shortcut Repair at the Reduplicant-Remnant Boundary 77 Chapter 6 Prosodic Adjustment 81

6.1 Onset Incorporation in K´hehe Unintensive Reduplication 81 6.2 Chumash CVC Reduplication and Apparent Countercyclicity 83 6.3 Augmentation of the Duplicant to a Heavy Syllable 88 6.4 C-Finality as a Secondary Prosodic Goal 105 6.5 Varieties of Syllable Prexation and Suffixation 111 6.6 The Onset-Coda Asymmetry and Moravcsik’s Generalization 113 6.7 Toward a Parametric Theory of Duplicant Adjustment 114

Chapter 7 Case Studies 117

7.1 Ndebele Unintensive Reduplication 117 7.2 Kinande Unintensive Reduplication 133

Contents

vii

7.3 Asheninca Campa Intensive Reduplication 140 7.4 Washo Plural Reduplication 144 7.5 Tohono O’odham Plural Reduplication 152 7.6 Consonant Copy in Temiar and Levantine Arabic 157 7.7 Sanskrit Intensive and Perfective Reduplication 161 7.8 Cyclicity and Multiple Reduplication in Lushootseed 173 7.9 Chaha Reduplication 177 Appendix A The NCC and the Retraction Condition 191 Appendix B Phonology On a Nonlinear Timing Tier 195

B.1 Segmental Phonology 195

B.2 Prosodic and Autosegmental Phonology 197 B.3 Linearization 199 B.4 Concatenative versus Nonconcatenative Morphology 201 Notes

205

References

213

Language index Author index Subjects index

221

223 225

Series Foreword

We are pleased to present the fty-second in the series Linguistic Inquiry Monographs. These monographs present new and original research beyond the scope of the article. We hope they will benet our eld by bringing to it perspectives that will stimulate further research and insight. Originally published in limited edition, the Linguistic Inquiry Monographs are now more widely available. This change is due to the great interest engendered by the series and by the needs of a growing readership. The editors thank the readers for their support and welcome suggestions about future directions for the series. Samuel Jay Keyser for the Editorial Board

Preface

The central goal of theoretical linguistics is to understand the nature of the mental computations carried out in producing the representations that the articulatory system acts on in producing speech. Although this book presents many new ideas about the computations that are put to use in reduplicative phonology, some fairly radical, the intent is quite conservative. I hope to provide evidence that the model of phonological computation developed by Chomsky and Halle in The Sound Patterns of English (1968) is fundamentally correct: surface forms are produced by the successive modication of underlying forms. I will call this model derivational phonology. The name is misleading, but in such common use that an attempt to change it might create even more confusion. It is misleading because the particular feature of the SPE model that distinguishes it is not that the computation employs derivations. All computations of any complexity go through a sequence of intermediate states, hence have intermediate representations and derivations. Rival theories, Optimality Theory (Prince and Smolensky, 1993), for example, compute the input-output relations by means of derivations that consist of progressive ltering of (very large) sets of phonological representations. What distinguishes derivational phonology is not that it employs derivations, but the nature of the intermediate states in the derivational computation. It proposes that intermediate states in the computation have the same general form as initial and nal states: the step-by-step derivation of the output proceeds through modication. A better name than derivational phonology might be “developmental phonology,” emphasizing that the issue is the progressive change of a single entity. Derivational phonology shares with developmental biology the understanding that certain features of the current state of an organism/representation can only be understood in terms of the state of the organism/representation at earlier stages in

xii

Preface

its development. A thorough and convincing demonstration of the adequacy of derivational phonology with respect to reduplicative phenomena has particular relevance at this time because many researchers still accept the claims of McCarthy and Prince 1995 that derivational phonology is incapable in principle of analyzing reduplication. Inadequacies in the analyses of reduplication that were available at the time their paper appeared gave some grounds for thinking that this might be so. But recent derivational analyses of reduplication, Raimy 2000a in particular, have shown that McCarthy and Prince’s criticisms of derivational phonology (at least insofar as reduplication is concerned) were only valid criticisms of particular proposals, not demonstrations of insoluble obstacles to a derivational analysis of reduplication. My hope is that this book will make this point more thoroughly and convincingly. Although particular reduplicative processes in many languages are discussed in chapters 1–6, which develop the general theory, the book closes with an extensive sequence of detailed case studies. The wide range of case studies is not simply intended to show off the accomplishments of Distributed Reduplication. It is intended to be evidence that the theory is true. It is a relatively easy task to show that some purported “general principles” successfully account for some particular phenomena in some particular language. It is much more difficult to show that general principles are truly general and can serve as the basis for analyzing widely different phenomena (within the range that UG makes possible) in very different languages. This book was completed in 2005, but production problems delayed its publication. I apologize to the authors of more recent work that I have not been able to revise the text to account for that work. I am not aware of any work that would necessitate major changes.

Acknowledgments

There are many people to thank for making this book possible. Above all, it would not have been possible without the ongoing support of Morris Halle. As the present book grew by successive iteration from a modest discussion paper written ve years ago, he carefully read and commented on each version. Arguing with him over that period about the matters the book discusses has been a real pleasure. Since there are several key aspects of the analysis that I have still not managed to convince him of, I have many more years of spirited argument to look forward to. Issues of computation and representation that Sam Gutmann and I have discussed over the last dozen years underlie the entire approach taken here. In addition to helping develop the perspective, he contributed many specic comments, as well as encouragement and support. Since I devote several pages in what follows to showing that he did not get it right, I am glad to have the opportunity at the outset to acknowledge the contribution of Eric Raimy. His dissertation showed that a derivational solution to the problems posed by reduplication was possible. It reminded us that most of the crucial advances in phonological theory are advances in understanding the structure of phonological representations. Thanks to audiences at MIT’s “Morph Beer” and Phonology Circle, the Chicago Linguistic Society, and the 2002 Graz Reduplication Conference for their interest in and comments on earlier versions of portions of this material. Thanks also to many individuals for helpful discussion and other assistance: Sylvain Bromberger, Terry Crowley, Stuart Davis, Justin Fitzpatrick, Larry Hagberg, Ken Hale, Sheldon Harrison, Mashudi Bin Kader, Michael Kenstowicz, Brett Kessler, Alec Marantz, Ngessimo Mutaka, Andrew Nevins, Eric Raimy, Galen

xiv

Acknowledgments

Sibanda, Suzanne Urbanczyk, and Alan Yu. I also owe a debt to the many people responsible for developing PSTricks, the Tex software that allowed me to depict the autosegmental representations that are fundamental to the theory that I propose. Herbert Voss, its current maintainer and developer, and Timothy Van Zandt, the original author of PSTricks, deserve special thanks. I have long thought that the difficulty in depicting complex strutures has had a negative impact on the development of theories of linguistic computation. Advances in typesetting tools will help free up linguists to pursue more thoroughly autosegmental approaches to phonology. I also owe a debt to Rei Fukui, who constructed the phonetic fonts (TIPA) which I relied on.

Chapter 1 Introduction and Overview

Copying is ubiquitous in morphophonology. Every time a lexical item enters a computation, its exponent must be copied from the lexicon to the active workspace. It would be remarkable if morphophonology reserved its copying capability for copying from the lexicon. Indeed, it does not. Many languages use copying as a way to mark word inection, similar in essence to the way that English phonology uses its vowel-lowering capability to modify sing to sang as a way to mark past tense. In Mokilese, for example, the verb root wadek ‘read’ inected for progressive aspect is wadwadek. In Manam, the adjectival form of the verb salaga ‘be warm’ is salagalaga. One obvious question is how the portion of the base that is doubled is chosen. A less obvious question is the nature of the copying process. Wilbur (1973) demonstrated that phonological rules sometimes apply in unexpected ways in reduplicated forms. Mester (1986) made signicant progress in showing how these apparent anomalies could result from the exceptional autosegmental structures involved in doubling a portion of the base, but this line of inquiry was cut short. In the early 1980s, work in reduplication was providing rich evidence for the important role of prosody in various morphological processes (McCarthy and Prince 1986). With respect to reduplication, this work was directed at the question of what was copied and provided important insights. Unfortunately, the project of understanding the autosegmental structure of the representations involved in copying was never integrated with the project of understanding the various roles that prosody can play in morphology. A close analysis of the copying process was largely abandoned in favor of prosodic analysis. In my view, this was a mistake. Prosody does play a crucial role in many reduplication processes, and it is impossible to fully understand the phenomenon of reduplication without understanding the ways prosody can affect structure building.

2

Chapter 1

But the effect of prosody is to shape the construction of autosegmental representations. Without understanding the unusual autosegmental representations that are built in carrying out reduplication, it is impossible to understand how phonology computes reduplicated forms. My hope is that this book can show how to continue the development of the analysis of copying that was cut short by the development of Prosodic Morphology in a way that allows the insights of Prosodic Morpholoty to be incorporated into the theory. Ultimately, of course, the value of the analysis of phonological phenomena is determined by the light shed on basic questions of structure and the algorithms that modify these structures in the course of phonological processing. As we will see, reduplication offers important evidence about the nature of autosegmental representations. In particular, the phenomenon of reduplication gives evidence that the No Crossing Constraint of autosegmental phonology is an interface condition, not something inherent in the structure of autosegmental representations. In place of a traditional introduction, this chapter will sketch out the main features of the theory that will be developed in detail in chapters 2–6, so that the reader will have some idea of where the project is heading and what the underlying assumptions are. This chapter is therefore mainly an overview. My hope is that starting from an overview will allow the reader to better appreciate the point of view and the relevance of the various components of the theory that are developed in the upcoming chapters. All of the examples and claims made in the overview will be revisited in later chapters, which will provide details, argumentation, and references to prior work. The presence of an inectional morpheme in underlying word structure is often expressed by the concatenation of a prex or suffix characteristic of the morpheme. This mode of morpheme realization, concatenative morphology, is often fairly transparent, in that the surface word can be broken up into pieces that correspond to pieces of the underlying syntax. But there is another important mode of morpheme realization. Some inectional morphemes are realized not by concatenation of a prex or suffix, but by a transformation of the stem in a fashion characteristic of the morpheme. This is nonconcatenative morphology. German, for example, uses noun root umlaut to realize ¨ ‘apple/apples’) in some contexts. a plural morpheme (Apfel/Apfel Many languages have morphemes whose realization involves doubling a portion of the stem. The presence of the morpheme in under-

Introduction and Overview

3

lying word structure is inferred from the appearance at the surface of the reduplication pattern characteristic of the morpheme. Mangarayi (Merlan 1982), for instance, uses reduplication to mark pluralization. (1)

Root gal.ugu waNgij jimgam

Plural g-al.-al.ugu w-aNg-aNgij j-img-imgam

‘poor things’ ‘children’ ‘knowledgeable people’

The copying process is simple to characterize: “starting to the left of the rst vowel, copy up to the left of the second vowel.” The material produced by copying (boldface above), called the reduplicant, is adjacent to the original, often called the base. (Here and throughout this book, unless there is a statement to the contrary, hyphenation and/or boldface in reduplicated forms is intended only to help the reader understand the copying process, not to indicate phonological or morphological boundaries, which may or may not be present.) The Mangarayi example is typical. Aside from a few cases in which roots have inherent reduplication, reduplication is the surface manifestation of nonconcatenative inectional morphology. The issues that need to be addressed under this view of reduplication are the modication that the stem undergoes at the point of morpheme realization, and the transformations that this modied stem undergoes in generating a surface form. Ilocano uses what Hayes and Abad (1989) call “light reduplication” in conjunction with the prex Pagin- to build a form with the meaning ‘to pretend to . . . ’. (2)

d´aPit s´aNit trab´aho

‘to sew’ ‘to cry’ ‘to work’

Pagin-da-d´aPit Pagin-sa-s´aNit Pagin-tra-trab´aho

‘to pretend to sew’ ‘to pretend to cry’ ‘to pretend to work’

The prex is concatenated and the portion of the stem from its left edge up to and including its rst vowel is copied to the left. Realizing a morpheme by a combination of affix concatenation and stem transformation is not uncommon. In the sell/sold tense alternation in English, for example, the presence of the past-tense morpheme is manifested not only by concatenation of the suffix -d , but also by lowering the vowel of the root. In other tense alternations, sing/sang for instance, the presence of the past-tense morpheme has a purely nonconcatenative manifestation: only vowel modication occurs, with no prex or suffix. This is analogous to Mangarayi pluralization (1) in which the stem is transformed, but no prex or suffix is concatenated.

4

Chapter 1

Copying is often assumed to be a unitary operation, but it can be decomposed into component suboperations. This decomposition is crucial to what follows, since much of the complexity of reduplicative phenomena will be shown to be the effect of phonological operations that intervene between the various components of copying. Three distinct steps in the copying operation are proposed here. The rst step is the direct effect of morpheme realization. The next two steps take place in the phonology, as reactions to the initial modication of the stem by the morphology. How is copying decomposed? The most remarkable example of a copying-like process in the biological world is the transcription of genes to RNA.1 In the simplest case, a gene is a subsequence of a long DNA sequence. More complex cases involve multiple subsequences with intrusions that are truncated when transcription takes place. Transcription is a two-step process. First, a marker, called a “transcription factor,” is attached to the DNA at the beginning of the gene 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 the attachment of a transcription factor to the DNA and transcribing marked DNA are independent mechanisms, carried out by different biological systems. I will attempt to establish in this book that reduplication, and copying generally, is organized in roughly the same way, as a multistage process. In place of the attachment of transcription factors to the DNA, junctures (called t-junctures, or transcription junctures) are inserted into the timing tier by the morphology. The transcription mechanism, which operates in the phonology, is triggered by these t-junctures to carry out transcription, just as the DNA transcription mechanism is triggered by transcription factors to carry out its biological copying. The t-junctures are interpreted as instructions that trigger and guide duplication and/or truncation. The decomposition of copying into one stage in which junctures are inserted into a linear string and later one in which operations actually carry out the reproduction is familiar from another source: repetition blocks in music. Part of the standard repertoire of signs and symbols and . They are used in writing music scores are repeat marks: instructions to be used in the translation of the score from an abstract representation to a concrete physical performance. There is a strong similarity between a music score and a phonological representation.

Introduction and Overview

5

Goldsmith (1976), for example, describes an autosegmental representation as an “orchestral score.” Both a music score and an autosegmental representation are abstract representations that can be, but need not be, translated into a physical performance.2 The initial operation of the full copying process is Juncture Insertion, which is the immediate consequence of morpheme realization. In the Mangarayi examples (1), for instance, the morphology inserts [ to the left of the rst vowel, and ] to the left of the second vowel. Two further phonological operations complete the copying process, as shown in (3), using one of the examples in (1). These operations are the response of the phonology to the introduction of t-junctures into the representation. Since the theory distributes the responsibility for reduplication over multiple suboperations, it is called Distributed Reduplication (DR). (3)

Distributed reduplication × × × × × ×

Juncture Insertion

−→

(morphology)

×[× × ×]× × j i m g a m

j i m g a m Transcription

−→

(phonology)

× × × × × × × × × j

NCC Repair

−→

(phonology)

i m g a m

× × × × × × × × × j i m g i m g a m

Transcription rst copies the timing tier and its phonemic associations, removing the t-junctures. If it is assumed that t-junctures are uninterpretable at the interface with phonetics, transcription can be regarded as a repair operation. Transcription produces violations of the No Crossing Constraint (NCC). It is further assumed that the NCC is an interface condition (following Sagey 1988), so a repair operation of some kind must follow that produces representations that are acceptable at the phonology-phonetics interface. NCC Repair of some sort (phoneme ssion in this case) must therefore follow Transcription. Aside from unpackaging various components of the copying process into separate operations, there are two key innovations in the proposed architecture: (1) the delimiters of the base ([ and ]) are considered to be real phonological objects, which are therefore subject to manipulation by the phonology; and (2) there is a stage in the derivation in

6

Chapter 1

which geminate structure links reduplicant and base phonemes. Because t-junctures can be manipulated by the phonology, a delimiter can be shifted to adjust the content of the base, for example. This will prove to be crucial in understanding how prosodic conditions can affect reduplication and allows some key insights of Prosodic Morphology (McCarthy and Prince 1986) to be incorporated into DR. Because Transcription produces a representation with multiple geminates, we expect that the well-known conditions on rule application to geminates will affect rules that apply after Transcription, but before NCC Repair. This will prove crucial in understanding what otherwise appear to be anomalous results of rule application to reduplicative forms (so-called overapplication and underapplication). The overall architecture can be pictured as follows: (4)

Juncture Insertion

Transcription

Rules can intervene here that adjust the prosodic shape of the projected reduplicant to a target prosodic condition.

NCC Repair

Segmental rules that intervene here can be affected by conditions on applying rules to geminates.

There is a long tradition in studies of reduplication, beginning with Marantz’s (1982) pioneering work, that views reduplication as essentially prexal or suffixal. An underspecied affix is concatenated with the stem (or perhaps inxed into the stem) and rules of association in the phonology ll out the unspecied aspects of the affix using the raw material provided by the stem. So, for instance, the derivation of the Mokilese progressive form wadwadek would be (5)

morphology

wadek

−→

phonology

-wadek

σ

−→

wad σ-wadek

The underspecied affix is called a reduplicative template or simply a template. In (5), the template is a syllable (specied to be bimoraic). Enough material is then associated in the phonology with the template to ll it with a bimoraic syllable. There is a large literature devoted to questions of how templates are specied and how association is carried out. DR does not view the morphological operation that initiates the reduplication process as concatenative morphology. Although the effect of reduplication in examples like (5) is to produce a string that looks

Introduction and Overview

7

like a prex at the surface, its origin in the morphology is not prexal. Its origin is Juncture Insertion. DR therefore views the morphology of reduplication as of the sing/sang variety, not of the like/like-d variety. DR follows Kitagawa 1987 and Raimy 2000a in this. Kitagawa rejects affixal accounts and proposes that “a morphological rule of reduplication three-dimensionally copies [my emphasis] the relevant CV skeleton and its association lines directly from the base.”3 Raimy proposes that the morphological operation leading to doubling at the surface is an alteration in the precedence relations between the timing slots in the stem. I turn now to consider the prosodic shape of the reduplicant. Ilocano plural reduplication, which was analyzed by Hayes and Abad (1989), illustrates the effect of target prosody. (6)

Root kald´ıN p´usa ˇjy´anitor

Plural kal-kald´ıN pus-p´usa ˇjyan-ˇjy´anitor

‘goats’ ‘cats’ ‘janitors’

One might try to characterize the base as extending from the left edge of the stem to the rst timing slot following the leftmost vowel. That is less than elegant, but does suffices for the examples in (6). Hayes and Abad demonstrate, however, that the copying process in Ilocano plural reduplication must be characterized in terms of the prosodic weight of the reduplicant: something like “starting at the left edge, copy enough material to form a heavy syllable.” CVC syllables are bimoraic in Ilocano. Hayes and Abad’s argument that the specication of the reduplicant as a heavy syllable shapes the reduplication process will be put off until chapter 6 since it relies on the analysis of various complexities and special cases that require some preparation. But I assume it here for the purposes of the overview. DR analyzes (6) in the following way. Juncture insertion by the morphology is the same as in light reduplication; [ is inserted at the left edge of the stem and ] after the rst vowel. But a Prosodic Adjustment operation applies before Transcription to adjust the base so that a bimoraic reduplicant is produced. σμ σμμ JncIns PrAdj (7) × × × × −→ [ × × ] × × −→ [ × × × ] × p u s a

p u s a

p u s a

8

Chapter 1

In this case, prosodic adjustment consists of shifting the ]-juncture to the right. Transcription and NCC Repair follow, yielding pus-pusa as desired. Although the reduplicant does not exist until Transcription creates it, its projected makeup is known before Transcription applies and the computation can use this information. The computation is autosegmental. The prosody of the projected reduplicant and the prosody of the stem are computed on different planes. σμ σμμ σ σ σ σ (8)

[× ×]× × p u s a

PrAdj

−→

[× × ×]× p u s a

The moraic count of syllables on the stem tier is not shown because it is irrelevant to the computation. In fact, the syllable structure of the stem is irrelevant to the prosodic adjustment rule. The two tiers in (8) are shown only to illustrate how reduplicant prosody can be computed in parallel with stem prosody, not to indicate that prosodic adjustment in Ilocano plural reduplication depends on the stem syllable structure. The representations (8) bear some similarity to the structures proposed in paraxal theories of reduplication (Clements 1985; Mester 1986; Uhrbach 1985). Parax theory assumes that reduplication is template affixation, but introduces the idea that in addition to prexation, suffixation, and inxation, there is a fourth category of affixation, parallel affixation, or paraxation.4 The template is rst lled out, then the parallel structure is linearized. DR builds the projected reduplicant not as a parallel structure, but as a substructure of the stem. The parallel computation is not parallel words, but two parallel prosodic analyses of the same material. I can now illustrate the effect of geminates on the application of phonological rules that intervene between Transcription and NCC Repair. One of the best known examples in the study of reduplication is the interaction of nasalization and reduplication in Malay (Onn 1976). Malay uses total stem reduplication for both plural and intensive inection (often with idiomatic meanings). The reduplicated form of aN˜an ‘reverie’ is ˜aN˜an˜aN˜an ‘ambition’. Malay has a rule of rightward nasalization, that nasalizes vowels that follow nasals (under locality conditions which will be discussed later). The great interest in the example, which Onn highlighted, is the nasalization of the initial vowel

Introduction and Overview

9

in the reduplicated form. The initial a does not follow a nasal in either the unreduplicated or reduplicated form. Naively, one might expect aN˜anaN˜an or aN˜an˜aN˜an, depending on the ordering of nasalization and reduplication. There does not appear to be any way for the initial vowel to nasalize. The architecture in (4) clears up the mystery. The morphology inserts [ at the left edge of the stem and ] at the right edge. Transcription copies the timing slots and their associations. (9)

× × × × a N a n

JncIns

−→

Trscr

[ × × × × ] −→ × × × × × × × × a N a n

a N a n

After Transcription, nasalization applies before NCC Repair. (10)

Nasalization

−→

× × × × × × × × ˜a N ˜a n

NCC Repair

−→

× × × × × × × × ˜a N ˜a n ˜a N ˜a n

One of the timing slots associated with the initial a in the nal representation in (9) immediately follows a timing slot linked to a nasal. The other timing slot linked to a does not. It is known from work on the application of rules to geminates that although the application of some rules to geminates requires all the timing slots associated with the geminate to satisfy the structural condition for rule application, other rules apply if any of the timing slots associated with the geminate satisfy the structural condition. Assuming that the Malay nasalization rule requires only one of the timing slots linked to a to follow a timing slot linked to a nasal, the derivation (10) results. There would be a different outcome if nasalization required that all timing slots associated with a geminate vowel follow timing slots linked to nasals. If nasalization did not apply again later in the derivation, and intervened between Transcription and NCC Repair as in (9) and (10), then aN˜anaN˜an would result. Instead of a surprising nasalization of the initial vowel, there would be a surprising failure to nasalize a medial vowel. Wilbur (1973), in her inuential study of anomalies in rule application to reduplicative forms, rst called attention to effects of this kind, calling the two kinds of deviations from naive expectations rule overapplication and rule underapplication. Accounting for overapplication and underapplication has been a central problem for theories of reduplication. DR explains many instances of underapplication and

10

Chapter 1

overapplication as effects of conditions on rule application to geminates. Other instances, as we will see, have their origin in operations that apply after Juncture Insertion, but before Transcription. A closer look at the application of phonological operations on structures with crossing violations (called crossed structures in what follows as a descriptive convenience) will reveal some subtleties. Nasalization is generally thought to be a feature-spreading operation. Although the NCC is not a derivational constraint on Transcription, the NCC (or something like it) does appear to constrain feature spreading, as a derivational constraint. A thorough discussion of feature spreading in the structures produced by Transcription is therefore required. Chapter 2 undertakes this, examining the No Crossing Constraint, constraints on feature spreading, and the application of phonological rules in structures with “long-distance geminates.” McCarthy’s work on nonconcatenative Semitic morphology, McCarthy 1986 in particular, plays a central role. It will be shown that known constraints on feature spreading imply that feature spreading can take place in crossed structures only in very special environments, accounting for why attested instances of overapplication are rare. This brief overview will close with a sketch of how some other aspects of Prosodic Morphology are adapted to DR. One of the accomplishments of Prosodic Morphology was to provide an answer to the puzzling absence of attested examples of syllable copy reduplication. Patterns 1 and 2 below are widely attested. But Pattern 3, in which the initial syllable is copied, is unattested. (11)

Stem gin.dal gi.dal gi:.dal

Pattern 1 gi-gin.dal gi-gi.dal gi-gi:.dal

Pattern 2 gin-gin.dal gid-gi.dal gi:-gi:.dal

*Pattern 3 gin-gin.dal gi-gi.dal gi:-gi:.dal

Moravcsik (1978), in her important worldwide survey of the reduplication patterns known at that time, had concluded that prosodic constituency does not play a role in delineating the base, the material that is copied in reduplication. Later work uncovered reduplication patterns showing that this conclusion was too strong, but the failure to nd instances of syllable copy reduplication showed that Moravcsik’s conclusion did contain an important kernel of truth. If prosodic constituents of the base could be freely singled out for reduplication, syllable copy reduplication would be expected. Marantz (1982) highlighted the example of plural reduplication in

Introduction and Overview

11

Yidiny (Nash 1979, 1980) as a counterexample to Moravcsik’s conclusion. It is exemplied in (12). (12)

Stem gin.dal.ba mu.la.ri

Plural gindal-gin.dal.ba mula-mu.la.ri

‘species of lizard’ ‘initiated man’

Determining the substring of the stem that is copied (i.e., its rst two syllables) requires reference to the prosodic structure of the stem, contradicting Moravcsik’s claim that prosodic properties of the stem are not used in delimiting the base of reduplicative copying. Marantz provided an analysis of Yidiny plural reduplication by introducing the important idea of a prosodic template, but he was then unable to explain why syllable reduplication does not occur. He concluded that it “leaves us with a mystery.” Prosodic Morphology explained the absence of cases of syllable copy reduplication, coupled with attested cases of foot reduplication, in the following way. First, following Nash 1980 and Broselow and McCarthy 1983, Yidiny reduplication was analyzed not as partial reduplication of the full stem, but total reduplication of a prosodically delineated substem of the stem. Importantly, it was assumed that only “wordlike” substems were possible. Prosodically, this implied foot-length substems, ruling out the specication of the initial syllable as a prosodic substem and explaining why total reduplication of a prosodic substem can produce Yidiny-type reduplication, but not syllable copy reduplication. Patterns 1 and 2 above were analyzed in terms of a prosodic template, partially following Marantz, but modifying his proposal for how the template was lled out by material from the stem. Marantz had assumed that syllables from the stem are used to ll out the foot template of the reduplicant. McCarthy and Prince assumed that the prosodic structure of the stem was invisible to the operation that lled out the template with stem material. Pattern 1 is then obtained by lling out a monomoraic syllable template and pattern 2 is obtained by lling out a bimoraic syllable template. The key to the Prosodic Morphology account is that prosodic shape is only relevant on the output side. Prosodic constituency on the input side, aside from the possible presence of a prosodic substem, is irrelevant to template satisfaction. DR adapts these ideas. First, it accepts the idea that reduplication can be dened with respect to a prosodic subword of the stem. More narrowly, an initial or nal foot of the stem can be designated as the domain to which Juncture Insertion applies. Second, it proposes that

12

Chapter 1

the landmarks in the stem that can be used to dene juncture insertion sites do not include prosodic constituent boundaries, but are restricted to notions like “before/after the leftmost/rightmost vowel” and “leftmost/rightmost edge.” Third, it proposes that the projected reduplicant can be adjusted so that it has desired prosodic characteristics. This allows the idea of template satisfaction to be incorporated into DR without the introduction of underspecied affixes. This adjustment is carried out by a Prosodic Adjustment rule that applies before Transcription. Crucially, prosodic adjustment is only an option in DR. Many reduplicative processes (Mangarayi and Yidiny plural reduplication, and Ilocano light reduplication, for example) do not employ prosodic adjustment. Also crucially, Prosodic Adjustment can attend to lowlevel prosodic issues, such as well-formed syllable structure, not just issues of prosodic weight. Most of what is to follow has already been previewed. Chapter 3 provides a full examination of the morphology that underlies DR, a version of Distributed Morphology. Chapter 4 develops the theory of truncated reduplication, in which the base is copied, subject to truncation in either the base or reduplicant. This is related to Steriade’s (1988) idea that reduplication consists of total stem reduplication coupled with truncation. The interaction of copying and truncation is complex, with a rich array of possibilities. It will turn out that many instances of metathesis as well as inxation are best analyzed as varieties of truncated reduplication. Chapter 5 discusses the range of variation in juncture insertion, with particular attention to showing why syllable copy reduplication is impossible. Chapter 6 develops the theory of prosodic adjustment. There is a detailed discussion of heavy-syllable prexal reduplication in the related Austronesian languages Mokilese, Ponapean, Agta, and Ilocano. The range of variation in juncture insertion coupled with the range of variation in prosodic adjustment is examined and it is demonstrated that there is a good t between the attested reduplication patterns and the predictions of DR. Chapter 7, which is by far the longest chapter, is a series of case studies. Most of the languages with well-studied complex reduplicative processes are discussed in some detail: Asheninca Campa, Chaha, Kinande, Lushootseed, Ndebele, Sanskrit, and Washo. The test of a theory is not the excellent job it does on its “poster-child” language (to borrow Donca Steriade’s colorful phrase), but its ability to make sense

Introduction and Overview

13

of a phenomenon as the manifestation of a computational ability that is put to use in different ways in different languages. The purpose of chapter 7 is to show that reduplicative phenomena in a wide variety of languages can be analyzed in detail on the basis of the proposals in chapters 2–6. Appendix B is a critique of Raimy 2000a. What I have called Distributed Reduplication is based on ideas of McCarthy and Steriade, mentioned above, as well as on the important insight of Odden and Odden 1985 that reduplication can be a multistage process.5 The only idea that DR shares with Raimy 2000a is the idea that the initial step in reduplication is nonconcatenative modication of the stem, but its implementation of this idea is much closer to Kitagawa 1987 than to Raimy. Nevertheless, Raimy’s return to core structural questions was the direct inspiration for the development of DR. His analysis therefore deserves special attention. Indirectly, the appendix serves to justify the different approach taken by DR.

Chapter 2 Transcription and Rule Application in Crossed Structures

It will be useful to have a name for the representations with longdistance geminates produced by Transcription, since such structures will be referred to frequently.1 They will be called crossed structures. Transcription is relatively straightforward. The only issue is directionality. Transcription can be to the right or the left. (13)

[ × ×] × ×

Right Transcription

−→

t a k i [ × ×] × × t a k i

× × × × × × t a

Left Transcription

−→

k i

× × × × × × t a k i

The newly created timing slots in (13) are boxed. If the initial representation has only timing slots and phonemic associations, which is all that is portrayed above, there is no distinction between the resulting structures. They are both × × × × × × k i t a Although the two outputs are identical, there is an argument to be made that (13b) is a simpler operation and that the default direction of transcription in this case is to the left. If more complex structures are considered, with morphological or prosodic structure, the argument from simplicity is much more decisive. If less impoverished structure is considered, there is a distinction between the structures that result from left and right transcription.2 Consider (14).

16

Chapter 2

σ (14)

a.

σ

[× ×]× × ×

Right Transcription

−→

t a l k i σ b.

σ

[× ×]× × ×

σ

σ

× × × × × × × t a

Left Transcription

−→

l k i σ

σ

× × × × × × ×

t a l k i t a l k i Right Transcription in (14a) is highly disruptive, requiring syllable reorganization, but Left Transcription is not. More for conceptual reasons and the sake of concreteness than because there is strong evidence one way or the other, I will assume that the unmarked choice is the one that is least disruptive to the structure of the stem. This choice is xed for the affix, so that if one direction of transcription produces a simpler outcome in most cases, that choice is assigned to the affix. If the duplicant, the material demarcated by the t-junctures, is aligned with one edge of the stem and not the other, the direction of transcription is toward that edge, at least as the unmarked choice. The notion “disruptive,” of course, is not clearly dened, so this should be taken as a preliminary statement. There can be conicts between disrupting morpheme structure and disrupting prosodic structure. Further discussion will be postponed until the issue presents itself concretely. 2.1 The No Crossing Constraint Some readers may be reluctant to consider representations like (15) seriously because they massively violate the the No Crossing Constraint (NCC), the ban on crossed association lines. (15)

× × × × × × × ×

a N a n If the role of the NCC in phonology is considered carefully, however, good grounds can be found for at least entertaining the possibility that structures like (15) are present at some underlying level. Constraints play various roles in derivational phonology. Derivational constraints restrict the application of rules. Rule application that would otherwise result in a violation of the constraint is blocked. Halle and Idsardi’s (1995) analysis of footing and stress is a good example of the extensive use of derivational constraints to control the application of

Transcription and Rule Application in Crossed Structures

17

a simple rule system. Constraints on the distribution of foot delimiters interact in that theory with general rules inserting foot delimiters to account for the stress facts of a large variety of languages. Most researchers, however, have not considered the NCC to be a derivational constraint. Summing up the rule formalism of tonal phonology that he proposes, Goldsmith (1990, 47) says, “If a rule is formulated to add a single association line, it can, in principle, cause a line-crossing situation. In this case, . . . the line that the rule adds remains, but the line that formerly existed is taken to be the offending line, and is automatically erased.” On this view, the NCC does not block rule application. Rules can apply even though their output violates it. A subsequent repair operation brings the structure into compliance with the NCC. For Goldsmith, repair is immediate (i.e., automatic), so that the structure with crossing violations is transitory. Nevertheless, it is clear that Goldsmith does not consider the NCC to be a derivational constraint. Much closer to the concerns of this chapter is the position taken by McCarthy in his groundbreaking work on nonconcatenative Semitic morphology. Example (16) is an example of the kind of morphology that McCarthy proposed for Semitic root-template association. It is (42) from McCarthy 1986. root affix (16)

/sb/root + μ

⇒

× × × × × × s i b

e

The affix μ species a template and vocalism. The combination of root and affix is partitioned into two units: the timing slots derived directly from the root consonants and the timing slots derived from the template vocalism. The CVCCVC template requires four consonants. The two consonants of the root are associated with the rst two consonant positions, left to right, and the last two are lled by multiple association of the rightmost root consonant. McCarthy implemented the idea by supposing that the stem and affix phonemes are placed on distinct autosegmental tiers. This provided a means of avoiding the NCC violation in (16). The idea is that the derivation proceeds as in (17), with the root and affix segments initially on distinct tiers, as shown in (17.1). The phonemes above and below the line should be viewed as being on different tiers. An operation called

18

Chapter 2

Tier Conation subsequently applies, eliminating split phoneme tiers, but creating NCC violations. Violations of the NCC are then removed by the operation Fission. Crucially, the fact that tier conation results in crossing violations does not block its application. For McCarthy 1986, as for Goldsmith 1990, the NCC is not assumed to be a derivational constraint. e i (17) 1.

Morphology:

/sb/root + template →

× × × × × × s

2.

Tier Conation:

→

b

× × × × × × e

s i b 3.

Fission:

→

× × × × × × s i

b

e b

McCarthy did not propose an explanation for why Tier Conation applies in (17.2). It was necessary to assume that it did in order to explain the fact that at late stages in the morphophonology, nonadjacent consonants do not act as if they are occurrences of the same phoneme. McCarthy implies that ssion applies more or less immediately to remove the NCC violations caused by Tier Conation, taking a position similar to Goldsmith’s. But he does not formally distinguish between output conditions that demand repair of violations at some point before the output interface and conditions that demand immediate repair. The picture he paints is that some phonology takes place while the root and affix tiers are split, then Tier Conation takes place at some point for some reason, with repair following immediately in order to bring the representation into compliance with the NCC. It is equivalent, and much more straightforward, to assume that there are no split tiers, and that NCC violations may persist in the phonology, repaired eventually but not necessarily immediately. It seems to me that the more straightforward approach was not taken because the importance of the NCC in restricting tone and feature spreading in some environments made it appear to be a strict constraint on derivation. The closest that McCarthy was willing to come to admitting NCC violations into phonology was to stipulate that the NCC could be temporarily vi-

Transcription and Rule Application in Crossed Structures

19

olated, but violations had to be immediately repaired. Like Goldsmith, McCarthy does not consider the question of immediate repair versus eventual repair. Certainly, no evidence was presented that repair must be immediate. Sagey (1988) argues that the NCC is an output condition on phonology. Whereas Goldsmith and McCarthy assume that NCC repair is immediate, undertaken as soon as NCC violations are created, Sagey does not. DR follows Sagey on this point. NCC violations are repaired at some point in the derivation, but not necessarily immediately after they are created. This has the important consequence that representations like the nal representation in (9) can persist long enough for phonological processes to take place while the structure has crossed association lines before the NCC violations are repaired by ssion. This is crucial in accounting for apparently anomalous nasalization in reduplicated forms in Malay and other similar examples, which will be examined later. In this book, I assume that the NCC is an interface condition, satised at the phonology-phonetics interface. The relevant point is that the NCC violations in (17) are not necessarily repaired immediately, so that phonological rules may apply before repair. As an interface condition, the NCC is not a constraint on rule application, but a constraint on possible grammars. I assume that the lookahead capabilities of rule application in derivational phonology are limited to what is needed for the imposition of derivational constraints. This allows the results of the computation of a potential output to gure in the conditions on rule application, but disallows looking further ahead into the consequences of rule application in determining if a rule applies. It follows that derivational steps (prior to the nal step) cannot be directly inuenced by interface conditions, except insofar as those requirements are incorporated into the design of the grammar. Grammars that produce outputs which which violate the NCC at the interface are rejected, not rule applications which produce outputs which will eventually violate the NCC. Marantz (1982, 446) briey considers reduplicative structures with crossing violations but quickly rejects the idea. His example (21) is repeated in (18). (We now understand that it would contain a geminate, but that is irrelevant to the crossing issue.) The structure in (18) is just a CV-slot version of the structure proposed here. Marantz rejected (18), saying (p. 446) that “association lines would cross in violation of the

20

Chapter 2

basic condition on autosegmental phonology [association lines never cross].” t a k k i (18)

C V C - C V C C V

Marantz adds in a footnote (note 8, p. 446) the argument that if crossed associations were allowed, then special stipulations would be needed to ensure that phenomena like mirror-image reduplication are impossible. As argued above, “the basic condition on autosegmental phonology” is not usually considered a derivational constraint. The second concern carries no weight. Derivational morphophonology does not operate by free generation and ltering.3 Transcription is not free association ltered by a system of well-formedness conditions on possible associations, but a core algorithm of the language faculty. The same mechanism that ensures that dog doesn’t come out of the lexicon as god ensures that total reduplication of dog produces dog-dog, not god-dog.4 It is hard to escape the conclusion that the essential obstacle to adopting (18) was the presumed inviolability of the constraint against crossed associations. Mester (1986) also comes close to the DR proposal, but is stopped by the NCC as well. In his framework, rather than (18), there would be the structure (19), with a later linearization operation aligning the ×-slots.5 × × × t a k k i (19)

× × × × ×

The structure in (19) is equivalent to t a k k i (20)

× × × ×  × × × × Tier1 Tier2

The step that DR takes is to insist that there is only one timing tier.6 Although Mester is able to make signicant progress in understanding a number of complex autosegmental effects on the basis of the multiple timing-tier representations that he proposes, it is impossible to explain phenomena like Malay an˜em→˜an˜em˜an˜em on this basis. For

Transcription and Rule Application in Crossed Structures

21

Mester, the relevant representation would be × × × × a n e m (21) × × × × Unless the linking of a to two timing slots persists past the point that the two timing tiers are aligned, there is no way to explain a→˜a. 2.2 Rule Application in Structures with Long-Distance Geminates Crossed structures can persist long enough for phonological processes to apply to them before NCC repair, but processes applying in crossed structures are constrained in two important ways by the special characteristics of these structures. First, geminate inalterability can prevent rule application that would otherwise be expected. Second, constraints on feature spreading are particularly restrictive because multiple linking gives multiple opportunities for intervention effects. 2.2.1 Geminate Inalterability The effects of geminate inalterability in Biblical Hebrew (BH) are wellknown and provide a good illustration of the issue. See Kenstowicz 1994, 410 for a particularly clear discussion. There is a spirantization process in BH that applies to postvocalic stops, but is blocked on geminate stops. For example, yiktob ‘write (imperfect)’ → yixtov, with both postvocalic stops spirantizing, but the postvocalic geminate in sappir ‘sapphire’ is unaffected by spirantization (*saffir). This is pictured in (22). (22)

a. × × × × × ×

× × × × × ×

k t o b p i r s a (k→x) (b→v) (*p→f ) In (22), an arrow points from the timing slot whose phonemic association provides the environment for rule application to the timing slot whose associated phoneme is the target of rule application. This notation will be used throughout this chapter, and only in this chapter, to help guide the reader through the complexities of rule application in crossed structures. Based on the work of Hayes (1986) and Schein and Steriade (1986), y

i

b.

22

Chapter 2

Kenstowicz concludes that the crucial factor is that only one of the timing slots associated with p in (22b) is in the proper environment for spirantization (i.e., postvocalic). Kenstowicz says that “two outcomes are possible a priori. We might expect the rule to “overapply” even though just one of the legs of the geminate satises it. Alternatively, the rule might be suspended even though one portion of the multiply linked representation does satisfy it” (p. 413). Kenstowicz does not use the term in this context, but we can say that the rule “underapplies” if the rule does not apply even though one leg of the geminate does satisfy the condition for rule application. I will call a timing slot and an associated phoneme an occurrence of the phoneme. The structural condition of a rule refers to occurrences of phonemes, not directly to phonemes. This is what gives rise to ambiguity and requires additional specication in order to determine how the rule applies to phonemes that are linked to multiple timing slots. Some phoneme-altering rules require that all occurrences of the phoneme satisfy the structural condition of the rule; others require only that at least one occurrence of the phoneme satises the structural condition. If there is an occurrence of a phoneme that satises the condition for the application of some rule and that phoneme has another occurrence that does not, then the rule will either overapply or underapply. Rules that underapply in this situation are said to be subject to geminate inalterability. Based on the preceding discussion, I conclude that spirantization in Biblical Hebrew requires all the timing slots associated with the potential spirantizing stop to be postvocalic (i.e., to follow a timing slot associated with a vowel). That is, spirantization in BH is subject to geminate inalterability. Kenstowicz points out that this is typical of the interaction of spirantization with geminate structures across languages: “Many languages spirantize postvocalic or intervocalic stops. In virtually every case, a geminate consonant resists spirantization” (p. 412). In Southern Paiute spirantization applies after reduplication, producing piNwa→piviNwa. There is nothing surprising here. First, piNwa→pipiNwa. Then after ssion, spirantization applies. (23)

[× ×]× × × → × × × × × × × → × × × × × × × p i Nw a

p i Nw a

p

i p i (p→v)

N w a

Transcription and Rule Application in Crossed Structures

23

If spirantization happened to be ordered so that it applied only to the intermediate structure above, the situation in (24) would result. (24)

× × × × × × × p

i

N

w

a

One of the occurrences of p is postvocalic, but one is not. Following Kenstowicz’s conclusion that spirantization is always subject to geminate inalterability, we would expect that spirantization would fail to apply in (24). DR therefore solves a problem that Kiparsky (2003) called attention to. He noted that there are no languages in which spirantization overapplies in examples similar to the Southern Paiute examples just discussed. In the analysis developed here, this is no surprise. It simply reects Kenstowicz’s observation that spirantization is quite generally subject to geminate inalterability. Kiparsky points out that this result is not consistent with Correspondence Theory, which has no mechanism other than stipulation to keep spirantization from overapplying to the reduplicant. In the theory proposed here, the following are structurally identical in relevant respects: (25)

b. × × × × × × ×

a. × × × × × × s

a

i p (*p→f )

p i (*p→v)

r

N

w

a

If progressive nasalization in Malay were subject to geminate inalterability, then nasalization of the initial a would not occur in (26) because only one of the occurrences of the initial a is in a nasalizing environment. (26) ×

×

×

×

×

×

×

×

a n ˜a m However, because Malay progressive nasalization is not subject to geminate inalterability, a does nasalize in (26). One of its occurrences is in a nasalizing environment. This is sufficient, even though another of its occurrences is not.

24

Chapter 2

2.2.2 Constraints on Feature Sharing If the NCC is conceived of as a constraint that tolerates temporary violation provided that repair follows (either immediately or eventually), it cannot block undesirable feature spreading. A feature could simply spread over any arbitrary intervening feature and, if a violation of the NCC results, the resulting structure could be repaired by feature ssion. Nevertheless, something like the NCC does appear to be at work in constraining feature spreading, at least in cases where the intervening feature has certain special properties. Since a precise understanding of at least certain aspects of the constraints on feature spreading will be necessary in order to understand how these constraints operate in the crossed structures created by reduplication, I consider them here. In cases in which features of a certain kind block certain varieties of feature spreading, the NCC is often held responsible. Halle (1995), based on work of Calabrese (1995), takes a nuanced position with respect to the application of the NCC in constraining feature spreading. In the case of a contrastive intervening feature, the NCC is taken to be a derivational constraint, blocking rule application.7 In the case of a noncontrastive intervening feature, the NCC can be freely violated, but violations are repaired. In effect, Halle takes the NCC to be a derivational constraint with respect to contrastive features and an interface condition with respect to noncontrastive features. He is not explicit about the timing of NCC repair (eventual or immediate), since it is not relevant to his concerns. Typically, spreading rules are stated in two parts. The rst part species the source, directionality, and target. This is given by statements like “spread [+back] to the left from vowels to consonants.” The second part gives the locality conditions on the spreading. One way of doing this is to specify the phonemes that are opaque to the spreading. I will assume that the locality conditions on spreading are given by constraints of the form (27)

* × ... ◦

× ...

×

α

◦

if α is opaque to F spreading.

[F]

That is, sharing F over α is blocked (derivationally) if α is opaque to F-spread.

Transcription and Rule Application in Crossed Structures

25

The symbols ◦ in (27) represent root nodes. Timing slots are not associated directly with features, but indirectly via root nodes. There is controversy about whether root nodes themselves are linked directly to segmental features or whether there are various intervening nodes. This depends on assumptions about feature geometry that are not directly relevant to the considerations here. I take the association to be direct, for reasons of simplicity of exposition. Nothing here will depend on this assumption. If a timing slot is associated with a root node that is associated with a feature F, then I will say that the timing slot has the feature F. This gives a general framework for considering feature spreading rules. Halle’s specic proposal was that a phoneme α is opaque to Fspreading if it has a contrastive feature on the same tier as F. In the case of 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 (27). (28)

Feature sharing between timing slots is blocked if an intervening timing slot is associated with a root node that is opaque to F-sharing.

It is implicit in (28) that the intervening timing slot is not also linked to the shared feature. There is no intention to prevent a feature from being shared by a string of consecutive timing slots. Principle (28) is obviously related to the NCC, particularly if the opacity is caused by the presence of features on the same tier as the shared feature. But the relation between (28) and the NCC is indirect. The NCC is an interface condition, a constraint on grammars, while Principle (28) is a constraint on rules. Grammars can adopt various strategies for ensuring that they produce output representations that satisfy interface conditions. One option is to repair defective representations by operations like Fission. Another is to ensure that rules do not produce defective outputs by imposing derivational constraints like (28). In this, Principle (28) is more like Fission than it is like the NCC itself. It is one (of many) strategies that can be adopted for ensuring that the NCC is satised at the interface. We will see later that although Fission is the most common way that the NCC violations that Transcription produces are repaired, other strategies are possible. Now that a precise statement of the form that constraints on feature spreading take has been proposed, the analysis of Malay nasalization in reduplicated forms can be considered more carefully. I will assume that

26

Chapter 2

progressive nasalization spreads [+Nas] to the right, targeting vowels and nasal consonants. Spreading a [+Nas] to consonants that are already [+Nas] has the consequence that a form like an˜em will have a single [+Nas] feature, shared between the nal three root nodes, as shown here: (29)

×

×

×

×

◦

◦

◦

◦

[−Nas]

[+Nas]

{a}

{N}

{˜a}

{n}

Locality is not an issue in spreading [+Nas] in the crossed structure (30) produced in the process of totally reduplicating (29). All timing slots come to share a single [+Nas] feature. (30)

×

×

×

×

◦ ×

◦

◦

◦

[−Nas]

[+Nas]

{a→˜a}

{N}

{˜a}

{n}

×

×

×

×

There are somewhat more complex reduplicated forms with unexpected nasalization that do raise the issue of locality. Onn (1976) gives the four examples in (31). (31) a. b. c. d.

root hama waNi aNan aNin

surface ham˜@ waN˜ı aN˜an aN˜en

‘germ’ ‘fragrant’ ‘reverie’ ‘wind’

h˜am˜@h˜am˜@ w˜aN˜ıw˜aN˜ı ˜aN˜an˜aN˜an ˜aN˜en˜aN˜en

‘germs’ ‘fragrant (intense)’ ‘ambition’ ‘unconrmed news’

The analysis of (31d) is the same in relevant details as the analysis of (31c). The analysis of (31a) and (31b) depends on the particularities of the locality of progressive nasalization in Malay. According to Onn (1976, 69), glides (w, y, h, and P) are transparent to nasal spread, but other consonants are not.8 He gives derivations like mewah→m˜ew˜ah and mayaN→m˜ay˜aN to illustrate the transparency of glides to progressive nasalization. There are no underlying nasal vowels in Malay. The analysis of waNi→w˜aN˜ıw˜aN˜ı is given in (32).

Transcription and Rule Application in Crossed Structures

(32)

×

×

×

×

◦

◦ ×

◦

◦

[−Nas]

[−Nas]

[+Nas]

×

×

27

×

×

{w} {a→˜a} {N} {˜ı} Six timing slots come to share a single [+Nas] feature. The six timing slots are not consecutive. The span is broken up by a timing slot associated 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 is unexpected vowel nasalization in the reduplicated form, there is a single nasal span broken up at most by intervening glides. The theory outlined here predicts this. Unfortunately, Onn did not provide examples to show that nasalization does not occur if a nonglide nonnasal consonant intervenes between the source of nasalization and the unexpectedly nasalized vowel. Kiparsky (2003) has tested such examples with a number of native speakers of the dialect that Onn describes and the results agree with the predictions made here. There is no nasalization. An analysis of one of Kiparsky’s examples is given in (33). It shows why harum→harum-h˜arum (*h˜arum-h˜arum). The locality condition on spreading disallows (33). (33)

×

×

×

×

×

◦

◦ ×

◦

◦

◦

×

×

×

×

×

[−Nas] [−Nas] [−Nas] [+Nas] [+Nas]

{h} {a→˜a} {r} {˜u} {m} This is disallowed because the consonant r intervenes in the disconnected span of 4 timing slots that would share a single nasal feature. It should be noted that blocking nasal spreading in the complex structure created by Transcription raises the possibility of underapplication of nasalization. Why isn’t the result harum→harum-harum, with neither a nasalized at the surface? If nasalization did not apply again after Transcription, this would be the result. Presumably, the observed nasalization in harum→harum-h˜arum occurs after NCC repair. Inkelas and Zoll (2002) point out that regressive velar assimilation

28

Chapter 2

at the reduplicant-stem boundary is common, but it never overapplies. For example, reduplication of the form kan→kaNkan is common, but kan→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 of Transcription. The medial vowel blocks the required place assimilation. (34)

×

×

×

◦

◦

◦ ×

[Velar]

{k}

×

×

×

[Coronal]

{a}

{n→N}

In (34), 4 timing slots come to share a velar place feature. But two instances of a intervene. Place sharing across a is not possible. Otherwise, we would expect nak→Nak. Place assimilation as shown in (34) is therefore impossible. Inkelas and Zoll’s intention is to completely discredit the idea of overapplication in reduplication, since the theory of reduplication that they propose has no explanation for it. But overapplication does occur, although only in very special circumstances, as predicted by DR. A number of conditions must be met: (1) NCC repair must be delayed long enough for phonological processes to take place in the complex structure produced by Transcription; (2) whatever phonological processes do apply to the unrepaired output of Transcription must not be subject to geminate inalterability; and (3) because of the crossed structures produced by Transcription, feature-spreading operations can apply in only relatively unusual situations. It should be no surprise that examples of overapplication are rare. Several additional examples are given in chapter 7 and one is discussed below. All of them involve feature-changing operations, not spreading rules. Malay nasalization is the only example I know of in which a spreading rule overapplies. 2.2.3 Korean Consecutive Reduplication Chung (1999) discusses various kinds of reduplication in Korean. One particularly interesting variety, which he calls “consecutive reduplication” (p. 170), provides another good example of the kind of rule overapplication that we have already seen in Malay. It is restricted to the stratum of Sino-Korean words, but there are many examples. See

Transcription and Rule Application in Crossed Structures

29

Kitagawa (1987) for discussion of a similar phenomenon in Mandarin Chinese. Both monosyllabic and bisyllabic roots reduplicate with intensive semantics. Reduplication of monosyllabic roots is unremarkable. The reduplication of bisyllabic roots is much more interesting: (35) a. b. c. d. e. f. g.

Root kikwe kimyo hy@nsak cason sipi sikak kuc@l

Reduplicated ki-ki-kwe-kwe ki-ki-myo-myo hy@n-hy@n-sak-s’ak ca-ca-son-son si-si-pi-pi si-si-kak-k’ak ku-ku-c@l-c@l

‘very strange’ ‘marvelous’ ‘all forms and colors’ ‘generation after generation’ ‘judgment’ ‘hourly’ ‘every phrase and sentence’

Each of the two syllables reduplicates independently. Since the initial syllable is reduplicated (see (35c) in particular), this looks like a counterexample to Moravcsik’s claim that there is no initial syllable reduplication. Unlike initial-syllable reduplication, there are several attested cases of nal-syllable reduplication. In chapter 6, I will show how this asymmetry follows from the assumptions of DR. What appears to be initial-syllable reduplication in (35) is reduplication of the remnant of nal syllable reduplication. In chapter 6, the relevant juncture insertion rules are discussed in detail and derivation leading up to the multiply linked structure produced by Transcription, the lefthand side of (38) below, is justied. Here I concentrate on the way phonological rules apply in the multiply linked structures produced by transcription. The only phonology at work in the examples in (35) is consonant tensing, which applies to a syllable-initial consonant that is preceded by an obstruent, producing s→s’ in (35c) and k→k’ in (35f). The phonological changes in the examples in (36) are much more interesting. The changes have been boldfaced. (36)

a. b. c. d.

hi.lak yu.lak u.lyaN lwe.lak

hi-hi-naN-nak yu-yu-naN-nak u-u-nyaN-nyaN nwe-rwe-naN-nak

‘rejoicing’ ‘quite willingly’ ‘very lonely’ ‘broad-minded’

Various phonological rules are involved in producing the boldfaced changes in (36). The crucial rule is (37a). (37)

a. Onset l→n following a noncoronal consonant coda. high vowel, else l→n. b. Word initially, l deletes /

30

Chapter 2

c. Onset l→r following an open syllable. d. k→N / nasal (regressive nasal assimilation). The context for l→n is not present in the unreduplicated forms. But after reduplication, only the initial l of the second conjunct of the reduplicated second syllable is in the required environment. A naive copy theory of reduplication would, for example, incorrectly predict: hilak → hi-hi-lak-lak → hi-hi-raN-nak This assumes that regressive nasalization can apply after onset l→n. For naive copy theory, the onset n in the rst conjunct of the reduplicated second syllable is a mystery. The context for l→n is present only in the second conjunct. This is a classic overapplication effect. The DR account parallels the account of the overapplication of nasalization in Malay. Suppose that (37a) is not subject to geminate inalterability and applies to the output of Transcription, before crossing violations have been removed and before it can be bled by (37a). Then σ σ σ σ (38) × × × × × × × × × × h

i

l

a

k σ

σ → ×

×

×

×

×

σ ×

h

i

n

a

× k

×

σ ×

×

The operation in (38) is not spreading, so the locality of spreading is not an issue. Various assumptions about the timing of the various operations involved will produce hihinaNnak. The main requirement is that regressive assimilation must have an opportunity to apply after NCC repair. It could be, for example, that l→n and NCC repair are cyclic, and that (37b–d) are postcyclic. Although other assumptions are possible, this arrangement of the rules would also ensure that there is no chance of overapplication of either word-initial l→n or postvocalic l→r in (39), which is what onset l→n after a noncoronal coda would produce in (36d).

Transcription and Rule Application in Crossed Structures

(39)

31

σ σ σ σ × × × × × × × × × × × × l w e

n a

k

Overapplication of (37b) would produce nwe-rwe-naN-nak, while overapplication of (37c) would produce rwe-rwe-naN-nak. Both possibilities are avoided if NCC repair is cyclic and these two rules are postcyclic. 2.3 The Retraction Condition Appendix A is devoted to a detailed discussion of the substance of the NCC. It would be a diversion to undertake that discussion here. The appendix is provided for interested readers. The formulation of the NCC that the discussion leads to is given in (40), which is an interface condition. A segment is taken to mean an element on a phonemic (as opposed to prosodic) tier. A set S of timing slots is called connected if there is no × ∈ S such that ×1 < × < ×2 , with ×1 and ×2 in S. (40)

Retraction Condition (RC): The set of timing slots that are associated with a segment is connected. In the remainder of this book, I will assume that the RC is an output condition on the phonology. The RC is based on the proposal that autosegmental representation is the privilege of phonology, not available either in the lexicon or phonetics. Without autosegmental linking, relationships between tiers can only be expressed directly by temporal simultaneity. The RC guarantees that the output of phonology can be transparently translated into a phonetic representation. These representations are not autosegmental, but composed of parallel tracks implicitly related by temporal simultaneity rather than explicitly linked by autosegmental associations. The RC is both weaker and stronger than the proposal that “association lines do not cross.” Representation (41a), with the medial timing slot unlinked, violates the RC but does not have crossed association lines. Representation (41b), on the other hand, has crossed association lines, but does not violate the RC. (41)

a. ×

× α

×

b. ×

×

c. ×

×

β

α

α

β

32

Chapter 2

In fact, I assume that the phoneme tiers have no intrinsic order, so that if α and β are phoneme segments, representations (41b) and (41c) are identical.9 This has various consequences, which seem to me to be correct: there are no oating segments, and there are no junctures on phoneme tiers. Segments must be associated with the timing tier, and only segments appear on phoneme tiers. Since the phoneme tier is unordered, RC repair (which I will call NCC repair or NCCR to maintain the familiar terminology) is straightforward phoneme ssion: (42)

× × × × × × × × ˜a n ˜a m

NCCR

−→

×

×

×

×

×

×

×

×

˜a ˜a n n ˜a ˜a m m

If the phoneme tiers were ordered, not only ssion operations but multiple metathesis operations would be necessary to carry out RC repair. Since the phoneme tiers are unordered, the righthand side of (42) is equivalent to: (43)

× × × × × × × × ˜a n ˜a m ˜a n ˜a m

Chapter 3 The Morphophonology of Reduplication

Reduplication is one of the ways the presence of a morpheme in the structure of a word is made manifest. The aim of this section is to sketch the theory of morphology that I assume in sufficient detail so that it is clear how reduplication ts into a general theory of word formation. The general framework is Distributed Morphology, Halle and Marantz 1993. In the computation of the phonological form of a complex word, the input to the morphophonology is a tree whose terminal nodes are bundles of morphological features. I will call this structure an m-structure. The m-structure could be the direct output of the syntax, or the output of the syntax could be modied even before the morphophonology begins to convert the m-structure into phonological form. Since that issue is not directly relevant in what follows, I leave it without further comment other than to indicate the issue. The terminal nodes of the m-structure, which do not have phonological content, are called morphemes. Morphemes are given phonological expression by means of vocabulary items, which do have phonological content (perhaps null).1 Vocabulary choice at each terminal node determines a lexical item that is inserted into the structure at that node. I ignore here the question of how the choice is made (see Halle and Marantz for a proposal) and concentrate on the process that uses vocabulary items to associate a phonological form with the m-structure. For the sake of simplicity, assume that one of the sisters of every node in the m-structure is a morpheme and that the computation can determine the root node in some way. The conversion is carried out sequentially, starting at the root node and working up the tree. A simple example will illustrate the core process. Suppose, for example, that the input to the morphophonology is [[ α β ] γ ], with β the root. The lexicon is consulted and a vocabulary item that is appropriate to realize β in the particular context in which it appears is determined.

34

Chapter 3

The vocabulary item is then “inserted” into the tree: α

(44)

β

γ

LexIns

−→

α

β

γ

× × n u I use a special arrow → in this chapter to clearly distinguish lexical insertion from instances of phonological rule application. The process continues, working up the tree, until all of the morphemes have been given phonological expression. Suppose α is realized by a prex with exponent re- , and γ is realized by a suffix with exponent -d . Assuming for simplicity that no phonological rules apply, (44) continues: (45)

LexIns

−→

α

γ

β

LexIns

−→

× × × ×

α

β

γ

× × × × ×

r e n u r e n u d The derivation (44–45) consists of a sequence of lexical insertion operations. We know that phonological rules can also apply. The simplest architecture would be for all of the lexical insertion operations to rst apply, lling up the m-structure morpheme by morpheme, then for phonological operations to apply to the resulting structure, as pictured in (46). (46)

Syntax Lexical insertion Phonology

After each lexical insertion, the process cycles back to another lexical insertion operation until all the morphemes are realized. The architecture is considerably more complicated, however. First, the phonology does not wait until all the morphemes have been realized. After a lexical item is inserted, rules of the general phonology may apply before further lexical insertion takes place. The system of rules that applies in this way is called cyclic phonology. Some lexical items trigger cyclic phonology, but some do not. The ones that do not are simply inserted and the derivation moves on to the next lexical insertion operation. Second, apart from cyclic phonology, morphophonological

The Morphophonology of Reduplication

35

rules can be triggered by the insertion of particular vocabulary items. After each lexical insertion operation, the morphophonological rules that are triggered by the inserted vocabulary item rst apply and then the cyclic rules apply (unless the lexical item is specically marked as “noncyclic”).2 Halle (1990) calls the rules triggered by lexical insertion readjustment rules (see Halle and Marantz 1993 for discussion). Of course, particular vocabulary items might not trigger any readjustment rules. The architecture is schematized in (47). (47)

Syntax Lexical insertion Readjustment rules Cyclic phonology Postcyclic phonology

The dashed line is the option that noncyclic affixes take, skipping cyclic phonology and cycling back to the next lexical insertion operation. (The case in which the outermost affix is noncyclic is not pictured.) This architecture derives from Pesetsky 1979 and much work that followed in what came to be known as Lexical Phonology. What is distinctive here is that both lexical insertion and readjustment are cyclic. Cases in which the choice of vocabulary item depends on the prosodic shape of the stem require the phonology to assemble the stem before the affix is inserted. Here and throughout this book, stem has only a structural meaning. For lexical insertion, the stem is the sister of the morpheme that is being realized, including its associated morphophonological structure. This remains the stem until the next lexical insertion operation. Rules that intervene between lexical insertion operations can be sensitive to the stem/nonstem distinction established by the last lexical insertion operation. Specically, both rules of the cyclic phonology and readjustment rules can refer to the stem. 3.1 Readjustment English past-tense morphology provides some useful illustrations of the effect of readjustment rules. Consider the derivation of sold, the past tense of sell. Throughout this book, the initial lexical insertion step in

36

Chapter 3

which the root is realized is often omitted for reasons of space, as it is below. In this chapter only, I use some special notation in order to make crystal clear the difference between various kinds of transformations. As noted before, lexical insertion is denoted by →. Readjustment will be denoted by ⇒ and phonological transformations by →. LexIns

[[sell]]

(48)

PAST

Readj

[[sell]]

−→

PAST

[[sell]]

=⇒

PAST

× × ×

× × × ×

× × × ×

s

s

s

E

l

E

l

d

o

l

d

The readjustment operation is triggered by the past tense affix -d. It backs the root vowel of a small class of verb roots (sell, tell ). [[sell]] denotes a morpheme, not a vocabulary item. The capital letters and double brackets are intended as a warning that what is referred to is simply the bundle of morphological features that the vocabulary item with phonology sEl is used to realize. As far as the computation is concerned, it is an accident that there is a single vocabulary item that is used to realize the morpheme [[sell]] independent of the environment it appears it. By way of contrast, the morpheme [[go]] is realized by a vocabulary item with exponent wEnd in the context of the past morpheme.3 In other contexts, it is realized by a different vocabulary item, which has the exponent go. From the standpoint of Distributed Morphology, there is nothing internal to [[go]] that determines whether it is realized by a vocabulary item with exponent go or one with exponent wend, just as there is nothing internal to the past morpheme that determines whether it is realized by a vocabulary item with exponent -d, by one with exponent -t, or by one with a null exponent. The choice of which suppletive allomorph is used to realize a particular morpheme is determined by the context that the morpheme appears in, not on the basis of an internal property of the morpheme. The term lexical insertion is a possible cause for confusion because its technical use departs from the colloquial use of “insertion” in case the vocabulary item that is inserted has a null exponent. Although no exponent is inserted, two things do happen when a morpheme is realized by a vocabulary item with a null exponent. First, since the computation must keep track of which morphemes have already been realized, the morpheme is marked as having undergone lexical insertion. Second, the readjustment rules associated with the vocabulary item are triggered. They are triggered by the vocabulary item, not by

The Morphophonology of Reduplication

37

the morpheme it realizes. Different realizations of the same morpheme might trigger different readjustment rules. The lexical insertion of the vocabulary item with exponent -d in (48) had two effects on the stem: an exponent was inserted and a readjustment rule was triggered. Morpheme realization in that case had both a concatenative aspect (concatenation of the suffix -d ) and a nonconcatenative aspect (application of a vowel ablaut rule). In the derivation (49) of sang, the past tense of sing, morpheme realization has an entirely nonconcatenative effect. LexIns

[[sing]]

(49)

PAST

Readj

[[sing]]

−→

PAST

[[sing]]

=⇒

PAST

× × ×

× × ×

× × ×

s

s

s

I

N

I

N

a

N

The default past-tense suffix -d was chosen for insertion in past in (48), but in (49), the verb belongs to the class of verbs that require the null past-tense suffix. Realization is marked here and for the rest of this chapter by underlining the morpheme in case it has a null exponent. If there is an exponent, realization is implicit. The possibility of readjustment rules does not exhaust the possibilities of interaction between the morphological structure of a word and phonology. A cyclic, postcyclic, or lexically triggered rule itself can be morphologically conditioned. That is, morpheme structure or the presence of certain lexical items can gure in the structural conditions of applicability of the rule. Within the word-formation architecture sketched above, morphemes can have various effects on the ultimate phonological form of words they appear in. It would be tedious to attempt a complete catalog, but some are worth reviewing. Lexical insertion can contribute an exponent and trigger readjustment rules. Morphemes can condition the vocabulary choice for other morphemes. The exponent of a vocabulary item that is used to realize a particular morpheme can eventually delete, but leave an effect via the phonological rules it conditions. Finally, as just mentioned, morphemes can condition the application of phonological or readjustment rules. With this understanding of the architecture of the computation of the phonetic form of words from m-structures, we can consider the place of reduplication in the architecture.

38

Chapter 3

3.2 Juncture Insertion is a Variety of Readjustment An affix that induces reduplication does so by triggering rules that insert t-junctures into the timing tier. Phonological rules, sensitive to the presence of these junctures, subsequently carry out transcription. Hayes and Abad (1989, 357) give a good example from Ilocano; see paradigm (50). According to them, what they call “light reduplication” is used with the prex si- to mean ‘covered with, lled with’: (50) a. buneN b. pandiliN c. liNPet

(type of knife) ‘skirt’ ‘perspiration’

si-bu-buneN si-pa-pandiliN si-li-liNPet

d. roPot e. ´jyaket

‘leaves, litter’ ‘jacket’

si-ro-roPot si-´jya-´jyaket

‘carry a buneng’ ‘wearing a skirt’ ‘covered with perspiration’ ‘covered with litter’ ‘wearing a jacket’

The m-structure is (n cov), with the morpheme cov some kind of prepositional element that combines with a noun root. The lexical item that realizes cov has a prexal exponent si- and its insertion triggers tjuncture insertion, as shown in the illustrative derivation of (50d), given in (51). (51)

cov [[roPot]]

LexIns

−→

cov [[roPot]]

Readj

=⇒

cov [[roPot]]

×××××

×××××××

× ×[× ×]× × ×

r o Po t

s i r oP o t

s i r oP o t

When the lexical item that realizes cov is inserted, rules that insert tjunctures into the verb root are triggered. I will return later to examine the structural description of such rules in detail. The present discussion is meant only to examine the place of these rules in the morphophonological derivation. Transcription (Trscr below) follows at some later point. Presumably, transcription is to the left so that disruption of the morpheme structure is minimized. Compare (51) with (48); sEl→sEld→sold, also lexical insertion followed by readjustment. The major difference is that the phonological effects of juncture insertion are produced later, in the phonology. It is not unusual for readjustment to be followed by a phonological modication or even a cascade of phonological changes. Later in this chapter I discuss a readjustment rule in Erromangan verbal morphology that cre-

The Morphophonology of Reduplication

39

ates modied roots in the context of certain tenses, roughly parallel to past-tense modication in English. The derivations of modied roots in Erromangan show that readjustment often has substantial phonological aftereffects; see the examples in (69) later in this chapter. The derivation (51) continues in the phonology: (52)

Trscr

−→

cov [[roPot]]

NCC Repair

−→

cov [[roPot]]

×××××××××

×××××××××

s i

s i r o r o P o t

r o P o t

The copied timing slots in the terminal representation in (52) are not associated with morphemes. One might argue that they come to be associated with some morpheme. While plausible, it seems reasonable to expect that the extra computation would be carried out only if there was some reason to do so. Since I do not know of any compelling reasons to the contrary, I will tentatively assume that the reduplicant timing slots are not associated with a morpheme. Juncture insertion rules will be examined in great detail in coming chapters. The following description of the Ilocano rule is a starting point:  (leftmost in stem) (53) 1. ∅ → ]  V × (leftmost in stem) 2. ∅ → [ In (53.1), “V” is taken to be a predicate on timing slots, true if the timing slot is associated with a vowel. (This will later be modied so that the predicate is true of nuclei, whether vocalic or not.) The structural condition in (53.1) therefore species the insertion of a ]-juncture into the timing tier following a timing slot in the stem associated with a vowel. The leftmost instance of this conguration is chosen as the insertion site. It will prove to be useful to write the rules (53) in a bipartite form, with the rules on the left and the domain to which they apply on the right.   , ∅ → [ Left × ; stem (54) ∅ → ] Left V In the interest of compact notation, leftmost (or  rightmost) application is sometimes specied by a subscript on the “ ” symbol that introduces the environment for rule application. It will become clear in chapter 5 that the variation in the rules which reduplicative affixes trigger is best

40

Chapter 3

accounted for if a domain is factored out and the rules written relative to the domain, as in (54). The domain is not always the full stem. The domain of some reduplicative affixes is a morphological or prosodic subword of the stem. The general format is (55)

juncture insertion rules

;

rule domain

Some further simplication is possible on the basis of several natural default choices: (1) rule application is leftmost, unless specied otherwise, and (2) the domain of juncture insertion rules is taken to be the stem, unless specied otherwise. Furthermore, I assume that there is a process by which unpaired [- and ]-junctures which remain after juncture insertion are closed by inserting ] or [ at the edges of the domain. This is called Default Closure (DC). In view of default leftmost juncture insertion rule application, Default Closure, and default domain selection, (53) simplies to  (56) ∅ → ] V This rule occurs so commonly that it is useful to have a name for it: C∗ V juncture insertion, or simply the C∗ V-rule. The Ilocano affix discussed above can then be described simply by saying that the exponent of cov is prexal si- and that it triggers the C∗ V-rule. 3.3 Klamath Distributive and Intensive Reduplication Klamath distributive and intensive reduplication and their interaction illustrate many of the ideas above. (The examples in this section are all from Zoll 2002, which proposes an Optimality Theory analysis.) (57)

Klamath distributive reduplication Root Reduplicated a. qlin ‘choke’ qli-qlin b. pag-a ‘bark’ pa-pag-a

Surface qli-ql@n pa-pg-a

Klamath has a rule (Root Vowel Reduction, RVR) that reduces the initial vowel of the root if that vowel is not word initial. Typically, prexes trigger RVR. The surface forms in (57) are the result of RVR applied to a reduplicated form. The vowel deletes in open syllables, as in (57b), and reduces to schwa in closed syllables, as in (57a). The lexical item that realizes the distributive morpheme and induces reduplication has a null exponent and triggers C∗ V-juncture insertion.

The Morphophonology of Reduplication

41

The derivation of (57a) is given in (58), with the distributive morpheme denoted by dis. T&R (transcription and repair) denotes the compound operation consisting of Transcription and NCC repair. This abbreviation will often be used when Transcription directly follows NCC repair. (58)

[[qlin]]

T&R

−→

dis

[[qlin]]

DIS

[[qlin]]

−→

dis →⇒

[[qlin]]

dis

××××

[ × × ×] ×

q l i n

q l i n

[[qlin]]

dis

RVR

−→

[[qlin]]

×××××××

×××××××

q l i q l i n

q l i q l @ n

dis

(To save space, lexical insertion and readjustment have been combined, as should be clear from the notation.) RVR applies because the rootinitial vowel is not word initial. RVR is a typical morphologically conditioned rule. It must be sensitive to the morpheme structure so that the initial vowel of the root can be located. It is a phonological rule in form, but it is neither purely phonological nor a readjustment rule (i.e., directly triggered by the insertion of a lexical item). Two assumptions are crucial to (58): transcription must be to the left, and RVR must apply after Transcription. If transcription were to the right, the environment for RVR would not be created. If RVR applied after transcription, but before NCC repair (ssion), the outcome would be ql@ql@n (overapplication) or qliqlin (underapplication), depending on whether RVR is blocked by geminate inalterability. Only one of the occurrences of the vowel is a root vowel. The reader might be skeptical of the possibility of ql@ql@n, in which the prex causes its own reduction via root vowel reduction. But we will later see a very similar interaction in Chaha (p. 181). Transcription to the left in this example is not an arbitrary stipulation. Transcription to the right would produce the complex structure in (59), with the copied material intruding into the morpheme. This kind of transcription is probably not excluded by general principles and may in fact occur in some languages. But all other things being equal, simplicity favors maintaining morpheme contiguity.

42

Chapter 3

[[qlin]]

(59)

dis

× × × × × × × q l

i

n

Klamath has another reduplicative affix that realizes a morpheme with intensive semantics, denoted by int below. Two examples are given in (60). (60)

Klamath intensive reduplication Root Reduplicated ‘op (as a sh)’ Wit-Wit a. Wit b. kesp ‘pant for breath’ kesp-kesp

Surface Wit’-Wit’

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, unrelated to reduplication. Two things stand out in (60): (1) the entire root is doubled and (2) there is no vowel reduction. We will see that these things are related. The lexical item that realizes int has a null exponent and triggers the readjustment rule:  (61) ∅ → [ × Recall that juncture insertion rules apply leftmost by default, Default Closure applies, and the default domain is the whole stem. The morphology of (60a) is straightforward: (62)

[[wic]]

INT

[[wic]]

−→

[[wic]]

INT

INT

→⇒

[[wic]]

×××

[× × ×]

W i t’

W i t’

INT

In distributive reduplication, discussed above, transcription was to the left in order to avoid splitting the root. Total reduplication does not interact with direction of transcription in the same way, since either left or right transcription is to the morpheme boundary. In Klamath, the absence of vowel reduction in the totally reduplicated forms in (60) is evidence that copying is to the right. The derivation (62) continues:

The Morphophonology of Reduplication Trscr

−→

(63)

43

NCC Repair

[[wic]]

[[wic]]

−→

INT

××××××

INT

×××××× W i t’ W i t’

W i t’

RVR does not apply because the root vowel remains word initial. The second vowel is not a root vowel. Zoll observes that the vowel of the rst conjunct of the intensive form does reduce when it is not word initial.4 She gives various examples, of which the following is particularly interesting because it combines intensive and distributive reduplication: (64)

‘be stiff’ ‘stiff’ ‘stiff-distributive’

Wiˇc Wiˇc-Wiˇc-l’i Wi-W@ˇc-Wiˇc-l’i

The derivation of the distributive intensive form is given in some detail below, simplied in the interest of brevity by ignoring the -l’i suffix. The affix that realizes int is assumed to be cyclic. There are three cycles of lexical insertion. (65)

[[wic]]

[[wic]]

DIS

[[wic]]

−→

INT

DIS INT

××× W i ˇc INT

→⇒

[[wic]]

DIS INT

T&R

→

[× × ×] W i ˇc DIS

→⇒

[[wic]]

[[wic]]

DIS INT

×××××× W i ˇc W i ˇc INT

DIS

[× ×]× × × × W i ˇc W i ˇc

T&R

→

[[wic]]

INT

DIS

RVR

→ WiW@ˇcWiˇc

×××××××× W i W i ˇc W i ˇc

Whereas the root vowel was the word initial vowel in (63), it is not in (65), so RVR applies in (65) but not (63).

44

Chapter 3

3.4 Erromangan: Evidence for Readjustment The discussion earlier in this chapter of English past-tense inection assumed without discussion that the alternation between sEl and sol in sell/sol-d is a consequence of a readjustment rule triggered by the lexical item that is used to realize the past-tense morpheme. The modied form sol is generated in the morphophonology from sEl, not supplied by the lexicon. A widespread alternate view is that exceptions to regular past-tense morphology in English are suppletive, with irregular forms supplied by the lexicon, not constructed by rule in the morphophonology. See Yang 2002 for convincing arguments that the irregularities are the consequence of exceptional rules, not suppletion. Yang’s argument is indirect, comparing the predictions about language learning that the two proposals lead to with the facts. One way to furnish evidence of a different kind that, for example, sol is derived from sEl in the morphophonology would be to nd some operation which applies to sEl before it is transformed to sol that leaves some trace of sol’s previous form. Unfortunately, the morphophonology of the English past-tense system is not complex enough to provide evidence of this kind. The Oceanic Austronesian language Erromangan, which is the subject of an excellent study by Crowley (1998), provides the kind of example that is missing in English. It has an extensive system of root alternation associated with various tenses that is comparable to Germanic root alternations in the past tense. Unlike English morphology, however, the morphology of Erromangan is sufficiently complex for signicant processes to intervene between the point at that the root is realized and the point at which the tense morpheme which conditions alternation is realized. Erromangan has total root reduplication, with intensive semantics. The intensive morpheme responsible for the reduplication can intervene between the root and the morpheme that conditions root modication. The interaction provides crucial evidence that verb-root alternation in Erromangan is the result of modication in the morphophonology, not suppletion. To understand the evidence, we rst need a description of root alternation and of reduplication. Tense and agreement morphology is largely prexal. Roots have a basic root form and a modied form. The modied form is used following future-tense subject prexes; following certain present-tense, conditional, and past habitual prexes; and following what are called

The Morphophonology of Reduplication

45

“echo subject markers.” Some examples are given in (66). The orthography is what Crowley calls Erromangan “practical orthography,” except that I have substituted N and G (which have their IPA values) for the practical orthography’s g and c in order to make the forms more readable. The basic root is used with the distant past prex, but the modied root must be used with the future prex.

3sg:distpast-basic root

(66) a. b. c. d.

y-aruvo yi-tovop y-ehri y-oGol

‘he/she sang’ ‘he/she laughed’ ‘it broke’ ‘he/she dug’

3sg:fut-modied root Go-naruvo Go-ntovop G-ahri G-aNkol

‘he/she will sing’ ‘he/she will laugh’ ‘it will break’ ‘he/she will dig’

As noted above, total root reduplication is used to realize an intensive morpheme. For example: amon ‘hide’ > amonamon ‘hide all over’, avan ‘walk’ > avanavan ‘walk all over’, alou ‘run’ > alowalou ‘run all over’, and omol ‘fall’ > omolomol ‘fall all over’. We are now in a position to test the hypothesis that root modication occurs in the morphophonology. The root omol has the modied form amol. Crowley gives the following example: (67)

cw-amol-omol 3pl:fut-fall (intensive) ‘they will fall all over’

One version of the root is modied and one is not. Modication must therefore occur after reduplicative copying. It follows that root modication must follow lexical insertion. A full derivation is given in (68), with readjustment operations indicated by a double arrow. (68)

root ⊕ int ⊕ 3sg:fut (order of morpheme realization) a. root omol omol ⇒ [omol] → omolomol b. int cw-omolomol ⇒ cwamolomol c. 3sg:fut

Like most grammars, Crowley’s concentrates on exceptional cases, so only one full example of a reduplicated root in a context that calls for the modied form is given. He is very clear, however, that only the form of the root adjacent to the tense prex occurs in the modied form (Crowley 1998, 79, 143). To solidify this analysis, I now turn to discussing root modication in some detail and working out the readjustment rules that generate the modied root.

46

Chapter 3

3.4.1 Modied Root Formation Before I specify the rules that transform basic roots into modied roots, we rst need to distinguish between strong and weak verbs in Erromangan, since modied root formation is affected by this distinction. There is a subclass of verb roots in Erromangan, called strong roots, that behave exceptionally in some contexts. Class membership affects both modied root formation and certain rules that simplify vowel sequences over prex boundaries. Roots that are not strong are called weak roots. Approximately one-fourth of the verb roots are strong. There are large phonologically dened classes of roots that are predictably weak or strong: all roots that begin with a glide, a high or low vowel, or an alveolar consonant are weak and all roots beginning with a labial consonant are strong. The only remaining roots, and therefore the only roots whose class membership is not determined phonologically, begin with a mid-vowel (e or o in Erromangan). About two thirds of them are strong. Class membership of these roots is not phonologically conditioned: owi ‘plant’ is weak, but owi ‘leave’ is strong, for example. According to Crowley, there is “no possible semantic conditioning factor that can be determined, and transitivity is not a factor either” (p. 81). The modied root of all but a handful of strong roots is formed by rst applying an operation that Crowley calls Nasal Accretion (NA), then prexing a. NA is (V)C . . . → (V)NC . . . , with NC forming a homorganic cluster. The effect of NA is prexing if the root is Cinitial, otherwise inxing. Not all consonants C can form a homorganic NC cluster: only stops and voiced obstruents. The set of such consonants is Σ = {p, t, k, v, r, G}. NA does not apply if such a cluster cannot be formed. The modied root of the strong root oGol, for example, is formed by oGol→oNGol→aoNGol. As mentioned above, V-initial strong roots are subject to special vowel-reduction rules at the prex-root boundary. In particular, their initial vowel deletes after a, so a concomitant of a-prexation to a V-initial root is the deletion of the root-initial vowel. So oGol→aNGol. In addition to initial-vowel deletion, several other rules of regular phonology can apply to the output of this process: fricatives in NC clusters become stops and various rules of triconsonantal cluster reduction apply. Some derivations of modied strong roots are given in (69). The names of the different processes are Crowley’s. To save space, morphologically conditioned deletion of the root-initial vowels is here com-

The Morphophonology of Reduplication

bined with a-accretion. (69) Readjustment nasal aaccretion accretion omol a omol oGol oNGol a oNGol vaN mvaN amvaN oGhi oNGhi a oNGhi evGah emvGah a emvGah evsor emvsor a emvsor eiti a eiti

47

Phonology

despirant- cluster ization reduction aNkol ampaN aNkhi ampGah ampsor

aNhi amsor

‘fall’ ‘dig’ ‘eat’ ‘see it’ ‘defecate’ ‘wake up’ ‘tie it’

Note that the alternation oGol/aNkol that was given in (66d) is not as dramatic as it might have seemed. The alternation that is attributable to readjustment is replacement of the initial vowel by aN. The further change is standard Erromangan phonology. English root adjustment in the past tense deletes at most the nal rhyme of the root. Erromangan modied root adjustment deletes at most the initial phoneme of the root (this is exceptionless). Although evsor >amsor results in the loss of the rst two phonemes, the loss of v is a phonological aftereffect of readjustment. The modied root of a weak root is formed by rst prexing n if the root is V-initial, then applying NA. Since NA does not apply to nasals, the result is that n-prexation bleeds NA in the case of V-initial weak roots. a-prexation does not occur for weak roots. Modied weak roots are illustrated by (66a–b) above (aruvo >naruvo, tovop >ntovop). These rules correctly generate all but a handful of Erromangan modied roots. We return to the handful of exceptions to (69) below. 3.4.2 A Contrary View I view the modied reduplicated form cw-amol-omol, (67), as a spectacularly clear demonstration of the fact that root modication is in the morphophonology. Surprisingly, Inkelas and Zoll (2000), who are responsible for calling attention to the signicance of Erromangan reduplication, come to a different conclusion. They see it as evidence that, contrary to widely held views, reduplication is not an instance of copying in the phonology. The basis for this conclusion is their view of readjustment. Unfortunately, they neither make their view clear nor defend it. This section is devoted to trying to make their view explicit, and showing that it is untenable. What they say is “Crowley (1998:84)

48

Chapter 3

proposes a long list of phonological steps that convert basic stems into their modied counterparts. Several of these steps are unnatural, and in any case the rules have exceptions, in which case listing at least some of the modied forms is inevitable. . . . In at least some cases the reduplicant and the base consist of different suppletive allomorphs of the same basic root morpheme.” It seems to me that their complaint that what Crowley proposes (and I reproduced above) is “a long list of phonological steps,” some of which are “unnatural” steps, has little merit. The analysis given above closely follows Crowley’s. Readers will have to judge for themselves how unnatural it is because Inkelas and Zoll furnish no arguments. One widely held but little discussed view of readjustment rules is that they cannot have lexical exceptions other than those that are the result of suppletive allomorphy taking priority over rule application. For the purposes of the following discussion, this will be called the Lexically Insensitive Readjustment (LIR) theory. See Lieber 1992 for an explicit proposal.5 On the one hand, I&Z appear to hold this view, since “rules have exceptions” is taken to imply “listing . . . modied forms.” On the other hand, I&Z justify allomorphy on the basis of exceptions to Crowley’s “long list” of rules. But Crowley’s rules are a proposal for how the exceptional case of strong roots should be analyzed. If the LIR theory is taken seriously, then one is committed to the view that all modied strong roots are stored as suppletive allomorphs of the basic root. It is not principled to hold that strong roots are exceptional but not suppletive, but more ne-grained exceptions within the class of strong roots must necessarily be suppletive because they are exceptional. Presumably, consistent advocacy of the LIR would lead to the conclusion that the weak-root rule is the regular rule and all exceptions are consigned to suppletive allomorphy. This is a difficult proposal to defend. It offers no clue about why it is that the modied roots of strong verbs are almost all predictable from the basic root. A suppletive analysis says it is an accident. A readjustment analysis has a simple answer. They are all formed in the same way, by rule, from the basic root. The absurdity of listing all the modied roots of strong verbs in the lexicon may explain I&Z’s inconsistent adherence to the LIR theory. The LIR theory has further unpleasant consequences. There are a number of phonological processes in Erromangan that affect vowel sequences at morpheme boundaries. Rules of midvowel backing and glide formation produce u-esomsaG 2pl:imp-breathe > wosomsaG ‘you

The Morphophonology of Reduplication

49

all breath!’. The process, however, is morphologically conditioned and does not apply if the root is strong (esomsaG is a weak root). For strong roots, the initial vowel of the root deletes. For example, u-empGu 2pl:imp-dance > umpGu ‘you all dance!’. Under the LIR theory there would have to be suppletive allomorphy, with mpGu (for use before back vowels) added to the basic root empGu and the modied root ampGu as a suppletive variant of the root. This compounds the problem of explaining the regularities in allomorphy. Every V-initial strong root must have a suppletive variant that differs only in that it has lost its initial vowel. Massive and transparent regularity in the lexicon would have no explanation. It explains nothing to propose that there are “redundancy rules” in the lexicon that maintain the requisite phonological similarity unless such rules play some more integral role in the theory than answering the question of why the forms are necessarily so similar. There are eight strong roots whose modied roots are not formed by the regular process of NA and a-prexation. But once lexically sensitive readjustment is part of the architecture of morphophonology, there is no obstacle to analyzing all the exceptions as exceptional readjustment, not suppletion.6 This is not to say that “anything goes.” Readjustment rules “satisfy the same formal constraints as ordinary phonological rules” (Halle and Marantz 1993, 129). The boundaries of the rulemaking abilities (and proclivities) of language users are not known with precision; but we can safely conclude that a rule deriving wEnt or wEnd from go would never be innovated, so there is no suggestion that readjustment can or should take over all the functions of suppletive allomorphy. I conclude from the considerations in this section that amol-omol gives no evidence that calls into question the fundamental role that copying in the phonology plays in reduplication. But it does give powerful evidence for morphologically conditioned readjustment.

Chapter 4 Truncated Reduplication

I rst consider string copy as a general process. One way the computation might be organized is illustrated by the successive stages of the following computation: 1. 2. 3. 4. 5. 6.

 

k ka kat kat

k ( k (k ( k ( k ( k k

a a a a a a a

t t t t t t t

) ) ) ) )

First, the string to be copied is delineated by some symbols (parentheses are used above) and the location to which it is to be copied is marked with some symbol ( above). Then markers  and  are inserted in order to keep track of which elements have already been copied.1 The actual copying is done in steps 3 through 5. At each step, the  symbol is shifted to the right of an element to be copied, and that element is copied to the immediate left of the  symbol. Finally,  can no longer shift to the right and the computation terminates by removing the bookkeeping symbols. String copy is a mechanism that the language apparatus needs independently of reduplication. Strings must be copied from the lexicon to an active workspace, for example. I suppose that morphology and phonology co-opt this mechanism for reduplication. Morphology inserts junctures into the timing tier and phonology carries out copying as shown above. The duplication junctures [ and ] are used as end markers in place of parentheses and the target location of copy is immediately outside one of the end markers, so that no explicit  symbol is needed.

52

Chapter 4

Copying can be either to the left or the right. The full reduplication of kat to kat-kat, with copying to the left, is then computed derivationally as shown in (70), with names given to the three suboperations that were proposed above. [× × ×]

(70)

k a 1.

[ × × × ]

Init

k a 2.

CopyShift

CopyShift

Cleanup

t

× × ×[ × × ×] k a

5.

t

× ×[ × ×  × ] k a

4.

t

×[ ×  × × ]

CopyShift

k a 3.

t

t

× × × × × × k a

t

A mirror-image version is used for copying to the right. 4.1 Truncation at the Leading Edge (Reduplicant Truncation) The  and  symbols above are inserted in the transcription process in order to keep track of the progress of the computation. They are bookkeeping symbols; nevertheless, they are real phonological objects.2 I suppose that morphology exploits the existence of the truncation junctures that copying uses to carry out its computation by taking it on itself to insert truncation junctures in addition to duplication junctures. For example, there might be a readjustment rule: × × × k a t

readjustment

=⇒

[× × × ] k a t

As far as phonology is concerned, this looks just like an intermediate

Truncated Reduplication

53

representation in a sequence of copying operations and is treated as such by Transcription. The derivation initiated by the morphology continues: CopyShift

−→

×[× × × ]

CopyShift

−→

× ×[× × ×]

k a t

Cleanup

−→

× × × × ×

k a t

k a t

So [kat ]→kat-kat, but [k at ]→at-kat. The boldface, hyphen, and explicit  are simply aids to the reader, not part of the representation.3 I call this leading-edge truncation. The truncate, the  . . .  material, is at the edge of the duplicant toward the direction of transcription, so it is called a “leading-edge” truncate. It appears to be truncated from the copy. In fact, it is simply not copied.  and  will be called truncation junctures. The transcription junctures (t-junctures) include not only the duplication junctures, but the truncation junctures as well. The general rule is (71)

Leading edge truncation Trscr

[u v ] −→ vuv 

Trscr

[v u ] −→ vuv

u and v are strings of timing slots. Madurese plural reduplication (Stevens 1968, 34) was discussed by Marantz (1982, 451). Some examples follow: (72)

Root buwaP maen estre

Plural waP-buwaP-an en-maen-an tre-estre-an

‘fruits’ ‘toys’ ‘wives’

The plural forms can be generated as shown in (73), with subsequent -an suffixation. (73)

[buwaP] → waP-buwaP [maen] → en-maen  [estre] → tre-estre  

In chapters 5 and 6, I show how these pretranscription structures are generated by morphological rules that insert t-junctures and phonological rules that adjust the structures before transcription applies. Marantz (1982, 451) also discussed a Chukchee reduplicative pattern that is similar to the Madurese pattern, but in which the partial copy of the stem is to the right. Some examples are given in (74). The meaning associated with the morpheme that induces the reduplication

54

Chapter 4

is the absolutive singular. (74)

Root nute inu jilPe

Absolutive singular nute-nut ‘earth, ground’ inu-in (part of a reindeer leg) jilPe-jil ‘gopher’

The reduplicated forms can be generated as shown below: (75)

[nute] → nute-nut [inu] → inu-in [jilPe] → jilPe-jil

The derivation of the pretranscription forms is given in chapter 6. 4.1.1 Truncated Prexes and Suffixes There are many examples of leading-edge truncation coupled with prexation or suffixation. Yoruba has a nominalizing affix that produces the following pattern: (76)

Root lo. d dun

Nominal form li-lo. d di-dun

‘to go’ ‘to be tasty, sweet’

The data is from Marantz (1982, 449), who attributed the examples to Doug Pulleybank. Marantz analyzed Yoruba nominalizing reduplication as “prexing a CV reduplication skeleton whose V is xed to i.” The considerations above suggest a different analysis: i is a prex, but juncture insertion rules apply after the prex has been concatenated with the stem, as shown in (77). prex

(77) × × × −→ × × × × d u n

JncIns

=⇒

[× × ] × ×

i d u n

i d u n Trscr

−→ × × × × ×

(d-i-dun)

i d u n The specication of the morphology of affixes of this type is greatly simplied if the exponents of lexical items are permitted to contain t-junctures.4 If the Yoruba NOM-prex is taken to be [i - , juncture insertion reduces to  ∅→] V

Truncated Reduplication

55

Recall that juncture insertion rules apply leftmost in the domain, unless otherwise specied, and that the domain is the stem, unless otherwise specied. I consider the resulting simplication to be sufficient justication for the assumption that t-junctures can enter the computation not only as the result of insertion by rule, but also directly as elements of the exponents of certain affixes. In the discussion of Chaha in chapter 7, it will be further proposed that t-junctures can be embedded not only in the exponents of affixes, but also in roots. Affixes similar to the Yoruba nominalizing affix are widespread. The following data is from Agta, (Healy 1960). (78)

Root wer bag pesuk PassaN

‘creek’ ‘g-string’ ‘peso’ ‘small’

Diminutive wala-wer bala-bag pala-pesuk Pala-PassaN

‘small creek’ ‘small g-string’ ‘a mere peso’ ‘very small’

The similarity between the diminutive affix responsible for (78) and the nominalizing affix responsible for be clear. The expo (76) should 5 V. nent is [ala - and it triggers ∅ → ] An illustrative derivation follows in (79). Recall that the domain of the juncture insertion rule is the stem wer, not [ala wer. prex

(79)

× × × −→ [× × × × × × w e r a l a w e r

]-insert

=⇒

[× × × ×] × × a l a w e r

Trscr

−→ × × × × × × ×

(w-ala-wer)

a l a w e r What is called “tasty-shmasty reduplication” is the most prominent reduplication process in English and has therefore received a great deal of attention. Examples are well-known: oil-smoil, money-shmoney, Bush-Shmush, and so on. An interesting point is that the reduplicated form has the prosody of a two-word sequence. The suffix -#shm ], word boundary, coupled with the juncture where # is an embedded  ∅ → [ V, produces derivations like the following: insertion rule (80)

× × × × × →⇒ t a s t i

×[ × × × × # × ×] t a s t i sh m (the derivation continues on the next page)

56

Chapter 4 Trscr

−→ × × × × × # × × × × × × t a s t i

(tasti- #shm-asti)

sh m

4.1.2 Reduplicative Prexes and Suffixes in Arrernte Pensalni and Breen (1999) discuss a number of reduplicative processes in Arrernte, an indigenous Australian language, which furnish good examples of reduplicative prexes and suffixes. Pensalni and Breen give an analysis in the Prosodic Morphology framework, which requires the assumption that Arrernte has the unusual property that its syllables are necessarily onsetless. An analysis in the terms developed here is much more straightforward. The affixes are easily described: Habitative: Frequentative:

Exponent -e”n ] -ep ]

Attenuative:

[elp -

(81)

Juncture insertion  ∅ → [ Right V  ∅ → [ Right V  ∅→] Vsecond

The context specication / Vsecond , “before the second vowel,” is provisional. The vocabulary for writing juncture insertion rules will Vsecond will no longer be a be constrained in chapter 5 so that / possible structural description for a juncture insertion rule. Writing the desired rule in Arrernte will depend on the assumption that juncture insertion rules can be relativized to a subword (either prosodic or morphological) of the stem. That is, the domain of the juncture insertion rules associated with an affix can be specied to be a subword of the stem. In Arrernte attenuative reduplication, the  domain is taken to be V. the initial bisyllabic foot and the rule is ∅ → ] Right A few examples of frequentative reduplication are given below, along with derivations.  V (82) Frequentative: - ep ] ; ∅ → [ Right a. eN-em ‘is standing’, eNepeN-em ‘keeps standing’ eN →⇒ [eN ep ] → eNep-eN b.

ater-em ‘is ghting’, atereper-em ‘keeps ghting’ 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

Truncated Reduplication

57

A few examples of habitative reduplication, along with derivations, are given in (83). The morpheme realized by the habitative affix makes a noun from the verb. The semantics are variable, generally meaning something like “object habitually associated with the action denoted by root.”  V (83) Habitative: - e”n ] ; ∅ → [ Right a. aNk ‘eat’, aNke”naNke”n ‘food’ aNk →⇒ [aNk-e”n ] → aNk-e”n-aNke” n b.

at wer ‘ght’, at were”nere”n n at wer →⇒ at w [er-e”n ] → at w-er-e”n-ere”

A few examples of attenuative reduplication follow in (84). The meaning is generally “beginning to do something” or “doing something unintensively.”  Vsecond (84) Attenuative: [elp - ; ∅ → ] a. itir-em ‘thinking’, itelpitir-em ‘half thinking’ itir →⇒ [elp it ]ir → it-elpitir b.

emp war. -em ‘making’, emp welpemp war. -em ‘starting to make’ emp war. →⇒ [elp emp w ]ar. → emp w-elpemp war.

The juncture insertion rule cannot apply to a monosyllabic root because there is no second vowel. Consequently, Default Closure must apply. prex

DC

Trscr

ar. −→ [elp ar. =⇒ [elp ar. ] −→ ar. -elpar. The root ar. therefore gives ar. -em ‘looking’ and ar. elpar. -em ‘starting to look’. 4.2 Trailing Truncation The viewpoint that has been adopted to this point is that duplication junctures, the procedure for copying duplicants, and leading-edge truncate, come almost for free since phonology must carry out copying irrespective of reduplication. The (left) transcription rules that have been proposed are given in (85).

58

(85)

Chapter 4

∅

1.

Init

2.

CopyShift

3.

Cleanup



 →  [

×

. . . × → × . . . ×

[×∗ ] → ×∗

It is implicit in CopyShift that the newly created timing slot (to the left of ) has the same phoneme associations as the original timing slot. This transcription mechanism will transcribe [k  at ] to the left as at-cat and [k at ] to the right as kat-k. If it attempts to transcribe [k at ] to the left, however, it takes two steps and halts, unable to proceed. Init

[k at ] −→



[k at ]

CopyShift

−→



k[k at ]

This is of no concern to string copy, because this conguration does not arise. Transcription extends the core mechanism (85) by innovating rules that allow forms like [k at ] to be transcribed. The core mechanism is extended with the rules (86). (86)

4.

CopyDelete

5.

-Cleanup



. . . × → × . . .    → ∅ 

Compare (86.4) with (85.2). CopyDelete is like CopyShift in that they both insert a new timing slot to the left of  (assuming left transcription) and associate it with the same phoneme that the source timing slot is associated with. But CopyDelete deletes the original timing slot. A truncate that is not a leading-edge truncate will be called a trailing edge truncate. A leading-edge truncate is not copied, giving the appearance that it has been truncated from the reduplicant. Trailing-edge truncates are actually copied, using (86), but are truncated (i.e., deleted) in the duplicant. An example from Ndebele, [lim a ] → lima-lim, which forms an unintensive stem of lim ‘cultivate’, is given in (87). Trailing truncation provides a simple mechanism whereby material can be inserted into the duplicant in such a way that the reduplicant is affected, but not the remnant. We will see in chapter 6 that many reduplicative processes follow juncture insertion with adjustment of the reduplicant to a characteristic prosodic shape before transcription applies. Material added to satisfy the prosodic demands is often added as a trailing truncate, so that after transcription it appears in the reduplicant, but not in the remnant. In the Ndebele example, juncture insertion produces [lim ], which is then adjusted to [lim a ] so that

Truncated Reduplication

(87)

59

[ l [l  l[ l li[ l lim[ l lima[ l lima[ l lima l 

1. 2. 3. 4. 5. 6. 7.

Init CopyShift CopyShift CopyShift CopyDelete -Cleanup Cleanup

i i i i i i i i

m a] m a] m a] m a] m a] m ] m ] m

the reduplicant is bisyllabic in the expansion [lim a ] → lima-lim, as shown in (87). The consequence is that the stem lim generates an unintensive stem lima-lim, which is then subject to further inectional morphology in the usual way. In Mokilese, a bimoraic reduplicant is obtained from [wi ] by truncated timing slot epenthesis at the right edge, coupled with spreading the vowel. Again, transcription is to the left. σμ (88) [ × ×]

Truncated Lengthening

−→

w i

σμμ Trscr [ × ××] −→ × × × × ×  w i

(wi:-wi)

w i This variety of duplicant adjustment will be called First Conjunct Vowel Lengthening (FCVL). In (88), FCVL is used to augment the reduplicant to bimoraicity, leaving the remnant unchanged. Before leaving this section, I give the general result for future reference. The most common case is when the trailing truncate is at the trailing edge (the edge opposite the direction of transcription), but there are instances when the truncate is neither at the leading nor trailing edge. (89)

Trailing truncation Trscr

Trscr

[u v w ] −→ uvwuw [w v u ] −→ wuwvu  Trailing-edge truncation (an important special case) Trscr

Trscr

[u v ] −→ uvu [v u ] −→ uvu



60

Chapter 4

u, v, and w are strings of timing slots, with u nonempty. If u were empty, the truncate would be a leading-edge truncate and transcription would proceed differently. Note that (89) is not the statement of rules. It is a theorem, derived from the elementary transcription rules (85) and (86). 4.3 Embedded Duplicants Surprisingly, the three rules for simple copying (Init, CopyShift, and Cleanup) are able to transcribe forms with one duplicant embedded within another. Perhaps even more surprisingly, various languages put this to work in their reduplicative systems. There is one potential ambiguity in applying the transcription rules. Each duplicant establishes a target location for copying, so there are two potential targets. The ambiguity is always resolved in favor of the containing duplicant. We will see this exemplied shortly. 4.3.1 Consonant Copy in Semai I begin with the examples below, from Diffloth 1976. They are identied as continuative aspect forms. (90) a. b. c.

Root bP@l dNOh kmrPE:c

Continuative bl-bP@l dh-dNOh kc-kmrPE:c

‘painful embarrassment’ ‘appearance of nodding’ ‘short, fat arms’

It is very easy to describe the result of reduplication. The initial consonant is doubled and the nal consonant is inxed between the two copies of the (formerly) initial consonant. But we need to see how the computation is carried out. Assume transcription to the left. Initial-consonant doubling is easy to generate: [b ]P@l→b-bP@l, for example. Final-consonant copying to the left edge is also easy to generate: [bP@ l ]→l-bP@l. Surprisingly, if these two mechanisms are naively combined without regard to their possible interaction, they produce the desired outcome without interfering with each other: [[b]P@l ] → bl-bP@l



The step-by-step derivation is given in (91).

Truncated Reduplication

61

(91) 1. 2. 3. 4. 5.

Init CopyShift Cleanup CopyShift Cleanup

[[ b ] [[ b ]  b[[ b] b[ b bl[ b b bl 

P P P P P P

@ @ @ @ @ @

l] l] l] l] l] l

Semai continuative reduplication can therefore be specied by    (92) ∅ →  ×#; ∅ → ] #; ∅ → ] × Default Closure and left transcription do the rest. 4.3.2 Transcription Tricks that Languages Play Semai reduplication exploits the simple transcription rules to copy both ends of a string without copying the middle, and does it with the relatively simple set of juncture insertion rules (92). Semai is not the only language that exploits embedded duplicants to produce compex patterns. The logic of the derivation (91) easily shows that if u, v, w, and x are strings of timing slots, then Trscr

(93)

[u [v ]w x ] −→ vxuvwx 

Sanskrit perfective reduplication uses this trick to copy the obstruent onset and high vowel from certain roots. tu-stau is produced from stau ‘praise’, for example. After juncture insertion forms [stau ], the high vowel is isolated, [stau ]→[sta u ], then the obstruent nucleus is picked out, [s [t ]a u ]. Transcription, according to (93), produces tu-stau. I return to consider Sanskrit perfective reduplication in detail in section 7.7.2. The reader should appreciate the subtlety of the Sanskrit mechanism. It exploits a copying mechanism which is designed to copy strings of consecutive elements. It nevertheless manages to carry out a complex variation on simple string copy: rst skip some material, then copy some, then skip some more, then copy some more. Another much more common embedded duplicant trick, called double reduplication, is used in several languages. If u is a string of timing slots, then Trscr

(94)

[[u ]] −→ uuu



It is easy to see why this is true. The inner duplicant is transcribed rst, so [[u ]]→u[u ]. But then [u ]→uu. Progressive reduplication in Mokilese doubly reduplicates monosyl-

62

Chapter 4

labic roots; see section 6.3.3.2. If necessary, Kinande uses double reduplication to generate a sufficiently weighty reduplicant in cases like swa→swaswa-swaa. See example (255) in section 7.2. 4.4 Transposition, Metathesis, and Inxation Earlier, I gave formulas, (71) and (89), that showed how a duplicant with a leading-edge or trailing truncate was transcribed. Now I consider a duplicant [u v ] with both a leading-edge and trailing truncate. Supercially, nothing surprising happens. The leading-edge truncate fails to copy, as it does whether or not there is a trailing truncate, and the trailing truncate copies with source deletion, as it does whether or not there is a leading edge truncate. The result, however, is unexpected. Trscr

Trscr

(95)

[u v ] −→ vu 

[u v ] −→ vu

and

This is an important result. It shows that reduplication provides a mechanism for transposition (metathesis) and inxation. 4.4.1 Choctaw Passive Choctaw passive morphology was discussed by McCarthy and Prince (1995). They give the following examples: (96) a. b. c. d.

Stem abani apisa hok¸ci tak¸ci

‘barbecue’ ‘set a date’ ‘plant’ ‘tie’

Passive albani alpisa holk¸ci talk¸ci

Surface albani aìpisa holok¸ci talak¸ci

Certain late operations (epenthesis, vowel spreading, and l-mutation) produce the surface forms. Transparently, l is inxed after the initial vowel. A simple way to accomplish this is to prex l, then permute the prex and the initial C∗ V of the stem. This can be realized very simply by means  of a reduplicative . prex. The exponent is [l - and it triggers ∅ → ] V prex

(97)

tak¸ci −→ [ltak¸ci

JncIns

=⇒

[lta]k¸ci DC

=⇒ [lta]k¸ci

Trscr

→

ta-l-k¸ci

Right transcription is portrayed above so that inxation in the usual sense is the result, but there is no direct evidence that transcription

Truncated Reduplication

63

is to the right. Transcription to the right minimizes the amount of copying that needs to be done, but transcription to the left is less disruptive of morpheme structure. Left transcription would copy ta to the left and delete the source in the root, leaving the remnant k¸ci. The morpheme structure after transcription would be ta (not associated with a morpheme) + l (the exponent of the passive) + k¸ci (the remnant of the exponent of the root). Consideration of prosodic structure plays no direct role in this analysis. McCarthy and Prince (1995, 347) propose a Prosodic Morphology analysis, which obscures the simplicity of the morphology: “Formally, l inxation is actually prexation under negative prosodic circumscription of an initial light syllable σμ , requiring Law-of-Parsing mediated restructuring of an initial heavy σ (Urbanczyk 1992). The morphological rule, restricted in this way, is expressed by O / Φ(σμ , Left), where O = ‘Prex l’.” Insisting on a prosodic basis for all inxing morphology leads to a relatively complex account of inxing passive morphology in Choctaw. This is evidence for an important conclusion I hope to establish by the weight of multiple examples, none of them conclusive in themselves: Prosodic Morphology misidenties as instances of instances of juncture insertion in the context V prosodic specication via syllable weight. Earlier, the specication of C∗ V-reduplication was made without explicit appeal to prosody. The Choctaw passive is another example of the same kind. Choctaw passive is exactly what it seems to be, “put l after the rst vowel.” Using the copying and truncation mechanisms available to the phonology, there is only one method for the morphophonology to accomplish the required inxation. l is rst “put” at the edge of the stem by concatenation (as the exponent of a morpheme), then transposed to the post-V position by reduplication. 4.4.2 Arrernte Rabbit Talk Arrernte furnished several good examples of reduplicative prexation and suffixation. The mechanism made extensive use of truncation. Arrernte also furnishes an excellent example of how juncture insertion and transcription accomplish metathesis. According to Pensalni and Breen (1999, 7), from whom the data is taken, “Rabbit Talk is a language game that involves transposing the initial portion of a word to the end of the word, not unlike the Pig Latin of English.” Examples follow:

64

Chapter 4

Arrernte Rabbit Talk a. emen enem ‘plant food’ b. ekwen. et’ek en. et’ekekw ‘to put in’ aNkwar ‘no’ c. araNkw d. itirem iremit ‘thinking’ e. ulket. et.ulk ‘perentie’ (lizard) For polysyllabic words, the translation from Arrernte to Rabbit Talk is accomplished by two rules:   ∅→[ × , ∅ →  Vsecond (98)

An illustrative derivation is given in (99). For the sake of concreteness, right transcription is assumed, but there is no evidence favoring right over left transcription. JncIns

(99)

itirem

=⇒

DC

Trscr

[it irem =⇒ [it irem ] −→ iremit

There is allomorphy, with prexal y- inserted in the context of monosyllabic stems (which have no second vowel). For example, iNk → yiNk and emp → yemp. 4.4.3 Varieties of Yaqui Reduplication Haugen (2003) discusses various reduplicative verbal affixes in Yaqui, an Uto-Aztecan language spoken in northern Mexico and Arizona. One of the affixes is usually associated with iterative aspect, but is also used with a continuative meaning as well as various idiosyncratic meanings. I will simply identify it as “iterative” below. Two others discussed here are associated with a habitual meaning. There is a fourth affix, not discussed here, that is associated with a habitual meaning. It is identical to Diyari plural reduplication (bisyllabic reduplicant), which is discussed in section 5.2. The affixes are specied as follows: (100) a. b. c.

Prex [×[×(none)

JncIns  ∅→] V  ∅→] V  ∅→] V

Name iterative geminating (habitual) light syllable (habitual)

× is a bare timing slot. Transcription is to the left. All three affixes trigger the same stem readjustment. The effect of these affixes is illustrated in (101) for some core cases: yena ‘smoke (tobacco)’, hine ‘use as a cover’, and vusa ‘awaken’.

Truncated Reduplication

(101) prex JncIns DC Trscr spreading

65

Iterative yena [×yena [×ye]na ye-×yena yeyyena yeyyena

Geminating hine [×hine [×hi]ne [×hi]ne hi-×ne hinne hinne

Light syllable vusa vu]sa [vu]sa vu-vusa vuvusa

There are complications for iterative and light syllable reduplication that will be discussed in detail in Section 6.4.2. According to Haugen, the choice of affix between light syllable reduplication and geminating reduplication as realizations of habitual varies from speaker to speaker for some verbs and is sometimes unstable for a single speaker. He cites hinne/hihine and himmaPako/hihimaPako (himaPako ‘cut wood’) habitual alternations. Viewed at the surface level, the connection between light-syllable reduplication and geminating reduplication is obscure at best. But the analysis of geminating reduplication given above makes the connection clear. It is the alternation between the prex [×- and a null prex. The stem adjustment rule stays the same.

Chapter 5 Sources of Variation

A wide variety of reduplicative surface patterns are found in the world’s languages. In part, the differences are due to different affix specications: different exponents and different juncture insertion rules. But these differences do not account for the full range of variation. The surface pattern associated with a reduplicative affix is the result of a number of different factors, whose effect is felt at various stages in the derivation. (102)

1. 2. 3. 4. 5.

The exponent of the affix. The juncture insertion rules that the affix triggers. Pretranscription modication of the duplicant. The direction of transcription. Posttranscription operations.

Not a great deal can be said about either (102.1) or (102.4). Obviously, different exponents produce different patterns, particularly since the exponents of reduplicative affixes can contain embedded t-junctures. But since there is little that can be said of a general nature about the range of possibile exponents other than to describe what occurs, (102.1) will be left without further comment. We saw in the discussion of Klamath in chapter 3 that the possibility of root-vowel syncope, which affects the surface pattern, depends on the direction of transcription. Nothing further will be said here about the variation that can be attributed to (102.4). There are only two choices. Discussion of (102.3) is the focus of the next chapter. Modication of the duplicant before transcription is very important in some languages, playing a major role in determining the surface pattern. This requires extensive discussion, which chapter 6 is devoted to. This chapter will focus primarily on the possibilities for (102.2). Secondarily, there is some discussion of (102.5), particularly those aspects of posttranscrip-

68

Chapter 5

tion modication that are special to reduplication. 5.1 Possible Juncture Insertion Rules I begin by putting aside the question of possible domains of juncture insertion rules, which is taken up in the next section, and simply assume that juncture insertion rules apply in some domain. The goal is to show that a very small inventory of juncture insertion rules suffices for an empirically adequate theory. Insertion sites for junctures are located with respect to the syllable nuclei in the domain, as well as the edges of the domain. To make this idea precise, we need the notion of the “nuclear skeleton” of the domain. Consider, for example, the representation in (103), where the nuclei ν are explicitly represented on their own plane. I adopt here the original proposal of Clements and Keyser 1983, which placed nuclei and syllables on parallel tiers.1 The nuclear skeleton, which is embedded in the full representation, is shown at the right. ν ν ν ν σ σ (103)

× × × × ×

Nuclear skeleton: × × × × ×

p O d o k Morphological associations are not shown in (103), but it is assumed that the full representation and its nuclear skeleton also share the same morphological associations. The starting point for establishing the inventory of juncture insertion rules is the proposal that juncture insertion  rules are operations must be possible on the nuclear skeleton. Rules like ∅ → ] ν juncture insertion rules, but there is an ambiguity that must be claried. Insertion is into the timing tier, but ν is an autosegment, not mean? If ν is on the timing tier. What does the environment ν linked to a single timing slot, there is no ambiguity: the meaning is “to the right of the timing slot linked to ν.” In the case of a long nucleus (a long vowel or long diphthongal nucleus), an ambiguity of the kind familiar from considerations of geminate inalterability arises. Consider ν ν (104) × × × × × ×

Sources of Variation

69

The environment ν can be interpreted narrowly to mean “immediately to the right of the set of timing slots linked to a ν,” or interpreted more broadly to mean “immediately to the right of one of the timing slots linked to a ν.” The broad interpretation species the locations in (105a) and the narrow interpretation species the locations in (105b). ν ν ν ν b. × × × × × × (105) a. × × × × × × I assume that both interpretations are available, with the broad interpretation the default and the narrow interpretation (105b) a marked option. In chapter 7, the marked option plays a central role in the analysis of Sanskrit and Tohono O’odham, an indigenous North American language. Nuu-chah-nulth, another indigenous North American language, from the Wakashan family, has a number of affixes that trigger characteristic reduplication patterns. (See Kim 2003 for many examples and classication.) The affix -Pal∼uk ‘to look after’ (p. 142 in Kim) induces C∗ν-reduplication of the stem. (106) ˇcapx. ‘man’ nuukw ‘song’

ˇca-ˇcapx. Pal∼uk ‘to look after a man/husband’ nuu-nuukwPal∼uk ‘to look after songs (as a caretaker)’

Along with concatenation of the suffixal exponent, a ]-juncture is inserted to the right of the nucleus (i.e., to the right of all of the timing slots associated with the nucleus).2 (107) ˇcapx. Puukw

Suffix/JncIns

DC

Trscr

ˇca]px. Pul∼uk Puu]kwPul∼uk

[ˇca]px. Pul∼uk [Puu]kwPul∼uk

ˇca-ˇcapx. Pul∼uk Puu-PuukwPul∼uk

The notation / ν will be used in what follows when the narrow when the broad interpretation is interpretation is intended and / V intended. More generally, the symbol V appearing in the environment of a juncture insertion rule will represent the predicate that is true of timing slots linked to a nucleus, regardless of the phonemic association of the timing slot. In the same way, the symbol C will stand for the predicate that is true of timing slots not linked to a nucleus. We are now in a position to propose that juncture insertion rules are always drawn from the array in (108).

70

(108)

Chapter 5

Possible juncture insertion rules ⎧ ⎫ ⎧ ⎫ V⎪ [⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪     ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ leftmost ⎬ ⎨ ] ⎬ before ⎨ × ⎪ , in the domain of ρ Insert ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ rightmost ν⎪ after ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ ⎭ ⎩ ⎭ C 

One other general issue of rule application is important for some reduplicative processes. I assume: (109)

Balancing an unpaired duplication juncture, forming a nontrivial duplicant if possible, takes precedence over leftmost or rightmost application.

The signicance of (109) is illustrated by the reduplicative process used to realize plurality in Mangarayi, which has been of interest for a long time. The data is from Merlan 1982. (110) a. b. c.

Root gal.ugu waNgij jimgan

Plural gal.al.ugu waNgaNgij jimgimgan

‘poor things’ ‘children’ ‘knowledgeable people’

The affix has a null exponent and triggers readjustment:   (111) ∅ → [ V; ∅ → ] V Take (111) to be an ordered list of rules. Consider, for example, its V applies to the stem application to (110c), jimgan. First, ∅ →  [ V applies. The leftmost and produces j [imgan. Then ∅ → ] application that forms a nontrivial duplicant is chosen. The result is therefore j [img ]an, rather than j []imgan. Right transcription then produces jimg-img-an. 5.1.1 Creek Plural-Adjective Reduplication Riggle (2004) discusses an unusual pattern of reduplication that is used to realize adjective plurality in Creek, based on data from Haas 1977, Booker 1980, and Martin and Mauldin 2000. (112) a. b. c. d.

Singular fay´atk-ii c´amp-ii low´ack-ii pol´ook-ii

Plural fayat-fa-k-´ıi cam-ca-p-´ıi low´ac-lo-k-ii poloo-po-k-´ıi

‘crooked’ ‘sweet’ ‘soft’ ‘round’

This is a particularly clear example of the usefulness of Default Closure

Sources of Variation

71

in simplifying the specication of complex reduplication processes. There are two juncture insertion rules:   ∅ → ] Right ×, ∅ →  V Transcription is to the right. Derivations follow: (113) a. b. c. d.

fayatk camp lowack polook

JncIns fayat]k cam]p lowac]k poloo]k

DC [fayat]k [cam]p [lowac]k [poloo]k

TrScr fayat-fa-k cam-ca-p lowac-lo-k poloo-po-k

A special rule is used if the root ends in certain consonant clusters.3 5.1.2 Some Common Juncture Insertion Rules It is useful to have names for some of the common juncture insertion rules and the kind of reduplication they generate (if there are no other t-junctures except those inserted by Default Closure).  (C∗ V-reduplication) (114) C∗ V-rule: ∅ → ] V  ∗ (C∗ν-reduplication) C ν-rule: ∅ → ] ν  ∅→] × (initial timing slot reduplication) ×-rule:  ∗ × -rule: ∅ → [ × (total reduplication) 5.2 Possible Domains of Reduplicative Affixes The juncture insertion rules that a reduplicative affix species are determined in a composite fashion. Rules drawn from the possibilities enumerated in (108) are specied, but a domain of application of these rules that is smaller than the whole stem can also be specied. For example, a rule like  ∅ → ] Right V has a very different effect if its domain of application is the initial foot of the stem or is the entire stem. If an affix does not designate a special domain, the rules are assumed to apply to the stem. In other words, the stem is taken to be the default domain for juncture insertion rules. The designated (or default) domain is called the domain of the affix. The domain of a reduplicative affix can be specied on the basis of the prosodic structure of the stem or on its morphological structure. In all

72

Chapter 5

the examples that I am aware of in which the morphological structure of the stem is used in domain specication, a morphological constituent of the stem is designated as the domain of the affix. Most morphologically based stem designation is designation of the root, embedded in an inected stem. The theory of prosodically based contraction that I adopt derives from Broselow and McCarthy 1983. They argue that several varieties of reduplication are most insightfully analyzed (in the template concatenation framework that their theory is framed in) by supposing that “reduplicative morphemes may be prexed not only to morphological but to phonological constituents, as is argued for Warlpiri prexing reduplication by Nash (1980)” (p. 53). This idea was further developed by McCarthy and Prince within the framework of Prosodic Morphology as the notion of “positive prosodic circumscription.” (See the discussion in McCarthy and Prince 1995). A key insight of Prosodic Morphology was that only special kinds of prosodic constituents could serve as the locus of affixation: those that had wordlike characteristics. It was established that if a language imposed a prosodic minimality condition on words, the minimality requirement was that there be sufficient material to form a well-formed foot. In other words, the prosodic constituents that are “wordlike” in the appropriate sense are feet. Since I do not know of any counterexamples, I will assume that specication of a prosodic constituent as the domain of a reduplicative affix is restricted to specifying it to be the leading-edge foot: the initial foot in left to right footing, or the nal foot in right to left footing. The Yidiny reduplicative plural affix (Dixon 1977; Nash 1979) species the initial bisyllabic foot (shown as a grouping of syllables below) as its domain, a null exponent, and triggers the ×∗-rule. This leads to derivations like (115). σ ) σ ( σ σ ) σ ( σ ×∗-rule (115) × × × × × × × × −→ [ × × × × × × × × g i n d a l b a g i n d a l b a ( σ σ ) σ −→ [ × × × × × ×] × × → gindal-gindalba g i n d a l b a DC

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

Sources of Variation

73

The Diyari plural affix (Austin 1981), like the Yidiny plural affix, species the initial bisyllabic foot as its domain and a null exponent. Unlike Yidiny, it triggers the rule.  ∅ → ] Right V This leads to derivations like σ ) σ ( σ σ ) σ ( σ (116) × × × × × × × × → × × × × ×] × × × N a n k a n t i N a n k a n t i ( σ σ ) σ −→ [ × × × × ×] × × × → Nanka-Nankanti N a n k a n t i DC



∅ → ] Right V

applies rightmost in the domain of the affix, the initial foot. Both the Yidiny and Diyari reduplicative plural affixes are discussed in McCarthy and Prince 1986. 5.2.1 Ulwa Possessive The original analysis and source of subsequent data is Hale and Blanco 1989. Other analyses include Bromberger and Halle 1988 and McCarthy and Prince 1995. In Ulwa, various possessive morphemes are realized by affixes that, descriptively, induce inxation of an exponent characteristic of the morpheme. The position of the inx is determined by the foot structure of the stem. Footing is mora based and left to right binary. The paradigm for the 3sg possessive is given in (117). The initial foot is parenthesized. (117)

Ulwa possessive inxation Stem Possessed a. (bas) bas-ka b. (sa na) sa na-ka c. (suu) lu suu-ka-lu d. (bas) kar na bas-ka-kar na e. (si wa) nak si wa-ka-nak f. ka (ras) mak ka ras-ka-mak

‘hair’ ‘deer’ ‘dog’ ‘comb’ ‘root’ ‘knee’

In (117f), all that is relevant to the analysis here is the location of the right boundary of the initial foot. The affix designates the initial foot as its domain, has the suffixal

74

Chapter 5

exponent -ka] and triggers  (118) ∅ → [ Right × The juncture insertion rule is relativized to the designated domain of the affix, but suffixation is to the stem. Suffixation and prexation are strictly concatenative in DR. There is no inxation per se, although from time to time the term will be used descriptively to refer to a process with a certain surface form. A typical derivation is shown in (119). Transcription is taken to be to the left. Only the initial foot is shown. (119)

(σ σ) ( σμ σμ ) σμμ suffix × × × × × × × −→ × × × × × × ×× ×] s i w a n a k s i w a n a k k a (σ σ) DC −→ × × × ×[ × × ×× ×] → × × × ×[× × ×× ×] s i w a n a k k a s i w a n a k k a [-Ins

Trscr

−→ × × × × × × × × × s i w a

(siwa-ka-nak)

n a k k a

Note that exponent of the affix is a suffix, not an “inx” of any kind. It is the t-junctures contained in the prex and the juncture insertion rule that its insertion triggers that cause ka to be transposed in the phonology and surface as an “inx.” Note also that (118) inserts a [-juncture after the rightmost timing slot of the domain. 5.2.2 Manam Final Foot Reduplication Manam has an affix that induces total reduplication of the nal foot. Footing is quantity sensitive and right to left. The affix takes the nal bimoraic foot as its domain, has a null exponent, and triggers the ×∗rule. Two derivations follow, one with a bisyllabic nal foot and one with a monosyllabic nal foot. The designated domain is boxed. sa la ga

mala boN

×-rule

sa [laga

ma la [boN

Default Closure

sa [laga]

mala [boN]

Trscr

salaga-laga

malaboN-boN

(120)

Sources of Variation

75

5.2.3 Exclusion of Initial Unfooted Material There are many examples in which prosodically defective material at the edges of a word is excluded from the foot structure. It can therefore be excluded from the domain that a reduplicative affix designates. 5.2.3.1 Bella Coola Bagemihl (1991, 598, 609), gives the following examples of what he calls CV reduplication in Bella Coola: (121)

a. b. c. d. e.

qayt tqn­ k stn­ t’ksn­ st’xwm­

qa-qayt t-qn­ -qn­ k s-tn­ -tn­ t’k-sn­ -sn­ st’-xwm­ -xwm­

He argues convincingly that Bella Coola syllable structure admits at most one obstruent into the onset and that the inxing pattern (121) is a consequence of this syllable structure. Nonvocalic sonorants are often syllable nuclei. The examples (121) illustrate two features  of the discussion above. must be interpreted First, V in the juncture insertion rule ∅ → ] V as being true of syllable nuclei in general, not just vowels. Second, there is domain selection. The domain of the affix that generates (121) excludes the unsyllabied initial obstruents. Derivations of some of the examples in (121) are given in (122). The domain is the initial foot, which excludes the unsyllabied initial obstruents. There is no minimal word condition in Bella Coola. The crucial point is that Juncture Insertion (JncIns) and Default Closure (DC) are relativized to the selected domain. st’ xwm­

qa]yt

t qn­ k t qn­ ]k

Default Closure

[qa]yt

t [qn­ ]k

st’ [xwm­ ]

Trscr

qa-qayt

t-qn­-qn­ k

st’-xwm ­ -xwm­

qayt

(122) C∗ V-rule

st’ xwm­ ]

5.2.3.2 Orokaiva The stress system of Western Aranda, as Halle and Vergnaud (1987, 48) showed, gives good evidence that UG makes available the option of excluding initial onsetless syllables from the computation of foot structure. Western Aranda has a bimoraic word minimum, so this exclusion is only available in that language when at least two moras are not excluded. In a language whose metrical

76

Chapter 5

structure employs this device, initial onsetless syllables can be excluded from the domain of a reduplicative affix. Orokaiva (Healy, Isoroembo, and Chittleborough 1969; McCarthy and Prince 1986) has a reduplicative affix whose domain is the initial foot, whose exponent is null, and that triggers the C∗ V-rule. It realizes repetitive verbal aspect. Like Western Aranda, the initial foot excludes an initial onsetless syllable, if there is one. Illustrative derivations follow: (123) C∗ V-rule Default Closure Trscr

hiri ke

u huke

hi]ri ke

u hu]ke

[hi]ri ke hi-hirike

u [hu]ke u-hu-huke

5.2.3.3 Imdlawn Tashlhiyt Berber Dell and Elmedlaoui (1988) discuss the following paradigm from Imdlawn Tashlhiyt Berber: (124)

Imperfective gemination Root Imperfective a. mrz mmrz b. frn ffrn c. rkm rkkm d. kSm kSSm

‘wound in the head’ ‘sift’ ‘rot’ ‘enter’

They argue that the contrast between initial gemination and second-slot gemination is a precise reection of syllable structure, with (124a–b) having initial syllables with onsets, (124c) having an initial onsetless syllable, and (124d) having an initial unsyllabied obstruent. Obstruents can be nuclei word internally, but not in word-initial or word-nal position. The syllable structures are σ σ σ σ σ σ σ b. × × × c. × × × d. × × × (125) a. × × × m r z f r n r k m k S m Descriptively, the rst onset geminates. Under the assumption that these stems have metrical structure (there is no word minimality) and that the leftmost left foot boundary is aligned with the leftmost syllable that has an onset, this pattern is generated by an null affix thatdesignates the initial foot as its domain . and triggers the ×-rule, ∅ → ] ×

Sources of Variation

σ(σ) × × × r k m

×-rule

−→

77

σ(σ) σ(σ) DC Trscr × × ] × → × [ × ] × −→ × × × × r k m

r k m

r

k m

Transcription is to the left, the default for left-edge duplicants.4 This is one of the few examples in which the representation that transcription produces does not violate the NCC. The imperfective geminate is the direct product of duplication, with no NCC repair. Note also that the example demonstrates fairly clearly the inadequacies of templatic approaches to reduplication. A template can neither specify what is to be duplicated nor predict that the output is a geminate rather than simply a doubled consonant. One nal point should be noted. Ideally, the correlation between the reduplication patterns that are found in the world’s languages and the reduplication patterns that are easy to describe would be transparent. To a considerable extent, there is such a correlation. The two most widespread patterns are total reduplication and C∗ V-reduplication, both of which have very simple rules. But the rule that species initial gemination is just as simple and it is rare. Why should this be so? Juncture insertion must mesh with the rest of the phonology of the language. If transcription produces a representation that violates the syllable well-formedness conditions of a language, then substantial phonological support is needed to transform the output of transcription into an acceptable form. The result is computational complexity and an opaque reduplication process. The implicit phonological support that may be necessary means that the complexity of a reduplication process cannot be evaluated simply on the basis of the simplicity of the juncture insertion rules involved. The reason that initial gemination is rare is not because the juncture insertion rules are complex, but because it is rare to nd 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 processes due to the variation in juncture insertion, factored into two components, domain selection and rule specication relativized to that domain. In most cases, NCC repair is simple phoneme ssion. There are a number of languages, however, in which special rules are used to repair the transcribed structure, adding further possibilities to the range

78

Chapter 5

of possible reduplicative processes. What I call “shortcut repair” is used by several languages and is the main example of a specialized process of NCC repair that has a signicant impact on the surface reduplication pattern. Kolami echo reduplication (Emeneau, 1955; McCarthy and Prince, 1995), is illustrated by the examples (126) from McCarthy and Prince. (126)

a. b. c. d.

pal kota saa maasur

pal-gil kota-gita saa-gii maasur-giisur

‘tooth’ ‘bring it’ ‘go’ ‘men’

The pattern is generated by the suffix -gi ] , coupled with the juncand a special mechanism that some ture insertion rule ∅ → [ V languages (including Kolami) use for eliminating crossing violations at the reduplicant-remnant boundary. Examples (126a–b), with a short nal vowel, are straightforward. suffix

(127)

× × × × −→ × × × × × ×] k o t a g i k o t a JncIns

−→

Trscr

× × [ × × × ×] −→ × × × × × × × × k o t a g i k o t a g i

Now consider the case of a long nal vowel, (126c–d). For saa, for example: Trscr

(128)

× × × → × × [ ×  × ×] −→ × × × × × × s a s a g i s a g i

Fission would yield sa:gia, with vowel hiatus, not the desired saagii. The desired result follows if we assume that crossing violations can be removed not only by Fission, but also by (129), which not only eliminates a crossing violation but also eliminates vowel hiatus. (129)

× ...

×

V1

V2

×

If (128) continues by eliminating the crossing violation using (129) the result is the desired:

Sources of Variation

(130)

79

× × × × × × s

a

g i

Note that no vowel is lost in the nal representation. At most an occurrence of a vowel is lost, but one occurrence remains. Informally, the operation can be looked at as hiatus elimination licensed by a kind of recoverability of deletion. The repair mechanism used by Kolami at the reduplicant-remnant boundary occurs often enough to merit a designation, Shortcut NCC Repair, and some discussion. The usual context in which it occurs is given in (131), where a CVC reduplicant has been formed to the left of the remnant. To make the shortcut more visible, the rst step has carried out partial NCC repair, leaving only the crossing violation at the boundary between the reduplicant and remnant unresolved. This is simply to make the shortcut operation clearer. The operation removes a crossing violation whether the other crossing violations have been removed or not. (131)

× × × × × × ... → C1 V C2

× × × × × × ... C1 V

C1 V C2

Shortcut Repair

−→

× × × × × × ... C1 V

C1 V C2

In Ponapean heavy-syllable reduplication, which is analyzed in detail in chapter 6, Shortcut Repair applies if C1 and C2 are sonorant and C1 is coronal, otherwise Fission applies. (132)

Ponapean durative reduplication Fission Shortcut Repair liro:ro ‘contract’ *lir-liro:ro lil-lirliro nur ‘protective’ *nur-nur nun-nur tep ‘begin’ tep-tep tepitep *tet-tep

Hausa pluractional reduplication (Newman 2000) is the clearest illustration of (132), since in many contexts either Fission or Shortcut Repair can be used, resulting in surface variation. The process copies a CVC to the left.5 If C2 is nonsonorant and noncoronal, Shortcut Repair is obligatory, otherwise there is free variation between Fission and Shortcut Repair. In the examples (133), the surface forms, in which

80

Chapter 5

standard Hausa phonological rules apply after transcription, are shown in parentheses. (133)

Hausa pluractional reduplication Fission Shortcut Repair fita fit-fita fif-fita (>fi˜rf`ıta ) (>fiff`ıta ) gas¯a gas-gas¯a gag-gas¯a (>ga˜rg`as¯a ) (>gagg`as¯akir¯a ) say¯a say-say¯a sas-say¯a (>s`ais. ay¯a` ) (>s`assay¯`a ) tambay¯a tam-tambay¯a tat-tambay¯a (>t`ant`ambay¯a` ) (>t`att`ambay¯`a )

‘go out’ ‘roast’ ‘buy’ ‘ask’

Chapter 6 Prosodic Adjustment

After readjustment inserts t-junctures and Default Closure applies, creating a duplicant, the duplicant can be modied in the phonology before transcription applies. In many languages, prosodic goals are imposed on the duplicant and rules specied for adjusting the duplicant so that the goals are satised. Later in the section, the set of possible prosodic targets and set of possible adjustment rules will be constrained so that the range of options made available by the theory is limited, corresponding to the range of options that are attested. Before I address those issues, however, several examples will be discussed. 6.1 Onset Incorporation in K´hehe Unintensive Reduplication In (134), ita, lima, and i:mbila are verb roots, and ku- is an innitival prex. The examples are from Odden and Odden 1985. (134) Reduplicated a. kuce´enga ‘to build’ ku-ceenga-ce´enga b. kulima ‘to cultivate’ ku-lima-lima c. kwiita ‘to pour’ d. kwiimbila ‘to sing’

‘to build a bit’ ‘to cultivate a bit’ kwiita-kwiita ‘to pour a bit’ kwiimbila-kwiimbila ‘to sing a bit’

Odden and Odden say that “any high vowel before another vowel becomes a glide, with compensatory lengthening of the following vowel.” In (134c–d), ku +i→kwii. In the theoretical framework that Odden and Odden had available to them at the time, Marantz’s template concatenation theory, (134a–b) forced the conclusion that the template is concatenated before the prex applies. They pointed out that (134c–d) then pose a major problem. Example (134c), for example, must be derived from something like

82

(135)

Chapter 6

ku-

-ita

An empty box represents the template. The operations that ensue are devoted to furnishing the melodic content of the template. There is no plausible motivation for copying the initial ku to the right edge of the template, which is what would be required to produce kwiita-kwiita. Odden and Odden’s solution was to move the template off the timing tier and to separate the operation of template concatenation from the operations that provided the template with melodic content. They proposed that the melody was not lled in immediately, but was lled in after syllabication fused the prex with the root. They further proposed that the operations that lled in the melody were formulated in such a way that material attached syllabically to the root was associated with the template along with the root itself. This analysis has various problematic aspects, particularly concerning the precise meaning of “concatenation of a template,” but the idea that the immediate changes in the representation induced by the morphology are only a prelude to the ultimate effects that these changes lead to is an important insight that DR has incorporated. The DR account of the K´hehe phenomenon has two main ingredients. First, prosodic goals can be associated with duplicants and certain operations specied for use in adjusting the duplicant so that those goals are satised. Second, transcription can occur signicantly later than juncture insertion. Particularly if the duplicant does not satisfy the prosodic goals, transcription can be put off until the goals come to be satised or transcription is forced for some independent reason. In (136), well-formed syllable structure is a duplicant goal, with Vinitial duplicant syllables taken to be ill-formed. [-Left is a specied adjustment operation. The cycles in the derivation of (134c) are shown in (136). (136)

root

unint nonfin

ita [ita] ku [ita] → [kwiita] → kwiita-kwitta

Transcription is suppressed in the unint-cycle, either because phonology dictates that it be suppressed if the duplicant does not satisfy the prosodic goal or because the phonology always delays transcription until a later point in the derivation. The unint-cycle could be noncyclic, for example. After ku is concatenated, [-Left is used to bring the duplicant into satisfaction of the goal. Transcription takes place later.

Prosodic Adjustment

83

Different languages use different adjustment rules to accomplish the same prosodic objectives. K´hehe uses [-Left to eliminate the onsetless syllable in the examples above. In Asheninca Campa, discussed in detail in section 7.3, [-Right is used in some environments. ∗σ σ σ σ σ PrAdj (137) ×[× × × × × ×] −→ × ×[× × × × ×] → no-sampi-sampi n o s amp i

n o s amp i

The prex n- is a rst-person marker, osampi ‘paddle’ is a verb root, and reduplication realizes a kind of intensive prex, usually glossed as ‘more and more’. Note that the result is not nosampi-nosampi, which is what [-Left would produce. Asheninca Campa generally uses [-Right, as above. But if this conicts with satisfaction of a second duplicant prosodic goal, that the duplicant be at least bisyllabic, [-Left is used. In (137), the duplicant is contracted, but in similar V-initial cases in Asheninca Campa with a short root, the duplicant is augmented. For example, with the same prex as (137) and the verb root asi ‘cover’:prosodic goalbisyllabicity ∗σ σ σ σ PrAdj (138) × [ × × ×] −→ [ × × × ×] → nasi-nasi n a s i

n a s i

In K´hehe, the duplicant is always augmented, even with long roots, as shown by (134d). 6.2 Chumash CVC Reduplication and Apparent Countercyclicity CVC-reduplication in Chumash is associated with intensive, distributive, repetitive, or continuative semantic force. CVC is descriptive. Chumash has no long vowels, so CVC reduplication is heavy syllable reduplication. A few typical forms, taken from Inkelas and Zoll 2000, which is the source for all the examples in this section, are given below. The primary sources are Applegate 1972, 1976. (139) a. b. c. d. e.

s-kitwon k-ni-ˇc’eq s-ikuk s-iˇs-expeˇc k-su-towiˇc

Reduplicated s-kit-kitwon k-ni-ˇc’eq-ˇc’eq sik-s-ikuk s-i-ˇsex-ˇs-expeˇc k-ˇsut-ˇsu-towiˇc

‘it is coming out’ ‘I’m tearing it up’ ‘he is chopping, hacking’ ‘they two are singing’ ‘I’m doing it fast’

84

Chapter 6

The verb root is on the right and s- and k- are agreement prexes. Various other prexes, called inner prexes and exemplied by ni- , iˇs- , and su- above, can appear between the root and the agreement prexes. Subject agreement originates outside the VP and at least some of the morphemes realized by CVC reduplication (distributive, for example) appear to be verbal modiers that occur in the VP domain. This leads to the expectation that the order of combination of the morphemes is (root + ρ) + agr, where ρ is the reduplicative morpheme. Note that in (139c), the subject-agreement prex is reduplicated. This is the K´hehe problem again. Theories that view reduplication as template prexation and assume that the morpheme order is (root + ρ) + agr must explain how (140) produces sik-sikuk. (140)

-ikuk

s-

McCarthy and Prince (1995) gave such an account in the framework of what they call Correspondence Theory (CT). They proposed, in effect, that two constructs are added to the structure by the realization of ρ, a reduplicant (as above) and a new construct, a base. The result is something like (141)

- ikuk

sredup

base

It is then the responsibility of the phonology to not only furnish a melody to the reduplicant, but to ensure that the reduplicant and the base suitably correspond to each other. McCarthy and Prince give an Optimality Theory analysis, but the idea of correspondence could be worked out equally as well in a standard derivational framework. The derivation (142) would ensue.1 Operations that progressively bring the template into compliance with the goal condition alternate with operations that bring the reduplicant and base into correspondence. Below, CVC and B = R above the arrows indicate the desideratum that motivates the operation. (142)

Correspondence Theory derivational solution root ρ AGR

ikuk ikuk s ik ik uk

CVC

→

CVC

→

ik ikuk

B=R

sik ik uk

→

ik ik uk

B=R

→

sik sik uk

It is widely believed that CT is the crucial ingredient in McCarthy and

Prosodic Adjustment

85

Prince’s solution to the problem because imposing correspondence is the only way for a change in the reduplicant to produce a change in the base. McCarthy and Prince say: “The base then, copies the reduplicant. In serial theories, this is an impossibility” (p. 60). The belief that CT is essential is without foundation. What is crucial in the McCarthy and Prince account is the introduction of constructs into the representation that the phonology can manipulate to produce the desired result. Although it is possible to rework the CT account as a serial account, DR offers a simpler and more explanatory alternative. The derivation is given below, with a box used rather than junctures, in order to facilitate  comparison with the CT account. ρ triggers ∅ → ] V , which in the notation used here boxes the material up to and including the rst vowel. Other material is incorporated into the box because of the CVC goal shape as it becomes available. The key point is that transcription of the duplicant does not occur until the CVC goal is achieved. (143)

Distributed Reduplication solution Root ikuk ρ

i kuk

AGR

s ik uk

CVC

→

CVC

→

ik uk sik uk

Trscr

→

sik-sikuk

A dashed box has been used since the duplicant in DR is much more like what is called the base in CT than it is like the reduplicant in CT or in Marantz’s original conception. Crucially, it does not have the properties of a prex or suffix. The duplicant, like the “base” of CT, is a grouping imposed on the stem that the phonology can refer to and manipulate. But there is no construct “reduplicant” that the phonology can manipulate. In DR, the notion of reduplicant is descriptive, referring to the results of transcription. There is no notion of base-reduplicant correspondence, simply the fact that the reduplicant is formed by transcription. In (142), the base introduced by ρ persists and is affected by morphophonological events that take place after the prex s- is later concatenated. Similarly, the duplicant introduced by ρ in (143) persists and is affected by later morphophonological events. The incorporation of phonemic material from a higher morpheme into the phonetic expression of a lower morpheme appears to be highly countercyclic. DR explains why this apparent countercyclicity is only apparent. The structural change caused by reduplication does not im-

86

Chapter 6

mediately produce a reduplicant. It introduces only a “promissory note” that material will be copied at some point in the derivation. It does not promise precisely what will be copied, only that the initially demarcated material will be taken as the starting point for determining what is to be copied. Languages are free (within prescribed limits) to make modications of the initially specied material. Languages are free (up to a point) to determine when the copying takes place. 6.2.1 Inxation Accounts One response to the apparent contradiction between the compositional order (ikuk + ρ) + agr and reduplicant sik is to deny that ρ is inside subject of agreement. Inkelas and Zoll (2000) work out an analysis along these lines, proposing that ρ is structurally higher than agr but is realized by prexation to an appropriate substem of the stem, which is established on mixed morphological and prosodic grounds. Prexation to a substem is inxation if the left edge of the substem and the left edge of the stem do not coincide, hence the use of “inxation analysis” to describe the approach. The substem that is the target of prexation is dened in the following way. Inkelas and Zoll rst note that certain prexes can reduplicate: (144)

a. k-su-towiˇc b. k-su-pˇseP

‘I’m doing it fast’ ‘I’m putting out the re’ c. k-sili-pi-wayan k-sili-piw-pi-wayan ‘I want to swing’ d. k-xu-ni-y1w k-xun-xu-ni-y1w ‘I am looking all over for it’ k-ˇsut-ˇsu-towiˇc k-ˇsup-ˇsu-pˇseP

They partition the prexes into two subsets, the reduplicating prexes and the nonreduplicating prexes. The target substem is dened to be the smallest prosodically well-formed substem that contains the root and all reduplicating prexes. The target substem is set off from the presubstem material below by a period. (145)

root  ]] [[IS[[S-]] ρ

expeˇc iˇs-expeˇc s-iˇs-expeˇc si . ˇsexpeˇc →si

ˇsexpeˇc →si ˇsex ˇsexpeˇc

A version of the inxation analysis is available in DR. The last line above is replaced by

Prosodic Adjustment

(146)

ρ

si.ˇsexpeˇc

87 JncIns

→

siˇs e xpeˇc

CVC

→

si ˇsex peˇc

Trscr

→

si-ˇsex-ˇsexeˇc

Juncture insertion targets the leftmost vowel in a reduplicating prex or the root and adjusts the duplicant to CVC. Since little is known about Chumash clausal and VP structure, there is little evidence from Chumash that can be brought directly to bear in deciding the issue of morpheme order, so the inxation analysis (145) cannot be denitively rejected. But there are general considerations that make it an undesirable solution to the problem. The major weakness in the inxation account is that it forsakes any possibility of correlating the word syntax with the possibility that a particular inner prex reduplicates. The simplest system from the standpoint of compositional semantics would be for the reduplicative morpheme to be in a very local relation with the target of reduplication, ideally sisterhood. It seems to be the case, for example, that no agreement prexes are reduplicating. If they reduplicate, it is only because they furnish an onset to a reduplicating prex. If agr is hierarchically above int, there is an explanation for this. In the inxation analysis, it must be stipulated. It is important to note that an analysis that attempts to give a structural account of the fact that certain prexes do not reduplicate is compatible with the occurrence of nonreduplicating prexes inside reduplicating prexes. Causatives generally create something similar to a biclausal structure. This could create positions for morphemes realized by nonreduplicating prexes under the causative, which itself is realized by a reduplicating prex. It may not be an accident that all but one of the examples Inkelas and Zoll give of nonreduplicating prexes inside reduplicating prexes are with causative su-. The single other example has the reduplicating prex xul- of unknown identity. The inxing analysis gives up any hope of providing a structural explanation for the occurrence of nonreduplicating prexes inside the causative prex. While there are no decisive arguments against the idea of a high ρ and reduplicative inxation, the same cannot be said about the particular implementation of this idea that Inkelas and Zoll propose. In the Morphological Doubling Theory they employ, a fully formed duplicant is inxed, not a template. The theory proposes that there are three different phonologies at work. The rst two independently compute the stem and the reduplicant, and the third takes the reduplicant and

88

Chapter 6

stem as separate inputs and positions the reduplicant with respect to the stem. In example (147), it carries out inxation: true prexal inxation, not templatic inxation. (147)

Stem phonology s + iˇs + expeˇc → si.ˇsexpeˇc Reduplicant phonology s + iˇs + expeˇc → ˇsex Combination phonology ˇsex + si. sexpec → si-ˇsex-ˇsexpeˇc

Not surprisingly, in view of the vast expansion of the machinery available in this theory of reduplication, unattested and improbable reduplication patterns are easily generated. It is an accident, for example, that the reduplicant is positioned adjacent to the material that it is derived from. One could just as well have assumed that the reduplicant phonology computed the rst CVC of the stem, but positioned it just as in (147): (148)

Stem phonology s + iˇs + expeˇc → si.ˇsexpeˇc Reduplicant phonology s + iˇs + expeˇc → siˇs Combination phonology siˇs + si. sexpec → si-siˇs-ˇsexpeˇc

The result is that the reduplicant is positioned inside the portion of the stem that it copies. As far as I know, such reduplication is impossible. 6.3 Augmentation of the Duplicant to a Heavy Syllable The prototypical heavy-syllable reduplication pattern is (149)

a. b. c.

lantu → lan-lantu laanu → laa-laanu lanu → lan-lanu

Heavy-syllable reduplication has the effect of concatenating a heavysyllable surface prex to the stem, regardless of the weight of the initial syllable of the stem. In the term surface prex, prex is used in order to describe the surface effect, but surface is used to emphasize that the vocabulary item that produces the surface effect is not itself a prex. Heavy-syllable reduplication attracted a great deal of attention when reduplication was rst examined in a generative framework because the portion of the stem that is copied is not necessarily a natural constituent of the base. In (149c), it consists of the initial syllable of the stem plus the onset of the following syllable. This played an important role in Marantz’s (1982) CV template theory. Subsequently, Levin’s (1985) analysis of heavy-syllable reduplication in Mokilese and Hayes and

Prosodic Adjustment

89

Abad’s (1989) analysis of heavy-syllable reduplication in Ilocano were important in establishing the inadequacy of CV templates for analyzing reduplication and supporting McCarthy and Prince’s (1986) theory of Prosodic Morphology. Heavy-syllable reduplication in four closely related Austronesian languages is analyzed in this section: Ilocano, Agta, Mokilese (two dialects), and Ponapean. A comparison of the differences between these closely related languages is both instructive about the sources of variation from the core pattern given in (149) and, since there has been extensive work on reduplication in these languages, useful in illuminating the special features of the DR theory of reduplication. In addition to Levin’s work on Mokilese and Hayes and Abad’s work on Ilocano, work by McCarthy (1984) and McCarthy and Prince (1986) on Ponapean were important steps in the evolution of Prosodic Morphology. I begin with one of the Mokilese dialects, because the reduplication process in this dialect is the simplest. The source for a description of Mokilese is Harrison 1973, 1976. 6.3.1 Mokilese Progressive I begin with relatively simple cases before considering some complications. The analysis here takes the important analyses of Levin 1985 and McCarthy and Prince 1986 as a starting point. (150) a. b. c. d. e. f.

Root pOdok kasO wadek wia pouce dauli

Progressive pOd-pOdok kas-kasO wad-wadek wii-wia poo-pouce daa-dauli

‘is planting’ ‘is throwing’ ‘is reading’ ‘is doing’ ‘is connecting’ ‘is passing by’

Source [pOd]ok [kas]O [wad]ek [wii]a [poo]uce [daa]u

Assuming that the progressive forms are the result of transcribing representations with embedded t-junctures, the pretranscription sources can be deduced without difficulty and are given in the last column. The pretranscription sources can be generated as follows. The exponent of the progressive affix is null, but it triggers the C∗ V-rule. Default Closure follows. The duplicant is then adjusted before transcription applies. The mechanics of the adjustment are simply stated. ]-Right C, otherwise FCVL (First Conjunct Vowel applies in the context Lengthening) applies (see section 4.2). Two illustrative derivations

90

Chapter 6

follow: (151)

C∗ V-Rule DC ]-Right Trscr

kasO ka]sO [ka]sO [kas]O kas-kasO

C∗ V-Rule DC FCVL Trscr

wia wi]a [wi]a [wii]a wii-wia

The crucial steps of the second derivation are repeated here in more detail. (152)

[ × ×] ×

FCVL

−→

w i a

Trscr

[ × ××]× −→ × × × × × × w i

a

w i a

6.3.2 Goal-Driven Rules With an understanding of the mechanics of the adjustment process, it is possible to understand its coherence by introducing the idea of a goaldriven rule. In The Sound Pattern of English (SPE), Chomsky and Halle (1968) introduced rules of the form (153)

( structural condition , transformation )

The structural condition specied which representations the rule applied to and the transformation specied how a representation that satised the structural condition was to be transformed by the rule. Call such a rule an SPE rule. Chomsky and Halle went on to show how multiple SPE rules could be combined into rule schema. If r is a simple rule of the SPE variety, we write (154)

r

α→β

to mean that r can apply to α to produce β. Although derivational constraints are usually thought of as applying to all rules, it is easy to imagine a derivational constraint g associated with a particular SPE rule r. We could then dene a rule r | g that operated as follows: (155)

r|g

α→β

r

if α → β and β satisfies g

Sommerstein (1974) proposed another way that a condition g could combine with an SPE rule r. Call the combination g :: r. It operates in the following way:

Prosodic Adjustment

(156)

g :: r

α → β

91 r

if α does not satisfy g, α → β, and β does satisfy g

The condition g, the goal, plays a dual role as a rule trigger and an output constraint. A rule of this form will be called a goal-driven rule (GDR). Sommerstein conceived of goals as surface phonotactics, but the proposed rule format is much more general. I do not assume that goals are restricted to surface goals. The goals of cyclic syllabication, for example, are usually different from the goals of postcyclic syllabication. The latter are associated with surface phonotactics, but the former may not be. Unsyllabied nal consonants are typically freely allowed in cyclic syllabication, but often excluded or at least discouraged at the surface. Before Sommerstein’s proposal, the principal role allotted to a condition on representations in shaping the course of a derivation was as a prohibition on representations that did not satisfy the condition. Potential applications of a rule that produced an output that violated the condition were blocked. Crucially, derivational constraints require evaluating potential outputs before application of a rule is determined. GDRs use the same limited lookahead to determine rule application. But goals are different than general prohibitions on representations that violate some condition in two important ways. First, they are associated with particular rules. Second, it is assumed that some representations violate the goal condition. Indeed, the raison d’ˆetre of a goal-driven rule is to transform the current representation from one that violates the goal condition to one that satises it. Once elementary goal-driven rules are admitted into the architecture of derivational grammar, it is natural to extend the schema expansion ideas of SPE to goal-driven rules. Frampton (2001) shows how to carry this out. Rules of the form (157), with multiple ordered goals and multiple ordered rules, expand into a disjunctively ordered list of elementary GDRs. ⎛ ⎞ ⎛ ⎞ ⎜⎜⎜ G1 ⎟⎟⎟ ⎜⎜⎜ R1 ⎟⎟⎟ ⎜ . ⎟ ⎜ . ⎟ (157) ⎜⎜⎜⎜ .. ⎟⎟⎟⎟ :: ⎜⎜⎜⎜ .. ⎟⎟⎟⎟ ⎝ ⎠ ⎝ ⎠ Gn Rm It is possible to show that the rst elementary rule R j in the expansion that applies is one that produces the maximal possible improvement with respect to the goals. Maximal improvement is understood in the familiar way. Evaluation with respect to a goal G j , if it does not result

92

Chapter 6

in a tie, outranks evaluation with respect to any number of lower ranked goals. If no application of an R j produces an improvement, the complex rule (157) does not apply. If applications of R j and Ri would both produce a maximal improvement with respect to the goals, then the higher-ranked rule is the one that applies. Since maximal constraint satisfaction is so familiar from its use in Optimality Theory, I will take this to specify the application of a goal-driven rule in this book, rather than derive it from schema expansion. Frampton (2001) works out a theory of syllabication as iterative application of a goal-driven rule.2 One crucial advantage that iterative goal-driven rules have over rule schema without goals is that the rules R1 , . . . , Rm are freed from applying in strict order. The order in which they apply is determined primarily by their effects of the representation, not their ranking in the list of rules. It is also shown in Frampton (2001) that foot structure can be built in a straightforward way by iterative application of a goal-driven rule. Iterative goal-driven rules also appear to best organize the various operations that are used to adjust duplicants to achieve a desired reduplicant prosodic characteristic.3 As an illustration, consider the Asheninca Campa examples mentioned earlier and the duplicant adjustment rule:     Polysyllabic rst conjunct [-Right (158) :: Well-formed rst-conjunct syllable structure [-Left How is n [osampi ] transformed by this rule? The possibilities are (159)

Input n [osampi ]

Rule [-Right [-Left

Output no [sampi ] [nosampi ]

Since both possibilities fully satisfy the goals, the highest-ranking rule applies and the result is no [sampi]. Now consider the input [asi]. The possibilities are (160)

Input n [asi ]

Rule [-Right [-Left

Output na [si ] [nasi ]

[-Left achieves satisfaction of both goals but [-Right does not, because the rst conjunct is not polysyllabic. Consequently, the lower-ranked rule applies and the output is [nasi ]. It is usually most natural to write a goal in bipartite form:

Prosodic Adjustment

(161)

93

substructure ; condition

The rst term singles out some substructure of the representation and the second term species a condition that this substructure may or may not satisfy. In these terms, the Mokilese rule for adjusting the duplicant can be written   ]-Right (162) rst conjunct ; bimoraic syllable :: FCVL There is an imprecision in the terminology here that has been (and will continue to be) exploited in the interest of simpler and more suggestive terminology. The rst conjunct, as it has been dened, is a portion of the posttranscription structure. The GDR (162) applies to the pretranscription structure. The crucial point is that the conjunct that transcription will produce from a duplicant can be identied (often trivially) and manipulated in the pretranscription structure. Suppose transcription is to the left. If there are no embedded duplicants the rst conjunct will be identical to the residue in the duplicant of the left-edge truncate, if there is one; otherwise it will be identical to the entire content of the duplicant. The second conjunct will be identical to the residue in the duplicant of the truncates that are not at the left edge. A nuanced terminology could be introduced, with “conjunct” distinguished from “conjunct of the duplicant”. I will refrain from this and rely on the reader to determine the correct referents of uses of the terms “rst conjunct” and “second conjunct.” If it is known that transcription is to the left, then the rst conjunct is the reduplicant. Since we can talk about adjusting the rst conjunct of the duplicant, if transcription is to the left (for example), we can then talk about adjusting the reduplicant when referring to alterations of the duplicant. Obviously, mirror considerations apply to right transcription. The rule ordering in (162) is intrinsic. FCVL is always applicable to the output of the C∗ V-rule, since it always produces a V-nal duplicant and application of FCVL would always create a σμμ-duplicant, which is the goal. ]-Right applies only in special cases. In the case of bisyllabic wia, consider: σμ σμ σμ ]-Right (163) [ × ×] × −→ [ × × ×] w i a w i a This is not a valid adjustment, because the output does not satisfy the

94

Chapter 6

σμμ-goal. The alternative is σμμ σμ FCVL (164) [ × ×] × −→ [ × ××]× w i a w i a This is a valid adjustment, since the goal is achieved. The syllable structure shown in (163) and (164) is rst-conjunct syllable structure, which is independent of the syllable structure of the stem. The two syllable structures should be thought of as lying on separate tiers: σμ σ (165)

μμ

σμ − stem syllable structure − first conjunct syllable structure

[ × ××]× w i

a

The examples that have surface diphthongs (pouce, dauli) are less clearcut. McCarthy and Prince (1986) speculate that at the relevant level of syllabication the surface vowel-glide sequence syllabies to a bisyllabic vowel-vowel sequence. If so, then the same considerations that applied to wia also apply to examples like pouce, which appear at the surface with a diphthong. The “relevant level” here is the syllabication of the duplicant, considered independently from the syllabication of the word it is embedded in. Alternatively, it could be that the vowel sequences are diphthongs at the relevant level, but a somewhat modied GDR itself is responsible for excluding diphthongs from the surface prex. The modied GDR would be   ]-Right ; *Diphthong (166) rst conjunct ; σμμ :: FCVL Derivational constraints on the adjustment rules, if any, are written following the rules, separated by a semicolon. In this formulation, the exclusion of diphthongs from the duplicant is built into the GDR that establishes the material to be copied by transcription. The use of ]-Right in prosodic adjustment to bimoraicity is very common, but a language can choose not to use it. Consider, in this context, how reduplication in Mokilese would change if the rule ]-Right in (162) were lost. Then FCVL would always apply, and the surface pattern generated would be

Prosodic Adjustment

(167) a. b. c. d. e.

Stem pOdok kasO wadek wia pouce

95

Progressive pOO-pOdok kaa-kasO waa-wadek wii-wia poo-pouce

This pattern does in fact occur as a recent innovative dialect of Mokilese discussed by Blevins (1996).4 It is a straightforward example of diachronic change by rule loss. Prosodic adjustment has lost one of its adjustment options. 6.3.3 Complications in Mokilese Progressive Reduplication There are a few complications. Only one of them requires any amendment to the analysis above. The others simply require an explanation of the interaction of some particularities of Mokilese phonology with reduplication. We need to start with one of these particularities, however, because an understanding of it is needed before we can discuss the more serious complication. Al6.3.3.1 CP-sequences That Are Not Morpheme Internal though the Austronesian languages studied in this section occasionally have morpheme-internal CP sequences, such sequences are generally eliminated if not morpheme-internal. This affects reduplication of Pinitial roots, since the surface prex that reduplication creates will often be C-nal, creating a CP sequence that is not morpheme internal. Three different mechanisms are used for repairing CP sequences that are created derivationally. In Mokilese, P is deleted and the consonant spreads to the bare timing slot that remains, creating a geminate. The rule is (168)

×

×

C

P

In Agta (obligatorily) and Ilocano (optionally), the timing slot associated with the glottal stop is simply deleted. Resyllabication then takes place. In Ponapean, the glottal stop is replaced by a y-glide. Resyllabication then takes place, vocalizing the glide. The different strategies employed are illustrated in (170). The starting point in each case is the product of heavy-syllable reduplication of Palu, given in (169).

96

(169)

Chapter 6

σ σ σ ××××××× P a l P a l u

(170)

Varieties of CP-elimination Mokilese-A Pal-Palu CP-elimination Pallalu Resyllabication

Agta, Ilocano Pal-Palu PalPalu Palalu

Ponapean Pal-Palu Palyalu Palialu

Note that σμμ-reduplication in Agta, Ilocano, and Ponapean can be opaque. At the surface, the initial syllables are monomoraic in the examples above. As a consequence of CP → CC in Mokilese, we have (171) a. b.

Stem Ponop Palu

Progressive Pon-Ponop >Ponnonop Pal-Palu >Pallalu

‘is preparing’ ‘is walking’

I will use CP-reduction as a general term for the process, which takes different forms in the different languages. 6.3.3.2 Double Reduplication The surface pattern of the progressive forms of monosyllabic roots in Mokilese is strikingly different from the pattern for bisyllabic roots. (172) a. b. c.

Root Pir caak pa

‘laugh’ ‘bend’ ‘weave’

Progressive PirPir-Pir >Pirrirrir caacaa-caak paapaa-pa

*Pir-rir *caa-caak *paa-pa

It is clear that reduplication is iterated. The issue is how this is enforced by the morphophonology. The discussion of Ponapean in section 6.3.6 gives a clue. There, it will be clear that the goal weight of the reduplicant depends on the prosodic characteristics of the stem. A kind of prosodic dissimilation is enforced. If the initial foot of the stem consists of a single heavy syllable, a σμ-reduplicant is generated, otherwise a σμμ-reduplicant is generated. Since juncture insertion automatically creates a monomoraic duplicant, the effect is that prosodic adjustment is annulled if the initial foot of the stem consists of a single heavy syllable. Double reduplication of monosyllabic roots in Mokilese can

Prosodic Adjustment

97

be viewed as prosodic dissimilation as well: polysyllabic roots generate a monosyllabic reduplicant and monosyllabic roots generate a polysyllabic reduplicant. I assume that a goal Complementarity is set for the  rst conjunct of the duplicant and the adjustment operation ∅ → ] ] is available. This much is relatively straightforward, but it is less clear why, for example, pa→papa-pa or paapa-pa is not generated. The σμμ-goal is imposed on each syllable of the reduplicant. There are several ways to implement this as a GDR. The following is as simple as any: ⎤ ⎡  ⎢⎢ ]-Right  ⎥⎥⎥ Complementarity ⎢⎢⎢ ⎥⎥⎥ :: ⎢⎣ FCVL  (173) First Conjunct ; μμ ⎦ σ -syllables μμ ∅→] σ ] Derivations of the three examples in (173) are given below. The adjustment steps are labeled by the goal that drives them. Default Closure has been combined with the juncture insertion operation that triggers it in the interest of compactness. Although Complementarity is ranked above σμμ-syllables, double reduplication (∅ → ] / ] ) does not initiate the adjustment process because double reduplication is constrained to apply only if the rst conjunct is already bimoraic. If the root is already polysyllabic, Complementarity is already satised and there is no motivation for double reduplication, so the derivations for the examples already considered are unaffected. (174) JncIns σμμ-syllables Complementarity σμμ-syllables Trscr

caak [ca]ak [caa]k [[caa]]k caacaa-caak

rir [Pi]r [Pir] [[Pir]] PirPir-Pir (→Pirrirrir)

pa [pa] [paa] [[paa]]† [[paa]a] paapaa-pa

† This is somewhat tricky. [pa  a] expands to paa-pa which satises σμμ-syllables, and [[pa  a]] expands to paapa-pa, which does not. [pa  a]→[[pa  a]] therefore achieves satisfaction of Complementarity at the expense of eliminating satisfaction of σμμ-syllables. Complementarity is more highly ranked than σμμ-syllables, so satisfying Complementarity, if possible, takes precedence. It is worth noting that the phenomenon of double reduplication in Mokilese sharply illuminates one of the limitations of OT approaches

98

Chapter 6

to morphophonology. It is clear that what is happening in Mokilese double reduplication is that an operation is applying twice. OT has no means for even talking about something happening twice, because it rejects the idea of things happening in sequence. 6.3.3.3 Prenasalized Geminates discussion. (175)

Stem Pandip

Progressive Pan-dandip

One last example requires some

‘is spitting’

The analysis above appears to predict (176)

[Pan ]dip→Pan-Pandip >Pannandip

Raimy (2000b), however, argued that nd is a prenasalized geminate. I adopt his proposal. The derivation is then straightforward. σ σ σ σ T&R (177) [ × × × ] × × × −→ × × × × × × × × × P a nd P a ðd P a nd i p i p CP-reduction

−→

σ σ σ × × × × × × × × × P a nd a nd i p

Order on the phoneme tier is not covertly used in the last step. Replacing P with a link to d is the only way to satisfy the syllable wellformedness conditions in Mokilese. 6.3.4 Agta In Agta, CP-reduction is simply deletion of the P-slot. C becomes an onset after resyllabication. (178)

Agta plural and diminutive Root Gloss a. Patu ‘dog’ b. balataN ‘girl’ c. Pulu ‘head’ d. Puffu ‘thigh’

Reduplicated Pat-Patu >Patatu bal-balataN Pul-Pulu >Pululu Puf-Puffu >Pufuffu

Gloss ‘puppy’ ‘little girl’ ‘heads’ ‘thighs’

6.3.5 Ilocano Given the preceding discussions, most of the facts of Ilocano heavy-

Prosodic Adjustment

99

syllable reduplication follow directly. I begin with the straightforward examples. (179)

Ilocano plural and verbal progressive a. kaldiN ‘goat’ kal-kaldiN b. jyanitor ‘janitor’ jyan-jyanitor c. pusa ‘cat’ pus-pusa d. trabaho ‘to work’ trab-trabaho e. saNgit ‘to cry’ saN-saNgit f. Paso ‘dog’ Pas-Paso, Pas-aso g. dait ‘to sew’ daa-dait >daadaPit

‘goats’ ‘janitors’ ‘cats’ ‘is working’ ‘is crying’ ‘dogs’ ‘is sewing’

Note the variant forms in (179f). CP sequences arising from transcription are only optionally eliminated. When they are, the derivation proceeds as in Agta. Note also that ]-Right does not apply in (179g). This is the same restriction on ]-Right that was seen in Mokilese-A. Juncture shift is prohibited from bringing a vowel sequence into the duplicant. Instead, FCVL is used to augment the duplicant to a heavy syllable. The root-internal glottal stop in (179g) is inserted by a late rule, distinct from cyclic glottal stop insertion at the left edge of vowelinitial stems. Two particularities of Ilocano heavy-syllable reduplication need to be discussed. First, as in Mokilese, there is a quirk in prosodic adjustment to a bimoraic syllable duplicant when the stem is monosyllabic. (180)

a. b. c.

trak bas nars

‘truck’ ‘bus’ ‘nurse’

traa-trak (*trak-trak) baa-bas (*bas-bas) naa-nars (*nar-nars)

‘trucks’ ‘buses’ ‘nurses’

In these examples, ]-Right does not apply for some reason, so FCVL is used instead to augment the duplicant to a heavy syllable. The conditions under which ]-Right is blocked are clear, but the motivation for blocking ]-Right with monosyllabic stems is not. Second, there is variation for polysyllabic stems with a consonantglide complex onset. (181) a. bwaya ‘crocodiles’ buu-bwaya, bway-bwaya, bu-bwaya b. pyano ‘pianos’ pii-pyano, pyan-pyano, pi-pyano c. dwa ‘two’ duu-dwa (no variation) The rst two examples, with a three way variation in the surface form, are from Boersma and Hayes 2001, who undertake a major study of

100

Chapter 6

how variation of this kind can be analyzed in OT terms. Hayes and Abad 1989, from which (181c) is taken, notes that there is no variation for monosyllabic stems. First, we consider the three-way variation in (181a–b). Following Hayes and Abad, I assume that the y-glides in (181) are melodically identical to i-vowels. Some of the variation in (181a–b) can be attributed to varying underlying syllabication. Consider (181b), for example, where y/i orthography will be used to make the syllabic role clearer. σ σ σ σ σ b. × × × × × (182) a. × × × × × p i a n o p y a n o The two possibilities lead to the derivations below. σ σ σ σ σ FCVL (183) a. × × × × × → [× ×]× × × −→ [ × ×  × ] × × × p i a n o p i a n o p i a n o → pii pi a no → pii pya no b.

σ σ σ × × × × × → [× × ×]× × p y a n o p y a n o

]-Right

−→

σ [× × × ×]× p y a n o → pyan pya no

Note that ]-Right is blocked in (183a) because its application would produce vowel hiatus in the duplicant. The third variation, pipyano in (181b), is the result of complex onset reduction that is independent of reduplication, possible after an open syllable. σ σ σ σ σ σ Onset Reduction −→ × × × × × × × × (184) × × × × × × × × p i p y a n o p i p y a n o σ σ σ Compensatory Shortening −→ × × × × × × × p i p y a n o Compensatory shortening, parallel to keep + t→kept in English verbal morphophonology, follows resyllabication.

Prosodic Adjustment

101

Finally, we can address the invariability of the reduplicated form of a monosyllabic stem with a consonantal-glide onset, (181c) above. Ilocano has a bimoraic minimal word phonotactic. This rules out underlying monomoraic (185a), so the reduplicative morpheme combines only with bimoraic (185b). σ μ σμ σμ b. × × × (185) a. * × × × d w a d u a Consequently, only duu-du a→duu dwa is generated. 6.3.6 Ponapean I 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, I temporarily bypass the complexities of V-initial roots (P-initial at the surface) and restrict my attention to C-initial roots. The puzzle is that the duplicant is sometimes bimoraic, sometimes monomoraic. (186)

a. b. c. d. e. f. g. h.

pa mi mem kaN duup miik mand leNk

paa-pa mii-mai mem-mem kaN-kaN du-duup mi-miik ma-mand le-leNk

‘weave’ ‘exist’ ‘sweet’ ‘eat’ ‘dive’ ‘suck’ ‘tame’ ‘acrophobic’

*duu-duup *mii-miik *man-mand *leN-leNk, *len-leNk

McCarthy concluded that σμ-stems undergo heavy-syllable reduplication, and that σμμ-stems undergo light-syllable reduplication, with the crucial insight that word-nal consonants do not count in determining syllable weight. He called this phenomenon “complementarity.” In DR terms, the difference between (186a–d) and (186e–h) is that in the later examples there is no prosodic adjustment. Specifying the class of stems for which prosodic adjustment is annulled is not straightforward, as the following examples show: (187)

duupek a. nOOrok b. maasaas c. tooroor

duu-duupek nOO-nOOrok ma-maasaas to-tooroor

‘starved’ ‘greedy’ ‘cleared of vegetation’ ‘be independent’

102

Chapter 6

d. waantuuke wa-waantuuke ‘count’ e. liaan lii-liaan ‘outgoing’ McCarthy deduced that the conditions under which the duplicant is monomoraic are related to the foot structure. I do not follow his analysis, but exploit its core idea. If footing is moraic, right to left, and nal C is taken to be extraprosodic, as suggested by McCarthy, it is plausible that the foot structures of the examples in (188) are as given in the third column. Suppose also that the initial foot is taken to be the domain of the reduplicative affix. The derived prex that is actually produced by heavy-syllable reduplication is given in the last column. (188) a. b. c. d. e. f.

Root

Reduplicated

Foot structure kaN kaN-kaN (ka)N mand ma-mand (man)d duupek duu-duupe (duupe)k maasaas ma-maasaas (maa)(saa)s waantuuke wa-waantuuke (waan)(tuuke) liaan lii-liian (liaa)n

Initial foot ka man duu pe maa waan li aa

Surface prex kaN ma duu ma wa lii

When the data is organized in this way, it is easy to see that there is no prosodic adjustment when the initial foot is a heavy syllable. This accounts for all of the examples above. What this accomplishes is to avoid reduplicated forms in which the two initial feet are prosodically identical. The motivation for this is obscure. If stress were iambic, it might be seen as a way to avoid stress clash. But footing patterns that admit heavy-light syllable feet are rarely coupled with iambic stress. Since the Ponapean footing and stress system is not understood well, I will leave unresolved the question of why Ponapean reduplication is organized so that sequences of σμμ feet at the left edge are avoided. Note that sequences of two heavy syllables can be generated (duu-duupek, for example), but only when the second heavy syllable is footed with a light syllable to its right. How precisely is Complementarity implemented? It was suggested above that prosodic adjustment was “annulled” in a particular environment. Formerly, there is a derivational constraint on duplicant adjustment. It is easier to state if the initial foot of the stem is taken to be the domain of the affix. Then prosodic adjustment is blocked if the domain consists of a heavy syllable. The adjustment rule is

Prosodic Adjustment

103

 (189)

rst conjunct ; bimoraic ::

 ]-Right ; *σμμ (σμμ) domain FCVL

If this is the correct analysis, it is the rst example we have seen in which prosodic adjustment is sensitive to the prosody of the stem. In most cases, prosodic adjustment is blind to the prosodic structure of the stem, simply using its phonetic material for its own purposes. This is not to say that juncture insertion, the starting point for prosodic adjustment, is blind to the prosodic structure. There are two (limited) ways in which juncture insertion depends on the stem prosodic structure. First, the prosodic structure determines the nuclear structure, which furnishes the insertion sites for junctures. Second, prosodic structure can determine a domain to which juncture insertion applies. I turn now to a different matter. Ponapean has a number of verb roots with an initial nasal geminate or prenasalized geminate. The rst portion of the geminate surfaces as a syllable nucleus. Two derivations follow, with dur the durative morpheme: (190)

σ σ σμ σμμ DUR PrAdj  ⇒ [× ×]× × × −→ [× × ×]× × a. × × × × × → P m e d P m e d P m e d σ σ σ σ σ σ T&R CP-Repair −→ × × × × × × × × −→ ×××××××× P m P m e d resyllab.

−→

P m y m e d

σ σ σ ××××××××

(Pmmimmed)

P m i m e d σ σ σμ σμμ DUR PrAdj b. × × × × →  ⇒ [ × × ] × × −→ [ × × ×] × P Nd

a

P Nd a P Nd a σ σ σ σ σ σ T&R CP-Repair −→ × × × × × × × −→ × × × × × × × P Nd P Nd a σ σ σ resyllab. −→ × × × × × × × P Nd

i Nd

a

P Nd (Pndinda)

y Nd

a

104

Chapter 6

Finally, there is an intricate process at the boundary of the reduplicant and remnant that must be accounted for. In the examples below, taken from Davis 2003, the root has the form CVC. . . . According to the analysis developed up to this point, we expect a reduplicated form: (191)

C1 VC2 -C1 VC2 . . .

The expectation is fullled only if C1 = C2 and both are sonorant. (192) Case 1: C1 = C2 , sonorant rer ‘tremble’ rer-rer rer-rer lal ‘make a sound’ lal-lal lal-lal Case 2: C1 = C2 , both coronal sonorants lirooro ‘protective’ lir-lirooro lil-lirooro nur ‘contract’ nur-nur nun-nur Case 3: C1 = C2 , obstruent tit ‘build a wall’ tit-tit tin-tit kak ‘able’ kak-kak kan-kak Case 4: C1 = C2 , C1 a coronal obstruent, C2 a coronal sonorant tar ‘strike (of sh)’ tar-tar tan-tar tilep ‘mend a roof’ til-tilep tin-tilep Case 5: otherwise tep ‘kick’ tep-tep tepe-tep tep ‘begin’ tep-tep tepi-tep ker ‘ow’ ker-ker kere-ker Rehg and Sohl (1981) give an account of (192) that depends crucially on two reduplication-specic rules operating at the reduplicant-remnant boundary.5 The reduplication specicity suggests that the anomalies may be an aspect of NCC repair. Indeed, if it is assumed that Shortcut Repair (see section 5.3) occurs at the reduplicant-remnant boundary except in Case 5, it is possible to account for all the data in (192). I assume that Shortcut Repair applies if C1 and C2 are identical, or if they are both coronal and C2 is sonorant. I examine the cases in (192) in turn. First, note that in Case 1, NCC repair produces a geminate coda-onset sequence, rather than a sequence of identical consonants. The Case 2 examples require no comment, since they are the expected result of Shortcut Repair. In the remaining cases, Ponapean syllable-structure repair rules come into play. Ponapean requires codas that are not word nal to be sonorants that share place with a following onset. If a coda-onset sequence is a sonorant-geminate sequence, as it is in the Case 2 examples, no repair is

Prosodic Adjustment

105

needed. If it is an obstruent geminate, the syllable structure is repaired by nasalizing the geminate, producing a prenasalized geminate. For example; (193)

× → ×

×

×

n t

t

Shortcut repair produces such a geminate in the Case 3 and 4 examples. Syllable repair accounts for the surface form. A typical derivation is given in (194). σ σ σ σ (194) × × × × × × → × × × × × × → × × × × × × t a t a r t a n t a r t a r In Case 5, Shortcut Repair does not apply. NCC repair via phoneme ssion produces a violation of the syllable well-formedness conditions that cannot be repaired by prenasalization. Instead, the repair is vowel epenthesis, as illustrated in (195). σ σ σ σ σ (195) × × × × × × → × × × × × × → × × × × × × × t e p

t e p t e p

t e p Vep t e p

Determining the quality of the inserted vowel is a complex issue, subject to some lexical specication. It is usually, but not always, a copy of the anking vowels (which are identical). See Rehg and Sohl 1981, 91, for discussion. The different outcomes for tep ‘kick’ and tep ‘begin’ in Case 5 in (193) demonstrate the need for at least some lexical specication. 6.4 C-Finality as a Secondary Prosodic Goal The core of weight-based stress theory is mora count, but secondary prosodic distinctions play a role in some stress systems.6 The same is true of reduplicative prosodic adjustment. C-nality is a secondary prosodic goal in Lardil, Yaqui, and Nuu-chah-nulth. In all three cases, coda consonants do not contribute to syllable weight, but reduplicants are C-nal when possible. 6.4.1 Lardil Iterative Reduplication The key data is given in (196).7

106

(196)

Chapter 6

Lardil iterative reduplication Root Iterative a. kele kele-kele b. la laa-la c. Naali Naal-Naali d. pareli parel-pareli

‘cut’ ‘spear’ ‘thirst’ ‘gather’

On the basis of minimal word and stress considerations, Wilkinson (1988) showed that Lardil footing is mora based, with CV(C) syllables monomoraic and CVV(C) syllables bimoraic. The affix that realizes iterative has a null exponent, species the initial moraic foot as its do× ). After juncture insertion main, and triggers the ×∗-rule ( ∅ → [ and default closure, but before prosodic adjustment, the structures of the examples in (196) are (197)

a.

( σμ σ μ ) [ × × × ×]

( σμ σμ ) σμ b. [ × × × ×] × ×

k e l e

p a r e l i

( σμμ ) σμ c. [ × × ×] × ×

( σμ ) d. [ × ×]

n a a l i

l a

The foot structure is marked with parentheses on the syllable tier. The duplicant is then adjusted to bimoraicity by a prosodic adjustment rule very similar to the prosodic adjustment rule used in the Austronesian languages analyzed above. The primary prosodic goal is the same and the repair rules are the same. The one difference is that there is a secondary goal, C-nality. Transcription is to the left, the default for left-edge duplicants. The prosodic adjustment rule is   bimoraic ]-Right (198) First Conjunct ; :: C-nal FCVL Duplicant adjustment produces (199)

σμ σμ a. [ × × × ×] k e l e

σμ σμ b. [ × × × × ×] × p a r e l i

σμμ c. [ × × × ×] × n a a l i

σμμ d. [ × ×  ×] l a a

Prosodic Adjustment

107

Here, rst-conjunct syllable structure is given. In (197) it was stem syllable structure. The secondary prosodic goal is unattainable in (199a, d), given the available adjustment rules. The primary weight goal is attained throughout. 6.4.2 C-Finality in Yaqui Light-Syllable Reduplication In Section 4.4.3, various Yaqui reduplicative verbal affixes were introduced. Consideration of one variety, light-syllable reduplication, was restricted to stems whose initial syllable is CV, which is doubled by the reduplication process. Some examples are repeated here: (200) a. b. c. d.

Stem vusa hewite koarek chi ke

Reduplicated vu-vusa he-hewite ko-koarek chi-chike

‘awaken’ ‘agree’ ‘wear a skirt’ ‘comb one’s hair’

If a more complete range of stems is considered, there are very interesting complications. (201)

Stem Reduplicated CVC initial syllable a. chukta chuk-chukta b. chepta chep-chepta c. bwalkote bwal-bwalkote d. vuite vui-vuite CVV´ initial syllable e. wa´ate wa-w´aate f. ka´ate ka-k´aate ´ CVV initial syllable g. k´aate k´a-kate h. w´aate w´a-wate

‘cut with a knife’ ‘jump over’ ‘soften, smooth’ ‘run’ ‘want’ ‘walk’ ‘build a house’ ‘remember’

We rst peel away the aspects of (201) that are the consequence of general features of Yaqui phonology and unrelated to the reduplication process itself. Accent is on the second mora in Yaqui unless the root is lexically specied as initial accenting, in which case it is on the initial mora. Reduplicated forms inherit the accentual type of the stem. This is responsible for the accent placement in (201e–h). Long vowels shorten unless one of their occurrences is accented, producing a short vowel in the remnant in (201g–h). Finally, the glide status of i in (201d) was

108

Chapter 6

conrmed by Demers, Escalante, and Jelinek (1999), on the basis of spectrographic analysis, so (201d) is CVC initial. What remains for the theory of reduplication to account for is the core paradigm: (202) a. b. c.

Stem chike chukta kaate

Reduplicated chi-chike chuk-chukta ka-kaate

On the basis of (202a) and (202b), we can surmise that juncture insertion is the C∗ V-rule and that (202b) must be the result of prosodic adjustment. On the basis of their analysis of the accentual system, Demers, Escalante, and Jelinek (1999) concluded that CV(C) were light and CVV(C) syllables were heavy. Prosodic adjustment is therefore to C-nality, just as in Lardil. The problem is to account for why ]-Right does not apply in (202a). Prosodic adjustment can be sensitive to the prosodic structure of the stem that the duplicant is embedded in. Examples are rare, but Ponapean provides one example. Yaqui provides a second. One can imagine various constraints on adjustment that would exclude (202a). I will suppose, tentatively, that a condition that will be called Onset Permanence (OnsPerm) is at work. It prevents adjustment from recruiting onsets for other syllabic roles. OnsPerm will play a role later in analyzing Sanskrit reduplication. We can then write the prosodic adjustment rule as (203)

First Conjunct ; C-nal :: ]-Right ; OnsPerm

The two examples of C-nality as a rst-conjunct goal that were considered above are from languages in which coda consonants are nonmoraic. I will assume that this is not accidental and follows from the primacy of weight in duplicant adjustment. Specically, I assume the following: (204) Primacy of weight in prosodic adjustment a. Duplicant adjustment never alters prosodic weight except to meet a duplicant weight goal. b. Prosodic weight goals always take precedence over other duplicant goals. In effect, weight changes are never accidental. Taken together, these assumptions prevent adjustment to C-nality from playing a role in du-

Prosodic Adjustment

109

plicant adjustment in languages in which coda consonants add syllable weight. 6.4.3 Moravcsik’s Generalization Reconsidered Consider a hypothetical language “Yaqui-A,” which differs only from Yaqui in that the reduplication process corresponding to Yaqui lightsyllable reduplication is based on the C∗ν-rule rather than the C∗ V-rule. The result is rst-syllable reduplication. (205) Yaqui-A (hypothetical variant of Yaqui) PrAdj C∗ν-rule a. [chi]ke chi-chike b. [chu]kta [chuk]ta chuk-chukta c. [kaa]te kaa-kaate Duplicant adjustment is blocked in (205a,c) by OnsPerm. No examples of rst-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 unusual in two respects. In the rst place, it employs adjustment of the reduplicant to C-nality, one of only a few languages to do so. In the second place, duplicant adjustment is sensitive to the prosodic structure of the stem that the duplicant is embedded in, one of only a few languages in which duplicant adjustment has this property. Normally, duplicant adjustment incorporates phonemic material into the duplicant without regard for the prosodic structure that that material is found in. Yaqui and Ponapean are the only two examples of this kind that I am aware of. Use of the C∗ν-rule is also rare. Nuu-chah-nulth, Tohono O’odham, and Sanskrit are the only examples in this book. Bear in mind that, for obvious reasons, unusual reduplication processes are signicantly overrepresented among the languages considered in this book. A language like hypothetical Yaqui-A would therefore have to combine an array of highly marked options, so it is not a surprise that no reduplication process like the one considered in hypothetical YaquiA has, as yet, been uncovered. It may have occurred to some readers that the OnsPerm constraint might also permit rst-syllable reduplication in a language in which coda consonants are moraic. Consider now the possibility of a hypothetical “Yaqui-B,” which differs from Yaqui in that coda consonants are moraic. Suppose that reduplication is based on the C∗ V-rule, and

110

Chapter 6

that a σμμ-goal is imposed on the duplicant, with ]-Right the only repair rule. These assumptions would lead to the following paradigm: (206)

Yaqui-B (hypothetical variant of Yaqui) PrAdj C∗ V-rule a. [chi]ke chi-chike b. [chu]kta [chuk]ta chuk-chukta c. [ka]ate [kaa].te kaa-kaate

The duplicant does not satisfy the σμμ-goal in (206a), but duplicant adjustment (limited to ]-Right) cannot apply because of OnsPerm. As far as I know, no such reduplication system is attested. The reason, I believe, is that FCVL is universally available as a repair option. If so, then (206a) [chi ]ke would be adjusted to [chi i ]ke→chii-chike. In place of (206), the paradigm would be (207)

Yaqui-C (hypothetical variant of Yaqui) PrAdj C∗ V-rule a. [chi]ke [chii].ke chii-chike b. [chu]kta [chuk]ta chuk-chukta c. [ka]ate [kaa].te kaa-kaate

If more languages are uncovered that impose OnsPerm on duplicant adjustment and in which coda consonants are moraic, a paradigm like (207) might be found. These considerations go a long way toward clarifying the status of Moravcsik’s Generalization. Moravcsik was clear about the fact that she was proposing a hypothesis about mechanism, not an empirical generalization. She realized that in certain cases, reduplicants were always identical to prosodic constituents of the stem— CV reduplication in a language in which all syllables were CV, for example. The issue for her was the characteristics of the stem that are available in the description of the rules that generate the reduplicant. Moravcsik concluded that it was possible to specify reduplicants without recourse to stem prosodic constituency. The conclusion reached here is that Moravcsik was partially correct. The complication that DR introduces is that the surface shape of the reduplicant is determined by multiple factors: mainly rules of juncture insertion and rules of duplicant adjustment. The former rules have the characteristic that Moravcsik pointed to: they do not refer to prosodic constituency, although the domain in which they apply can be determined by foot boundaries. As Yaqui and Ponapean demonstrate,

Prosodic Adjustment

111

duplicant adjustment can be sensitive to the prosodic structure of the stem.8 6.5 Varieties of Syllable Prexation and Suffixation It should be unnecessary to say at this point (but will be repeated so that there are no grounds for confusion) that the terms prexation and suffixation are used in this section only descriptively. Morphologically, there is no suffix, which is a term relevant to the concatenation of the exponent of an affix. The surface affix in the cases discussed in this section is the product of transcription, not concatenation. 6.5.1 Heavy Syllable Suffixation of Left-Edge Material (Chukchee) Chukchee absolutive singular reduplication was analyzed in section 4.1 as shown in (208). (208)

nute inu jilPe

[nute] → nute-nut [inu] → inu-in [jilPe] → jilPe-jil

The derivation of the pretranscription structures was left unexplained at that point, but we can nish the discussion now. The juncture insertion rules and the duplicant adjustment rule are given in (209). The exponent of the reduplicative morpheme is null. Transcription is to the right.   (209) a. ∅ → [ × (×∗-rule); ∅ →  V b. Second Conjunct ; Bimoraic :: -Right The derivations of the two pretranscription forms in (209) is given here: (210) JncIns DC PrAdj

jilPe [jilPe [jilPe] [jilPe]

nute [nute [nute] [nute]

inu [inu [inu] [inu]

6.5.2 Final-Syllable Prexation (Madurese) It is remarkable that although initial-syllable prexation is not attested, nal-syllable prexation is. Madurese plural reduplication was analyzed in example (73) as

112

(211)

Chapter 6

buwaP maen estre

[buwaP] → waP-buwaP [maen] → en-maen  [estre] → tre-estre  

The juncture insertion rules and the duplicant adjustment rule are given in (212). The goal is rst-conjunct WFSS (well-formed syllable structure). The exponent of the reduplicative morpheme is null. Transcription is to the left.   × (×∗-rule); ∅ →  V (right) (212) a. ∅ → [ b. First Conjunct ; WFSS :: -Left The derivation in (213) is typical. (213)

××××× b uw a P

JncIns

−→

DC

∗σ

[× × ×× × → [× × ×× ×] b uw a P b uw a P σ -Left −→ [× ×× × ×] → waP-buwaP b uw a P

The goal is unattainable for [ma en ], which cannot be adjusted to rstconjunct WFSS by -Left. Although the effect is duplication of the nal syllable, the content of the duplicant is not determined on that basis. Juncture insertion isolates the rime of the nal syllable. The rst conjunct is augmented in order to eliminate the onsetless syllable in the rst conjunct of the duplicant. 6.5.3 Final-Syllable Suffixation (Kaingang) Whereas Madurese has nal-syllable prexation, Steriade (1988), discusses Kaingang nal-syllable suffixation, based on Wiesemann 1972. (214) a. b.

Stem jem˜ı jengag

Reduplicated jem˜ı-m˜ı jengag-gag

Pretranscription je[m˜ı] jen[gag ]

The affix  that induces this pattern has a null exponent and triggers V. Prosodic adjustment to well-formed duplicant Right syllable structure occurs, using the adjustment rule [-Right. An illustrative derivation follows:

∅ → [

Prosodic Adjustment

JncIns/DC

→

(215)

113 ∗σ

σ [-Left × × × ×[× ×] −→ × × ×[× × ×] → jengag-gag j e n g a g j e n g a g

Just as in Madurese, the nal syllable is isolated. But the content of the duplicant is not determined on that basis. 6.5.4 Korean Consecutive-Syllable Reduplication The following reduplication pattern was discussed in section 2.2.3. (216) a. b. c.

Stem cason kikwe hy@nsak

Reduplicated ca-cason-son ki-kikwe-kwe hy@n-hy@nsak-s’ak

Pretranscription [ca][son] [ki][kwe] [hy@n][sak]

Note that the internal junctures in (216) fall at the syllable boundaries. It is possible that these Sino-Korean words are bimorphemic in a sense that would allow the juncture insertion rules to correctly position the junctures. Juncture insertion can be sensitive to morpheme boundaries, but not syllable boundaries. It is interesting, nevertheless, to consider whether duplicant adjustment can account for (216), without an appeal to polymorphemic stems. It is sufficient to suppose that the affix that induces Korean consecutive-syllable reduplication has a null exponent  V (right). Prosodic adjustment of the duplicant and induces ∅ → ][ to well-formed syllable structure follows, using juncture shift. The crucial derivational step is (217)

∗σ σ σ σ PrAdj [ × × × × ×][× × ] −→ [ × × × ×][× × × ]

h y @ n s a k

h y @ n s a k → hy@n-hy@nsak-sak

6.6 The Onset-Coda Asymmetry and Moravcsik’s Generalization Systematically examining the various examples of duplicant adjustment has had a dual purpose. The rst, of course, was to show that reduplication processes discussed in earlier chapters are consistent with the restricted framework of juncture insertion and duplicant adjustment that has been proposed. The second was to show that the possibility of duplicant adjustment did not inadvertently predict the occurrence of

114

Chapter 6

unattested reduplication patterns. In the Kaingang, Madurese, and Korean examples, the nal syllable was isolated by rst isolating the nal VC∗ , then adjusting it to wellformed syllable structure. The initial syllable cannot be isolated in this way. The initial CV∗ can be isolated, but it has perfectly wellformed syllable structure. A coda consonant can be attached to CV∗ only on the basis of weight. This yields the heavy-syllable (prexal) reduplication I analyzed in Mokilese and related languages, on the one hand, or heavy-syllable (suffixal) reduplication I analyzed in Chukchee. Heavy-syllable prexal reduplication (Mokilese) comes from isolating initial C∗ V as the rst conjunct of the duplicant and adjusting it to be a bimoraic syllable. Heavy-syllable suffixal reduplication (Chukchee) comes from isolating initial C∗ V as the second conjunct of the duplicant and adjusting it to bimoraicity. 6.7 Toward a Parametric Theory of Duplicant Adjustment The discussion of duplicant adjustment above assumed implicitly that duplicant adjustment is a specic process, not simply a descriptive name for whatever rules happen to alter the prosodic structure of the duplicant or one of its conjuncts. A specic form for the rule that carries out prosodic adjustment was proposed, a goal-driven rule. The purpose of this section is to make these assumptions explicit, rene them, and provide some justication for them. The main justication for thinking that prosodic adjustment is more than merely a name for what happens, is the narrow range of possibilities encountered. Consider again the rule that was initially identied as empirically adequate to account for the Mokilese reduplication paradigm.  C, otherwise FCVL (218) ∅ → ] Suppose that the otherwise clause was dropped, and the rule reduced to C. The resulting surface paradigm would be ]-Right / (219)

Root pOdok kasO wia kookO

Reduplicated pOd-pOdok kas-kasO wi-wia ko-kookO

Prosodic Adjustment

115

As far as I can determine, such a paradigm is unknown in a language that, like Mokilese, distinguishes the syllable weight of C∗ V and C∗ VC syllables. This demands some account. A reasonable conjecture is that over and above the specic rules that alter the prosodic structure of the target of adjustment (one of the conjuncts of the duplicant), there are prosodic conditions that those rules subserve. Given plausible assumptions about the availability of default adjustment rules, paradigm (219) would then be ruled out because there are no coherent prosodic conditions that govern the adjustment. The matter is entirely analogous to the process that builds syllable structure. Various rules (projecting nuclei, attaching codas, and so on) are bound together in a way that serves a collection of well-formedness conditions imposed on syllable structure. It will be clear in the case studies that follow that prosodic adjustment must employ an array of rules and that it must be iterative. Goal-driven rule schemata offer a means of combining various subrules into a single complex rule that can be iterated. Prosodic goals organize the process, but the specic adjustment rules designated for prosodic adjustment have a major impact on the form that adjustment takes. We have already seen this in the Mokilese example. If juncture shift is available to build a heavy-syllable rst conjunct, the standard surface pattern results. If it is not, a different surface pattern results: the pattern of the innovative dialect of Mokilese illustrated in (167). If we assume that C-nality in languages in which coda consonants do not contribute to syllable weight can be taken to be a syllablestructure condition for reduplication, then all the examples of prosodic adjustment considered up to this point have the general form  syllable structure conditions ! " (220) target ; weight condition :: adjustment rule list; constraints on adjustment The various case studies in the next chapter will not require any revision of (220). Restricting the prosodic adjustment rules to the form (220) does little to constrain the possibilities for reduplication if other rules can carry out similar operations. I will assume that the prosodic adjustment schema (220) is uniquely privileged to manipulate t-junctures (insert them and shift their position). Prosodic adjustment is, in some sense, an extension

116

Chapter 6

of the morphology of t-juncture insertion into the phonology. Manipulating t-junctures is not an option for arbitrary phonological rules. The situation is very much the same with, for example, parameterized theories of footing such as Halle and Idsardi 1995. If other rules were capable of manipulating footing delimiters, outside the context of the tightly parameterized theory developed there, the force of the theory would be lost. The hardest aspect of (220) to constrain is the class of possible constraints on repair rules. In Yaqui, prosodic adjustment was constrained by OnsPerm, which prevented stem onsets from assuming other syllabic roles in the reduplicant. In Ponapean, prosodic adjustment was blocked from creating a heavy reduplicant if the initial foot of the stem consisted of a heavy syllable. There is no obvious natural class. Both restrictions are plausible, but it is not easy to nd the common thread. A condition like “the duplicant cannot be heavy if the initial syllable of the stem is light” would immediately allow syllable copy reduplication. While it is not hard to draw some distinction between this condition and the condition that constrains Ponapean prosodic adjustment, the distinction remains ad hoc without a denition of “possible condition on repair,” or at least some notion of markedness. I am unable to provide this. Ultimately, this is needed to account for acquisition. If proposal (220) about the form of prosodic adjustment is correct, then it is a major residue of Prosodic Morphology in DR. The restriction of the domain of a reduplicative affix to a foot is the second major residue. Recall, however, that DR differs sharply from Prosodic Morphology in that while prosodic conditions may play a major role in some reduplicative processes, the DR analysis of most reduplicative processes involves neither domain designation nor prosodic adjustment.

Chapter 7 Case Studies

At this point, all the tools are in hand, so I proceed to analyze a number of complex examples. The rst three, Ndebele and Kinande unintensive reduplication and Asheninca Campa intensive reduplication, were chosen not only because of their intrinsic interest, but because they have been extensively analyzed in very different frameworks. McCarthy and Prince (1995) claim that derivational phonology is an inadequate framework for understanding the complexities of Asheninca Campa intensive reduplication. Inkelas and Zoll (2000) claim that Ndebele unintensive reduplication demonstrates that apparently duplicated phonological material is not in fact the result of copying in the phonology. I will show that both of these claims are unfounded. Three prominent examples from Raimy 2000a are included (from Tohono O’odham plural reduplication, Temiar continuative reduplication, and Chaha intensive reduplication) so that the reader can assess the differences with and similarities to his analysis. A fairly thorough treatment of Sanskrit verbal reduplication is included so that readers can make a similar assessment with respect to Steriade’s (1988) wellknown analysis of perfect and intensive reduplication in Sanskrit. 7.1 Ndebele Unintensive Reduplication This section relies on Hyman, Inkelas, and Sibanda 2005, HI&S in what follows, and Sibanda 2004 for data, description, and insight into the morphophonology of Ndebele.1 Both works analyze Ndebele unintensive reduplication in the framework of Inkelas and Zoll’s (2000) Morphological Doubling Theory (MDT). The MDT analysis will be discussed later in this section. In Ndebele, verbal reduplication has unintensive semantics. It is translated as ‘a bit’ or ‘here and there’. Roughly, it is the difference

118

Chapter 7

between ‘he ate his food’ and ‘he picked at his food’. Some preliminary examples are given in (221), with the unintensive forms on the right. A guide to the glosses in this section and the next (on Kinande) is given in the footnote below.† If the root is at least bisyllabic, there is a single unintensive form, as in (221.1). If the root is shorter, there is variation, as in (221.2–3). Consonantal roots have ve different variant unintensive forms. (221) 1. uku-zi-nambith-el-a nonfin-om-taste-appl-fv ‘to taste for’ 2. uku-zi-lim-el-a nonfin-om-cultivate-appl-fv ‘to cultivate for’ 3. uku-zi-dl-el-a nonfin-om-eat-appl-fv ‘to eat for’

ukuzi-nambi-nambitha

a. ukuzi-lime-limela b. ukuzi-lima-limela a. b. c. d. e.

ukuzi-dlela-dlela ukuzi-dleyi-dlela ukuzi-dlayi-dlela uku-zidle-zidlela uku-zidla-zidlela

Since the semantics of unintensive modication is not the issue being explored here, neither a gloss nor a translation will be given for unintensive forms. Both can be inferred with sufficient accuracy from the corresponding unreduplicated form. Similarly, because they do not affect the process under discussion, accent and tone are not indicated. For the remainder of this section, references to reduplication in Ndebele will always be to unintensive reduplication. The striking thing about the possibilities in (221) is that an object agreement prex and an applicative suffix or some material from it, can be incorporated into the reduplicant under certain conditions. HI&S view reduplication in Ndebele as a process that applies to a stem that has already been inected by all of the affixes that interact with reduplication. Producing the results above requires the † om

object agreement prex appl sm subject agreement prex pass nonfin nonnite prex subj tm tense (general) prex perf fs Final Suffix (to be explained) fv Final Vowel (to be explained)

applicative suffix passive suffix subjunctive suffix perfect aspect suffix

Case Study: Ndebele Unintensive Reduplication

119

reduplication process to be sensitive to the morphological structure of the stem that it operates on. The assumption here is quite different; juncture insertion and duplicant adjustment are blind to the morphological structure of the representation to which they apply. All the variation in (221) can be attributed to the timing of duplicant adjustment and the particular adjustment operations used at different points in the derivation. I rst specify juncture insertion. It takes the initial bisyllabic foot of the stem as its domain, if there is such a foot, otherwise the entire stem. Optional extrametricality applies to initial onsetless syllables, as will be clear when V-initial stems are considered. Juncture insertion is ∅→[ × and Default Closure closes the duplicant by inserting ] at the right edge of the domain. Call the unintensive morpheme UnInt and the morpheme which is realized as an object agreement prex AgrO . If UnInt is realized before AgrO , the account of (221.1) is straightforward. However, if UnInt is realized before AgrO , an explanation of how zi is incorporated into the duplicant in (221.3d–e) is needed. The decomposition of reduplication into a number of separate operations in DR provides a way to solve this problem. The realization of UnInt by juncture insertion, prosodic adjustment, and transcription are different operations. Transcription can occur in a later cycle, with adjustment possible at various points between the formation of the duplicant and its elimination by transcription. I will assume that in Ndebele, transcription is delayed until rst conjunct bisyllabicity is satised. The following derivation, for example, will be justied in the coming pages. Each row shows one cycle, named either by the morpheme which is realized or the affix which is introduced. (222)

Root UnInt

appl om fv

dl [dl] [dl]el→[dle]l zi[dle]l→[zidle]l→zidle-zidlel zidle-zidlela

The duplicant incorporates material as it becomes available, using either [-Left or ]-Right juncture shift. When sufficient material has been incorporated to satisfy bisyllabicity, transcription occurs. The adjustment [dl ]el→[dle ]l is driven by a secondary goal, well-formed rst conjunct syllable structure. In several of the variations in (221.2) and (221.3), epenthetic material

120

Chapter 7

appears in the reduplicant. Consider the reduplicant dlayi in (221c), for example. The derivation that will be justied is (223)

Root UnInt

appl om fv

dl [dl] [dl]el zi[dl]el zi[dl]ela→zi[dla]ela→zi[dlayi]ela →zi-dlayi-dlela

No adjustment is made until quite late in the derivation. At that point, right-edge truncated epenthesis is used to ll out the rst conjunct to bisyllabicity. First, a is epenthesized, driven by a WFSS (WellFormed Syllable Structure) goal. Then yi is epenthesized, driven by a bisyllabicity goal. I assume that adjustment is not obligatory, at least in the early stages of the derivation. In the AgrO -cycle above, for example, zi [dl ]el is left unadjusted. Of course, adjustment and transcription cannot be deferred indenitely. A key issue in understanding the cyclic structure of Ndebele reduplication is to understand when and why it is eventually forced. I will return to this shortly, but rst examine the epenthetic adjustment operations. a -epenthesis at the right edge of the duplicant, which will be called a-Ep, is very much like what was called First Conjunct Vowel Lengthening (FCVL) in the analysis of Mokilese and other Austronesian languages in chapter 6; see example (88) for instance. FCVL is used for adjustment to bimoraicity in Mokilese, while a-Ep is used for adjustment to WFSS in Ndebele. The rules are compared in (224). (224)

a. Mokilese (innovative dialect, see p. 95) σμμ σμ PrAdj [ × × ] × × × −→ [ × ××]× × × → pOO-pOdok p O d o k b.

p O

d o k

Ndebele σ σ σ PrAdj [ × × × ] −→ [ × × × ×] → lima-lim l i m

l i m a

In Mokilese, a nal timing slot is epenthesized, then truncated and its phonemic value lled in by spreading. In Ndebele, a nal timing

Case Study: Ndebele Unintensive Reduplication

121

slot is epenthesized, truncated, and its phonemic value is lled in with epenthetic a. In addition to a-Ep, the derivation (223) also assumes that there is a similar syllable epenthesis operation. It will be called yi-Ep. We will see later that Ndebele also uses yi-epenthesis in certain nonreduplicative contexts, driven by a minimal weight condition. To justify the derivations proposed in (222) and (223), we rst need some understanding of the structure of inected verbs in Ndebele, at least that part involved in reduplication. 7.1.1 The Structure of Inected Verbs Inected Ndebele verbs have the following surface structure:   nal suffix (225) . . . - (om)-root-(inner suffix)∗ -a If there is an object-agreement prex, it appears adjacent to the root. Other prexes can occur, including subject agreement and various tense markers, but they will not be relevant to what follows. Many of the inner suffixes are associated with argument-structure modication. In this section we will be concerned mainly with the applicative and passive inner suffixes. Multiple inner suffixes can follow the root.2 There is at most a single nal suffix. Perfect, recent past, subjunctive, and negative nal suffixes occur. In the absence of a nal suffix, the stem would end in a consonant, since roots and inner suffixes are Cnal. In this case, a nal a is appended to the stem. I take this a to be epenthetic, inserted so that the form satises the syllable phonotactics (which require V-nal syllables).3 Any of the inner suffixes is subject to reduplication, in whole or in part, if the root is sufficiently short. In some of the (221.3) variations, an object agreement prex reduplicates. Final suffixes are never reduplicated, in whole or in part, as (226) illustrates. (226)

a. b.

lim-e cultivate-subj lim-ile cultivate-perf

*lime-lim-e, lima-lim-e *limi-lim-ile, lima-lim-ile

I will show below that under plausible assumptions about where UnInt can appear in the hierarchical structure, this pattern of prex and suffix reduplication can be explained if the syntax of the unreduplicated inected verbs is

122

Chapter 7

(227) (realized by prexes) In AgrO . . . En Root

E1

Infl → nal suffix AgrO → om Ei

→ inner suffix

This is only a rough sketch. The heads AgrO and In have been labeled and their realizations described, but their syntactic features have not been identied. Since it would take the discussion far outside the focus of this book, a deeper understanding of the syntax will be left unexplored. In what follows I will describe the AgrO phonology as the om-cycle and the In phonology as the fs-cycle. The fs-cycle is important whether or not there is an overt suffix. It plays a special role in the prosody and a bisyllabic minimum weight condition is imposed on its output, independently of reduplication.4 Inected verb forms are almost always at least bisyllabic, but imperative forms of consonantal roots require augmentation since they are not suffixed and appear without an om. For example, lim-a ‘cultivate!’ is not augmented, but yi-dl-a ‘eat!’ and yi-m-a ‘stand!’ are augmented. An epenthetic yi syllable is prexed, if necessary, in order to meet the fs-cycle minimal weight requirement. Since the fs-cycle is the locus of a minimal weight requirement, it is plausible to suppose that it is the rst cycle in which the morphophonology produces a surface-like form. Full syllabication is carried out and both minimal weight and the appearance of nal-a are responses to this syllabication. It is plausible to identify the output of the fs-cycle as what is called a phase in phasal theories of spellout.5 If the output of the fs-cycle is in some sense a surface form, even though it eventually appears word internally, it is natural to demand that all t-junctures must be removed at this point in the derivation. It is therefore reasonable to assume that adjustment of duplicants to reduplicant bisyllabicity is obligatory in the fs-cycle and that the duplicant adjustment operations available in the fs-cycle are adequate to accomplish this. Summarizing and extending the considerations to this point, I will assume that (228)

1. Adjustment is optional prior to the fs-cycle. 2. Transcription is obligatory if the duplicant satises rst conjunct bisyllabicity, otherwise does not apply.

Case Study: Ndebele Unintensive Reduplication

123

3. Adjustment to bisyllabicity is obligatory in the fs-cycle. 4. fs-cycle adjustment uses a-Ep and yi-Ep; otherwise adjustment uses JncShift. This determines the overall organization of the proposed derivations (222) and (223), which derive (221.3c–d). The individual adjustment steps are all driven by WFSS, assuming that UnInt appears directly above the root. The derivations of (221.2a–b) and (221.3b) proceed in a similar fashion. We turn now to consider (221.3a,e), which raise new issues. First consider (221.3a), with reduplicant dlela. Earlier, I assumed that UnInt is realized before the AgrO , accounting for the fact that object prexes enter the duplicant only by adjustment, not by initial juncture insertion. In the examples (222) and (223), UnInt was realized immediately after the root. UnInt is an adverbial modier. The syntactic position of adverbs is often quite free, with highly nuanced differences in the resulting semantics. If UnInt can appear higher in the structure of the complex verb than immediately above the root, with little difference in semantic interpretation, there is a simple account of (221.3a). The derivation is (229)

Root

appl UnInt

om fs

dl dlel [dlel] zi[dlel] zi[dlel]a→zi[dlela]a→zi-dlela-dlela

I will therefore suppose that although UnInt must be below the AgrO in the syntactic structure, it is not further constrained syntactically. Last, consider (221.3e), with reduplicant zidla . The problem is the motivation for the adjustment step zi [dl ]el→[zidl ]el , which appears to be involved in generating the output form. It is not justied by the prosodic goals assumed to this point because the output is neither bisyllabic nor satises WFSS. In the formalism that has been proposed, adjustments must be driven by an immediate improvement of the representation with respect to the goal that drives the iterative adjustment rule. Intuitively, zidl is “more bisyllabic” than dl even though neither form has well-formed syllable structure. This idea can be formalized by taking the rst-conjunct prosodic goal to be (230), with ‘minimally monosyllabic’ taken to mean that the rst-conjunct contains at least one

124

Chapter 7

syllable. (230)

⎤ ⎡ Bisyllabic ⎥⎥⎥ ⎢⎢⎢ BiSyl = ⎢⎢⎢⎣ Minimally monosyllabic ⎥⎥⎥⎦ WFSS

This formally expresses the idea that zidl is closer to being bisyllabic than dl . Under this assumption, [zidl ] is preferred to [dl ], which motivates zi [dl ]el→[zidl ]el . The condition for transcription in Ndebele is maximal satisfaction of BiSyl. Since I know of no evidence to the contrary, I will assume that BiSyl is not a special Ndebele innovation, but is universally taken to be the default goal for duplicant adjustment to bisyllabicity. This completes the analysis of the unintensive forms in (221). In the analysis proposed here, the fact that nal suffixes and Final Vowel a do not reduplicate is a consequence of the fact that the fs-cycle makes available epenthetic adjustment operations and these operations are preferred to JncShift. Since adjustment goals can be achieved by the epenthetic adjustment operations, JncShift is never used in the fs-cycle. Consequently, nal-suffix material is never used to augment the rst conjunct of the duplicant. 7.1.2 V-Initial Roots The initial glides in the imperative forms w-os-a ‘roast!’ and y-akh-a ‘build!’ are epenthetic glides that are inserted late to avoid a V-initial surface form: w before a round vowel, otherwise y. Duplicant adjustment also uses glide insertion. Some examples follow in (231). All the reduplicated forms except (231.3b) and (231.4b) have a epenthetic glide introduced by prosodic adjustment of the duplicant. (231)

1. w-os-a 2. y-akh-a 3. si-z-os-a

‘roast!’ ‘build!’ ‘we roast them’

4. si-z-akh-a

‘we build them’

w-osa-w-os-a y-akha-y-akh-a a. si-z-osa-w-os-a b. si-zosa-z-os-a a. si-z-akha-y-akh-a b. si-zakha-z-akh-a

si- is a subject-agreement prex and z- is an object-agreement prex. The reduplicated forms in (231) can be derived if it is assumed that (1) in addition to BiSyl, prosodic adjustment tries to build a duplicant

Case Study: Ndebele Unintensive Reduplication

125

with a C-initial second conjunct, and (2) truncated glide epenthesis at the left edge of the duplicant is available as an adjustment rule. Presumably, the C-initial goal is motivated by hiatus avoidance. The rst conjunct always ends in a vowel so hiatus will occur unless the second conjunct is C-initial. These assumptions give derivations like (232). Recall that if copying after truncation is to the left, truncated left edge material appears in the second conjunct and is not copied to the rst conjunct. (232)

1. 2. 3. 4.

[os] [osa] [wosa] osa-wos

juncture insertion motivated by BiSyl motivated by C-Initial transcription

The derivation could continue to eventually produce z-osa-w-os-a, the reduplicated form in (231.3a), by prexing an object-agreement prex and suffixing a nal vowel. Alternately, if there is no object-agreement prex to supply an initial consonant to the posttranscription form, as in (231.1), (232) could continue to produce w-osa-w-os-a, by prexing an epenthetic glide and suffixing a nal vowel. The derivations of (231.1) and (231.4a) are similar. The remaining forms achieve a Cinitial second conjunct by incorporating an object-agreement prex rather than epenthesizing a consonant. If longer V-initial stems are considered, there is one further source of variation. The inected verb in (233) is imperative. (233)

y-ebolek-a ‘borrow!’

ye-bole-boleka / y-ebo-yeboleka

The variation is most simply attributed to optional initial vowel extraprosodicity, the exclusion of the initial V from the domain to which juncture insertion applies. This is another means of avoiding a V-initial second conjunct. Recall that juncture insertion rst inserts [ at the left edge of the domain and Default Closure inserts ] at the right edge. This produces the following variation: (234)

ebolek

[ebo]lek → [yebo]lek → ebo-yebolek e[bole]k → e-bole-bolek

The forks above give the two alternatives that result from the option of initial vowel extraprosodicity. Initial glide insertion and nal-vowel epenthesis follow, producing the variation in (233). There is further variation if there is an inner suffix, because the position of UnInt in the word syntax can affect the reduplicated form.

126

(235)

Chapter 7

a. y-esul-a ‘wipe’ b. y-esul-el-a ‘wipe for’ c. y-os-ek-a

y-esu-yesula/ye-sula-sula ye-sula-yesulela/y-esu-yesulela/ ye-sule-sulela ‘roastable’ w-osa-woseka/wo-seka-seka/ w-ose-woseka

The inner suffix -ek is akin to the suffix -able in English. Derivations (with some steps combined) of the unintensive forms in (235) are given in (236). For (235b–c), which have an inner suffix, there are four possible derivations. UnInt can be realized either before or after the inner suffix, and initial-vowel extraprosodicity can either apply or not. (236)

a.

esul

b.

esul

[esu]l → [esu]la → [yesu]la → esu-yesula e[sul] → e[sul]a → e[sula]a → e-sula-sula

[esu]l → [esu]lela → [yesu]lela → esu-yesulela e[sul] → e[sul]ela → e[sula]ela → e-sula-sulela

[esu]lel → [yesu]lela → esu-yesulela esul → esulel

e[sule]l → e[sule]la → e-sule-sulela c. os→[os]→[os]ek→[os]eka→[wosa]eka→osa-woseka [ose]k → [wose]ka → ose-woseka os → osek

o[sek] → o[seka]a → o-seka-seka Note that although there are four possible derivations in (236b), there are only three different outcomes. o-seka-oseka in (236c) provides good supporting evidence for the assumption made earlier that UnInt can appear higher in the syntactic structure than immediately above the root. If UnInt merged directly with the root, it would be necessay for [os ]ek to somehow be adjusted to o [sek a]. The only plausible basis for this would be extraprosodicity of initial o. However, although initial onsetless syllables can (optionally) be taken to be extraprosodic for the purpose of juncture insertion, they cannot be extraprosodic in evaluating the rstconjunct BiSyl goal. If the rst syllable of [ose ], for example, did not contribute to the syllable count for BiSyl, we would expect the adjustment [ose ]k→[ose yi ]k to be possible in order to obtain rstconjunct satisfaction of BiSyl. But it is not; o-seyi-seka is not possible in (236c).

Case Study: Ndebele Unintensive Reduplication

127

7.1.3 Duplicant Adjustment Summarized Before the summary, one last adjustment rule needs to be discussed. It was noted above that epenthetic yi-prexation is used to augment the fs-form to bisyllabicity in imperative forms like yi-dl-a ‘eat!’. Call this left-edge epenthesis yi-Ep (as opposed to yi-Ep, introduced above). It is also available to adjust the duplicant in special circumstances. In example (237), there is a subject-agreement prex and a nal suffix (perfect). (237)

ba-m-ile ‘they stood’

ba-mayi-mile, ba-yima-yimile

yi-Ep is used only if there is no object agreement prex, as in (237). As a duplicant adjustment operation, it is constrained to apply in the context # [ . I assume that like all the other epenthetic adjustment rules, it is an fs-cycle rule. The fs-cycle of the forms in (237) is given in (238). (238)

a. [m]ile→[yim]ile→[yima]ile→yima-yimile b. [m]ile→[ma]ile→[mayi]ile→mayi-mile

The full iterative duplicant adjustment rule can now be given:  ⎫ ⎞ ⎛⎧ ∅ → a  ] ⎪ ⎪ ⎪ ⎟⎟⎟ ⎜⎜⎜ ⎪  ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎟⎟⎟ ⎜ ∅ → yi  ] ⎬   ⎜⎜⎜ ⎨  ⎟⎟⎟ ⎪ First conjunct: BiSyl ⎜⎜⎜ ⎪ ⎪ ⎪ ∅ → yi #[ ⎪ ⎟⎟⎟ (239) :: ⎜⎜ ⎪ ⎪ ⎪ ⎪ ⎪  ⎭ Second conjunct: C-Init ⎟⎟ ⎜⎜⎜ ⎩ ∅ → y, u  [ FS-cycle ⎟ ⎟⎠ ⎝⎜ JncShift The epenthetic adjustments are unordered and available only in the fscycle. Juncture Shift is ordered after all of the epenthetic adjustments. The effect of this ordering is that JncShift never applies in the fs-cycle because it is always bled by an epenthetic operation. There is much that is arbitrary about (239). This is as it should be. Many variations on (239) are possible, even given the general framework of bisyllabic reduplication. This will become much clearer after the discussion of Kinande unintensive reduplication in the next section. It will be shown to be a variation of (239), but with many minor differences in the array of adjustment operations and their ordering. First, Juncture shift is much more limited in Kinande. [-Left is not available at all, so object-agreement prexes are never reduplicated. ]-Right is allowed, but only to the right edge, so suffixes are either fully reduplicated or not reduplicated at all. Second, JncShift is ordered before epenthetic adjustment, so epenthesis is never used to augment

128

Chapter 7

the duplicant if incorporating a nal suffix into the duplicant is possible. Third, syllable epenthesis is not available as an adjustment, so some other mechanism for augmenting consonantal roots in Kinande is required, if necessary, in order to produce a bisyllabic reduplicant. Kinande uses double reduplication. Comparison with Asheninca Campa (section 7.3) also reveals the arbitrary character of the adjustment choices made in Ndebele. Rather than a-Ep at the right edge, Asheninca Campa uses nontruncated a-Ep at the right edge to obtain a bisyllabic reduplicant. (240)

Asheninca Campa [kow] → [kowa] → kowa-kowa Ndebele [lim] → [lima] → lima-lim

Note the hiatus problem that the V-nal form in Asheninca Campa will cause with V-initial suffixes. It is overcome by epenthesizing a consonant. The morphology certainly affects how the reduplication process plays itself out. It cannot be otherwise. The word syntax determines the framework within which the juncture insertion rules and the duplicant adjustment operations apply. The Ndebele adjustment rules, however, are sensitive only to the phonological structure of the forms they apply to. None of them mention morphological structure. 7.1.4 Applied Passives Ndebele is one of several Bantu languages that allow either object in double-object constructions to raise to the surface subject position in passives. (241)

a. abantwana b-a-phek-el-w-a ukudula children sm(3pl)-tm-cook-appl-pass-fv food ‘food was cooked for the children’ b.

ukudula kw-a-phek-el-w-a abantwana food sm(3sg)-tm-cook-appl-pass-fv children ‘food was cooked for the children’

Subject agreement conrms that the the left-edge nominal is the surface subject. The unintensive sentences corresponding to (241) are   pheka (242) a. abantwana b-a-phek-el-w-a ukudula pheke

Case Study: Ndebele Unintensive Reduplication

b.

129

⎧ pheka ⎫ ⎪ ⎪ ⎪ ⎪ ⎬ ⎨ pheke -phek-el-w-a abantwana ukudula kw-a-⎪ ⎪ ⎪ ⎪ ⎭ ⎩ phekwa

The appearance of the reduplicant phekwa in (242b) is unexpected, since it does not appear to copy the material to its right. HI&S give an account of the appearance of phekwa in (242b) based on the idea that surface phek-el-w in (242a) is the underlying form, but that surface phek-el-w in (242b) is derived from underlying phek-w-el . The forms derived from underlying phek-el-w are called passivized applicatives and those derived from phek-w-el are called applied passives. If this is translated into a cyclic framework, the claim would be that there are derivations that proceed as follows: (243)

phek pass phek-w appl phek-w-el → phek-w-el-w

To construct an account on this basis, it must be the case that this order of morpheme realization is possible in (242b), but not in (242a). Consonantal roots give evidence that supports these claims. There is passive suffix allomorphy, with the passive realized as -iw directly after consonantal roots, otherwise as -w. This producess lim-w-a ‘be cultivated’, but dl-iw-a ‘be eaten’. The forms dl-el-w-a ‘be eaten for’ and lim-el-w-a ‘be cultivated for’ are as expected. But dl-iw-el-w-a is also possible. The most straightforward explanation for the possibility of surface dl-iw-el-w-a and the impossibility of surface lime-w-el-w is illustrated in (244). Hiatus avoidance prevents w deletion in (244b), but not in (244a).6 (244)

a.

b.

lim

pass lim-w appl lim-w-el → lim-w-el-w → lim-w-el-w dl

pass dl-iw appl dl-iw-el → dl-iw-el-w

This gives evidence that lim-el-w can be derived from an applied passive, as well as evidence that transposition is copying plus deletion, a fundamental idea in DR. The succesion of autosegmental representations is instructive. It is given in (245).

130

Chapter 7

(245) a. × × × × × → × × × × × × dl i w e l dl i w e l b. × × × × × × → × × × × × × × → × × × l i mw e l

l i mw e l

×××

l i mw e l

NCC Repair applies to the output in (245a) at some point. Assuming that this is correct, consonantal roots therefore furnish a means to disambiguate structures of the form [Root] ⊕ appl ⊕ pass from those of the form [Root] ⊕ pass ⊕ appl. According to Sibanda (personal communication), the grammatical sentences below all equivalently mean ‘food was eaten for the children’. (246)

a.

ukudula abantwana b. ukudula *abantwana

kw-a-dl-el-w-a b-a-dl-el-w-a kw-a-dl-iw-el-w-a b-a-dl-iw-el-w-a

abantwana ukudula abantwana ukudula

In the passivized applicative, either the direct object (ukudula ‘food’) or the applied object (abantwana ‘children’) can appear in the surface subject position. The applied passive, on the other hand, allows only the direct object in subject position.7 We therefore have evidence that (241b) can be either an applied passive or a passivized applicative and (241a) must be a passivized applicative.8 The (241b) form with reduplicant phekwa is derived as follows: (247)

phek pass phek-w UnInt [phek-w] appl [phek-w]-el† fs [phek-w]-el-a → [phek-wa]-el-a → phekwa-phek-w-el-a → phekwa-phek-el-w-a

Crucially, w does not transpose out of the duplicant at †. Transposition is itself reduplicative. Transposing w and el requires the formation of [w el ]. But this would create impossible overlapping duplicants, blocking w transposition. After transcription eliminates the t-juncture, w transposes to the right. 7.1.4.1 The MDT Analysis of Applied Passives The HI&S 2003 analysis of applicativized passives is based on Inkelas and Zoll’s (2000)

Case Study: Ndebele Unintensive Reduplication

131

Morphological Doubling Theory (MDT). They are driven to this kind of analysis because they reject a morpheme-by-morpheme view of how morphophonological form is constructed. On the one hand, they have an underlying applicativized passive form phek + pass + appl and on the other the surface form phek-el-w. Without a cyclic analysis, there is no stage in which there is a phonological form phek-w that reduplication could operate on. They must therefore derive both phekelwa and the reduplicant separately from underlying form phek + pass + appl. This is morphological doubling. What HI&S needs from the separate phonologies which generate the reduplicant and base is illustrated below. (248)

phek + el + w

a. 

pheka pheke



phekelwa

b.

phek + w + el ⎧ pheka ⎫ phekelwa ⎪ ⎪ ⎪ ⎪ ⎬ ⎨ phekwa ⎪ ⎪ ⎪ ⎪ ⎭ ⎩ pheke

There is no difficulty in formulating rules that derive the reduplicants in (248a) and the rst two reduplicants in (248b). But it is far from clear why pheke appears in (248b) and why phekwe does not. HI&S is forced to impose unmotivated stipulations. To get the reduplicant pheke, it proposes that the passive suffix can be skipped in deriving reduplicants.9 Since there is no way to connect this to right transposition of w in the derivation of the stem in the other phonology, this is purely a stipulation, not an explanation of anything. Then, to prevent phekwe, HI&S claims that “the morphotactics prohibit the [e] of applicative -el- from following the passive -w- on the surface.” If it could be argued that this were the driving force behind passive displacement to the right, this would be an explanation. But it is not, as dl-iw-el-w demonstrates. This form not only violates the purported phonotactic, but it shows that displacement to the right does nothing to eliminate its violation, since applicative e continues to follow w in the output. In summary, MDT manages to produce (248b), but only by making two crucial stipulations. The disconnect between skipping the passive suffix in the reduplicant cophonology and transposing the passive suffix to the right in the base cophonology is particularly telling against MDT. As one might suspect, the problem with separating the derivation of the reduplicant from the derivation of the base is that steps in the derivation

132

Chapter 7

of the base (i.e., rightward w displacement) become disconnected from features of the derivation of the reduplicant (i.e., “skipping over” w). 7.1.5 Overapplication of Perfective Imbrication There is perfect affix allomorphy, with the morpheme which carries the perfect feature realized by both -ile (the long perfect suffix) and -e (the short perfect suffix). This morpheme will be called Perf in what follows. Most roots can take the long suffix and some of these also take the short suffix. There are also some roots that take only the short suffix. The short suffix induces stem ablaut, either as the result of some sort of fusion of the il portion of the suffix with the root, or as a stem readjustment rule historically derived from such fusion. If the nal vowel of the stem is a, it raises to e, as in (249c), which is boxed below. This is called imbrication in the Bantu literature. (249) a. b. c.

Root dabul lim dumal

‘tear’ ‘cultivate’ ‘become depressed’

Perfect dabul-ile dabul-e lim-ile *lim-e *dumal-ile dumel-e

(These perfect forms will actually occur with various prexes, which are irrelevant to the issue at hand.) We expect the unintensive perfect form duma-dumele if the reduplicant duma is transcribed before the suffix -ile has a chance to induce imbrication. This form occurs, but two other forms also occur: dume-dumele and remarkably duma-dumalile, with no imbrication of a verb root that in unreduplicated forms always takes the short perfect suffix with imbrication. The unexpected forms can be understood in the following way. I proposed earlier that transcription is obligatory if the duplicant satises rst-conjunct bisyllabicity. But nothing was said about NCC repair. Presumably, it is not forced by any general principles until the fs-cycle. Suppose that NCC Repair following Transcription is not obligatory until the fs-cycle, which produces a surface form. Then, as an option, the stem at the point that Perf is realized can be [[dumal]]

(250)

× × × × × × × × × d u m a l

Case Study: Kinande Unintensive Reduplication

133

If the short perfect suffix -e is concatenated and imbrication applies to (250), vowel ablaut occurs in both the reduplicant and root, producing dume-dumele. This is overapplication. But this outcome assumes that imbrication is not blocked by Geminate Inalterability. If it is, we would expect either that -e is concatenated and imbrication does not apply, producing duma-dumale, contrary to fact, or that realization of Perf by the short suffix is blocked. If Perf is realized by the long suffix when realization by the short suffix is blocked, duma-dumalile would result. We can therefore understand the three-way variation as variation in the timing of NCC repair, and variation in the application of Geminate Inalterability to imbrication, either yielding overapplication or blocking the use of the short suffix (thereby forcing the use of the long suffix). 7.2 Kinande Unintensive Reduplication Kinande and Ndebele are related Bantu languages whose inected verbs have a similar morphological structure, with similar unintensive verbal reduplication processes, and with similar syllable structure. Comparison is therefore useful in understanding the loci of parametric difference. Kinande forms with unintensive inection are usually translated into English with a meaning of ‘hurriedly’ or ‘here and there’. Kinande unintensive reduplication was analyzed in Hyman and Mutaka 1990, a key work on Bantu reduplication.10 This section is heavily indebted to it and to Mutaka 2000.11 The goal will be to understand Kinande and Ndebele reduplication as variants of each other. There are minor differences in the syntax and morphology, and more extensive differences in the phonology of adjustment to bisyllabicity. Syntactically, I assume the verb structure relevant to unintensive reduplication is the same, except that the position of UnInt is more restricted in Kinande. It must appear as a sister of the root, or directly above the causative morpheme. This will be discussed more fully in section 7.2.1. UnInt realization is ∅ → [ ×, just as in Ndebele, but there is no domain selection in Kinande. Consequently, UnInt realization is just α → [α] (after Default Closure). On the phonological side, the general structure of cyclic adjustment with obligatory transcription in the fs-cycle is the same. But there are differences in how duplicant adjustment operates. First, it is obligatory. Second, extrametricality is treated differently. In Ndebele,

134

Chapter 7

extametricality was relevant to juncture insertion, but not duplicant adjustment. Precisely the opposite is true in Kinande. Finally, and most signicant, although there is some overlap, there are substantial differences in the array of adjustment operations that iterative duplicant adjustment can call on. I begin with some examples. The abbreviations used in the glosses were given on p. 118. (251)

a. tu-nyi-sw-er-e tunyi-swere-swere sm-om-grind-appl-subj, ‘let’s grind for me’ b. eri-hum-ir-a eri-huma-humira nonfin-beat-appl-fv, ‘to beat for’ c. eri-nyi-sw-a erinyi-swaswa-swa nonfin-om-grind-fv, ‘to grind for me’

The Kinande applicative suffix is -er/-ir, subject to vowel-height harmony. There are some striking differences between the examples in (251) and corresponding examples in Ndebele: there is no variation; nal suffixes are sometimes reduplicated (251a); object-agreement prexes are never reduplicated; there is no syllable epenthesis; and there is double reduplication (251c). It will be most efficient to give the full array of duplicant adjustment operations at the start, then illustrate with relevant examples. Consideration of details of the goal shape that drives duplicant adjustment will be put off until V-initial roots are considered. Until then, it suffices to assume that the goal is BiSyl, as in Ndebele. ⎞ ⎛ ]-Rightmost ⎟⎟⎟ ⎜⎜⎜ ⎟⎟⎟ ⎜⎜⎜ ⎟⎟⎟ ⎜⎜⎜ [V → V[ ⎟⎟⎟ (252) ⎜⎜⎜ # a-Ep $ ⎟⎟⎟ ⎜⎜⎜ ⎠⎟ ⎝⎜ ∅ →[  [ FS-cycle

As in Ndebele, special adjustment rules are active only in the fscycle. Because transcription must occur in that cycle, ample means to adjust duplicants to rst-conjunct bisyllabicity must be available. Except for the last one, the repair rules are a sparse subset of the rules available in Ndebele, which allows arbitrary juncture shift, syllable epenthesis, and glide insertion. Notably, however, the juncture shift adjustment operations in Kinande are more highly ranked than the special fs-cycle operations, unlike in Ndebele, in which the special fs-cycle operations bleed the juncture shift operations. The Kinande juncture

Case Study: Kinande Unintensive Reduplication

135

shift operations are much more limited than those in Ndebele. ] can shift to the right, but only the right edge (hence the name ]-Rightmost). [ can shift, but is used only to exclude an initial vowel from the duplicant. Recall that “V” in this context designates a timing slot linked to a vowel. The use of the adjustment operations (252) will be illustrated by going through the derivations of the forms in (251) in some detail. (253)

unint [sw] appl [sw]er → [swer] (adjustment is obligatory) om nyi[swer] subj nyi[swer]e → nyi[swere] → nyi-swere-swere

In the last cycle, note that although epenthetic adjustment bleeds juncture shift in the fs-cycle in Ndebele, it does not in Kinande. This accounts for the fact that there is nal suffix reduplication in Kinande, but not in Ndebele. In (254), [hum ]ir→[humir ] is not possible because the rst conjunct of the resulting duplicant would be overweight. ]-Rightmost is always to the right edge. Because arbitrary JncShift is available in Ndebele, [hum ]ir→[humi ]r would be possible. (254)

unint [hum] appl [hum]ir fv [hum]ira → [huma]ira → huma-humira

In (255), note that the object agreement prex cannot be used to augment the duplicant because [-Left is not a Kinande adjustment operation. Most interestingly, since Kinande does not have syllable epenthesis repair, another means must be found for producing a bisyllabic reduplicant from a monosyllabic duplicant. Double reduplication is used. Another [-juncture is inserted at the left edge and closed by Default Closure. (255) unint

[sw] om nyi[sw] fv nyi[sw]a → nyi[swa] → nyi[[swa]] → nyi-swaswa-swa I now consider a group of examples that illustrate the treatment of extrametricality. am denotes an aspectual marker and nonfin denotes the innitival marker. (256)

a. eri-ambal-a sc-dress-fv, ‘to put on clothes’

eri-a-mbala-mbala

136

Chapter 7

b.

mO-tu-a-Ot-ire

tm-sm-tm-light-fs, ‘we lit (a re)’

c. tu-kabi-Ot-a sm-am-light-fv, ‘to grind for me’

mO-tu-a-O-tire-tire tu-kabi-OtaOta-Ota

The crucial example is (256b), since this requires [Ot]ire → [Otire] → O[tire] We can understand the rst adjustment as driven by BiSyl, with the initial syllable taken to be extrametrical. What then drives the second adjustment? It can be understood as a secondary preference for a Cinitial rst conjunct. We therefore suppose that the iterative adjustment rule is driven by   BiSyl First conjunct ; C-initial BiSyll must be understood as excluding onsetless syllables. The fs-cycle steps in the derivations of the (256) examples are given below. Since there is no domain selection, initial juncture insertion is always at the left edge. (257)

a. [ambal]a → [ambala] → a[mbala] → a-mbala-mbala b. [Ot]ire → [Otire] → O[tire] → O-tire-tire c. [Ot]a → [Ota] → [[Ota]] → OtaOta-Ota

The reader might expect the result in (257c) to be O-tata-ota, given the preference for C-initial reduplicants. Iterative duplicant adjustment does not look beyond the immediate representation and potential transformations of it by the adjustment rules, so [Ot ]a→[Ota ]→[[Ota ]] proceeds to satisfy the most highly ranked goal, BiSyl, before C-initial could have any effect on the derivation. At that point, [-Right does not produce a C-initial rst conjunct, only [O [ta ]]. The hiatus in the output of (257c) is resolved later in the derivation. Kinande resolves hiatus by deleting the leftmost vowel, so OtOtOta (with whatever other changes the phonology dictates) surfaces. The fact that juncture insertion is α → [α] creates a problem in reduplicating long roots in Kinande since the rst conjunct of the duplicant might be overweight. ambul is not overweight, because of extrametriality, and reduplicates as shown in (256a) and (257a). bugul, on the other hand, is overweight and does not reduplicate. In H&M’s corpus of 218 overweight verb roots, approximately half undergo total stem

Case Study: Kinande Unintensive Reduplication

137

reduplication (with unintensive meaning), but do not take suffixes except nal-a.12 For example, consider eri-bindula-bindul-a ‘to change’, but *tu-bindule/bindula-bindul-e ‘let’s change (subjunctive)’. About one-third do not reduplicate at all. The remainder are roots that have a pseudostructure as (V)CVC root plus suffix, like gambul, with pseudostructure gamb + ul, even though -ul is not an active suffix in contemporary Kinande. These roots reduplicate as if the pseudostructure were real, which it must be in some appropriate sense. There is one peculiarity: they do not reduplicate if there is an inner (applicative, passive, and so on) suffix, but do reduplicate with an inectional suffix.13 For example: eri-gamba-gambul-a ‘to talk’, but *eri-gamba-gambul-ir-a ‘to talk to’, with an applicative suffix; and tu-gamba-gambul-e ‘let’s talk’, with a subjunctive nal suffix. 7.2.1 Passives and Causatives The passive suffix -w and causative suffix -y both transpose to the right, like the passive suffix in Ndebele. First consider the passive: (258)

nonfin-beat-pass-fv

eri-hum-w-a ‘to be beaten’

a. b.

eri-humwa-humwa p eri-huma-humwa

eri-hum-ir-w-a ‘to be beaten for’ nonfin-beat-appl-pass-fv

c. d.

eri-humwa-humirwa eri-huma-humirwa

The derivations are given below (with some obvious steps omitted). (259)

a.

unint [hum] pass [hum]w † fv [hum]wa → [humwa] → humwa-humwa

† There is no adjustment because it would not contribute to satisfying any of the goals of duplicant adjustment. b.

unint appl pass fv

[hum] [hum]er [hum]erw [hum]erwa → [huma]erwa † → huma-humerwa

† ]-Right must be to the right edge, so it cannot apply here. The causative facts are slightly different.

138

(260)

Chapter 7

eri-gul-y-a

‘to sell’

nonfin-buy-caus-fv eri-gul-ir-y-a ‘to sell to’ nonfin-buy-appl-caus-fv

a. b. c. d.

eri-gulya-gul-y-a eri-gula-gul-y-a eri-gulya-gul-ir-y-a eri-gula-gul-ir-y-a

So far, I have assumed that unint must merge with the root. In order to account for (260c), I will assume that unint can also merge directly above the causative morpheme. The order of morpheme composition in (260c) can therefore be [Root] ⊕ cause ⊕ unint ⊕ appl. The two outcomes (260c,d) come from different syntactic structures. (261)

a.

caus unint appl fv

gul guly [guly] [guly]ir† [guly]ira → [gulya]ira → gulya-gulyira → gulya-gul yirya

† Just as in the similar case in Ndebele (section 7.1.4, p. 130), y transposition to the right at † is blocked and is therefore delayed until after duplicant transcription. b.

unint caus appl fv

gul [gul] [gul]y [gul]yir → [gul] yiry [gul]irya → [gula]irya → gula-gulirya

7.2.1.1 The -es Causative Suffix Kinande has two causative sufxes. In addition to the short causative suffix -y, there is a long causative suffix -es (or -is, subject to vowel height harmony). All causative forms have -y. Some must have both -y and -es, others can have either only -y or both -y and -es, and others have only -y (gul above, for example). The paradigm of roots that optionally take the -es/is is illustrated in (262a). It should be compared with (260), repeated here in (262b). On the left below are the unintensive long and short causative forms of gend. On the right are unintensive causative, and unintensive applied causative forms of gul, which takes only the short causative suffix.

Case Study: Kinande Unintensive Reduplication

(262)

a. eri-genda-gend-y-a eri-gendya-gend-y-a eri-genda-gend-es-y-a eri-gendya-gend-es-y-a

139

b.

eri-gula-gul-y-a eri-gulya-gul-y-a eri-gula-gul-ir-y-a eri-gulya-gul-ir-y-a

Based on (262), I assume that the causative suffix -es of gend-type verbs interacts with the y causative in exactly the same way that the applicative suffix does. -y is realized inside -es, but transposes to the right past -es, just as it transposes to the right over the applicative suffix. The reduplicated forms for roots that require -es/is is illustrated in (263). (263) eri-huma-hum-is-y-a, *eri-humya-hum-is-y ‘to cause to beat’ I assume that for this class of verb roots, -y transposition over -es has been grammaticalized in some way and that -y is realized outside of -es. Therefore, humya-hum-is-ya cannot be generated, since the only possible source is [hum ]-es-y. Reduplication of the causative stem hum-es-y is impossible because it is overweight. For sufficiently light roots, however, a long causative stem (i.e., one with both -es and -y) can be reduplicated. Mutaka (2000, 37 n. 4) gives several relevant examples. For the root Ot ‘light’ (a re, for example), there are two unintensive perfect causative forms. (264)

mo-tu-a-o-tesa-t-es-ir-y-e, mo-tu-a-o-tesya-t-es-ir-y-e ‘we made (s.o.) light hurriedly’

The variation comes from the variation in the hierarchical position of

unint, either a sister of the root, or directly over cause. The following partial derivations should be sufficient to understand the source of the variation.14 (265)

a.

unint caus1 caus2 perf

Ot [Ot] [Ot]es → [Otes] [Otes]y † [Otes]yire → [Otes] yirye → [Otesa]irye → O[tesa]irye → O-tesa-tesirye

† ]-Right does not apply because it would not contribute to satisfying any of the goals of duplicant adjustment.

140

Chapter 7

b.

caus1 caus2 unint perf

Ot Otes Otesy [Otesy] [Otesy]ire † → [Otesya]ire → O[tesya]ire → O-tesya-tes yirye

† y does not transpose out of the duplicant, for the same reasons that it does not in (261a). Note that y transposes to the right to become an onglide of the nal vowel, even to a position inside the perfect suffix -ire, so -yire→- yirye. 7.3 Asheninca Campa Intensive Reduplication The version of Asheninca Campa that is analyzed here is the one which was analyzed in McCarthy and Prince 1995, based on Payne 1981 and Spring 1990.15 The insights into the morphophonology of Asheninca Campa are largely due to Payne, McCarthy and Prince, and Spring. The contribution here is mainly to show that McCarthy and Prince’s claim that a satisfactory analysis of Asheninca Campa reduplication is beyond the reach of derivational morphophonology is unfounded. Verbs usually appear with at least one prex (commonly a subjectagreement or future-tense marker) and one or more suffixes. Syllable structure is straightforward. The only codas that occur are nasals that form a homorganic cluster with a following onset. Onsetless syllables occur only initially. A few morpheme-internal diphthongs are permitted, as in the continuative suffix -wai, but hiatus over morpheme boundaries is eliminated. V1 V2 over a morpheme boundary is repaired by deleting one of the vowels if V2 is a root vowel, otherwise by epenthesizing an intervocalic t. In the deletion case, V2 is deleted if it is high and V1 is low, otherwise V1 is deleted. In the examples below, it is always a prex nal vowel that deletes. Epenthetic a is used to break up impermissible consonant clusters and to syllabify nal consonants. Some examples of boundary epenthesis and deletion follow. Epenthetic a and t are marked with a “dotunder” to aid the reader. (266)

no +asi >nasi ‘1st-cover’ (‘I/we cover’) asi +a:nc/ hi >asit•a:nc/ hi ‘cover-nonfin’ (‘to cover’) kow +wai +ak +i >kowa• wait•aki ‘want-cont-perf-nonfut’ (‘has continually wanted’)

Case Study: Asheninca Campa Intensive Reduplication

141

C-initial suffixes trigger stem augmentation to bimoraicity. Below, -a:nc/ hi is the nonnite suffix, and Payne glosses -piro as ‘verity’. The augment is boxed. (267)

a. b. c.

p +piro +a:nc/ hi >pa• a -piro-t•a:nc/ hi na +piro +a:nc/ hi >na ta -piro-t•a:nc/ hi naa +piro +a:nc/ hi >naa-piro-t•a:nc/ hi

‘to truly feed’ ‘to truly carry’ ‘to truly chew’

a-epenthesis at the p +piro juncture in (267a) is expected, but a long vowel is not. Syllable epenthesis in (267b) is completely unexpected. The stem naa is bimoraic in (267c) and there is no augmentation. V-initial suffixes do not induce augmentation. (268)

a. b. c.

p +a:nc/ hi >p-a:nc/ hi na +a:nc/ hi >na-t•a:nc/ hi naa +a:nc/ hi >naa-t•a:nc/ hi

‘to feed’ ‘to carry’ ‘to chew’

Examples like no +na +wai >nonawai (*nona ta wai) ‘I/we carry’ show that the issue is stem augmentation, not root augmentation. The subjectagreement prex no suffices for the bimoraicity condition. To account for this stem minimality condition, I assume that (1) all prexes are realized before any suffixes; (2) no stem syllabication occurs until all the prexes are realized, at the earliest; (3) C-initial suffixes induce stem syllabication; and (4) syllabication imposes a concomitant bimoraic minimal word requirement. Stem augmentation to bimoraicity, if required, is carried out by  & % ta-Epenthesis Root Stem ; Bimoraic :: Vowel lengthening Two derivations are given below. Since ta epenthesis is restricted to the right edge of the root, it cannot apply to pa• at the point † below, so lengthening applies: (269)

p-piro pa• -piro† pa• a -piro

na-piro na ta -piro

a-Epenthesis Stem Augmentation

One might speculate on why ta epenthesis and lengthening alternate in the way that they do, but since the facts are clear (and very sparse), I refrain from speculation. I now consider reduplication. Asheninca Campa has a productive verbal reduplication process used to signal “excess” and usually glossed ‘more and more’. It frequently occurs in conjunction with the continua-

142

Chapter 7

tive suffix -wai. I rst consider V-nal roots. no- is rst-person subject agreement and the suffix is the continuative. (270) a. b. c. d. e. f.

no-kawosi-wai no-koma-wai n-osampi-wai n-asi-wai no-na-wai no-naa-wai

‘I/we bathe’ ‘I/we paddle’ ‘I/we ask’ ‘I/we cover’ ‘I/we carry’ ‘I/we chew’

Reduplicated no-kawosi-kawosi-wai no-koma-koma-wai n-osampi-sampi-wai n-asi-nasi-wai no-na-nona-wai no-naa-nonaa-wai

Example (270c) justies the assumption above that transcription is to the right, since transcription to the left would produce a morpheme internal reduplicant, o-sampi-sampi. The reduplicant is polysyllabic throughout, incorporating the subject prex if necessary to achieve a polysyllabic reduplicant. Note that the reduplicant in (270a) is trisyllabic. The reduplication process is total root reduplication, with prosodic adjustment to polysyllabicity:   % & WFSS [-Right First Conjunct ; :: Polysyllabic [-Left Since not much is known about the syntax and semantics of Asheninca Campa, the intensive morpheme int cannot be located in the syntactic structure of the inected verb with any condence. Since it allows the most straightforward analysis, I will tentatively assume that int is realized after the prexes are realized, but before any suffixes  are × realized. It species the root as its domain and triggers ∅ → [ (leftmost in the domain). Several derivations are given in (271). Default closure is implicit. (271)

a. Root 1sg:agr

int

b.

Root 1sg:agr

int

c. Root 1sg:agr

int

osampi no-osampi→n-osampi n[osampi]→no[sampi]→nosampi-sampi asi no-asi→n-asi n[asi]→[nasi]→nasi-nasi na no-na no[na]→[nona]→nona-nona

I suppose that onsetless syllables are permitted word initially because

Case Study: Asheninca Campa Intensive Reduplication

143

they follow a word boundary, but WFSS does not allow onsetless syllables in the rst conjunct, where the initial syllable follows a duplication juncture. Prosodic adjustment in (271b) is therefore driven by WFSS. [-Left must be used because [-Right does not result in polysyllabicity. In (271a), on the other hand, either [-Left or [-Right gives full satisfaction of the adjustment goal, but [-Right is used because it is more highly ranked. Prosodic adjustment calls for syllabication of the rst conjunct. This has several effects. C-nal rst conjuncts violate WFSS and are eliminated by nal-a epenthesis. Syllabication entails augmentation to bimoraicity, so there will be augmentation of the rst conjunct if the root is sufficiently light and there are no prexes. (272) a. b. c.

kow-a:nc/ i p-a:nc/ h i na-t•a:nc/ h i h

‘to want’ ‘to feed’ ‘to carry’

Reduplicated kowa-kowa-t•a:nc/ h i paa-paa-t•a:nc/ h i nata-nata-t•a:nc/ h i

The derivations follow: (273)

a. Root

int nonfin b.

Root

int nonfin c. Root

int nonfin

kow [kow]→[kowa• ]→kowa-kowa kowakowa-a:nc/ h i→kowakowat•a:nc/ h i p [p]→[pa• ]→[paa]→paa-paa paapaa+a:nc/ h i→paapaat•a:nc/ h i na [na]→[nata]→nata-nata natanata+a:nc/ h i→natanatat•a:nc/ h i

Augmentation to bimoraicity is considered part of prosodic adjustment above, since it is a side effect of rst conjunct syllabication, hence na→nata and pa→paa. 7.3.1 Copying a Word Boundary There is one last feature of Asheninca Campa intensive reduplication to consider. Consider an unprexed VCV root like asi ‘cover’. (274)

asi-t•a:nc/ h i asi asi-t•a:nc/ h i

‘to cover’ ‘to cover more and more’

Surprisingly, the reduplicated form has an internal word break. Both

144

Chapter 7

the phonetic judgments of linguist eld workers and the stress pattern agree on this. See Spring 1990, 148 n. 3. Payne (1981, 146), in his careful and insightful study of Asheninca Campa, says that “one of the most striking elements of Asheninca reduplication is that it reduplicates a word boundary into some forms, but not into others. Preferring twosyllable verbs with initial consonants to reduplicate, it appears that in the absence of the initial consonant in VCV-type verbs, the rule actually copies the word boundary (italics added). Payne’s insight can be formalized in the following way. Suppose that word breaks are represented by timing slots which are associated with some kind of special marker; # is used in (275). Then (274) is the result of the regular application of prosodic adjustment—the rule actually copies the word boundary. Recall as well that the idea that word breaks can occur inside the duplicant was already introduced in the discussion of tasty-shmasty reduplication in section 4.1.1. ∗σ σ σ σ σ σ [-Left (275) ×[× × ×]× −→ [× × × ×]× → × × × × × × × × × # a s i #

# a s i #

# a s i

#

If a vowel is associated with a timing slot that follows a timing slot associated with a word boundary, it satises the conditions on wellformed syllable structure, so [-Left is motivated by WFSS. I take it to be evidence in favor of the theory advocated here that this “striking element” of Asheninca Campa reduplication is so immediately comprehensible in terms of the computational mechanisms which the theory makes available. 7.4 Washo Plural Reduplication Washo plural reduplication was analyzed in Broselow and McCarthy 1983 in their important study of internal reduplication. Their starting point was the analysis of Winter 1970, based on the data in Jacobsen 1964. It is set in the melody-copy framework inherited from Marantz 1982 and depends heavily on the manipulation of long vowels in ways that have since proved untenable.16 Washo reduplicative pluralization is also the subject of a recent study by Yu 2001, who adds much useful data and discussion of various approaches to Washo reduplication. I rely on Broselow and McCarthy and Yu for the data. There is allomorphy in Washo reduplicative pluralization. There is a

Case Study: Washo Plural Reduplication

145

default affix that realizes plurality, and a special affix that is used with monosyllabic stems and some lexically specied polysyllabic stems. (The use of the term affix here does not imply concatenative morphology, but simply designates a vocabulary item which realizes an inectional morpheme.) This ts a pattern that we have encountered several times before, with special treatment of monosyllabic stems. I will consider the two affixes in turn, beginning with straightforward examples of the effects of the default affix, which will be called ρCV , for reasons that will soon be clear. 7.4.1 CV-Reduplication First, some examples. (276) a. b. c. d. e. f. g. h.

Root ha˜nakmuwe baloxat daPa duweP moya PewˇsiP nent’us mokgo

Plural ha˜nakmuwewe baloxaxat daPaPa duweweP moyaya PeˇsiwˇsiP net’unt’us mogokgo

‘elks’ ‘bows’ ‘mother’s brother’ ‘to try to, to want to’ ‘shoulder’ ‘father’s brothers’ ‘old women’ ‘shoes’

The task for both the linguist and the Washo language learner is to deduce the pretranscription representations, and then the rules for generating these pretranscription representations. Examples (276f–h) are the most complex, and therefore the most revealing. The simplest way to generate this pattern is left transcription with leading-edge truncation.17 (277)

ne[nt’u]s → ne-t’u-nt’us mo[kgo] → mo-go-kgo Pe[wˇsi]P → Pe-ˇsi-wˇsiP

This suggests that the juncture insertion rules are   ∅ → ] Right V , ∅ → [ Right V and that prosodic adjustment of the rst conjunct results in truncation. Recall that the second rule will insert [ in such a way that a nontrivial duplicant is formed. Since an increasing sonority requirement on onset consonants is a common well-formedness condition, we can further

146

Chapter 7

speculate that the truncation is driven by a well-formed syllable structure desideratum for the rst conjunct. These considerations successfully derive all the examples in (276). Note that the juncture insertion rules are the mirror image of the Mangarayi juncture insertion rules (p. 70). Jacobsen (1964, 117) observed that initial two-consonant clusters that occur in indigenous Washo words are in fact much more restricted than simply by sonority sequencing. Only P followed by a voiced sonorant is permissible. If we assume that this is a syllable wellformedness condition on onsets, the following examples support the idea that the truncation in (277) is driven by rst conjunct syllable wellformedness. (278) a. b. c.

Root baPlew inPy1n iPdeb

Plural baPlePlew iPy1nPy1n idePdeb

‘Paiutes’ ‘varicolored’ ‘wrinkled’

The derivations are: (279) baPlew inPy1n iPdeb

JncIns ba[Ple]w i[nPy1]n i[Pde]b

PrAdj i[nPy1]n i[Pde]b

Trscr ba-Ple-Plew i-Py1-nPy1n i-de-Pde

In each case, the onset is truncated minimally to a permissible onset using left-edge truncation. These examples make the designation “CVreduplication” for this kind of Washo plural reduplication problematic. I will nevertheless retain it because it does describe the reduplicant accurately in the large majority of cases and does contrast nicely with the other type of reduplication, “VC-reduplication,” which will be discussed shortly. V-initial roots display a supercially different surface pattern. (280) a. b. c. d.

Root ileg ipes aNkas emc’i

Plural leleg pepes kaNkas c’imc’i

‘red’ ‘black’ ‘hollow’ ‘to wake up’

Winter (1970) realized that the examples in (280) were the result of vowel syncope of initial unstressed vowels in Washo. Stress is usually

Case Study: Washo Plural Reduplication

147

penultimate. Once this is taken into account, the analysis of the examples in (280) is no different than the analysis of the examples in (279). (281) a. b. c. d.

i[le]g i[pe]s a[Nka]s e[mc’i]

Trscr i-le-leg i-pe-pes a-ka-Nkas e-c’i-mc’i

Stress il´eleg ip´epes ak´aNkas ec’´ımc’i

Syncope l´eleg p´epes k´aNkas c’´ımc’i

There is stem ablaut associated with CV-reduplication, which is evident in (282). The pretranscription forms are in the last column. (282) a. b. c. d. e.

Root t’anu aˇsun albul alPmul amk’um

Plural t’a-no-no ˇso-ˇson  a-bo-lbol  a-Pmo-lPmol  a-k’o-mk’om

‘person’ ‘red’ ‘spherical’ ‘big and round’ ‘arched’

t’a[no] a[ˇso]n a[lbo]l a[lPmo]l a[mk’o]m

Under ρCV -reduplication, a high round vowel in the duplicant is lowered if the preceding vowel is low. I will call this ρCV -ablaut. This ablaut occurs after juncture insertion, but before transcription. Some derivations follow: (283) JncIns PrAdj Ablaut Trscr Stress Syncope

t’anu t’a[nu] t’a[no] t’a-no-no t’an´ono

albul a[lbu]l a[lbu]l a[lbo]l a-bo-lbol ab´olbol b´olbol

alPmul a[lPmu]l a[lPmu]l a[lPmo]l a-Pmo-lPmol aPm´olPmol Pm´olPmol

7.4.1.1 Underlying Bare Timing Slots Roots with a long penultimate vowel appear to be exceptions to the reduplication rule as developed above. (284) a. b. c. d.

Root wa:ˇsiw memde:wi Pa:t’u 1:l1l

Plural waˇsi:siw memdewi:wi Pat’o:t’o il1:l1l

‘Washo’ ‘deer’ ‘older brother’ ‘pure white’

148

Chapter 7

Halle (personal communication) suggests that long vowels can be represented in underlying forms as a single occurrence of the vowel followed by a bare timing slot. This immediately explains the examples in (284). The derivation (285) is typical. (285) × × × × × × → × ×[× × ×] × → × × × × × × × × wa

ˇs i w

w a

ˇs i w

wa ˇs i w → ×××××××× → ×××××××× w a ˇs i

ˇs i w

w a ˇs i

ˇs i w

ρCV -ablaut applies in (284c), as expected. Finally, one last group of examples of ρCV -reduplication requires discussion. (286) a. b. c.

Root ayam ayuk iyeb

Plural ya:m yo:k ye:b

‘to hit with an instrument’ ‘parent-in-law’ ‘to copulate’

We can account for these examples if we slightly modify the proposal that bare timing slots that are postvocalic are associated with the vowel, producing a long vowel. Suppose that this rule is blocked if the timing slot is also prevocalic, as a hiatus-avoidance mechanism. Suppose also that if this rule does not provide phonemic association to a bare timing slot, then phonemic association is provided by epenthetic y. Assuming that the y-glides in (286) come from underlying bare timing slots, which induce epenthetic y because long-vowel formation is blocked, the derivation of (286a), for example, begins (287)

× × × × a

a m

JncIns

−→

Trscr

× [ × × ] × −→ × × × × × × a

a m

a

a m

If we can nd a reason why the second bare timing slot deletes, the other steps in the derivation are clear. The rst bare timing slot is realized as y intervocalically, and the unstressed initial vowel syncopates as discussed above. Note that there is a “crossing violation” in the nal representation in (287) in the sense discussed in chapter 2. There are two timing slots associated with a particular phoneme, but there is an intervening timing slot that is not associated with that phoneme. This provides a motivation for deleting the the second bare timing slot. I

Case Study: Washo Plural Reduplication

149

assume that NCC repair, as a simpler alternative to ssion, simply deletes the intrusive timing slot. All of the examples in (286) are explained in the same way. ρCV ablaut applies as expected in (286b). This completes the discussion of the mechanics of Washo ρCV -reduplication. I now turn to illustrating VC-reduplication and discussing how the choice between ρVC and ρCV is made. 7.4.2 VC-Reduplication The juncture insertion associated with ρCV -reduplication does not produce a nontrivial reduplicant if the stem is monosyllabic. We have seen already that it is not unusual for reduplication rules to break down on monosyllabic stems. Recall that the Mangarayi reduplicative plural, which also needs a pair of nuclei in order to demarcate the duplicant, does not apply to monosyllables. The Arrernte Rabbit Talk transformation similarly needs a pair of nuclei. In that case, there is allomorphy, with a different affix combining with monosyllabic roots. Monosyllabic roots in Washo are therefore of particular interest. Some examples follow: (288) a. b. c. d. e.

Root ips iˇsl im akd sesm

Plural p´eps ˇs´eˇsl m´em k´akd ses´esm

‘up from a surface’ ‘to give’ ‘out from’ ‘slowly’ ‘to vomit’

In general, polysyllabic words are stressed on the penultimate syllable (s´aksag, bal´oxat, mal´osa˜n, and so on). Note, however, that in polysyllabic (288e) there is nal stress. If all reduplicated monosyllables have nal stress, then a simple account of (288) is possible. Assume for the moment that reduplicated monosyllabic roots have nal stress. This will be justied below. The pretranscription representations of the forms in (288) are given in (289). Like CV-reduplication, VCreduplication also triggers stem ablaut, but of a different variety. A high vowel in the duplicant is lowered to a mid vowel. In the examples (289), i→e. Why should the bisyllabic output of reduplicated monosyllables be stressed on the nal syllable? It cannot be the case that stress is assigned before duplication, necessarily nal for a monosyllabic form, and simply persists. Many examples like m´oya→moy´aya, (276e) above, show

150

Chapter 7

(289) a. b. c. d. e.

Root ips iˇsl im akd sesm

JncIns [ip]s [iˇs]l [im] [ak]d s[es]m

Ablaut [ep]s [eˇs]l [em]

TrScr ep-eps eˇs-eˇsl em-em ak-akd s-es-esm

Stress ep´eps eˇs´eˇsl em´em ak´akd ses´esm

Syncope p´eps ˇs´eˇsl m´em k´akd

that reduplication can shift the location of stress. If stress persisted, we would expect m´oyaya, not moy´aya. The desired result follows, however, if foot delimiters are inserted cyclically and persist, assuming that main stress is assigned to the leftmost element of the rightmost foot. The persistence of penultimate stress in m´oya→moy´aya, follows because cyclic binary footing can insert another foot delimiter after reduplication. (∗ ∗ (∗ ∗ (∗ (∗ ∗ ∗ (290) × × × × → × × × × × × → × × × × × × m o y a

m o

y a

m o

y a

With a monosyllabic root, like (290e), foot delimiter insertion cannot shift main stress off the nal syllable. (∗ ∗ (∗ (291) × × × × → × × × × × × s e s m

s

e s m

I conclude therefore that in the context of monosyllabic stems, the affix item which realizes plural has a null exponent and triggers the rules:   ∅→[ V, ∅ → ] C This affix will be called ρVC . There are a number of polysyllabic roots that are lexically stipulated to take the ρVC plural affix. Examples are given in (292), with the pretransciption forms in the last column. The roots in (292) are of two types. The rst type has the form (C1 )VC2 C3 V(C4 ), where the two vowels are identical. The second type is of the form (C1 )aC2 C3 VC4 , where V is a high unround vowel, i or 1. This is unlikely to be an accident. Some speculation is relegated to a note.18

Case Study: Washo Plural Reduplication

(292) a. b. c. d. e. f. g. h. i. j. k.

Root maPˇsaP mayNa ˇsawlam helme hesge aPˇsam baliP aˇsiw ac’im aˇsd1m ap’1l

Plural m-aP-aPˇsaP m-ay-ayNa ˇs-aw-awlam h-el-elme h-es-esge P-aPˇsam b-al-aliP ˇs-aˇsiw c’-ac’im ˇs-aˇsd1m p’-ap’1l

151

‘brother’s child’ ‘fawn’ ‘to be a girl’ ‘three’ ‘two’ ‘to lie’ ‘to shoot’ ‘clear’ ‘green, yellow’ ‘to hide’ ‘tail’

m[aP]ˇsaP m[ay]Na ˇs[aw]lam h[el]me h[es]ge [aP]ˇsam b[al]iP [aˇs]iw [ac’]im [aˇs]d1m [ap’]1l

7.4.3 The Residue The only examples from Broselow and McCarthy 1983 and Yu 2001 that are still unexplained are given in (293). The results of VC- and CV-reduplication (if applicable) are given in each case. a. b. c.

Actual VC CV Na:m Namim *Na:ma:m elˇs1m ˇselˇs1m *lelˇs1m *ˇs1lˇs1m emlu mumlu *memlu *lumlu

‘son’ ‘to sleep’ ‘to eat; food’

d.

a:ˇs

da:ˇs

‘in, into’

e. f. g.

iPib iPiˇs iPiw

PePb *PePib PePˇs *PePeˇs PwePw *PePew

h. i.

a:t’i a:m

t’a:t’i ma:m

(293)

*ˇsa:ˇs *PiPib *PiPis *PiPiw

‘cry, weep’ (empty stem) ‘to eat’ ‘upwards, uphill’ ‘happy’

I will leave the rst group without comment. They are presented in the interest of full disclosure. (293d) may be analogous to Arrernte Rabbit Talk use of a consonantal prex (y) in case the reduplication rule for polysyllabic stems fails to apply. Yu 2001 argues that the roots in (293e–g) are underlying monosyllables and the i that appears in the singular form is epenthetic. If so, they would be subject to VCreduplication, explaining why the vowel lowers.19 The most interesting examples are (293h–i). There are two potential explanations, both of which are plausible, but neither of which has

152

Chapter 7

supporting evidence. The problem is the assumption which was made above that long vowels in Washo are represented as vowel plus bare timing slot at the point that ρCV juncture insertion applies. The analysis of (293h–i) would go through easily if the long vowel is doubly linked and if unstressed long initial vowels syncopate. The derivation of (293i) would be JncIns

Trscr

(294) × × × −→ [× × ×] −→ × × × × × × a m

a m

NCC repair

−→

×××××× a m a m

a m Stress

−→ a:m´a:m

Syncope

−→

m´a:m

It could be that initial long vowels are treated differently than medial long vowels, or that long vowels in ρVC -reduplication are treated differently than long vowels in ρCV -reduplication. While plausible, this is not an explanation. Another more interesting solution revolves around the ambiguity in the VC juncture insertion rule of the structural condition / C  ∅→] C . In chapter 4, C was taken to be the contradiction of V. However, if C is taken to be false of a bare timing slot, a:m→ma:m can be derived.20 (295)

× × × a

m

JncIns

−→

Trscr

[ × × × ] −→ × × × × × × a

m

a

m

The result is a:ma:m, with the usual nal stress pattern of reduplicated monosyllables. If unstressed initial long vowels delete just like unstressed initial short vowels, the desired result is obtained. 7.5 Tohono O’odham Plural Reduplication In Washo, there is a class of polysyllabic roots that are lexically specied to reduplicate according to the rules for monosyllabic roots. Tohono O’odham has a much more complex system of morphologically conditioned exceptions to a regular reduplication process. Developing an analysis in which this morphological conditioning is straightforward and learnable is the primary concern. Although signicantly different than the analysis of Raimy 2000a, it is is indebted both to his approach to the problem and organization of the data. The examples are from

Case Study: Tohono O’odham Plural Reduplication

153

Saxton, Saxton, and Enos 1983 and Zepeda 1983. The impressive variation in Tohono O’odham reduplication is evident in the forms in (296). (296)

toon gimai bool daak gaso

too-ton gi-gimai bo-bol daa-dk ga-gso

‘knee’ ‘braggart’ ‘a ball’ ‘nose’ ‘a fox’

pad sikol looba gook ban

paPa-pad siPi-skol lo-lba goPo-gok baa-ban

‘badly’ ‘circular’ ‘dry goods’ ‘(a) two’ ‘coyote’

This includes all the varieties of reduplication classied by Raimy, except for a group of forms with an initial consonant cluster that will be discussed below (see his book for many more examples of each type). There does not appear to be any phonological basis for the variation shown in (296), so extensive lexical specication is required. The challenge that (296) poses to a theory of reduplication is to specify the variation in an economical and plausible fashion. I begin by identifying the regular pattern: (297)

Regular roots a. b.

toon gimai

JncIns [too]n [gi]mai

Truncation [too]n

too-ton gi-gimai

The plural morpheme has a null exponent and triggers C∗ν-juncture insertion. Transcription is to the left, the default for left-edge duplicants. The second conjunct is adjusted to CV, if necessary. Various morphologically conditioned adjustment rules can apply after juncture insertion: vowel lengthening or shortening, and degemination of a long vowel with P inserted to break the resulting vowel hiatus. Exceptional truncation is also possible, which truncates the second conjunct to C rather than CV. The effect of these morphologically conditioned adjustment rules is illustrated in (298). The exceptional adjustment rules are indicated by superscripts: 1 is lengthening, 2 is shortening, 3 is degemination, and 4 is exceptional truncation. Formally, there are four classes of irregular roots: XTrunc, XLong, XShort, and XDegem. Exceptional truncation is triggered after juncture insertion if the root is class XTrunc, lengthening if the root is class XLong, and so on. The four classes are not mutually exclusive. There are some roots that are in more than one class. Examples with multiple adjustments are given in (299).

154

(298)

Chapter 7

Irregular roots JncIns a. daak [daa]k b. gaso [ga]so c. ban [ba]n d. bool [boo]l e. gook [goo]k

[baa]n1 [bo]l2 [goPo]k3

Truncation [daa]k4 [ga]so4 [baa]n [goPo]k

daa-dk ga-gso baa-ban bo-bol goPo-gok

Irregular roots subject to multiple exceptional adjustments JncIns Truncation 2 a. looba [loo]ba [lo]ba [lo]ba4 lo-lba 1 3 b. pad [pa]d [paa]d [paPa]d [paPa]d paPa-pad 1 3 4 c. sikol [si]kol [sii]kol [siPi]kol [siPi]kol siPi-skol

(299)

All the varieties in (296) are included in (297–299). Some ordering among the exceptional adjustment rules is implicit. For the sake of concreteness, assume that the ordering is ⎡ Lengthening (XLong) ⎤⎥⎥ ⎢⎢⎢ ⎥ ⎢⎢ Shortening (XShort) ⎥⎥⎥⎥ (300) ⎢⎢⎢⎢⎢ ⎥ (XDegem) ⎥⎥⎥⎦ ⎢⎣ Degemination Exceptional Truncation (XTrunc) Some combinations of the morphologically conditioned adjustments (300) are incoherent (shortening followed by degemination) or unlearnable (shortening followed by lengthening). Most of the other combinations are exemplied in (296). There is no example of a root which is both XLong and XTrunc, which would produce (the ctitious example) pal→paa-pl. This would perhaps surface as papl because of compensatory shortening, which would make the combination unlearnable since it would be learned as a simple XTrunc root. So it is not entirely clear that this is a gap in the possibilities. There is also no example of an a root which is both XDegem and XTrunc, which would produce (the ctitious) paal→paPa-pl. I have no suggestions for accounting for this gap. Fitzgerald (2000) claims that the reduplication pattern CV1 V2 . . . → CV1 CV2 . . . is obligatory if the vowel sequence V1 V2 is one of {io, 1o, oa, ua}. For convenience in the discussion, call these vowel sequences Type F vowel

Case Study: Tohono O’odham Plural Reduplication

155

sequences and call a root with an initial Type F vowel sequence a Type F root.21 The following examples are from Fitzgerald 2000, 715. (301)

CV1 V2 . . . → CV1 CV1 V2 . . . (not Type F) a. kui ku-kui ‘mesquite tree’ b. hoiki ho-hoiki ‘to move reiteratedly’ c. ñ1a ñ1-ñ1a ‘look, see’ d. ñ1id ñ1-ñ1id ‘seeing’ CV1 V2 . . . → CV1 CV2 . . . (Type F) e. hiopˇcig hi-hopˇcig ‘to be full of body lice in one place’ f. ˇc1oˇj ˇc1-coˇj ‘boy, man’ g. doa do-da ‘to be healthy’ h. ˇcuama ˇcu-ˇcama ‘roast in ashes’

Raimy 2000a,125, however, gives examples to show that not all Type F verb roots behave as the ones in (301e–h) do. Assuming that the CVV sequence in the roots in (301) is bisyllabic, the reduplication pattern (301e–h) is the result of exceptional truncation. [ˇcu]ama

XTrunc

−→

[ˇcu]ama

Trscr

−→ ˇcu-ˇcama

In terms of the analysis above, the observations of Fitzgerald and Raimy can be summarized by saying that the default for Type F roots is the class of XTrunc verbs, but there are exceptions. Fitzgerald develops an OT account of the reduplication properties of Type F roots in terms of hiatus avoidance for these sequences in the output of reduplication. In the theory developed here, avoidance of such sequences may well be the reason that certain verb roots tend to be put in the XTrunc class. But this does not mean that hiatus avoidance must be reected directly in the form of the rule that achieves it. To the extent that the hiatus avoidance account is correct, it is an account of why these particular roots tend to be put in the XTrunc class. Exceptional truncation itself is quite independent of hiatus avoidance and many roots in the XTrunc class have long initial vowels, for which the issue of hiatus avoidance is moot. There is no imperative to include a mention of hiatus avoidance in the grammar. See Frampton 2002 for some discussion of the distinction between understanding the pressures that phonotactics may exert for including a particular rule in the grammar and understanding what the rule is and how it interacts with the rule system it is embedded in.

156

Chapter 7

Finally, there is one more set of forms to consider. There are some stop-liquid initial consonant clusters in Tohono O’odham. They are associated with a characteristic reduplication pattern that is not so far accounted for. (302)

a. b. c. d. e.

‘a tramp’ ‘iron’ ‘a nail’ ‘(the metal) lead’ ‘a truck’

tlaamba plaanja klaawo ploomo tloogii

t-la-lamba p-la-lanja k-la-lwo p-lo-lmo t-lo-logii

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 is unsyllabied, although I have no other evidence for this. All of the roots in (302) must also be assigned to the XShort class of roots and a few are also in the XTrunc class. A typical derivation is (303)

JncIns

DC

ploomo −→ ploo]mo −→ p[loo]mo XShort

XTrunc

Trscr

−→ p[lo]mo −→ p[lo]mo −→ p-lo-lmo

I have no explanation for why all these verbs are in the class XShort. The verb roots in question form a small class with relatively similar form, so it is perhaps not surprising that they all have the same irregularity. If there were no morphologically conditioned adjustment rules (i.e., no irregularities), we would expect tlaamba→t-laa-lamba for example, with regular prosodic truncation of the second conjunct to CV. In closing the discussion of Tohono O’odham, it should be noted that it is very simple to code the irregularities in the system presented above. There are four exceptional subclasses of verbs: shortening, lengtheningnating, and exceptionally truncating. Membership in each of these verb classes has a simple effect, the application of a morphologically conditioned adjustment rule. The Tohono O’odham learner must code the irregularities in some way and it is difficult to see how the coding could be any simpler. A complex system of irregularities like this presents a serious challenge to OT. It is not at all clear that OT approaches to reduplication can carry out this coding in a plausible fashion. If it were the case, for example, that a constraint hierarchy was possible in which the irregularities were determined by turning off and on particular constraints, a system of comparable complexity would be

Case Study: Consonant Copy in Temiar and Levantine Arabic

157

possible. But it remains to be seen if this can be done. If each class of irregularities requires constraint reranking, the complexity of such a system argues strongly against it. Recall that the irregularity classes overlap and a root can be in as many as three different classes. This complicates an attempt to account for the irregularities by reranking. To my knowledge, no OT analysis of the irregularities of Tohono O’odham reduplication has been published, so no comparison can be made at this point. 7.6 Consonant Copy in Temiar and Levantine Arabic Temiar is a Mon-Khmer language spoken in Thailand and Malaysia which is closely related to Semai, discussed earlier in section 4.3.1. It is discussed here along with Levantine Arabic because both realize certain morphemes as inxation of a copy of an edge consonant. The similarity between the two mechanisms was noted in Broselow and McCarthy 1983. 7.6.1 Temiar Verbal Reduplication Reduplication in Temiar has attracted a great deal of attention in the literature on reduplication. See in particular Broselow and McCarthy 1983, Sloan 1988, Gafos 1998a, 1998b as well as by Raimy 2000a and the references cited there. The interest stems from the fact that Temiar reduplication patterns are resistant to a plausible treatment in prosodic terms. Verb roots in Temiar have the form (C)CV(V)C. The paradigm of most interest is given below. (304) a. b. c. d.

Root g@l kO:w slOg sma:ñ

Continuative gl-g@l kw-kO:w s-g-lOg s-ñ-ma:ñ

‘sit down’ ‘call’ ‘sleep with’ ‘ask a question’

Examples (304a–b) involve exactly the same reduplication scheme as the Semai continuative examples discussed earlier (p. 60). Recall the Semai juncture insertion rules:  (305) 1. ∅ → ] # ×  2. ∅ → ] #  ×] 3. ∅ → 

158

Chapter 7

Under these rules, for example, g@l→g ]@  l ]→[[g ]@ l ] (the last step is Default Closure). As shown in section 4.3.1, this transcribes to gl-g@l. To generate s-g-slOg, however, the juncture insertion rules must generate s [lO g ], not [[s ]lO g ], as the Semai rules would do. This would follow if the CV(V)C portion of the root is the domain of the reduplicative affix in Temiar, but not in Semai. Application of (305.1) to slOg would be blocked because, on the one hand, it must apply in the domain of the affix, but on the other, it must apply in the context # . Since (305.1) does not apply, the result is slO g ], which Default Closure transforms to s [lO g ], as desired. Recall that Default Closure closes open truncation and duplication junctures at the edge of the domain. There is justication for the assumption that the CVC portion of the root is singled out as the domain. Benjamin 1976, the main source for all Temiar studies, drew a distinction between the nal stressed closed syllable and the unstressed “prenal” material. Although there appears to be agreement that there is a qualitative difference between the prenal material and the stressed nal CV(V)C, there has been ongoing debate in the literature about the status of the prenal material. Gafos (1998b) proposed empty syllable nuclei. Sloan (1998) proposed that the extension consonants remain unsyllabied. It is consistent with either proposal to assume that the extension is prosodically inferior in some way and that the nal CV(V)C is singled out as the domain of verbal affixation. Other verbal inections appear as inxes at the left edge of the root kernel. The inxes in (306) are boldface. (Simulfactive, causative, and continuative have been abbreviated.) (306) slOg g@l srlOg trg@l

(root) (root) (causative stem) (causative)

Simul. salOg gag@l sralOg trag@l

Caus. srlOg trg@l

Cont. sglOg glg@l srglOg trlg@l

The affixes are easily specied. The specications in (307) give more evidence that inxation is closely connected with reduplication. The domain for all the affixes is the nal CV(V)C of the stem. Note that several readjustment rules are blocked if there is prenal material because they both require the conguration #× and must apply to the domain of the affix. If there is prenal material, # × cannot be found in the domain. This is illustrated for the causative affix below.

Case Study: Consonant Copy in Temiar and Levantine Arabic

(307)

Suffix Continuative

∅

Simulfactive Causative

-a ] -r ]

159

Readjustment   ∅ → ] #× ; ∅ → ] #;  ∅→ ×]  ∅ → ] #×  ∅→t # ×

The effect of domain specication is pervasive. It blocks the readjustment rule associated with the causative for the root slog, and it determines the insertion sites for junctures in Default Closure. This is illustrated in (308). (308) slog g@l

Suffix slogr] g@lr]

Readjust tg@lr]

DC s[logr] t[g@lr]

Trscr s-r-log t-r-g@l

7.6.2 Levantine Arabic Perjorative Reduplication Broselow and McCarthy (1983) called attention to the similarity between continuative reduplication of CCVC (triconsonantal) roots in Temiar and intensive/pejorative reduplication of triconsonantal stems in Levantine Arabic. They give the following examples (and many more): (309)

Stem Reduplicated marat. marmat. ‘cut unevenly’ barad barbad ‘shaved unevenly’ ˇsah. at. ˇsah. ˇsat. ‘dragged unevenly’ Descriptively, a copy of the initial consonant is postposed to the position following the second consonant. The semantics above are pejorative, but sometimes the semantics are essentially intensive. I will call the morpheme int in what follows, simply for convenience. The affix has null exponent and triggers the juncture insertion rules  ∅→ × and ∅ → ] V (rightmost). The effect of these rules is illustrated below: (310)

⊕ INT

DC

Trscr

× × × × × −→ ×× ×]× × → [ ×× ×]× × −→ × × × × × × b a r a d b a r a d b a r a d ba r ad

Transcription in (310) is taken to be to the right. The transcription operation is considerably simpler if transcription is to the right, since truncation is then carried out by noncopying. The derivation (310) should be compared with the Temiar continuative:

160

Chapter 7 ⊕ CONT

DC

Trscr

(311) × × × × −→ × × ××] → ×[× × ×] −→ × × × × × s l o g s l o g s l o g l o g s The output forms are close to being mirror images of each other. Although this analysis does produce the correct results, it has one problematic aspect. No direct reference to the “second consonant” is made in the juncture insertion rules. The formalism developed to this point does not allow it. The insertion point of the ]-juncture, immediately after the second consonant, is identied as “before the nal V). Given the consonant oriented morphology of the vowel” ( / Semitic languages, a more direct way of locating this insertion site is desirable. The simplest analysis is that reduplication actually applies to the consonantal root, before vowels are inserted into the representation. (312)

⊕ INT

DC

Trscr

brd −→ br]d → [br]d −→ brbd

In order for this analysis to go through, the domain of the affix must consist of the initial consonant pair. It stretches usual notions of foot structure, but it is plausible that in a consonant oriented language that footing can take place in the absence of syllable structure, with each consonant taken to be a footable element. The initial consonant pair is an initial binary foot under this interpretation. Assuming that this is possible and that the domain of the intensive affix is this initialfoot, then . the juncture insertion rules are simply the ×∗-rule and ∅ →  × Intensive reduplication applies to biconsonantal roots. The biconsonantal root lf , for example, has the simple verb form lafaf and the intensive/pejorative form laflaf . We will return to these questions again in discussing Chaha, but suppose that lafaf derives from l [f ]. Suppose that the perjorative reduplication rules apply directly to the form l [f ], before vowel insertion. [l [f ] results, which Default Closure transforms to [l [f ]], with a nested duplicant. Following the discussion in section 4.3, transcription produces: [l [f ]] →[l f ]f →lflf . (313)

[ × [ × ]] → [ ×  × ] × → × × × × l

f

l

f

l

f

The resulting representation is identical to the representation produced by totally reduplicating lf . The interest in this result is that it may contribute to understanding at what point in the derivation biconsonantal roots are given a full triconsonantal form. It shows that pejorative redu-

Case Study: Sanskrit Intensive and Perfective Reduplication

161

plication, at least, can operate on quite abstract consonantal structures in which reduplication is marked, but has not yet been carried out. 7.7 Sanskrit Intensive and Perfective Reduplication This section should be read as a revision of Steriade 1988, which carefully unravels the relevant Sanskrit phonology. The differences are mainly over the reduplication process itself. Most of the phonological insights, organization of the data, and highlighting of crucial examples are due to Steriade. There are ve varieties of prexal reduplication: perfect stem, aorist, desiderative, present stem, and intensive. Perfect and intensive reduplication are considered in this section. Both are quite productive and exploit the full range of mechanisms that are employed in Sanskrit reduplication. Sanskrit reduplication is unusual and interesting because duplicant adjustment makes crucial reference to the segmental features associated with nuclear timing slots. Descriptively, perfective reduplication prexes a (C)V syllable to the verb root to create what is called the “perfect stem.” Intensive reduplication prexes a Ca X syllable to the verb root. Both types of reduplication interact extensively with a widespread process of low vowel syncope that affects many roots in certain prosodic environments. The syncopated root is called the zero-grade form of the root in the Sanskrit literature, contrasting with the full-grade form. 7.7.1 Low-Vowel Syncope The underlying vowel inventory in Sanskrit is {a, i, u }. Underlying verb roots are heavy monosyllables. Except for a few roots with long high vowels, the nucleus of underlying roots contains the low vowel a. For verb roots that are subject to low-vowel syncope, and whose reduced form has a vocalic nucleus, the vowel of the perfect prex is the nuclear vowel of the reduced form. This is shown in (314). In those roots which do not syncopate, the vowel of the perfect prex is a. Explaining this relation is the major challenge in analyzing Sanskrit reduplication. Underlying u surfaces as a v aspirant in syllable initial onset position and otherwise as a round labial glide before a. Sanskrit orthography uses “v” in both cases. A “y” is used for nonnuclear occurrences of underlying i. All sonorants can be nuclear in Sanskrit and there are many syncopated roots that have a nonvocalic nucleus (n, m, r, or l ).

162

(314)

Chapter 7

Full-grade a. suap ‘sleep’ svap b. suaj ‘embrace’ svaj c. miaks. ‘be situated’ myaks. d. tiaj ‘forsake’ tyaj e. uas ‘shine’ vas f. uas ‘clothe’ vas

Zero-grade Perfect stem sup su-svap sa-svaj miks. mi-myaks. ta-tyaj us u-vas va-vas

Roots of this kind invariably have the vowel a in the perfect prex. The puzzle is to explain why, for roots that syncopate and whose syncopated root has a vocalic nucleus (314a,c,e), the vowel of the perfect prex is the same as the nuclear vowel of the reduced root, and for those roots which do not syncopate (314b,d,f), the vowel of the perfect prex is invariably a. Solving this puzzle led Steriade to conclude that the computation of the perfect prex was accomplished by rst copying the entire root, then computing its zero-grade reduction, then identifying the eventual prex as a subpart of this zero-grade reduction and deleting the superuous material. For roots whose syncopated form has a vocalic nucleus, this gives the correct results. For roots whose reduced form has a syllabic liquid, the proposal makes the wrong prediction. Steriade predicts derivations such as those in (315). For roots with a syllabic nasal in the reduced form, tan in (315), the correct surface form is obtained because syllabic nasals in Sanskrit surface as the low vowel a. For roots with a syllabic liquid in the reduced form, mard in (315), Steriade’s analysis produces the incorrect outcome, as she notes. She is forced to stipulate a rule converting such syllabic liquids to a in certain reduplicative contexts. (315)

Root Full copy Zero-grade Truncation Surface form

tan tan-tan tn­ -tan ta-tan

mard mard-mard mr­ d-mard mr­ d-mard mr­ -mard

Even more serious than this empirical problem is the fact that the analysis relies on an incorrect assumption about the accentual conditions under which low-vowel syncope applies. Steriade assumes that low-vowel syncope applies in all unstressed syllables. Studies of Sanskrit stress, however, conclude that the environment for low-vowel syncope is an unstressed syllable that is followed directly by a stressed

Case Study: Sanskrit Intensive and Perfective Reduplication

163

syllable, see Halle 1997 (p. 292) for discussion. We can see this at work in one of Halle’s examples. The noun sv´asar ‘sister’ bears inherent stress on the rst syllable. When it combines with a stressed suffix, the singular dative suffix -´e, for example, low-vowel syncope applies and sv´as.re (sv´as are) results. (Only the leftmost stress surfaces.) When it combines with an unstressed suffix, the singular accusative suffix -am, for instance, low-vowel syncope does not apply. Instead, sv´a.sar.am→ sv´a.s¯a.ram. (Compensatory lengthening accompanies the recruitment of the root nal consonant as onset of the following syllable.) The deciding factor excluding sv´as.ram (parallel to sv´as.re) is the absence of inherent stress on the suffix, not simply the absence of stress on the nal syllable of the root. Without the assumption that low-vowel syncope applies in all unstressed syllables, the derivations Steriade proposes do not go through. According to Steriade 1988 (p. 123), the derivation of the 3sg perfect form bu-budh´us, from the root baudh ‘know’, is (316)

full copy low-vowel syncope truncation

baudh-baudh-´us bud-bud-´us bu-bud-´us

As we have seen, however, there is no justication for assuming that syncope applies in the initial syllable. An account of the interaction of low-vowel syncope and reduplication in the DR framework requires an underlying structural difference that leads, on the one hand, to the syncopated root or, on the other, to the correct reduplicant. The starting point is the observation that although a certain accentual environment certainly required, some condition on the possibility of resyllabication is also in force. The root pat ‘y’, for instance, syncopates if it is an environment in which the onset and coda can be incorporated into adjacent syllables, but not otherwise. So pat has the inected perfect form pa-pt-´us, with syncopation, but does not syncopate in the form pat-´ı-ta. The same is generally true of CaC roots. For another example, the root nau ‘praise’ syncopates to nu before Cinitial stressed suffixes, but not in forms like nau-´anti→na.v´an.ti, in which u is recruited as an onset to the following syllable. The root suap syncopates to sup and has the perfect stem susvap, while the root suaj does not syncopate and has the perfect stem sasvaj. How can a lexical difference like this be related the possibilities of resyllabication? I assume that the difference is in the nuclear structure of the two roots. suap has a long nucleus (the diphthong ua) and suaj

164

Chapter 7

has a short nucleus with u an onset glide. Furthermore: (317)

Resyllabication Condition Syncope can apply only if the postsyncope structure can be resyllabied without altering nuclear structure.

Since suap has a long nucleus, syncope simply shortens the nucleus. No resyllabication is necessary. The root suaj, on the other hand, has a short nucleus and a complex onset. Syncope removes the nucleus, leaving three unsyllabied consonants. There is no context that can absorb all three consonants into the existing syllable structure. The only possibility for resyllabication is promotion of u to be a new nucleus, but this alteration of nuclear structure is not allowed by (317). What accounts for the difference in nuclear structure? One might speculate that the phoneme features of u in suaj are different from the phoneme features of the u in suaj, somehow preventing it from becoming nuclear. Steriade argues convincingly that this cannot be the case. The crucial fact is that for roots like duais, with both a pre-a vocoid and a post-a vocoid, it is invariably the post-a vocoid which becomes the nucleus of the syncopated root. It is impossible to explain this if vocoids can bear an underlying feature which prevents them from becoming nuclear. I assume that roots with a potential falling diphthong syllabify with long nucleus, unless there is lexical marking to the contrary. Other roots syllabify with a short nucleus, again unless there is lexical marking to the contrary. Crucially, the lexical marking [±long nucleus] (or simply [±LN]) is a property of the root, not of particular phonemes in the root. In these terms, suap is marked [+LN] and suaj is unmarked for [LN]. The root duais is unmarked for [LN] and therefore syllabies with a long falling nucleus, surfacing as dves. Diphthong ai surfaces as e and diphthong au surfaces as o in Sanskrit. If it were marked [−LN], it would syllabify with a short nucleus. There is no marking that causes it to syllabify with a rising nucleus.22 In terms of lexical marking for [LN], I analyze the examples in (314) as shown in (318). In what follows (318), {i, u, r, l, m, n } will be used in lexical representations whose nuclei have not yet been established. After nuclei have been formed, {i, u, r­ , l­, m ­ , n­ } will be used for the nuclei and {y, v, r, l, m, n } for nonnuclei. This departs in various ways from standard Sanskrit studies usage, which is oriented toward phonetic form. So I will write, for example, miaks and mar­ d, whereas the usual way of writing these is myaks and mard. This allows the

Case Study: Sanskrit Intensive and Perfective Reduplication

(318) a. b. c. d. e. f.

suap [+LN] suaj miaks. [+LN] tiaj uas [+LN] uas

‘sleep’ ‘embrace’ ‘be situated’ ‘forsake’ ‘shine’ ‘clothe’

Full-grade suap svaj miaks. tyaj uas vas

165

Zero-grade sup miks. us

nuclear structure of tyaj, for instance, which has a short nucleus, to be distinguished from that of miaks, which has a long nucleus. Several additional examples follow: (319) Full-grade Zero-grade a. mard ‘crush’ mar­ d mr­ d nard b. nard [-LN] ‘bellow’ c. skand ‘leap’ skan­ d skn­ d d. siand ‘move on’ syan­ d syn­ d e. mand [-LN] ‘exhilarate’ mand With an analysis of low-vowel syncope in hand, we are now in a position to analyze perfective and intensive reduplication. 7.7.2 Perfect Reduplication Perfect reduplication realizes a perfect tense/aspect morpheme, applies to the root, and produces what is called the perfect stem. An agreement/voice suffix is concatenated with the perfect stem to build a surface verb form. Roots in Sanskrit are monosyllabic. Some examples are given in (320). The reduplicant (C)V-syllable is easy to describe. Its nucleus is always a nuclear vowel of the root, a high vowel if there is one. Its onset consonant is the leftmost obstruent in the onset of the root, if there is one, note (320b,i) in particular, otherwise the leftmost consonant in the onset of the root, if there is one. The leftmost obstruent will usually be the leftmost consonant because of onset sonority sequencing, but s-obstruent and C-obstruent sequences are possible. (320)

High vowel in the nucleus of the root Root Perfect stem a. baudh bu-baudh ‘know, wake’ b. stau tu-stau ‘praise’ c. smai si-smai ‘smile’ d. suap su-suap ‘sleep’

166

Chapter 7

e. miaks mi-miaks ‘glitter’ f. auc u-auc ‘please’ No high vowel in the nucleus of the root g. suaj sa-suaj ‘embrace’ h. kran­ d ka-krand ‘cry out’ i. skan­ d ka-skand ‘leap’ j. mar­ d ma-mard ‘rub, crush’ k. mnaa ma-mnaa ‘note’ Recall the discussion of Semai in section 4.3.1 (p. 60). Compare: (321)

Sanskrit: Semai:

smai bP@l

si-smai bl-bP@l

Both Sanskrit and Semai use a complex copying process in reduplication which is able to copy elements from the edges without copying the medial elements. I assume that the copying mechanisms are the same in Sanskrit and Semai, so that Sanskrit produces si-smai by generating the duplicant [[s ]ma i ], which is transcribed to the left. If there is an embedded duplicant, it is often easier to think of the transcription as carried out in two stages. [[s]mai]



→ s[smai]

→ sismai

Recall that embedded duplicants were discussed in section 4.3 in some detail and that discussion should be reviewed by the reader if the process of transcription above is not clear.  I assume that juncture insertion is ∅ → ] ν , right delimiter insertion at the right edge of the nucleus, and turn to working out the duplicant adjustment rule. This is a process of simultaneously deducing the adjustment operations and the goal schema that motivates and organizes the application of those operations. The starting point is deducing the underlying juncture structure of the output forms, which I will assume follows the Semai model when it is necessary to copy elements which are separated by elements that are not copied. I rst consider roots that have an aX nucleus, with X a high vowel, and work backward to see how the juncture structure is built. We need to identify both the adjustment operations that are employed and the goals that motivate them.

Case Study: Sanskrit Intensive and Perfective Reduplication

(322) a. b.

si-smai tu-stau

(a) [[s]mai] [s[t]au]

(b) [smai] [stau]

167

(c) [smai] [stau]

(d) smai stau

The (b) → (a) steps use × → [×], motivated by Ons (Onset) for smai and by ObsOns (Obstruent Onset) for stau. The more special goal ObsOn is more highly ranked than the general goal Ons, so that it can have an effect. Rules apply leftmost by default, so × → [×] will pick out the leftmost onset if driven by Ons, and the leftmost obstruent onset if driven by ObsOn. The (c) → (b) steps use ∅ → , presumably driven by a High Vowel Nucleus goal, which will be abbreviated HighV. This is what gives perfect reduplication its particular character and its connection with low-vowel syncope. The initial step, of course, is simple juncture insertion and default closure. Other examples indicate that ]-Left is also used for adjustment. (323) a. b.

su-suap ma-mar­ d

(a) [su]ap [ma]rd

(b) [sua]p [mar­ ]d

(c) suap mar­ d

[sua]p→[su]ap is driven by HighV, but [mar­ ]d→[ma]rd indicates that if a high vowel nucleus is not possible, a vocalic nucleus is preferred to a nonvocalic nucleus. The nucleus of the reduplicant is always short, so we take the relevant goal to be ShortV (Short Vocalic Nucleus). The more specic HighV is ranked more highly than the general ShortV. Summarizing the discussion to this point, we have the following rst approximation to the duplicant adjustment rule: ⎡ HighV ⎤ ⎥ ⎧ ]-Left ⎫ ⎢⎢⎢ ⎪ ⎢⎢⎢ ShortV ⎥⎥⎥⎥⎥ ⎪ ⎪ ⎪ ⎬ ⎥⎥⎥ :: ⎨ ∅ →  (324) First Conjunct ; ⎢⎢⎢⎢ ⎪ ⎪ ⎪ ⎪ ⎥ ⎩ ⎭ ObOns ⎥⎦ ⎢⎣ × → [×] Ons Now consider the derivation of su-svaj along the same lines as the derivations above. (325)

svaj → [sva]j → [sva]j → [[s]va]j

Trscr

→

sa-svaj

What motivates [sva ] j→[sv a ] j ? The way the adjustment appears to work is that when confronted with more than one prenuclear element, they are rst all truncated, then an onset is reduplicated. I take the driving force in truncating the onset cluster to be *Cmplx (*Complex Onset).

168

Chapter 7

The operation ∅ →  must be restricted so that it cannot truncate the onset cluster only partially, otherwise the derivation would produce [sva ] j→[s va ] j→va-svaj. There are various possibilities, but I will suppose that ∅ →  is constrained by *C, so that -insertion into the onset cluster is blocked. (This particular formulation is chosen so that the rule works correctly for intensive reduplication, as we will see later in this section.) The possibility of [sva ] j→[su ]aj (driven by HighV) must also be excluded. Recall that the vocoid u can be nuclear or not, depending upon the prosodic demands made on it. Recall also that in the discussion of Yaqui, the requirement that stem onsets reduplicate as onsets played an important role in determining the reduplicant. This was called Onset Permanence (abbreviated to OnsPerm) in Yaqui. Precisely the same constraint appears to be at work in Sanskrit. Stem onsets that are reduplicated must be onsets in the reduplicant.23 This prevents onset v from being promoted to nuclear u in [sva ] j→[su ]aj. It is useful to separate the goal schema for the nucleus and onset: ⎤ ⎡   ⎢⎢⎢ *Cmplx ⎥⎥⎥ HighV Perfect Nucleus = Perfect Onset = ⎢⎢⎢⎣ ObOns ⎥⎥⎥⎦ ShortV Ons Then we can write the duplicant adjustment rule (326), which generates all the forms in (320). ]-Left ⎫   ⎧ ⎪ ⎪ ⎪ ⎪ Perfect Nucleus ⎨ ⎬ ∅ →  ; *C (326) 1st Conjunct ; :: ⎪ ; OnsPerm ⎪ ⎪ ⎪ ⎩ ⎭ Perfect Onset × → [×] Some illustrative derivations are given in (327). The adjustment operations are not ordered, so there are two different derivations of the perfect stem of roots with a long vowel (mnaa above, for example).24 Both derivations produce the same perfect stem. (327)

JncIns [smai] HighV [smai] ShortV *Cmplx ObsOn *NoOns [[s]mai] Trscr si-smai

[skan­ ]d

[mnaa]

[ska]nd [mna]a [ska]nd [mna]a [s[k]a]nd [[m]na]a ka-skand ma-mnaa

[mnaa] [mnaa]

[[m]naa] ma-mnaa

Case Study: Sanskrit Intensive and Perfective Reduplication

169

Some readers might object that (327) is needlessly complex, given that Sanskrit chooses the perfective reduplicant is such a simple fashion: best onset and best vowel. While this may be straightforward from the standpoint of the descriptive linguist, the process is not at all simple from the standpoint of copying theory. It is no easier for Sanskrit to generate si-smai than it is for Semai to generate bl-bP@l, and Semai reduplication is usually considered to be fairly exotic. Reduplication is not free to pick and choose what to copy. Copying is organized around copying strings of consecutive elements, with particular kinds of truncation possible. smai→si-smai must be analyzed in these terms, and in these terms it is complex. 7.7.2.1 Sanskrit Specialization of NCC Repair The considerations above have analyzed perfect reduplication up to the point that Transcription has applied. There are some NCC repair effects as well. When copies of root consonants appear in the reduplicant, velar consonants are palatalized and aspirated consonants are deaspirated. Steriade (1988, 106) argues, convincingly to my mind, that this should be seen as dissimilation. How and when in the derivation is it carried out? The simplest assumption is that it is carried out at the point that ssion creates a separate reduplicant phoneme in the course of NCC repair. The repair of the form produced by transcription is illustrated in (328) for the roots skand ‘leap’, dhar ‘hold’, and kh¯ad ‘chew’. (328)

NCC repair, with dissimilation a. × × × × × × × → × × × × × × × → × × × × × × × s k a nd b.

c

s k a n d

c a s k a n d

× × × × × → × × × × × → × × × × × dh a r

d

dh a r

d a dh a r

c. × × × × × × → × × × × × × → × × × × × × kh a

d

c

kh a

d

c a kh a

d

In (328c) there is both palatalization and deaspiration. 7.7.3 Intensive Reduplication In case the nucleus of the root is a long falling diphthong or long a, intensive reduplication is straightforward:

170

(329)

Chapter 7

a. b. c. d. e.

baudh stau duais mar­ mnaa

bau-baudh tau-stauh dai-duais mar-mar maa-mnaa

This data is easily generated by the C∗ν-rule and adjustment of the onset exactly as in prefect reduplication. If roots with other nuclear shapes are considered, it is clear that there is adjustment of the rst conjunct to a CaX shape. (330)

a. b. c. d.

pat suap suaj kriid

paa-pat saa-suap saa-suaj kai-kriid

Example (330a) in particular suggests that FCVL (First Conjunct Vowel Lengthening) is an adjustment operation. ν (331)

[ × ×] × p a t

FCVL

−→

ν [ × ××]× → paa-pat p a t

Recall that FCVL was used for duplicant adjustment in Mokilese (section 4.2, p. 59), and Lardil la→la:-la (section 6.4.1, p. 106). Example (330d) suggests that truncated a-epenthesis (called a Ep in the discussion of Ndebele, section 7.1, p. 120) must also be an adjustment operation, since a appears in the reduplicant, but the remnant is left unchanged. Truncates that are not at the left edge undergo copy/delete under left transcription, so a  that is not at the left edge contributes to the reduplicant, but not the remnant. The use of these epenthesis operations produces derivations like the ones in (332), on the next page. Note at the places marked † that the nucleus initial high vocoid is demoted to an onset by the epenthesis operation. This is in the rst-conjunct syllable structure, not the stem syllable structure. The two exist in parallel, as explained in section 6.1. The reversion to a nuclear vocoid at the places marked ‡ is therefore not a violation of Onset Permanence, but simply a switch to showing the stem structure. The intensive duplicant adjustment rule is a simple variation of the perfect duplicant adjustment rule. The operations FCVL and a -Ep are added to the repertoire of adjustment operations, and the PerfNuc goal

Case Study: Sanskrit Intensive and Perfective Reduplication

(332)

pat suap JncIns [pa]t [sua]p a X-Nuc [paa]t [svaa]p† *Cmplx [svaa]p ObOns Ons [[s]vaa]p Trscr paa-pat saa-suap‡

svaj [sva]j [svaa]j [svaa]j

171

kriid [krii]d [kryai]d† [kryai]d [[k]ryai]d

[[s]vaa]j saa-svaj kai-kriid‡

schema is replaced by the a X-Nuc goal (true of nuclei with the form a X). Compare (333) with (326). ⎧ ⎫ FCVL ⎪ ⎪ ⎪ ⎪ ⎪   ⎪ ⎪ ⎪ ⎪ a X-Nuc ⎨ a -Ep ⎪ ⎬ (333) First Conjunct ; :: ⎪ ⎪ ⎪ ⎪ ⎪ ∅ →  ; *C ⎪ Perfect Onset ⎪ ⎪ ⎪ ⎪ ⎩ ⎭ × → [×] Onset Permanence plays no role, so it is omitted. Interestingly, one can think of a “bag” of operations (the one given in the rule above, with ]-Left added) which Sanskrit duplicant adjustment uses: the same bag (“bag of tricks,” if you will) for both perfect and intensive reduplication. It turns out, because of their particular needs, not as a stipulation, that perfect adjustment never calls on FCVL and a -Ep, and intensive adjustment never calls on ]-Left. The same constraint applies in both cases. It simply turns out that Onset Permanence plays no role in intensive reduplication. Since both perfect and intensive  reduplication are initiated by the same juncture insertion rule, ∅ → ] ν , the differences between the reduplication processes is therefore narrowly conned to the different rst conjunct nuclear goals that they adjust the duplicant to. 7.7.3.1 an, am, ar, and al Roots There are variations on the pattern (329d), which is only one of three patterns for this type of root. The intensied root gam ‘come’, for example, appears as gam-gam , gaa-gam , and gami-gam (abstracting away from secondary segmental changes).25 The processes at work appear to be ν ν ν (334)

a. [× × ×] → [× ×××] → [× ××]× → gaa.gam g a m g a m g am

172

Chapter 7

ν b.

ν

ν

[× × ×] → [× × ××] → gami.gam g am g am i

Sanskrit scholars discuss various “weakening” and “strengthening” processes. It is plausible from what we know about sonority that aX nuclei, with X nonvocalic, are less “strong” than other long nuclei. The operations above therefore produce a “stronger” reduplicant. Plausibly, given a better understanding of Sanskrit strengthening, it might be possible to integrate the operations above into the theory of prosodic adjustment, with some secondary lexical variation. 7.7.3.2 Low-Vowel Syncope and Intensive Reduplication There are no examples in the literary record, but the Sanskrit grammarians claimed that the intensive prex of those roots which we have identied as having a long rising nucleus (ia, ua, ra, and so on) depended on whether or not the root appeared in the full grade or zero grade. The root svap/sup, for example, was said to have the present active intensive forms s¯a´svapmi (1sg) and sausupm¯a´s (1pl). The zero-grade root appears before the stressed suffix. This is a classic timing problem. On the one hand, it strains conceptions of syntactic structure to suppose that the intensive morpheme combines with the stem after tense and subject agreement morphemes since the semantics of the intensive in Sanskrit is typical. Whitney (1889, 363), says: “The intensive conjugation signies the repetition or the intensication of the action expressed by the primary conjugation of the root.” On the other, it cannot be the case that intensive reduplication produces suap→saa-suap before tense and agreement suffixes are concatenated, because this would lead to saa-supm´as after syncope, not sau-supm´as. The resolution of the apparent contradiction must be that junctures are inserted early, when the syntactically low intensive morpheme is realized, suap→[sua ]p, but adjustment of the duplicant to the CX shape does not occur until later, after tense/agreement suffixation. One possibility is that the intensive affix is noncyclic. The morphological juncture insertion rule applies, but the phonological cycle is not triggered. Duplicant adjustment and transcription are part of the phonological cycle, so they do not apply. Two illustrative derivations are given in (335).

Case Study: Cyclicity and Multiple Reduplication in Lushootseed

(335)

Root JncIns Suffixation Stress, Syncope Adjustment Transcription Surface

suap [sua]p [sua]pmas [su]pm´as [sau]pm´as sau-supm´as sosupm´as

173

gr­ abh [gr­ a]bh [gr­ a]bhmas [gr­ ]bhm´as [gar­ ]bhm´as gar­-gr­ bhm´as jargr­ bhm´as

7.8 Cyclicity and Multiple Reduplication in Lushootseed This section is based on Urbanczyk 1996, 2000. The examples and basic generalizations are all hers. The theoretical framework, therefore the analysis, is different. Lushootseed has three kinds of verbal reduplication. The particular interest of Lushootseed for the general theory is that there are many examples of verb forms in which two different kinds of reduplication have applied. 7.8.1 Diminutive Reduplication In surface form, diminutive inection is a CV-prex whose C is the initial C of the stem and whose vowel is either the initial V of the stem or, in certain contexts, the vowel i. In several examples below, irregular P coda epenthesis in a stressed open syllable results in a CVP prex at the surface. I will assume that this is post-reduplication phonology. The examples below are from Urbanczyk 1996. (336) CV-inital stems (V = @) a. ˇc´al@s ‘hand’ w ‘bad’ b. s-duk C@-initial stems c. t@l´aw-il ‘run’ d. xˇ@c-bid ‘afraid’ CCV-inital stems ‘rock’ e. ˇc’ń∼ ’´aP f. c’q´ays@b ‘ower’ CV:-initial stems g. s-du:kw ‘knife’ h. bu:s ‘four’

ˇc´a-ˇcal@s s-d´ uP-dukw

‘little hand’ ‘riffraff’

t´ı-t@law’-il x ˇ´ıP-ˇx@c-bid

‘jog’ ‘a little afraid of it’

ˇc´ıP-ˇc’ń∼ ’aP ‘little rock’ s-c’´ı-c’q´ays@b ‘ower’ s-d´ı-du:kw b´ıP-bu:s

‘small knife’ ‘four little items’

174

Chapter 7

Particularly straightforward examples have been chosen above. The overall corpus is not as clearcut as these examples might indicate because there are a signicant number of irregularities and extensive vowel syncope. See Urbanczyk for a careful discussion of all of these issues. There are only a few examples in the data of diminutive reduplication of the last two varieties, but many examples of diminutive reduplication of C@-initial stems. Urbanczyk gives arguments that initial consonant clusters are broken up by what she terms a “voiceless schwa.” Something along those lines is forced by the Optimality Theory theoretical framework she adopts since that theory does not admit intermediate levels. But her arguments translate into arguments that at some intermediate level, the apparent CC cluster is actually C@C and the vowel later syncopates. I assume that the CCV-initial stems in (336) are C@CV-initial at the point of juncture insertion. The examples categorized as CCV-initial above therefore fall together with those classied as C@-initial. Pretranscription forms can be easily identied, assuming transcription is to the left. (337)

Exceptional cases C@ t@law → ti-t@law CV: bu:s → bi-bu:s Otherwise dukw → du-dukw

Pretranscription form [ti]@law [bi]u:s [du]kw

The pretranscription forms suggest that juncture insertion is initial consonant reduplication, since the initial consonant is the element common to the duplicant in all cases. Additionally, there must be prosodic adjustment with a CV target prosodic shape and both ]-Right and FCVE (First Conjunct Vowel Epenthesis) prosodic adjustment repair rules, with epenthetic vowel i. (338)

a. b. c.

[t]@law → [ti]@law [b]u:s → [bi]u:s [d]ukw → [du]kw

FCVE FCVE ]-Right

An explanation is needed for the choice of FCVE in (338a) and (338b). Stress in diminutive forms is initial, on the reduplicant. Stressed schwa, while it occasionally occurs in Lushootseed, is not common. The prosodic adjustment rule for diminutive reduplication is designed to avoid stressed schwa by employing FCVE rather than ]-Right when

Case Study: Cyclicity and Multiple Reduplication in Lushootseed

175

the stem-initial vowel is schwa. Technically, a *C@ constraint is imposed on the duplicant in prosodic adjustment. The idea that a rule is “designed” in a certain way merits some discussion because there is widespread confusion about this idea. If one views stressed schwa as something that the Lushootseed phonological system tries to avoid, we can ask at what level the solution is found. It could be that in those instances in which stressed schwa occurs, some corrective action is taken. Alternatively, it could be that operations which produce a stressed schwa in an environment in which they are liable to be stressed are disfavored (in the space of possible operations) and therefore modied so that they avoid producing schwas in those environments.26 The statement that the prosodic adjustment rule is designed to avoid stressed schwa should be taken to mean that Lushootseed solves the stressed schwa problem in reduplication by choosing its reduplication rules in a certain 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 reduplication (at the level of design) is responsible. Distributive reduplication, as we will see in the next section, produces du:k→du-du:k (for example). Since there are only two examples of distributive reduplication in the data with a long vowel, this matter will be left unresolved. Note the similarity to Mokilese reduplication. Juncture shift is used where possible and First Conjunct Vowel Epenthesis is used where juncture shift does not obtain the desired form. The same is true in Mokilese, except that only a timing slot is epenthesized, yielding a long vowel in the rst conjunct. 7.8.2 Distributive Reduplication Generally, distributive inection is accomplished by prexing a copy of stem initial CVC. (339)

‘y’ a. saqw’ b. g@lk’ ‘entangle’ c. p´ast@d ‘Caucasian’

saqw’-saqw’ ‘y here and there’ P@s-g´el-g@lk’ ‘all tangled up’ pas-p´ast@d ‘many white folks’

Note the modication of @ under stress in (339b). There are two environments in which the distributive prex is not CVC. If the stem is CV:-initial, the prex is CV. This suggests that  and prosodic adjustment is driven juncture insertion is ∅ → ] V by C-nality, with repair limited to ]-Right. In the case of CV:-initial

176

Chapter 7

stems, C-nality via ]-Right is impossible, so there is no adjustment. Urbanczyk (1996) makes the important observation that in the special case that the stem is C1 VC2 . . ., with C1 = C2 , then the surface reduplicant is C1 V, not C1 VC2 . (340) a. t’it’-@b ‘bathe’ b. w´ıw’su ‘little c. l@lw´aPs@d ‘sleeping platform’

t’i-t’it’-@b ‘bathe for a while’ w´ı-w´ıw’su ‘little (plural)’ l@-l@lw´aPs@d ‘sleeping platforms’

Urbanczyk also observes that geminates and sequences of identical phonemes are quite rare in Lushootseed and attributes the departure from the CVC shape requirement to avoidance of this conguration. The simplest way to incorporate this idea into the present analysis is to suppose that elimination of the sequence of identical phonemes is carried out as part of the NCC repair process. For (340c), for example: Trscr

(341)

[× × ×] . . . −→ × × × × × × . . . → × × × × × × . . . l @ l

l l @ l → × × × × × × ... → × × l @

l @ l

l @

l @ l × × × ... l @ l

7.8.3 “Out-of-Control” (OC) Reduplication First, some examples: (342)

P´ıb@ˇs P´ıb-ib@ˇs saqw’ saqw’-aqw’ dzz´@kw’ dz´@kw’-@kw’

‘walk, travel or journey over land by any means’ ‘pace back and forth, walk without achieving a destination’ ‘y’ ‘y around, wheeling in the sky’ ‘fall, topple over’ ‘totter, teeter back and forth, stagger’

The initial VC of the stem is copied. Transcription to the right avoids internal copying in the many examples in which the stem is CVC, so I assume that this is the direction of transcription. 7.8.4 Interactions Given the basics of diminutive, distributive, and OC reduplication, we are in a position to examine their interactions. It is anticlimac-

Case Study: Cyclicity and Multiple Reduplication in Lushootseed

177

tic. Once Urbanczyk’s insights into the phonology are incorporated into the analysis, the interaction of diminutive and distributive reduplication is exactly what is predicted. Both orders of application are possible. (343)

a. b. c. d. e.

b@d´aP b´ı-b@daP b@d-b@d´aP b´ı-b@d-b@daP b´ı-bi-b@daP

child,offspring young child children young children litter (of animals)

dim dist dim after dist dist after dim

Derivations are given in (344). (344)

b@daP [b]@daP ⊕dim PrAdj† [bi]@daP Trscr bi-b@daP ⊕dist [bi]b@daP PrAdj [bib]@daP Trscr bib-bib@daP NCCR‡ bibib@daP

b@daP ⊕dist [b@]daP PrAdj [b@d]aP Trscr b@d-b@daP ⊕dim [b]@db@daP † PrAdj [bi]@db@daP Trscr bi-b@db@daP

At the steps marked † above, *C@ blocks ]-Right, which forces FCVE. At the step marked ‡ above, exceptional NCC repair is used to avoid bb. There are a few examples in which OC reduplication combines with distributive or diminutive reduplication. (345)

b´ali b`alal’bali ˇc´al(a) ˇcaPˇc@l´al’, ˇciˇc@lal’

‘forget’ ‘suddenly [. . .] forgetting’ ‘chase, pursue, catch’ ‘almost caught’

Two derivations are given below: (346)

1. 2. 3. 4. 5.

⊕dist PrAdj Trscr ⊕oc Trscr

[ba]li [bal]i bal-bali b[al]bali bal-al-bali

⊕oc Trscr ⊕dim PdAdj Trscr

ˇc[al] ˇcal-al [ˇc]alal [ˇca]lal ˇca-ˇcalal

See Urbanczyk for discussion of various processes that yield surface ˇcaPˇc@l´al’ or ˇciˇc@lal’ from ˇcaˇcalal.

178

Chapter 7

7.9 Chaha Reduplication Chaha, one of the Semitic languages of Ethiopia, has been a rich source of phenomena which provide insights into autosegmental phonology. 7.9.1 Overapplication of Impersonal Labialization and Palatalization Chaha has impersonal verb forms alongside verb forms which are inected for person, number, and gender. Impersonal inection is realized by labialization of the rightmost labial or velar consonant in the root and palatalization of the rightmost consonant in the root, provided that it is a coronal obstruent. The following examples of the impersonal inection of second person masculine forms were taken from Hudson, 1995, 789, and Banksira, 2000, 207. (347)

2sg masc a. f1r@x b. k1ft c. g1r@z d. n1t’1r

Imper. f1r@xw (*f w1r@xw) ‘Tolerate!’ k1f wty (*kw1f wty) ‘Open!’ gw1r@Z ‘Age!’ ‘Separate (from the teats)!’ n1t’1r (*n1ty’1r)

Rounding is indicated by superscript w and palatalization by superscript y. Examples (347a–c) show that nal, medial, and initial consonants can be rounded. Examples (347a–b) show that rounding applies only to the rightmost labial or velar consonant. Example (347b) shows that both palatalization and labialization can apply. Example (347d) shows that palatalization is restricted to the rightmost (i.e., nal) consonant. In certain cases, labialization or palatalization appears to apply to a consonant that is not the rightmost potential target. (348)

Personal Impersonal ‘plant in the ground’ a. s@k@k s@kw@kw ‘chip the rim’ b. g@m@m g@mw@mw w y y ‘be wide’ c. b@t@t b @t @t d. s@k@k s@kw@kw ‘place a peg’ Insightfully, McCarthy (1983, 1986) realized that overapplication in (348) is related to the C1 C2 C2 consonant structure and proposed a theory in which the identical consonants were two occurrences of the same phoneme, a “long-distance geminate,” which is what such nonadjacent

Case Study: Chaha Reduplication

179

occurrences of the same phoneme came to be called. This structure was achieved by supposing that adjacent identical consonants were excluded from consonantal roots and that roots with fewer consonants than demanded by a template lled it by associating from left to right and multiply associating the nal consonant to ll out the template. If the template is CVCVC, then we expect: @ @ (349)

gm → × × × × × g

m

As discussed in chapter 2 (p. 17), McCarthy placed the vocalism on a separate tier in order to avoid an NCC violation. Combination with the affix which realizes the impersonal morpheme imp takes place while multiple association still persists, so that mutation occurs while the nal consonant is still spread over multiple timing slots. The result is apparent “overapplication” of labial mutation. @ @ @ @ (350)

× × × × × g

⊕ IMP

−→

m

× × × × × → g@mw@mw g

mw

Since only one of the timing slots linked to m is in the proper environment for labialization, it must be assumed (quite plausibly) that the labialization rule is not subject to geminate inalterability. There is another class of biconsonantal roots, the so-called 1212 roots, which exhibit similar overapplication. The following examples are from Hudson 1995, 794. (351) a. b. c.

Personal b1t@b@t s1b@s@b n1k’@n@k’

Impersonal bw1ty@bw@ty s1bw@s@bw n1kw@n@kw

‘dissolve’ ‘gather’ ‘shake’

Imitating the analysis of 122 roots leads to a derivation like: 1 (352)

@

@

sb → × × × × × × × s

b

1 ⊕ IMP

−→

@

@

× × × × × × × → s1bw@s@bw s bw

180

Chapter 7

In (352), there is no escape from crossing violations by putting the offending elements on a separate tier. Since the NCC was regarded as sacrosanct, McCarthy considered a representation like (352) to be impossible and proposed that 1212 roots were reduplicative, so that the overapplication in (351) was a reduplicative overapplication effect, rather than a manifestation of the presence of long-distance geminates (LDGs). It was untenable to claim that 122 overapplication was due to the presence of an LDG, but that 1212 overapplication was due to reduplication. There was resistance, however, to a uniform account, because 122 reduplication does not t easily into the mold of Prosodic Morphology, which was held to govern reduplication. Eventually, however, opinion appears to have shifted to favor a uniform account of 122 and 1212 roots, with overapplication attributed to reduplicative overapplication, with a correspondence relation holding between the source and target of reduplication (see Gafos 1998a for an extensive discussion). The long-distance geminates in (350) and (352) were abandoned. I think this was a mistake. From the standpoint of DR, the debate pitting LDGs against reduplication is a false debate rooted in the Correspondence Theory of reduplication which OT has adopted. Representations with LDGs like (352) and (350) are exactly what DR predicts as a consequence of reduplication. 7.9.2 Inherently Reduplicated Roots There has been ongoing discussion among experts in Semitic morphology 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) a template. Hudson (1995) proposed that reduplication was coded in the root as, for example: g [m]α α

[s b]α α

or

The notation for coding reduplication into the root proposed by Hudson is remarkably similar to the idea of inserted t-junctures that DR proposes as the trigger for reduplication. Exploiting Default Closure, we can write the roots in the DR formalism as: g[m

or

[sb

If we suppose that roots consist of consonants, with an optional embedded [-juncture, we have a uniform account of the full range of reduplicative roots which are found in various Semitic languages.

Case Study: Chaha Reduplication

181

McCarthy’s insight that identical adjacent consonants are barred from the root is maintained. (353)

a. C1 [C2 → C1 C2 C2

b. [C1 C2 → C1 C2 C1 C2

c. C1 C2 [C3 → C1 C2 C3 C3

d. C1 [C2 C3 → C1 C2 C3 C2 C3

I suppose that the vocalism is epenthetic, inserted in the process of syllabication, subject to morphophonological conditions on the choice of vowel. By morphophonological, I mean that both the morphological and phonological environments can inuence the choice of vowel. These assumptions lead to derivations roughly like the following: (354)

[sb (lexicon)

DC

Trscr

→ [× × → [× ×] −→ × × × × s b NCCR

−→

s b ×××× s bw s bw

s b Syllabication

−→

⊕ IMP

−→

×××× s bw

σ σ σ ××××××× s 1 bw @ s @ bw

There is some uncertainty in the order of the various operations. NCCR (NCC Repair) must follow combination with imp, but syllabication could be much earlier. It is difficult to nd evidence in Chaha, but there is evidence from other Semitic languages that syllabication is late. So-called Classical Arabic Metathesis has a relatively straightforward explanation under the assumption of late syllabication. 7.9.3 Chaha x-Dissimilation There is a dissimilation process in Chaha whose interaction with reduplicative roots gives further evidence of the presence of long-distance geminates. It is the subject of a study by Kenstowicz and Banksira (1999), K&B in what follows, who show that there instances in which it overapplies in reduplicative roots. Banksira (2000) gives extensive evidence that k does not occur in underlying representations and that surface k is the result of either a devoiced underlying g or the dissimilation rule:  . . . [−sonorant, +continuant] (355) x → k The [−sonorant, +continuant] phonemes are f, s, z, x, and Q. For the purpose of the present discussion, we will call such a phoneme a “triggering phoneme.” The precise conditions under which (355) applies

182

Chapter 7

will turn out to be important in understanding the interaction of xdissimilation and frequentative reduplication, which will be considered later in this section. For the present, I accept B&K’s conclusion that it is not restricted by adjacency (i.e., “. . . ” in (355) can be nonnull), at least if “x . . . trigger” is root internal. They demonstrate this with various examples: /xtf / → y@-kt1f ‘hash’ (x→k/ . . . f ) and /xBs/ → y-a-kBaBs ‘make dirty’ (x→k/ . . . s), for example. If the target, trigger and the intervening material are not all in the root, the conditions under which x-dissimilation applies are not so clear. That question will be taken up later, but for the moment, the root-internal case is sufficient. Rule (355) is not subject to geminate inalterability. x→k requires only that one timing slot associated with x be followed by a timing slot associated with a triggering phoneme. Consequently, geminate x will always dissimilate to geminate k. The rst timing slot of × × x is followed by a timing slot linked to x, a [−sonorant, +continuant] phoneme. Since adjacency is not required in (355), root internal LDG 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 overapplies in very much the same way that impersonal labialization/palatalization overapplies. A few representative illustrations are given below. See K&B for further examples, most of which are less transparent than the ones given here because of various interacting morphophonological processes. (356)

a. b.

/sx/ /xt/

s1k1k k@tk1t

‘drive a peg’ ‘crush’

(122 root) (1212 root)

Naively, one might expect /sx/→s1k1x, with the medial x dissimilating to k because there is a [−sonorant, +continuant] phoneme to its right, the nal x. But the fact is that the nal x also changes to k in (356a), even though there is no [−sonorant, +continuant] phoneme to its right. The analysis of 122 and 1212 roots given above makes it clear why there is overapplication. (357)

Trscr

a. ×[ × ] −→ × × × s x

s x

x -Dissim

−→

× × × → s1k1k s k

Case Study: Chaha Reduplication

b.

Trscr

183

[ × × ] −→ × × × × x t

x t

x -Dissim

−→

× × × × → k@tk1t k t

7.9.4 Frequentative Reduplication in Tigre and Chaha Frequentative reduplication is found in all the Ethiopian Semitic languages. Since various process, which will be discussed shortly, make the underlying reduplication less transparent in Chaha than it is in some of its sister languages, we begin the discussion with frequentative reduplication in Tigre, an Ethiopian Semitic language that is closely related to Chaha. The facts are from Rose 2001. 7.9.4.1

Tigre

(358) Root type A B C 122 Quadriliteral 1212 1233

Regular d@gm-a: w@ll@b-a: ba:r@k-a: m@zz-a: d@ng@s’-a: n@kn@k-a: d@nz@z-a:

Frequentative d@ga:g@m-a: w@la:l@b-a: b@ra:r@k-a: m@za:z@z-a: d@n@ga:g@s’-a: n@k@na:n@k-a: d@n@za:z@z-a:

‘tell’ ‘look both ways’ ‘bless’ ‘give responsibility’ ‘be scared’ ‘shake in hysterics’ ‘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 retained when reduplicated.” This supports the proposal that syllabication and vowel insertion are late, after the affix that generates frequentative reduplication combines with the stem. The morphology of the stem modication in (358) is clear. The penultimate consonant is geminated and a: is inxed into the geminate, not unlike simulfactive reduplication of CVC roots in Temiar which was discussed in section 7.6.1. In Temiar, g@l→gag@l, for example. I assume that reduplication is carried out using the same formal mechanism as in Temiar, with the nal consonant pair designated as the domain of the reduplicative affix. -aa] is suffixed and juncture insertion  ∅ → ] × applies. Nested duplicants result and transcription is carried out in two cycles, as in Section 4.3. For example:

184

Chapter 7

(359)

suffix

JncIns

DC

→ × × ×× ×] → × × ] ×× ×] → ×[[× ] ×× ×] d g m

inner transcription

−→

a

d g m

× ×[× ×× ×] d

g m

a

a

outer transcription

−→

d g m

a

× × × × × × d

g m a

The left edge of the domain has been marked with a dashed line. Note the crucial role that it plays in both juncture insertion and in Default Closure. Note as well that the nal representation is equivalent to: × × × × × × d

a

g m

The derivation (359) continues with NCC repair and syllabication to produce d@ga:g@m. Recall that the devices for specifying the domain are limited. The penultimate consonant must be located by designating a domain, which requires constructing a foot at the right edge of the root. I assume that consonants are moraic in the root cycle, so that the domain is simply a nal binary foot. (360)

∗ ∗ ∗ × × × d g m

footing

−→

∗(∗ ∗ × × × d g m

7.9.4.2 Chaha In place of aa, the inxed exponent of the frequentative morpheme is a in Chaha, but it generally shows up on the surface as @. In some cases, Chaha frequentative forms are similar to corresponding Tigre forms. The frequentative imperative of /sBr/ ‘smash’ is s1B@B1r, for example. In imperfect and perfect forms, however, the underlying structure is not transparent. Banksira (2000) carefully establishes the relevant facts about Chaha morphophonology which produce the opacity: (1) independently of reduplication, the morpheme realizing perfect aspect readjusts the root by geminating its penultimate consonant; (2) root geminates devoice if they are followed in the root by a sonorant; and (3) geminates in Chaha systematically degeminate (by deletion of one of the twins). This synchronic phonology of Chaha is

Case Study: Chaha Reduplication

185

claried by comparing Chaha with some of its sister Ethiopian Semitic languages. (361)

Gemination and devoicing in the perfect (root /Bdr/, ‘be rst’) Imperative Imperfect Perfect Ezha y@-Bd@r y1-B@d1r b@dd@r-@m Endegen y@-Bd@r y1-B@d1r b@tt@r-@m Chaha y@-Bd@r y1-B@d1r b@t@r-@m Ezha has transparent penultimate consonant gemination in the perfect. Endegen also has the geminate devoicing rule. In Chaha, this is opaque because of a late rule of degemination. Penultimate gemination in Chaha, which is triggered in all roots by the perfective morpheme, is also triggered by the imperfective morpheme if the root is quadriliteral. Banksira (2000, 32) gives the template -C1 C2 @C3 C3 (1)C4 for the imperfect form of quadriliterals. Triliteral roots are not subject to imperfect penultimate gemination. With respect to their perfective and imperfective morphology, forms produced by reduplication are treated in the morphophonology as if they were roots. Roots that become quadriliteral as the result of either inherent reduplication (called 1212 roots earlier) or frequentative reduplication are therefore subject to imperfect penultimate gemination. To illustrate, frequentative forms of the triliteral root /sBr/ are given in (362). Devoicing, B→p, is the surface sign of underlying gemination in Chaha. The nal sonorant r licenses geminate devoicing. (362)

Frequentative forms Root Imperative /sBr/ s1B@B1r

Imperfect t1-sB@p1r

Perfect s1B@p@r-x@-m

These forms are derived in (363), up to the point of morphologically conditioned syllabication. Because the frequentative modication of the triliteral root is subject to quadriliteral penultimate gemination in the perfect, the perfect forms are identical to the imperfect forms prior to syllabication. (363) Imperative: × × × × ×

Fission → ×××××

Devoicing → ×××××

a B r

s B a B r

s B a B r

s Imperfect:

×××××× → ×××××× → ×××××× s

a B

r

s B a B r

s B a p r

186

Chapter 7

Syllabication follows. The penultimate consonant devoices (B→p) in the context of the nal sonorant. Since the initial B does not devoice, it must be that devoicing follows NCC repair, as assumed in (363), so that B has already undergone ssion into separate phonemes by the time that devoicing applies. 7.9.4.3 The Interaction of Frequentative Reduplication and x Dissimilation K&B give the examples in (364). For each root, the rst line gives the regular forms and the second line gives the frequentative forms. (I assume that the vowel quality and syllable structure variation in the forms below is secondary to the issue at hand.) The forms in which x-dissimilation applies are marked with †. (364)

Imperative s1x@r t@-sxax@r

Imperfect y1-sx@r y1-t-s1kak@r†

/mxr/

m1x1r t@-mxax@r

y1-m@x1r m@k@r† ‘advise’ y1-t1-mkak@r† t@-mkak@r† ‘advise each other’

/rxB/

n1x@B t@-rx@x@B

y1-r@x1B n@k@B† ‘nd’ y1-t1-rk@k@B† t@-rk@k@B† ‘show up’

/sxr/

Perfect s@k@r† t@-skak@r†

‘get drunk’ ‘act naughtily’

To understand this pattern, consider the detailed structure of the six forms of the root /mxr/. (365)

Imperative × × ×

Imperfect × × ×

Perfect × × × ×

m x r m1x1r

m x r m@x1r

x r m m@k@r†

× × × × ×

× × × × × ×

× × × × × ×

m a x r t@-mxax@r‡

a x r m y1-t1-mkak@r†

a x r m † t@-mkak@r

Regular

Frequentative

The puzzle is to account for why x-dissimilation applies to the † forms but not the ‡ form. The key forms are reproduced below, with the relevant timing slots boxed.

Case Study: Chaha Reduplication

(366)

a. no dissimilation /mxr/ × × × × × m

a x r

187

b.

x→k /mxr/ × × × × × × m

a

x

r

The salient difference between (366a) and (366b) is that there are two adjacent occurrences of x in (366b), but not in (366a). Not only are the two occurences of x in (366a) nonadjacent, but the intervening material is the exponent the exponent of the frequentative affix. The precise role of adjacency in the x-dissimilation rule (355), repeated here as (367), is somewhat obscure.  . . . [−sonorant, +continuant] (367) x → k All of the examples of nonadjacent dissimilation given by B&K (see p. 181 for some examples) are root internal. Banksira (2000, 33) gives only two examples in which x-dissimilation potentially applies when the target and trigger are not both root internal and in both, the target and trigger are adjacent. In one, (368a), neither the target nor the trigger are in the root. x-dissimilation applies. In the other, (368b), the target is root internal and the trigger is not. x-dissimilation does not apply. (368)

a. s@p@rk@m ‘I have broken you (2sg.m)’ s@p@r x x@ m break 1sg.sub 2sg.m.obj perf b. naxx1nam (*naknam) ‘I sent her’ nax x1 n a m send 1sg.sub obj case 3sg.f.obj perf

It is consistent with the known examples to assume that nonadjacent x-dissimilation is blocked unless the trigger, target, and intevening material are all associated with the same morpheme. Example (368b) shows that even x-dissimilation under adjacency is not entirely free. It may be, for instance, that a root target requires a root trigger, in order to account for (368b). In spite of uncertainties about the precise environment in which xdissimilation applies, it is certainly plausible that it does not apply in (366a), but does apply in (366b), where adjacency holds. Note that under this analysis, the initial k in the forms in which dissimilation applies is the result of overapplication. The medial occurrence of x dissimilates, taking the initial and nal occurrences with it. This com-

188

Chapter 7

pletes the discussion of the x-dissimilation pattern in (365). The fact that the initial x does not dissimilate in the frequentative imperative is due to the intervention of the copy of the exponent of the frequentative morpheme, as shown in (366a). Dissimilation of the initial x in the frequentative imperfect and perfect (366b) is due to the overapplication of dissimilation of the medial occurrence of x. 7.9.5 Kenstowicz and Banksira’s Analysis of x-Dissimilation B&K is valuable in demonstrating that x-dissimilation overapplies in Chaha since clear examples of overapplication are rare. The analysis, however, is untenable. Since Chaha phonology is particularly opaque and B&K attempt an analysis in the framework of Correspondence Theory, which is known to break down in analyzing opaque processes, it is not surprising that Correspondence Theory proves inadequate to the task. Two illustrations will suffice to make clear the problem. First, B&K claim that x-dissimilation is due to a phonotactic constraint: (369) x . . . [+continuant, −sonorant] But the constraint cannot be a surface constraint. The 3sg Jussive form of triliteral root f xQ ‘escape’ is y@-fka (B&K, example 5a), for example. In the surface representation, there is no trigger for x-dissimilation. Derivationally, the surface form is no mystery. The Q-trigger for dissimilation merges with the vowel (in some fashion) after it triggers dissimilation, producing a. Q -assimilation x-dissimilation syllabication −→ f kQ −→ f k@Q −→ f ka f xQ Second, one of the objectives of the analysis is to account for the interaction of frequentative reduplication and penultimate devoicing in the perfective and imperfective. This is impossible unless there is a convincing account of penultimate devoicing (imperative s1B@r versus perfect s@p@r ‘break’) to base the analysis on. B&K’s account is not. Banksira 2000 has a carefully worked out derivational analysis of the origin of penultimate p that relies on morphologically conditioned gemination (template satisfaction), followed by geminate devoicing, followed by degemination. Rule ordering is crucial. The geminate that triggers devoicing is opaque, present only at an intermediate stage in the derivation. This analysis is not available to an OT account. B&K appear to forget this, saying that “the template for the perfective of the derived frequentative verbs requires the penultimate radical to be a geminate

Case Study: Chaha Reduplication

189

“which surfaces [my emphasis] as a stop if nonstrident.” In addition to providing an example of overapplication of the Malay variety, B&K intend their analysis to support McCarthy and Prince’s contention that generative phonology is descriptively inadequate. They say that “as observed originally by Wilbur (1973), the traditional derivational model in which reduplication is expressed by a copy rule applying at some xed point in the derivation is unable to describe this phenomenon adequately.” I have shown this claim to be false by providing a detailed derivational analysis of the interaction of x-dissimilation and reduplication. A copy rule (transcription) applies at some xed 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, I have shown that the Optimality Theory model that B&K’s analysis is intended to conrm is not even capable of adequately describing the core facts of x-dissimilation and penultimate devoicing. Instead of supporting the idea that derivational phonology is descriptively inadequate, B&K succeed only in giving further evidence that opaque processes in phonology are pervasive and that Optimality Theory is unable to properly account for them.

Appendix A The NCC and the Retraction Condition

Sagey (1988) argues that it follows from the logic of temporal simultaneity that the NCC is an interface condition. This is not convincing. Phonological representations, even at the interface with phonetics, are not representations of physical events. They are instructions for creating physical events. More accurately, they are once removed from such instructions. Phonological representations are blueprints for creating motor instructions. The explanation for the NCC as an interface condition must be found in the structure of representations employed by the motor system, not directly in the structure of the physical world. Output interface conditions reect the demands of the system that is being fed, in this case the motor system leading to articulation. In what way might the motor system cause the imposition of the NCC as an output condition on phonology? Under the plausible assumption that the lexicon is optimized for space efficiency, it is implausible to suppose that exponents are stored as autosegmental representations. I therefore suppose that exponents are stored as simple phoneme/juncture sequences and that timing slots are supplied in the process of transcription from the lexicon to the phonological workspace. Geminates are indicated by some diacritic that is removed in the process of transcription: perhaps a [+long] feature in the phoneme itself that is translated into autosegmental terms and removed in the process. Crucially, I assume that autosegmental representation is the privilege of phonology, used in its internal computation, but that autosegmental representation is unintelligible to both the input and output systems. Suppose further that the interface mapping from phonology to the motor system, which translates the output of phonology into the language of the motor system, has limited resources. In particular, suppose that copying is not available to the interface mapping. Autosegmental representations contain tiers of three types: the timing tier, phonemic

192

Appendix A

tiers, and prosodic tiers. The point is controversial, but I assume that the phonemic and prosodic tiers have rather different properties, with the timing tier serving as a kind of interface, and the only interface, between the phonemic and prosodic tiers. Phonemic tiers in autosegmental representations are, in many cases, easily converted into sequential representations by a simple retraction operation, with the help of a few special symbols (diacritics). (370)

× × × × × × → t i

k

t

i

k

:

a

:

a

This is just the reverse of the process that mapped lexical items to autosegmental representations. Multiple segmental tiers translate into multiple parallel (aligned) retractions onto the timing tier. See Eisner 1997, which works out such a representational theory for autosegmental phonology as a way to provide a representational foundation for Optimality Theory. The picture of the output of phonology that this gives is multiple streams of motor instructions, aligned in time (representational temporal simultaneity). The mapping (371), in which a separate nasal tier is assumed, suggests what is involved. N

(371)

× × × × × × → t i

k

N

t

i

k

:

a

(tiNka:)

:

a

Retraction onto the timing tier is not sensitive to linear order on the phoneme tier. Consider the tak¸ci→talk¸ci Choctaw passive example discussed earlier (p. 62). The structure produced by transcription retracts onto the timing tier without accessing phoneme-tier order. × × × × × × →

(372) l

t a

t

a

l

k

c¸

i

k c¸ i

I therefore assume that phoneme tiers are unordered. Some implications of this are that there can be no oating elements on phonemic tiers and no junctures on phonemic tiers. Both of these predictions are correct, as far as I can see. Note that if phoneme tiers are unordered, many standard formulations of the NCC are inadequate since they rely on comparison of phoneme order with the order of associated timing slots.

The NCC and the Retraction Condition

193

In general, the reduplication process produces representations that cannot be retracted. The retraction of representation (76), which is produced in Yoruba nominalization, for example, fails: (373)

×

× × × × →

d

i

u n

d u n i Two copies of d are needed for retraction to be successful, but the interface mapping cannot carry out copying. The requisite copying in (373) must be done in phonology, prior to the output interface. This copying is what I have been calling Fission. phonology

(374)

×

× × × ×

d u n i

Fission

−→

×

× × × ×

dd u n i 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 phonetics. This imposes output condition (375) on autosegmental representations. A set of timing slots is called connected if for any two timing slots in the set, any intervening timing slots are also in the set. (375) Retraction Condition (RC) The set of timing slots associated with a segment is connected. A few examples will clarify the import of the RC. First, note that (376a) and (376b) are equivalent with respect to the RC, since the sets of timing slots associated with the two phonemes are identical in the two examples. Both satisfy the RC. (376)

a. × × × β

α

b.

× × × α

β

In (377a), on the other hand, the set of timing slots associated with α is not a connected sequence of timing slots, so (377a) violates the RC in exactly the same way that (377b) does.

194

(377)

Appendix A

a. * × × ×

b. * × × ×

α β

α

The RC and the NCC impose different conditions. There are several advantages to the RC. Most importantly, it can be derived from the assumption that autosegmental representations are the privilege of phonology and plausible assumptions about the representations employed by the motor system and the mapping from phonological representations to motor-system representations. With respect to reduplication it also has the advantage that then melodic tiers can be considered unordered, with the order of segments in motor-system representations derived entirely from order on the timing tier. This has the important consequence that no RC repair is needed in representations like (372) and the output of inxation-type reduplication in general. Minimal RC repair (Fission) is needed in examples like (374) and similar products of transcription. Note that if phonemic tiers are considered to be ordered and NCC repair were needed in examples like (374), not only Fission but multiple metathesis operations would be required as well. Finally, note that the RC makes metathesis comprehensible as an RC-violating preposing operation followed by timing-slot deletion that removes the RC violation, as illustrated in (378). (378)

× × α β

prepose

−→

delete

× × × −→ × × α β

α β

Appendix B Phonology On a Nonlinear Timing Tier

The theory of juncture insertion and transcription that DR is based on was developed as a means of overcoming certain conceptual and empirical problems with the theory proposed by Raimy 2000a. In the end, DR retains nothing from Raimy’s theory, which proposed that over/underapplication effects in reduplication are the consequence of phonological rules applying to a structure with a nonlinear timing tier in which multiple timing slots can have a common successor and a single timing slot can have multiple successors. Instead, DR proposes that over/underapplication effects are the consequence of phonological rules applying to a structure with long-distance geminates, which derives from McCarthy’s early work on Semitic morphology. This section is devoted to outlining Raimy’s proposal, explaining what its problems are, and contrasting it with DR. Since this appendix is mainly critical, it is important to rst acknowledge my debt to Raimy for asking the right questions: What phonological structures does UG make available? What is their architecture and vocabulary? What operations on these structures are possible and how is their relative complexity evaluated? B.1 Segmental Phonology Raimy proposes that phonological adjacency is encoded explicitly in phonological structure by pointers rather than implicitly by adjacency in a linearly ordered structure. The relevant structures for the overapplication of nasalization in the Malay example anem→˜an˜em˜an˜em that was discussed in chapter 1 are given in (379). In the DR structure on the left, a has two occurrences. In the R-representation (as I will call these structures) on the right, there is only one occurrence of a, but it has two predecessors.

196

(379)

Appendix B

a.

× × × × × × × ×

b.

#

a n e m

×

×

×

×

a

n

e

m

%

R-Representation Theory (henceforth RRT) does not exclude the possibility of multiple occurrences of phonemes (geminates). But it adds the possibility that an occurrence can be in multiple environments, depending on the choice of predecessors and successors. The a in (379b), for example, has one environment in which it follows a nasal and one environment in which it is initial. Raimy proposes that just as rules can specify that all occurrences of a phoneme must satisfy the structural condition for rule application or require only that one occurrence satisfy the condition, rules can specify that a phoneme must satisfy the structural conditions for rule application in all the environments in which it appears, or require only that it satisfy the condition in one of the environments in which it appears.1 Unlike RRT, DR does not extend standard ideas about rule application. It is already explicit in the theory of rule application to geminates that some rules require all occurrences of a phoneme to satisfy the structural condition for rule application, while others require only that one occurrence satisfy the condition. In DR, phonemes can have multiple occurrences, but each occurrence is in a unique environment. DR does, of course, extend standard ideas about what kind of geminates occur. So that the reader can better understand what is at stake in the two theories, (379) is redone below in a way that is intended to help the reader refrain from reading more into the pictures than the representations justify. (380)

a.

× × × × × × × × e a m

b.

a

n m

#

× ×

×

n

%

×

e Although DR representations have usually been portrayed as (379a) rather than (380a), the reader should be aware of the fact that no intrinsic order between phonemes is assumed, so (380a) is equivalent to (379a). Similarly, RRT assumes no order among timing slots other than that specied by pointers, so (380b) is equivalent to (379b). Note that

Phonology On a Nonlinear Timing Tier

197

(380b) makes it clear why RRT must employ explicit start and stop symbols in representations. Without them, there would be no way to determine how the loop is entered or exited. In what follows, “R-representation” will be abbreviated to RR. Note that RRs introduce a left-right asymmetry. It is trivial to determine the phonemes that follow a given phoneme, but difficult to determine the predecessors, since there are pointers to successors but no direct way to access predecessors other than an exhaustive search through the representation. One possibility is that RRs should be doubly linked lists, with two kinds of pointers, successor pointers and predecessor pointers. Raimy (personal communication) rejects this, but it is not clear why the architectural asymmetry appears to have no effect on the phonology. Finally, a linearization algorithm plays the role in RRT that Transcription plays in DR. The RR (380), for example, is mapped to the RR: (381)

#

˜a

n

˜e

m

˜a

n

˜e

m

%

A linearization operation that is adequate for the full range of RRs that must be considered turns out to be quite complex. I will examine Raimy’s proposal in some detail below. Before considering linearization, one other problem deserves to be highlighted: the difficulty of building prosodic and autosegmental structure over a nonlinear timing tier. B.2 Prosodic and Autosegmental Phonology Segmental phonology can be extended to RRs, but prosodic phonology cannot be. Raimy does not even attempt it. This makes it impossible for an RR analysis of reduplication to incorporate insights of Prosodic Morphology that are central to many reduplicative processes. To choose one example of many, recall that the analysis of Asheninca Campa in section 7.3 relies on the ill-formed syllable structure of forms like ∗σ σ (382) [ × × × × ] a m i n The unsyllabied nal consonant triggers vowel epenthesis, [amin ]→ [amina ], and the ill-formed onsetless initial syllable triggers [-Right,

198

Appendix B

[amina ]→a [mina ], resulting in a duplicant with well-formed syllable structure. This duplicant adjustment is responsible for the contrast between amin-a:nc/ hi ‘look’ (nonnite) and the corresponding continuative form amina-mina-ta:nc/ hi, with t epenthesized to join the V-nal stem and the V-initial suffix. It is not at all clear how RRT can deal with phenomena of this kind in a way that makes clear the inuence of prosody on the reduplicative process. Although segmental phonology can be extended to RRs, Raimy has nothing to say about their prosodic structure. The analog of (382) in the RR analysis would be a

#

m

(383)

%

n i

What is the syllable structure of (383)? Does the “nal” n syllabify as the onset of the “initial” a ? These problems do not arise in DR because linear structure is maintained throughout. Transcription in DR produces representations with complex crossing structures. Supercially, RRT avoids this. But it is not at all clear that it actually does. What the NCC is directed at is ensuring that the linear order on tiers corresponds to the linear order of the timing tier. If the timing tier is not linearly ordered, it is certainly not obvious that the NCC can be construed in a way that makes phonological sense. So it may be that RRT avoids crossed structures, but only because the NCC is moot if the timing tier is nonlinear. Consider, for example, the RR in (384), where R and L are high and low tones. (384)

#

×

×

H

L

%

This is precisely the kind of behavior that the NCC is supposed to bar, a high tone which both precedes and follows a low tone, for example. I conclude from this that RRT avoids portraying the complex crossing structures that DR exploits, but does not avoid crossing violations in reduplicative structures. Although DR allows Transcription to violate the NCC, much as it has generally been assumed that the NCC does not block tier conation, crossing considerations in DR are still important for autosegmental spreading. Since the timing tier is always linear, it is relatively simple

Phonology On a Nonlinear Timing Tier

199

to determine if there are crossing violations in DR structures. The issue is not so clear in RRs. Raimy does not comment on the problem or offer a revision of the NCC suitable for RRs, so the comments here will be restricted to illuminating the problem. Consider the structure in (385), for example. Can the articulator of the velar spread regressively to the nasal as shown? k n

◦ ×

◦

#

×

(385) %

◦ a

◦ ×

◦

#

×

−→ ×

k

N

%

×

◦ a The analog of this question in the DR framework was addressed in chapter 2 (p. 28). The issue was overapplication of velarization in kan→kaN-kaN. The conclusion was that spreading is blocked. Once the crossed structure produced by Transcription is granted, nothing new is involved here. Spreading would necessitate sharing a velar articulator feature across a coronal articulator, an NCC violation. RRT can approach this question in various ways, depending on how the NCC is formulated. The intention here is not to develop that theory, but simply to point out that RRT does not avoid the complexities of crossing that I believe are intrinsic to reduplication. B.3 Linearization Raimy presents the linearization algorithm somewhat informally, so we need to make it somewhat more precise before it can be discussed. This runs the usual risk of misconstruing the author’s intention, but it cannot be avoided. Two different kinds of pointers are recognized in RRT: pointers already present in the lexicon and pointers introduced either in attaching the exponent of an affix to a stem or by readjustment rules acting on the stem. For convenience, call the former l-pointers (lexical pointers) and the latter m-pointers (morphological pointers). In pictorial representations, heavier arrows will be used to represent m-pointers. The linearization of (386) will illustrate the algorithm. It will be clear shortly why the pointers have been indexed.

200

Appendix B 8 3

(386)

#

1

b

2

4

a

b

5

a

6

s

7

%

9

The linearization of (386) is determined by choosing the optimal representation from among the set of linear RRs that correspond to a traversal (i.e., “follow the arrows”) of the graph (386) from # to %. Candidate linear RRs are given in (387). Correspondence is established by coindexing the pointers. The desiderata that determine optimality will be given shortly. (387)

#

1

#

1

c.

b

5

a

4

b

5

2

a

b

2

b

2

b

2

b

2

#

1

d.

#

e. f. .. .

a. b.

a

6

s

7

%

b

5

a

6

s

7

%

a

6

s

7

%

3

a

9

b

8

b

5

a

6

s

a

9

b

8

b

5

a

6

s

7

%

a

9

b

2

a

9

b

8

b

5

a

a

4

a

3

b

8

1

b

#

1

#

1

7

%

6

s

7

%

The ranked list of desiderata given in (388) determines the optimal element of the set (387). It was culled from various places in Raimy 2000a. (388) Optimal linearization 1. Traverse as many phonemes as possible. 2. Traverse as many of the m-pointers as possible. 3. Traverse m-pointers as early in linearization as possible. 4. Traverse as many of the l-pointers as possible. 5. Make the shortest possible traversal. The computation of the optimal element can be carried out by the usual sequential ltering method which is well-known from OT. (389)

candidate set ltered by 388.1 ltered by 388.2 ltered by 388.3 ltered by 388.4 ltered by 388.5

{a, b, c, d, e, f, . . .} {a, b, c, d, e, f, . . .} { a,  b, d, e, f, . . .} { d, e, f, . . .} {e, f, . . .} {e, f, . . . }

Phonology On a Nonlinear Timing Tier

201

The linearization is therefore (390)

#

b

a

b

b

a

s

%

This is a very peculiar operation to insert into a phonological derivation, which otherwise proceeds by progressive modication. Raimy’s algorithm does not compute (390) by modifying (386). Consider, for example, the representation below, which is equivalent to (390). (391)

#

b

a

b

a

s

%

b This is a recognizable transform of (386) and one can imagine various steps (ssion, pointer modication, pointer deletion) that could get from (386) to (391). The linearization algorithm does not accomplish anything like this, because it is based on the idea of correspondence rather than the idea of progressive modication. As far as I can see, it is extremely difficult to give an algorithm that transforms the RR into the linearized form by progressively modifying the input. Note that the criticism here is not directed at the use of optimality or economy criteria, or even criteria which choose a derivation on the basis of global economy. Principles of this kind could choose between alternate derivations of the linearized RR from the input RR. The criticism is that the linearized form is not derived from the input form by progressive modication. Depending on the reader’s commitment to derivational phonology, the criticisms made above carry more or less weight. But they should have at least convinced the reader that linearization introduces a major complication into the theory and retreats from the idea of derivation by progressive modication. This is important in itself because Raimy claims a conceptual advantage for his theory of RRs based on the claim that the theory requires no reduplication-specic devices. A reasonable characterization of a reduplicative structure in RRT is one that has a nontrivial linearization. Since the linearization algorithm has no other use than linearizing such representations, it is hard to avoid the conclusion that the linearization algorithm is reduplication specic. If the linearization algorithm were more or less straightforward and conceptually obvious, it would be fair to claim that nothing new is introduced into the theory. But linearization in RRT is far from being either obvious or straightforward.

202

Appendix B

B.4 Concatenative versus Nonconcatenative Morphology In closing these comments on RRT, one further comparison with DR is important to make. I will leave it to the reader and future work to decide on the relative advantages and disadvantages. In RRT, there is no difference in kind between prexation and suffixation, on the one hand, and inxation, on the other. In Distributed Reduplication, there is a difference in kind. Prexation and suffixation are concatenative, while inxation involves both concatenation and t-juncture insertion. Although there are no crucial examples that give evidence one way or the other, the difference is noteworthy enough that it is worth reviewing. Consider, for example, the combination of the stem taki and an affix with the exponent ga in Raimy’s theory. Prexation, forming gataki, is shown in (392a); suffixation, forming takiga, is shown in (392b); and inxation before the nal syllable, forming tagaki, is shown in (392c). The three RRs that result are (392)

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 theory between inxation, suffixation, and prexation, Raimy must account for the relative rarity of inxation. He attributes it to the greater complexity in specifying the source and target positions in the stem for the attachment of links between the exponent of the affix and the stem. The difference between inxation and concatenation is much more sharply drawn in DR. The DR structures corresponding to the RRT structures in (392) are (393)

a.

× × × × × × g a

b.

t

a k

i

× × × × × × t

a k

i

g a

Phonology On a Nonlinear Timing Tier

c. × × [ × × × × ] t

a k

i

g a

203

or

[ × × × × ] × × g a

t

a k

i

A prex or a suffix is joined to the stem by concatenation. An inx is an underlying concatenated prex or suffix, coupled with the insertion of t-junctures whose ultimate effect is to prepose a suffix, or postpose a prex, into the stem.

Notes

Chapter 1 1. I call it “copying-like” because it combines copying with translation. Bases are mapped to complementary bases. 2. The translation of a phonological representation/score into a physical performance is not direct. A phonological representation is rst translated into a phonetic score, which can undergo further processing before it in turn is translated into a physical sound signal. 3. Except that Kitagawa proposes multiple timing tiers (with later linearization) in order to avoid NCC violations, this is quite close to DR’s view of Transcription. 4. Kitagawa (1987) also builds a parallel structure, but it does not have a templatic origin. 5. Mester (1986) also uses this idea. In his analysis, phonological operations can apply after the parax structure is created, but before the parallel structures are combined into a linear structure.

Chapter 2 1. Transcription generally produces crossed structures, but there are circumstances in which it does not. If an initial consonant is reduplicated, for example, a geminate is produced, with no crossing. This occurs in Imdlawn Tashlhiyt Berber: σ σ σ σ [×]× × → × × × × f

r

n

f

r

n

2. Halle and Harris (2005), for reasons they do not explain, claim that there is no distinction between what is copied and the copy, so that the question of the direction of copying is moot. Aside from conceptual problems, we will see later (section 3.3) that there are empirical consequences to the choice of direction that give evidence that Left and Right Transcription are different operations. 3. See Frampton and Gutmann 1999, 2002 for a development of essentially the same point of view in syntax. We argue against Chomsky’s early Minimalist Program idea of massive overgeneration and ltering by economy conditions.

206

Notes

4. Certainly, one of the more bizarre aspects of OT’s conception of the language faculty is that a complex calculation is needed to move dog out of the lexicon and into a phonological computation. 5. There are antecedents of Mester’s theory in the work of Kitagawa (1987), Clements (1985), and Uhrbach (1985), as Mester discusses (pp. 171–172). 6. Note that (20) suggests a way to reconcile DR crossed structures with the NCC. One could restrict the NCC to intramorpheme crossing in the phonology, with later conditions ensuring that all crossing is eliminated at the interface. 7. This oversimplies the extensive discussion in Halle 1995. In some cases, only marked features block feature spreading. 8. McCarthy and Prince (1995) assume that glides themselves nasalize. They say that “Onn (1976) does not transcribe nasality in glides; we have altered his transcription in this respect.” If in fact glides nasalize, then the discussion in the text is besides the point. Onn, in my reading, is fairly explicit that they do not. He says that “nasalization in JM [Johore Malay] penetrates not only vowels but also other nonconsonantal segments as well, namely y, w, h, and P, and proceeds to nasalize the vowel or vowels following these nonconsonantal segments” (p. 70). 9. Coleman and Local (1991, 309) say that this assumption is implicit in various investigations, although I was not able to nd any clear statement of the idea in the sources I was able to obtain. They mention Clements and Keyser 1983, 11, Kaye 1985, 289, 301–304, Lowenstamm and Kaye 1986, Prince 1984, 235, Clements 1986, and Pulleybank 1986, 14. They are certainly correct that “[if] the timing relations between Autosegments are dependent of the ordering of objects in the skeleton . . . [then] the maximally parsimonious account is one in which autosegmental tiers are not explicitly ordered” (p. 309).

Chapter 3 1. If a particular morpheme is invariantly realized by a unique vocabulary item, it is immaterial whether it is associated with the morpheme early or late. At the level of detail presented here, this is a question of implementation. In any case, the term vocabulary insertion will refer to the point at which the exponent of the vocabulary item is integrated into the full structure. 2. It is possible to entertain the idea of a more complex theory, in which triggered rules are interspersed among the cyclic rules, but this possibility is not considered here. Halle and Marantz (1993, 129) explicitly note this possibility. “[Readjustment rules] satisfy the same formal constraints as ordinary phonological rules (and might even be ordered among the phonological rules; see the discussion of ‘allomorphy rules’ in Davis 1991).” 3. This assumes that wEnt is formed is the same way that past tense forms like sEnt, bEnt, bIlt, spEnt, and lEnt are formed: the past morpheme is realized by -t and readjustment deletes the nal consonant of the stem. 4. Zoll uses this fact to highlight weaknesses in the Correspondence Theory (McCarthy and Prince 1995) account of reduplication. 5.

Lieber’s proposal was made under the assumption that all lexical insertion and

Notes

207

readjustment preceded phonology. Under those assumptions, there was more reason to believe it. 6. The exceptions are handled in the following way. Four strong roots are marked as exceptions to Nasal Accretion. For example, evinte ‘look after’ produces a evinte, not a emvinte. All the other exceptions are restricted narrowly to the singular imperative. In place of the regular rules, two roots simply prex e: ve ‘go’ produces eve and mevyaG ‘dream’ produces emevyaG. So eve appears in modied root contexts in the singular imperative, but ampe appears in other modied root contexts. Two other roots simply delete the initial phoneme instead of applying the regular rules: velom ‘come’ produes elom and velaG ‘go/come a long way’ produces elaG. Readers who may be uncomfortable about a rule that applies in the context of only two roots should consider the alternative: that the striking similarity of velom and velaG to elom and elaG has no status in the grammar and the modied roots could just as well be anything else rather than almost identical to the basic root.

Chapter 4 1. It is possible to use only one symbol. I explicitly mark both ends of the already copied material in order to simplify the discussion and make it easier for the reader. As far as I know, nothing substantial is at stake. 2. Some linguists are troubled by this point of view, suggesting that if these bookkeeping symbols are real phonological objects, they should be subject to the same operations that other phonological objects are subject to: spreading, lengthening, or shortening, for example. This is misguided. These are operations that particular kinds of phonological objects are subject to, not phonological objects in general. Timing slots, for example, do not spread or lengthen or shorten. Insertion of a timing slot is used in vowel lengthening, and deletion of a timing slot is part of vowel shortening, for instance. But timing slots neither lengthen nor shorten. 3. At least until readjustment is completed and cyclic phonology is entered, the computation does keep track of the distinction between stem and nonstem, and the reduplication process keeps track of whether transcription is to the right or the left. But it is premature to make a commitment about how the computation carries this out. 4. This is slightly inaccurate, since I assume that t-junctures are timing tier junctures and assume that lexical entries are not autosegmental. More accurately, lexical entries contain symbols that translate into t-junctures (in an obvious way) when the lexical entry is translated into an autosegmental representation on entering the morphophonological computation. 5. Healy does not give any examples with initial consonant clusters and it is not clear . if they exist in Agta, so it is possible that the stem adjustment rule is ∅ → ] / ×

Chapter 5 1. I do not share their view that postvocalic consonants in heavy syllables are in the nucleus. Frampton (2001, 2003) contain some arguments in favor of a more thoroughly autosegmental view of syllable structure than what is currently assumed. The centrality of the nuclear skeleton in reduplication gives further evidence. 2.

McCarthy and Prince (1995, 126) note the reduplication pattern in Nuu-chah-nulth

208

Notes

(which they call Nootka) and a similar one in Nitinaht, a related language, based on Stonham 1990 and Shaw 1992. They account for these patterns by assuming that a “heavy syllable template” is imposed, but that penalties against syllables with codas outrank the penalties against nonmaximal copying. This account is implausible, since the possibility of a penalty on the presence of a coda outranking penalties against nonmaximal copying can easily produce unattested reduplication patterns. The following unlikely variant of total reduplication results if NoCoda is elevated over the constraint that penalizes nonmaximal copying: (i) a. da ti kun → da ti ku - da ti kun b. da tin ku → da ti - da tin ku c. dan ti ku → da - dan ti ku Max(B,R) forces as much material to be reduplicated as possible, but is stopped by the more prominent NoCoda. NoCoda is, in turn, outranked by the usual array of correspondence constraints that ensure that BR-correspondence imitates faithful copying. As far as I know, such a pattern is not attested. 3. If the stem ends in k followed by m, n, l, or s, there is metathesis of the nal two consonants combined with reduplication. So caniks ‘sideways’ → canis-cak, not canik-cas . It is worth noting that a close-to-ordinary process of reduplication and reduplicative metathesis can be combined to give the desired output. According to the rules worked out in chapter 4 for expanding embedded duplicants, [ca ni [k s ]] → [ca nis ]k → caniscak 4. It is sufficient for left transcription to be simpler for most stems (those which begin with a syllable with an onset, for instance) for the grammar to choose left transcription for all stems. It would be a complication for the direction of transcription to depend on the particular stem. 5. It is worth noting that a CVC is copied to the left even if the inital vowel of the stem is long, so ka:ma: → kamka:ma:/kakka:ma:. The source is [kaam ]maa. An account is given in chapter 6.

Chapter 6 1. CT computes the input-output relation s+ρ+ikuk → siksikuk by means of a much more abstract derivation whose intermediate stages are not phonological representations, but sets (very large sets) of phonological representations, which are progressively ltered by conditions imposed on the base, reduplicant, and their correspondence. 2. In fact, they were called defect-driven rules. But it is simply a question of point of view. For iterative rules, the process is sometimes most easily thought of as removing defects. For shaping the reduplicant to a desired form, the process is most easily thought of as achieving a goal prosodic shape. 3. Optimality Theory (Prince and Smolensky, 1993) takes the idea of goal-driven rules in a different direction, viewing all phonological change as driven by minimal constraint set violation (or maximal goal set satisfaction), eliminating transformations altogether, and proposing that there are no intermediate representations. For a useful discussion of the possibilities opened up by Sommerstein’s proposal, see Goldsmith 1990, 321–331. See Frampton 2002 for my view of Optimality Theory.

Notes

209

4. Blevins analyzes the transition from Mokilese-A to Mokilese-B as a diachronic instance of “the emergence of the unmarked,” the elevation of NoCoda in the penalty function hierarchy in particular. This account is implausible for the reasons given in chapter 5 note 2. 5. Davis (2003) avoids an appeal to reduplication specicity by using Sympathy Theory and proposals concerning the moraicity of geminates. 6. The discussion of syllable weight in Hayes 1995, 270 is useful. He suggests that the notion weight should be restricted to mora count, and that it is more useful to analyze what I call secondary weight distinctions as prominence distinctions. That may be correct. 7. Lardil roots are often written with a nal laminal consonant T. This consonant is supposed to surface when a V-initial suffix is concatenated; otherwise it is supposed to delete because the 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 by rule to eliminate vowel hiatus. I assume that this is the case. 8. Haugen 2003, 79 takes Yaqui to demonstrate that Moravcsik was simply wrong. This is an overly harsh judgment. There remains a kernel of Moravcsik’s insight that an adequate theory of reduplication must account for. Syllable copy is still an unattested form of reduplication, and even if some example is found, its rarity would still demand an account. Haugen is incorrect in claiming that “Yaqui illustrates a pattern of reduplication which has been repeatedly claimed not to exist: the pattern of so-called ‘syllable-copy’.” This is not entirely true; if the initial syllable is bimoraic, it is not copied.

Chapter 7 1. I am grateful to Galen Sibanda for being extremely helpful and patient in answering numerous questions via e-mail. 2. Sibanda 2004, 88 gives sik-is-el-an-isis-w ‘be caused to cut for each other well’. The gloss is cut-appl-caus-reciprocal-intensier-pass. 3. Perhaps more complex morphophonotactic conditions are relevant to forcing nala repair. In any event, a is epenthetic. It is not clear how much this position differs from the position of HI&S. They say “-a may be invoked, typically in nal position, to ‘ll out’ any kind of stem.” If the way that the morphophonology “knows” that something needs to be “lled out” is that some morphophonotactic is violated, the positions appear to be identical. Downing (1999) argues against this, pointing out that the Bantu languages generally use an epenthetic front vowel in loanword adaptation to the relevant phonotactics, but use nal-a just as in Ndebele. But it is not difficult to see why a different nal vowel is used in the derivational morphophonology than is used in roots. All the V-nal suffixes have a nonlow nal vowel. The use of nal-a can therefore be seen as dissimilatory: it distinguishes nal epenthetic vowels from suffixal vowels. A second line of objection comes from HI&S (note 5). It points out that in many Bantu languages, there are verb stems that would otherwise end with a tense ¸i (causative) or u¸ (passive) vowel that nevertheless add nal-a. In Ndebele, these tense vowels have become glides, so the argument is indirect. All that this demonstrates is that if the syllable phonotactics

210

Notes

do nothing more than demand V-nal syllables, then nal-a cannot be attributed to the syllable phonotactics. But in order to make a solid case that nal-a is not a repair forced by the syllable phonotactics, it must be shown that these phonotactics freely permit these tense vowels in nal position. Perhaps this can be done, forcing a reconsideration of the proposal that the appearance of nal-a is not driven by the phonotactics. But HI&S does not attempt this demonstration. I will therefore assume that a is the vowel of choice for repairing nal syllables throughout the verbal system. This extends to its use achieving well-formed reduplicant syllable structure. 4. Most work on Bantu morphophonology is oriented toward surface form and many subconstituents of inected-verb surface forms have been dened in the literature on various phonological, prosodic, and morphological grounds. The output of the FScycle corresponds roughly to what Downing 1999 calls the I-stem, which derives from Myers 1987. 5. See Marantz 2001, Marvin 2002, and Newell 2004, for discussion of word-internal phases (“spellout” in the terminology of the Minimalist Program). 6. The parallel with syntactic “movement” should be noted. The Copy Theory of Movement (Chomsky 1993) eliminates the idea of movement in place of copying the source to the target location. As in (245a), there are sometimes factors that force the source of copying to surface. See Boskovi´c 2001 and Nunes 2004 for many examples. See also Frampton 2004, which brings copy theory much closer to the picture in (245). In place of a copy, it is simply a question of the same object being merged in two different places in a tree. Multiple linking replaces copying. 7. Although forms like dliwelwa are called here “applied passives,” more serious investigation into the underlying syntax is needed to put this on a rm footing. The term should be taken as tentative, and will perhaps turn out to be essentially descriptive. As far as I know, there have been no proposals in the syntactic literature that the applicative can occur above the passive in the verbal structure and that this conguration prevents the applied object from appearing in the surface subject position. 8. Since (241b) can be either an applied passive or a passivized applicative, this partially contradicts the claim of HI&S that (241b) must be an applied passive. But it is sufficient for our purposes that it can be. 9. In fact, HI&S claims that any suffix can be skipped. Sibanda 2004 corrects this and restricts “skipping” to the passive suffix. 10.

See Downing 2000 for an Optimality Theory approach.

11.

Ngessimo Mutaka was very helpful in answering numerous e-mail inquiries.

12. The fact that the behavior of nal-a diverges from that of all suffixes is further evidence that it is epenthetic, not a suffix. 13. This peculiarity may not be a restriction on reduplication, but a restriction on combining pseudo–extension suffixes with actual inner suffixes. Kinande, like Bantu languages generally, has strong templatic conditions that affect surface suffix order. It is reasonable to speculate that these templatic conditions break down when faced with ordering nonexistent suffixes. Since unreduplicated roots of this type do accept extension suffixes, it would have to be the case that verbs of this type have the option of pseudostructure, but are not necessarily structured in this way.

Notes

211

14. The reduplicant tesa appears to propose a serious challenge to the analyses of Kinande reduplication by Downing (1999, 2000) and Inkelas and Zoll (2000). Downing relies heavily on the proposal that the reduplicant, if prosodic constraints prevent it from being a copy of the base, must be what she calls a “Canonical Verb Stem” (CVS). A CVS is simply an I-stem with a default nal vowel. But tesa, or even Otesa, is not a possible I-stem. Inkelas and Zoll face roughly the same problem. There is no reason for the reduplicant cophonology to delete y. 15. Payne 1981 is an impressive analysis of a single speaker’s language. Spring 1990 is used to ll in many details of the treatment of unprexed monosyllabic roots. Spring’s data comes from eldwork by her and Payne with a second speaker. Since the second speaker’s reduplication data showed signicant instability, there is some concern that the idealized version of Asheninca Campa that is hypothesized by combining the two studies might not correspond to a stable linguistic reality. See note 2 (p. 147) in Spring’s thesis for some discussion of the instability in her informant’s reduplication rules. 16. In DR, melody copying is incoherent, because copying requires linear order and the melodic tiers have no intrinsic order. 17. [k go ] → go-kgo, for example, under either left or right transcription. All other things being equal, however, left transcription with leading-edge truncation is preferred to right transcription with trailing-edge truncation because it involves less copying. Recall that trailing-edge truncation must copy and delete, whereas leading-edge truncation simply fails to copy. 18. For roots of the rst type, the difference between VC- and CV-reduplication is minimal: (C1 )VC2 VC2 C3 V(C4 ) versus (C1 )VC3 VC2 C3 V(C4 ). As noted, this is unlikely to be an accident. Unfortunately, without a theory of how verb irregularities of this kind are learned (the irregularity being vocabulary selection), it is impossible to transform this observation into a theoretical proposal. Roots of the second type might reect an earlier stage in the language in which the roots were monosyllabic, ending in a consonant cluster. Later, after i or 1 broke up the consonant cluster, irregular assignment to the VC-reduplicating class might have been made in order to minimize the change in the reduplicated forms. 19. If true, this would be related to the existence of a sizable number of roots of the form (C)aCiC or (C)aC1C that undergo VC- rather than CV-reduplication, which is the expected variety of reduplication for polysyllabic roots. 20. This construal of what it means for a timing slot to satisfy C would require a revision of the proposal that morphological juncture insertion (as opposed to juncture insertion as part of duplicant adjustment) is sensitive only to the nuclear skeleton. There is nothing in the nuclear skeleton that allows unassociated timing slots to be distinguished from associated timing slots. 21. The vowel inventory is {i, 1, u, o, a}. There are nine V1 V2 sequences that occur in Tohono O’odham that are not Type F: {ia, iu, 1a, 1i, 1u, ui, oi, ai, au}. 22. Alongside the preference for falling diphthongs is a secondary preference for diphthongs with smaller sonority differences, so au is preferred to an, for example. If the preference for falling diphthongs is combined with the preference for low-sonoritydifference diphthongs, with the rst criterion given priority, the following markedness scale on potential nuclear diphthongs is produced:

212

Notes

(i) 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 syllabies with a long nucleus. The following table uses the gures that Steriade gives. The “possible” column gives the count of the number of roots whose most preferred diphthong is of the indicated kind. The “actual” column gives the count of the number of these that have a zero grade reduction. (ii) {ai, au} {ar, al} {an, am} {ia, ua} {ra, la} {na, ma}

Possible 157 80 48 28 29 6

Actual 156 71 36 17 5 0

Percentage 99 89 75 61 17 0

23. Steriade (1988) has an extensive discussion of “transfer effects” in Sanskrit reduplication. OnsPerm, as formulated here, uses this idea. 24. The discussion would be a diversion, but there are reasons to believe that unordered rule systems are unlikely to be part of internal grammars. Essentially, random choice is incorporated into the computation. Computational devices that allow random choice are more complex and more computationally powerful than those that do not. 25. There is historical shift and a spotty record, so it is not clear to me for this particular root how much this reects a synchronic reality. There is evidence that there is synchronic variation of this kind, but perhaps only between different roots. 26. The “modication,” it should go without saying, is not under the conscious control of the language speaker, but under the automatic control of the language-learning apparatus that works in the complex environment which results in language change over time.

Appendix B 1. Formally, there is a four-way distinction: every environment of one occurrence, one environment of one occurrence, every environment of every occurrence, and one environment of every occurrence. It would be very difficult to nd relevant examples, but if examples showed the need for a four-way distinction, it would be an impressive conrmation of RRT.

References

Applegate, Richard. 1972. Inese˜no Chumash Grammar. Doctoral dissertation, University of California, Berkeley. Applegate, Richard. 1976. Reduplication in Chumash. In M. Langdon and S. Silver, eds., Hokan Studies, 271–283. The Hague: Mouton. Austin, Peter. 1981. A Grammar of Diyari, South Australia. Cambridge: Cambridge University 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: John Benjamins. Barker, M. A. R. 1964. Klamath Grammar. Berkeley: University of California Press. Benjamin, Geoffrey. 1976. An outline of Temiar grammar. In P. Jenner, L. Thompson, and S. Starosta, eds., Austroasiatic Studies, volume 2, 129–187. Honolulu: University Press of Hawaii. Blevins, Juliette. 1996. Mokilese reduplication. Linguistic Inquiry 27 (3): 523–530. Boersma, Paul, and Bruce Hayes. 2001. Empirical tests of the Gradual Learning Algorithm. Linguistic Inquiry 32 (1): 45–86. Booker, Karen M. 1980. Comparative Muskogean: Aspects of Proto-Muskogean Verb Morphology. Doctoral dissertation, University of Kansas, Lawrence.  Boskovi´c, Zeljko. 2001. On the Nature of the Syntax-Phonology Interface: Cliticization and Related Phenomena. Amsterdam: Elsevier. Bromberger, Sylvain, and Morris Halle. 1988. Conceptual issues in morphology. Unpublished, MIT. Broselow, Ellen, and John McCarthy. 1983. A theory of internal reduplication. Linguistic Review 3: 25–88. 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.

214

References

Chomsky, Noam. 1993. A minimalist program for linguistic theory. In K. Hale and S. J. Keyser, eds., The View from Building 20. Cambridge, MA: MIT Press. Chung, Chin Wan. 1999. Reduplication in Korean. Doctoral dissertation, University of Indiana. Clements, George. 1985. The problem of transfer in nonlinear morphology. Cornell Working Papers in Linguistics 7: 38–73. Clements, George. 1986. Compensatory lengthening and consonant gemination in LuGanda. In L. Wetzels and E. Sezer, eds., Studies in Compensatory Lengthening. Dordrecht: Foris. Clements, George, and Samuel Jay Keyser. 1983. CV Phonology: A Generative Theory of the Syllable. Cambridge, MA; MIT Press. Coleman, John, and John Local. 1991. The “No Crossing Constraint” in autosegmental phonology. Linguistics and Philosophy 14: 295–338. Crowley, Terry. 1998. An Erromangan (Sye) Grammar. Oceanic Linguistics Special Publication No. 27. Honolulu: University of Hawai‘i Press. Davis, Robert. 1991. Allomorphy in Spanish. Doctoral dissertation, University of North Carolina. Davis, Stuart. 2003. The interaction of nasal substitution and reduplication in Ponapean. ZAS Papers in Linguistics 32: 31–46. Dell, Fran¸cois, and Mohamed Elmedlaoui. 1988. Syllabic consonants in Berber: Some new evidence. Journal of African Languages and Linguistics 10: 1–17. Demers, Richard, Fernando Escalante, and Eloise Jelinek. 1999. Prominence in Yaqui words. International Journal of American Linguistics 65 (1); 40–55. Diffloth, Gerard, 1976. Expressives in Semai. In P. Jenner, L. Thompson, and S. Starosta, eds., Austroasiatic Studies, volume 2, 229–247. Honolulu: University Press of Hawaii. Dixon, Robert M. W. 1977. A Grammar of Yidiny. Cambridge: Cambridge University Press. Downing, Laura. 1999. Morphological constraints on Bantu reduplication. Linguistic Analysis 29: 6–46. Downing, Laura. 2000. Morphological and prosodic constraints on Kinande verbal reduplication. Phonology 17: 1–38. Eisner, Jason. 1997. Efficient generation in primitive Optimality Theory. Proceedings of the 35th Annual Meeting of the ACL and 8th Conference of the EACL, Madrid. Emeneau, M. B. 1955. Kolami: A Dravidian Language. University of California Pulications in Linguistics 12. Berkeley: University of California Press. Fitzgerald, Coleen. 2000. Vowel hiatus and faithfulness in Tohono O’odham reduplication. Linguistic Inquiry 31 (4): 713–722. Frampton, John. 2002. What kind of a thing is a language faculty? GLOT International 6 (1), 1–11.

References

215

Frampton, John. 2003. Root vowel syncope and reduplication in Sanskrit. In J. Cihlar, A. Franklin, and D. Kaiser, eds., Proceedings of the Annual Meeting of the Chicago Linguistic Society 39 (1), 75–91. Chicago: Chicago Linguistic Society. Frampton, John. 2004. The Remerger Theory of Movement: Evidence from Bulgarian multiple wh-phenomena. Unpublished, Northeastern University, Boston. Frampton, John. 2008. SPE extensions: Conditions on representations and defect driven rules. In B. Vaux and A. Nevins, eds., Rules, Constraints, and Phonological Phenomena. Oxford: Oxford University Press. Frampton, John and Sam Gutmann. 1999. Cyclic computation, a computationally efficient Minimalist syntax. Syntax 2 (1), 1–27. Frampton, John and Sam Gutmann. 2002. Crashproof syntax. In S. Epstein and D. Seely, eds., Derivation and Explanation in the Minimalist Program. Oxford: Blackwell. Gafos, Adamantios. 1998a. Eliminating Long Distance Consonantal Spreading. Natural Language and Linguistic Theory 16 (2): 223–278. Gafos, Adamantios. 1998b. A-Templatic Reduplication, Linguistic Inquiry 29 (3), 515–527. Gafos, Adamantios. 2003, Greenberg’s Asymmetry in Arabic: A consequence of stems in paradigms. Language 79 (2): 317–357. Goldsmith, John. 1976. An overview of autosegmental phonology. Linguistic Analysis 2: 23–68. Goldsmith, John. 1990. Autosegmental and Metrical Phonology. Oxford: Blackwell. Haas, Mary R. 1977. Nasals and nasalization in Creek. In Proceedings of the Third Annual Meeting of the Berkeley Linguistics Society, 194–203. Berkeley: Berkeley Linguistics Society, University of California, Berkeley. Hale, Ken, and Abanel Lacayo Blanco. 1989. Diccionario elemental del Ulwa (Sumu meridional). Cambridge, MA: Center for Cognitive Science, MIT. Halle, Morris. 1990. An approach to morphology. In M. Kadowaki and S. Kawahara, eds., Proceedings of NELS 20, 150–84. Amherst: GLSA, University of Massachusetts. Halle, Morris. 1995. Feature geometry and feature spreading. Linguistic Inquiry 26 (1): 1–46. Halle, Morris. 1997. On stress and accent in Indo-European. Language 73 (2): 275– 313. Halle, Morris, and James Harris. 2005. Unexpected plural inections in Spanish: Reduplication and metathesis. Linguistic Inquiry 36 (2): 195–222. Halle, Morris, and William Idsardi. 1995. In John Goldsmith, ed., The Handbook of Phonological Theory. Oxford: Blackwell. Halle, Morris, and Alec Marantz. 1993. Distributed morphology and the pieces of inection. In K. Hale and S. J. Keyser, eds., The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, 111–176. Cambridge, MA: MIT Press.

216

References

Halle, Morris, and Jean-Roger Vergnaud. 1987. An Essay on Stress. Cambridge, MA: MIT Press. Harrison, Sheldon P. 1973. Reduplication in Micronesian languages. Oceanic Linguistics 12: 407–454. Harrison, Sheldon P. 1976. Mokilese Reference Grammar. Honolulu: University of Hawaii Press. Haugen, Jason. 2003. Allomorphy in Yaqui reduplication. In L. Barragan and J. Haugen, eds., Studies in Uto-Aztecan: MIT Working Papers on Endangered and Less Familiar Languages 5, 75–103. Cambridge, MA: Department of Linguistics and Philosophy, MIT. Hayes, Bruce. 1986. Inalterability in CV phonology. Language 62: 321–351. Hayes, Bruce. 1995. Metrical Stress Theory. Chicago: University of Chicago Press. Hayes, Bruce, and May Abad. 1989. Reduplication and syllabication in Ilokano. Lingua 7: 331–374. Healy, Alan, Ambrose Isoroembo, and Martin Chittleborough. 1969. Papers in New Guinea Linguistics 9: 33–64. Healy, P. M. 1960. An Agta Grammar. Manila: SIL. Hudson, Grover. 1995. In John Goldsmith, ed., The Handbook of Phonological Theory. Oxford: Blackwell. Hyman, Larry, Sharon Inkelas, and Galen Sibanda. 2005. Morphosyntactic correspondence in Bantu reduplication. In K. Hanson and S. Inkelas, eds., The nature of the word: Essays in Honor of Paul Kiparsky. Cambridge, MA: MIT Press. Hyman, Larry, and Ngessimo Mutaka. 1990. Syllables and morpheme integrity in Kinande reduplication. Phonology 7: 73–119. Inkelas, Sharon, and Cheryl Zoll. 2000. Reduplication as morphological doubling. Rutgers Optimality Archive 412. Inkelas, Sharon, and Cheryl Zoll. 2002. Presentation, Graz Reduplication Conference. Jacobsen, William. 1964. A grammar of the Washo language. Doctoral dissertation, University of California, Berkeley. Kaye, Jonathan. 1985. On the syllable structure of certain west African Languages. In Didier Goyvaerts, ed., African Linguistics: Essays in Memory of M. W. K. Semikenke. Amsterdam: John Benjamins. 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, Eun-Sook. 2003. Theoretical Issues in Nuu-chah-nulth Phonology and Morphology. Doctoral dissertation, University of British Columbia. Kiparsky, Paul. 2003. Paradigms and Opacity. Unpublished book manuscript, Stanford University.

References

217

Kitagawa, Yoshihisa. 1987. Redoing reduplication: a preliminary sketch. In G. Huck and A. Ojeda, eds., Syntax and Semantics, Vol. 20, Discontinuous Geometry. Orlando, FL: Academic Press. Levin, Juliette. 1985. A Metrical Theory of Syllabicity. Doctoral dissertation, MIT. Lieber, Rochelle. 1992. Deconstructing Morphology. Chicago: University of Chicago Press. Lowenstamm, Jean, and Jonathan Kaye. 1986. Compensatory lengthening in Tiberian Hebrew. In L. Wetzels and E. Sezer, eds., Studies in Compensatory Lengthening. Dordrecht: Foris. Marantz, Alec. 1982. Re reduplication. Linguistic Inquiry 13 (3): 435–482. Marantz, Alec. 2001. Words. Unpublished, MIT. Marvin, Tatjana. 2002. Topics in the Stress and Syntax of Words. Dissertation, MIT. Martin, Jack, and Margaret Mauldin. 2000. A Dictionary of Creek/Muskogee, with Notes on the Florida and Oklahoma Seminole Dialects of Creek. Lincoln: University of Nebraska Press. McCarthy, John. 1983. Consonantal Morphology in the Chaha verb. In M. Barlow, D. Flickinger, and M. Wescoat, eds., Proceedings of the Second West Coast Conference on Formal Linguistics, Stanford Linguistics Association, Stanford, California. McCarthy, John. 1984. Ponapean reduplication. Unpublished, University of Texas. McCarthy, John. 1986. OCP effects: Gemination and antigemination. Linguistic Inquiry 17 (2): 207–263. McCarthy, John, and Alan Prince. 1986. Prosodic morphology. Unpublished manuscript, University of Massachusetts and Brandeis. Reprinted in J. Goldsmith, ed., 1999. Phononological Theory: The Essential Readings. Oxford: Blackwell, 1999. McCarthy, John and Alan Prince. 1995. Prosodic morphology. In J. Goldsmith, ed., The Handbook of Phonological Theory, 318–366. Oxford: Oxford University Press. McCarthy, John, and Alan Prince. 1995. Faithfulness and reduplicative identity. In J. Beckman, S. Urbanczyk, and L. Dickey, eds., Papers in Optimality Theory, 249–384. University of Massachusetts Occasional Papers in Linguistics 18. Amherst, MA: Graduate Linguistics Students Association, University of Massachusetts. (Available online on the Rutgers Optimality Archive website: http://roa.rutgers.edu/view.php3?id=568. Page reference is to the online version.) Merlan, F. 1982. Mangarayi. Lingua Descriptive Series 4: 1–242. Mester, Armin. 1986. Studies in Tier Structure. Doctoral dissertation, University of Massachusetts, Amherst. Moravcsik, Edith. 1978. Reduplicative constructions. In Joseph Greenberg, ed., Universals of Human Language, vol. 3: Word Structure, 297–334. Stanford, CA: Stanford University Press. Mutaka, Ngessimo. 2000. Interaction of imbrication and reduplication in Kinande. Journal of West African Languages 18, no. 1, 2000/2001: 33–47.

218

References

Myers, Scott. 1987. Tone and the structure of words in Shona. Doctoral dissertation, University of Massachusetts, Amherst. Nash, David. 1979. Yidiny stress: A metrical account. CUNY Forum 7/8 (NELS 9): 112–130. Nash, David. 1980. Topics in Warlpiri Grammar. Doctoral dissertation, MIT. Newell, Heather. 2004. The phonological phase. Montreal-Ottawa-Toronto Phonology Conference (MOT), Ottawa, Ontario, February 2004. Newman, Paul. 2000. The Hausa Language. New Haven, CT: Yale University Press. Nunes, Jairo. 2004. Linearization of Chains and Sideways Movement. Cambridge, MA: MIT Press. Odden, David, and Mary Odden. 1985. Ordered reduplication in K´hehe. Linguistic Inquiry 16 (3): 497–503. Onn, Farid. 1976. Aspects of Malay Phonology and Morphology: A Generative Approach. Doctoral dissertation, University of Illinois, Urbana. Payne, David. 1981. The Phonology and Morphology of Axininca Campa. Summer Institute of Linguistics Publications in Linguistics, No. 66, University of Texas, Arlington. Pensalni, Rob, and Gavan Breen. 1999. Arrente: A language with no syllable onsets. Linguistic Inquiry 30 (1): 1–25. Pesetsky, David. 1979. Russian morphology and lexical theory. Unpublished manuscript, MIT. Prince, Alan. 1984. Phonology with tiers. In M. Aronoff and R. Oehrle, eds., Language Sound Structure. Cambridge, MA: MIT Press. Prince, Alan, and Paul Smolensky. 1993. Optimality theory: Constraint interaction in generative grammar. Technical Report RuCCS TR-2, Rutgers University Center for Cognitive Science. Pulleybank, Douglas. 1986. Tone in Lexical Phonology. Dordrecht: D. Reidel. Raimy, Eric. 2000a. The morphology and morphophonology of reduplication. Berlin: Mouton de Gruyter. Raimy, Eric. 2000b. Deconstructing prosodic morphology. Talk given at the Eastern States Conference on Linguistics, 2000. Rehg, Kenneth, and Damien Sohl. 1981. Ponapean Reference Grammar. Honolulu: University of Hawaii Press. Riggle, Jason. 2004. Nonlocal reduplication. In K. Moulton and M. Wolf, eds., Proceedings of NELS 34. Amherst: GLSA, University of Massachusetts. Rose, Sharon. 2003. Triple take: Tigre and the case of internal duplication. Department of Linguistics, UCSD. San Diego Linguistic Papers, Issue 1. Paper 5. Sagey, Elizabeth. 1988. On the ill-formedness of crossing association lines. Linguistic Inquiry 19 (1): 109–118.

References

219

Sapir, Edward. 1922. The Takelma language of Southwestern Oregon. In Franz Boas, ed., Handbook of American Indian Languages, Bureau of American Ethnology, Bulletin 40, Part I2:1–296. Saxton, Dean, Lucille Saxton, and Susie Enos. 1983. English-Papago/Pima Dictionary, 2nd ed. Tucson: University of Arizona Press. Schein, Barry, and Donca Steriade. 1986. On geminates,. Linguistic Inquiry 17 (4): 691–744. Shaw, Pat. 1992. Templatic evidence of the syllable nucleus. In Amy Schafer, ed., Proceedings of NELS 23. Amherst: GLSA, University of Massachusetts. Sibanda, Galen. 2004. Verbal Phonology and Morphology of Ndebele. Doctoral dissertation, University of California, Berkeley. Sloan, Kelly. 1988. Bare-consonant reduplication. In Hagit Borer, ed., Proceedings of the Seventh West Coast Conference on Formal Linguistics, 319–330. Stanford, CA: CSLI, Stanford University. Sommerstein, Alan. 1974. On phonotactically motivated rules. Journal of Linguistics 10: 71–94. Spring, Cari. 1990. Implications of Axininca Campa for prosodic morphology and reduplication. 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, CT: American Oriental Society. Stonham, John. 1990. Current Issues in Morphological Theory. Doctoral dissertation, Stanford University. Uhrbach, Amy Jill. 1985. Some facts of reduplication in Bahasa Indonesia. Unpublished manuscript, University of Texas, Austin. Urbanczyk, Suzanne. 1992. Inxing and melodic circumscription. Unpublished manuscript, University of Massachusetts, Amherst. Urbanczyk, Suzanne. 1996. Patterns of reduplication in Lushootseed. Doctoral dissertation. University of Massachusetts. Urbanczyk, Suzanne. 2000. The bases of double reduplication. In R. Billery and B. Lillehaugen, eds., Proceedings of WCCFL 19, 518–531. Somerville, MA: Cascadilla Press. Whitney, William Dwight. 1885. The Roots, Verb-forms, and Primary Derivatives of the Sanskrit Language. Reprinted as: American Oriental Series 30 (1945). New Haven, CT: American Oriental Society. Wiesemann, U. 1972. Die phonologische und grammatische Struktur der Kaingang Sprache. The Hague: Mouton. Wilbur, Ronnie Bring. 1973. The Phonology of Reduplication. Bloomington: Indiana University Linguistics Club.

220

References

Wilkinson, Karina. 1988. Prosodic structure and Lardil phonology. Linguistic Inquiry 19 (2): 325–334. Winter, Werner. 1970. Reduplication in Washo: A restatement. International Journal of American Linguistics 36 (3): 190–198. Yang, Charles. 2002. Knowledge and Learning in Natural Language. Oxford: Oxford University Press. Yu, Alan. 2001. Washo internal duplication revisited. Unpublished manuscript. University of California, Berkeley. Zepeda, Ofelia. 1983. A Papago Grammar. Tucson: University of Arizona Press. Zoll, Cheryl. 2002. Vowel reduction and reduplication in Klamath. Inquiry 33 (3): 520–527.

Linguistic

Languages

Agta, 12, 55, 55n5, 89, 95–96, 98–99 Arabic, Classical, 181 Arrernte, 56–57, 63–64, 149, 151 Asheninca Campa, 12, 83, 92, 117, 128, 140-144, 140n15, 197 Bella Coola, 75 Chaha, 12, 41, 55, 117, 178–189 Choctaw, 62–63, 192 Chukchee, 53–54, 111, 114 Chumash, 83–88 Creek, 70–71 Diyari, 64, 73 Endegen, 185 English, 1, 3, 35–37, 39, 44, 47, 55, 63, 100, 126, 133 Erromangan 38–39, 44–49 Ezha, 185 German, 2, 44 Hausa, 79–80 Hebrew, Biblical, 21–22 Ilocano, 3, 7–8, 12, 38–40, 89, 95–96, 98–101 Imdlawn Tashlhiyt Berber, 15n1, 76– 77 Kaingang, 112–114 Kinande, 12, 62, 117–118, 127–128, 133–140, 137n13, 139n14 Klamath, 40–42, 67 Kolami, 78–79 Korean, 28–31, 61, 113 K´hehe, 81–83

Lardil, 105–108, 105n7, 170 Levantine Arabic, 157, 159–161 Lushootseed, 12, 173–177 Madurese, 53, 111–114 Malay, 8–9, 19–20, 23, 25–26, 26n8, 28, 30, 189, 195 Manam, 1, 74 Mangarayi, 3, 5, 12, 70, 146, 149 Mokilese, 1, 6, 12, 59, 61, 88–90, 93– 99, 95n4, 114–115, 120, 170, 175 Ndebele, 12, 58, 117–35, 121n3, 137– 38, 170 Nuu-chah-nulth, 69, 69n2, 105, 109 Orokaiva, 75–76 Ponapean, 12, 79, 89, 95–96, 101–105, 108–110, 116 Sanskrit, 12, 61, 69, 108–109 117, 161–172, 164n22, 168n23 Semai, 60–61, 157–159, 166, 169 Southern Paiute, 22–123 Temiar 117, 157–159, 183 Tigre, 183–184 Tohono O’odham, 69, 109, 117, 152– 157, 155n21 Ulwa, 73–74 Washo, 12, 144–152 Yaqui 64–65, 105, 107–110, 111n8, 116, 168 Yidiny, 11–12, 72–73 Yoruba, 54–55, 193

Authors

Abad, M., 3, 7, 38, 89, 100 Applegate, R., 83 Austin, P., 73 Bagemihl, B., 75 Banksira, D., 178, 181–185, 187–188 Barker, M., 42 Benjamin, G., 158 Blanco, A., 73 Blevins, J., 95 Boersma, P., 99 Booker, K., 70  129n6 Boskovi´c, Z, Breen, G., 56, 63 Bromberger, S., 73 Broselow, E., 11, 72, 144, 151, 157, 159 Calabrese, A., 24 Chittleborough, M., 76 Chomsky, N., 90, 129n6 Chung, C., 28 Clements, G., 8, 20n5, 32n9 Coleman 32n9 Crowley, T., 44 Davis, R., 35n2 Davis, S., 104, 104n5 Dell, F., 76 Demers, R., 108 Diffloth, G., 60 Dixon, R., 72 Downing 121n3, 122n4, 133n10, 139n14 Eisner, J., 192

Elmedlaoui, M., 76 Emeneau, M., 78 Enos, S., 153 Escalante, F., 108 Fitzgerald, C., 154 Frampton, J., 7n3, 20n3, 68n1, 91–92, 92n3, 129n6, 155 Gafos, A., 157, 180 Goldsmith, J., 5, 17–19, 92n3 Gutmann, S., 20n3 Haas, M., 70 Hale, K., 73 Halle, M., 16, 13n2, 20n3, 24–25, 24n7, 33, 35, 35n2, 49, 73, 75, 90, 116, 148, 163 Harris, J., 13n2 Harrison, S., 89 Haugen, J., 64, 111n8 Hayes, B., 3, 7, 21, 38, 89, 99–100, 105n6 Healy, A., 55, 76 Hudson, G., 178–179, 180 Hyman, L., 117, 133 Idsardi, W., 16, 116 Inkelas, S., 27, 47, 83, 86–87, 117, 130, 139n14 Isoroembo, A., 76 Jacobsen, W., 144, 146 Jelinek, E., 108 Kaye, J., 32n9 Kenstowicz, M., 21–22, 181–182, 188 Keyser, S., 32n9

224 Kim, E., 69 Kiparsky, P., 23, 27 Kitagawa, Y., 7, 8n4, 13, 20n5, 29 Levin, J., 88–89 Lieber, R., 48 Local, J., 32n9 Lowenstamm, J., 32n9 Marantz, A., 6, 10–11, 19, 33, 35, 35n2, 49, 53–54, 81, 85, 88, 122n5, 144 Martin, J., 70 Marvin, T., 122n5 Mauldin, M., 70 McCarthy, J., 1, 6, 10–11, 13, 17–19, 26n8, 43n4, 62, 65n2, 72–73, 76, 78, 84, 89, 94, 101–102, 117, 140, 144, 151, 157, 159, 178–181, 189, 195 Merlan, F., 3, 70 Mester, A., 1, 8, 20, 20n5 Moravcsik, E., 10, 11 Mutaka, N., 133, 139 Myers, S., 122n4 Nash, D., 11, 72 Newell, H., 122n5 Newman, P., 79 Nunes, J., 129n6 Odden, D., 13, 81 Odden, M., 13, 81 Onn, F., 8, 26, 26n8 Payne, D., 140, 140n15, 144 Pensalni, R., 56, 63 Pesetsky, D., 35 Prince, A., 1, 6, 11, 26n8, 32n9, 43n4, 62, 65n2, 72–73, 76, 78, 84, 89,

Authors 92n3, 94, 101, 117, 140, 189 Pulleybank, D., 32n9 Raimy, E., 7, 13, 98, 117, 152, 155, 157, 195, 200 Rehg, K., 101, 104 Riggle, J., 70 Rose, S., 183 Sagey, E., 5, 19, 191 Saxton, D., 153 Saxton, L., 153 Schein, B., 21 Shaw, P., 65n2 Sibanda, G., 117, 121n2 Sloan, K., 157 Smolensky, P., 92n3 Sohl, D., 101, 104 Spring, C., 140, 140n15, 144 Steriade, D., 12–13, 21, 112, 117, 161, 168n23 Stevens, A., 53 Stonham, J., 65n2 Uhrbach, A., 8, 20n5 Urbanczyk, S., 63, 173, 176 Vergnaud, J-R., 75 Whitney, W., 172 Wiesemann, U., 112 Wilbur, R., 1, 9, 189 Wilkinson, K., 106 Winter, W., 144, 146 Yang, C., 44 Yu, A., 144, 151 Zepeda, O., 153 Zoll, C., 27, 40, 47, 83, 86–87, 117, 130, 139n14

Subjects

[[. . .]], 36 C∗ν-rule, 71, 109, 153, 170 C∗ V-rule, 40, 63, 71, 75–77, 89–90, 93, 108–110 ×∗-rule, 71–72, 74, 106, 111-112, 160 [ , Left duplication juncture. See Duplication juncture  , Left truncation juncture,. See Truncation juncture ] , Right duplication juncture. See Duplication juncture  , Right truncation juncture. See Truncation juncture ×-rule, 71, 76–77 Allomorphy, 35n2, 48–49, 64, 129, 132, 144, 149 suppletive, 48–49 Autosegmental representation, as privilege of phonology, 191 Bare timing slot, 64, 95, 147–148, 152 Base of reduplication, 1, 3, 5–7, 10–12, 48, 84n1, 84–85, 88, 131, 139n14, 183, C-nality, 105–107, 108–109, 115, 175–176 Complementarity, 97, 101–102 Constraint, 10, 16–18, 20–21, 24– 25, 35n2, 49, 65n2, 91–92, 92n3, 139n14, 156–157, 188. See also No Crossing Constraint on possible grammars, 19, 25 derivational, 10, 16, 19–20, 24–25,

90–91, 94, 102 Correspondence Theory, 23, 43n4, 84, 84n1, 180, 188, 200 Crossed association lines, 16–17 Crossed structure, 10, 15, 21, 24, 26, 28, 198, 199 Crossing violation. See NCC violation CT. See Correspondence Theory Cyclic affix, 43 Cyclic Phonology, 34–35, 131 Cyclic rule, 30–31, 35, 35n2, 53n3, 91, 99, 120, 129, 133, 150 DC. See Default Closure Default Closure, 40, 42, 57, 61, 70–71, 75, 81, 89, 97, 106, 119, 125, 133, 135, 142, 158–160, 167, 180, 184 Default leftmost juncture insertion, 40, 167 Direction of transcription, 15–16, 42, 53, 59, 67, 77, 176 default, 15, 77, 106, 153 Distributed Morphology, 12, 33, 36 Domain of a juncture insertion rule, 12, 39–40, 42, 55, 56, 66, 68, 70– 72, 110, 119, 125. See also Domain of a reduplicative affix default 40, 42, 71 Domain of a reduplicative affix, 71–77, 102–103, 106, 116, 133, 136, 142, 156, 158–160, 183–184 Doubly reduplicated monosyllables, 61, 67, 96–98, 128, 134-135

226 DR, Distributed Reduplication Duplicant, 1, 53, 58, 62, 67, 72, 81–83, 85, 87–89, 92–94, 96–97, 99–102, 106, 108–116, 119–120, 122–128, 130, 132–140, 144–145, 147, 149, 152n20, 160–161, 166–168, 170– 172, 174–175, 198 embedded in another duplicant, 60– 61, 77n3, 93, 166 nontrivial, 70 Duplication juncture, 51–52, 57, 143, 158 unpaired 70 Embedded word boundary, 55 English past-tense morphology, 35, 44 Epenthetic vocalism, 181 Exponent (of a lexical item), 1, 33n1, 34, 36–38, 40, 54–56, 62–63, 67, 69, 73–74, 111, 184, 187–188, 191, 199, 202 null, 40, 42, 70, 72–74, 76, 89, 106, 111-113, 150, 153, 159 FCVE. See First Conjunct Vowel Epenthesis FCVL. See First Conjunct Vowel Lengthening Feature sharing, constraints on, 24–25 Feature spreading, constraints on, 10, 18, 21, 24–25, 24n7 First Conjunct Vowel Epenthesis, 174– 175, 177 First Conjunct Vowel Lengthening, 59, 89–90, 93–94, 97, 99–100, 103, 106, 110, 114, 120, 170–171 universally available 110 First conjunct 30, 43, 92–94, 97, 112, 123–125, 135–136, 142–143, 145, 167, 170–171, 175 bisyllabic 119-120, 122, 127, 132, 134 bimoraic, 93, 97, 103, 106, 114–115 C-initial, 108, 112, 125, 136 syllable structure, 92, 94, 107, 112, 119–120, 142–143, 146, 170 Fission, 5, 18–19, 22, 25, 32, 41, 77– 80, 105, 149, 169, 185–186, 193– 194, 201

Subjects Floating segments, nonexistence of, 32, 192 GDR. See Goal-driven rule Geminate 6, 8–10, 15n1, 19, 76–77, 95, 98, 103–105, 104n5, 176, 185, 188, 191, 196 long-distance, 10, 21, 178, 180–181, 195 Geminate inalterability, 21–23, 28, 30, 41, 68, 179, 182–183 and spirantization, 21 and spreading, 24 Goal-driven rule, 90–94, 92n3, 97, 114–115 Grammar, 49n6, 77n4, 91, 168n24 constraints on 19, 25 inuence of phonotactics on, 155 Hiatus avoidance, 78–79, 100, 125, 129, 148, 155 Immediate repair, 17-19, 24 versus eventual repair, 19 Inxation, 8, 12, 62–63, 86–88, 157– 158, 202 Juncture insertion rules insertion site, 12, 39, 68, 103, 159– 160 inventory of, 68 common, 71 Juncture insertion, 5–7, 10–12, 29, 33n1, 38–40, 42, 54–58, 61, 63, 67– 72, 74–75, 77–78, 82, 87, 96–97, 103, 106, 108, 110–113, 116, 119, 123, 125–126, 128, 134, 136, 145– 147, 149, 152–153, 152n20, 157– 160, 166–168, 171–172, 174–175, 183, 195, 202–203 Leading-edge, 53, 58 truncation, 52–54, 57–58, 60, 62, 145n17 Lexical insertion, 33–38, 41, 45, 48n5 Lexical item, 1, 34–35, 37–38, 40, 42, 44, 54, 192. See also Vocabulary item Lexical Phonology, 35 Lookahead, 19, 91 m-structure, 33-34, 37–38

Subjects MDT. See Morphological Doubling Theory Melody copy, 144, 144n16 Metathesis, 12, 32, 62–63, 71n3, 181, 194 Monosyllabic root/stem, 29, 57, 64, 96–97, 99–101, 135, 140n15, 145, 149–150, 150n18, 152, 165 Moraic consonant, 184 Moravcsik’s Generalization, 10–11, 29, 109–110, 111n8, 113 Morpheme, 2–5, 16, 20n6, 33–42, 33n1, 36n3, 44–45, 48, 53, 57, 63, 72–73, 84–85, 87, 95, 101, 103, 111–113, 119, 129, 131–133, 138, 140, 142, 145, 153, 157, 159, 165, 172, 179, 184–185, 187–188 Morphological Doubling Theory, 87, 117, 130–131 Morphological conditioning, 37, 41, 46, 49, 152–154, 156, 185, 188 Morphology, 1, 4–7, 9, 12, 17–18, 33, 35, 38, 42, 44, 51–54, 59, 62–63, 82, 116, 128, 133, 160, 183, 185. See also Distributed Morphology; Prosodic Morphology concatenative, 2, 6, 11, 145, 165, 172, 202 nonconcatenative, 2–3, 10, 13, 17, 37, 202 Semitic 10, 17, 180, 197 Motor system, 194 instructions to, 191–192 Music score, 4–5 NCC repair, 5–6, 8–9, 20–21, 24–25, 28, 30–32, 39, 41, 43, 77–79, 104– 105, 130, 132–133, 149, 152, 169, 176–177, 181, 184, 186, 194, 198199. See also NCC repair; RC repair Shortcut Repair, 77–80, 104–105 NCC. See No Crossing Constraint NCCR. See NCC repair No Crossing Constraint (NCC), 2, 5, 7n3, 10, 16–19, 20n6, 24–25, 31, 77, 169, 179–180, 191–194, 198 Noncyclic affix, 35–36, 82, 172 Nonmoraic coda consonant, 108

227 Nuclear skeleton, 68, 68n1, 152n20 Nuclear structure 103, 161-165, 170– 171 Occurrence of a phoneme, 18, 22–23, 41, 79, 107, 148, 178–179, 187–188, 195–196, 196n1 in multiple environments, 196, 196n1 Onset-coda asymmetry, 113 Onset Permanence, 108, 168, 170–171 Onsetless syllable, 56, 76, 83, 112, 119, 126, 136, 140, 142–143, 197 OnsPerm (constraint). See Onsert Permanence Opaque process, 77, 184–185, 188– 189, 196 Optimality Theory (OT), xi, 20n4, 40, 54, 84, 92, 92n3, 97–98, 100, 133n10, 155–157, 180, 188, 200 Overapplication, 6, 9–10, 22-23, 28, 199 of nasalization in Malay 8–9, 30, 195 of n→r in Korean, 30–31 of labialization in Chaha, 178–80 of x-dissimilation in Chaha, 181–182, 187–189 of RVR in Klamath, 41 of imbrication in Kinande, 132-133 Parax theory, 8, 13n5 Phoneme tier, unordered, 32, 98, 192 Phonology-phonetics Interface, 5, 18– 19, 20n6, 25, 191–193 condition, 2, 5, 19, 24–25, 31, 191 Phonotactic, 91, 101, 121, 121n3, 131, 155, 188, Possible conditions on repair, 116 Possible juncture insertion rules, 68, 70 Postcyclic rules, 30–31, 35 PrAdj (rule). See Prosodic adjustment Precedence relations, 7 Primacy of weight goals, 108 Prosodic Adjustment, 7–8, 12, 55n5, 58–59, 81–83, 89–90, 92–97, 99, 101–103, 105–116, 119–120, 122–128, 133–137, 139, 142–145, 152n20, 153–154, 156, 161, 166– 168, 170–176, 198 sensitive to stem prosody, 103, 108–

228 109, 111 restricting, 115 Prosodic subword, 11, 56 Prosodic circumscription, 63, 72 Prosodic minimality condition, 72 Prosodic Morphology, 2, 6, 10–11, 56, 63, 72, 89, 116, 180, 197 R-representation, 195, 197–199 RC. See Retraction Condition RC. See Retraction Condition Readj (rule). See Readjustment Readjustment, 35–39, 35n2, 36n3, 41– 42, 44-45, 47–49, 48n5, 52, 64, 70, 81, 132, 158–159, 199, Recoverability of deletion, 79 Reduplicant, 3, 6–8, 11-12, 23, 39, 48, 52, 58–59, 62, 64, 79, 84–88, 84n1, 92–93, 92n2, 96–97, 104, 109-110, 116, 118, 120, 121n3, 122-123, 128–129, 130–133, 135, 139n14, 142, 146, 149, 163, 165, 167–170, 172, 174, 176 -remnant boundary, 28, 77–79, 104 Reduplication consecutive syllable, 28–29 consonant copy, 60, 157 double, 61, 96 echo, 78 nal syllable prexation, 111 nal-syllable suffixation, 112 heavy syllable prexation, 12, 88–101 heavy syllable suffixation, 111 inherent (root) 180 initial consonant, 174 internal, 142, 144 leading-edge foot 72, 74 “light” (in Ilocano) 3, 38 multiple 173 syllable copy, 10–12, 111n8, 116 total, 42, 44, 71 truncated, 12, 51 Regressive velar assimilation, nonoverapplication of, 27–28 Repair, 5, 17–19, 25, 77, 79, 95, 103– 106, 110, 116, 140, 169. See also NCC repair; Shortcut repair immediate versus eventual, 17–18

Subjects Retraction Condition repair, 32, 194 Resyllabication Condition, 163–164 Retraction Condition, 31–32, 191–194 Retraction onto the timing tier, 192 Second conjunct, 30, 93, 111, 114, 125, 127, 153, 156 Shortcut Repair, 77–80, 104–105 Stem, dened 35 String copy, 51, 58, 61 t-juncture, 4–6, 16, 38, 53–55, 54n4, 67, 71, 74, 81, 89, 115–116, 122, 130, 180, 202–203 Template, 6, 8, 11–12, 17–18, 72, 77, 77n4, 81–82, 84, 87–89, 179–180, 185, 188 Temporal simultaneity, 31, 191–192 Tier Conation, 18, 198 Trailing edge, 58 truncation, 57–59, 62, 145n17 Transcription, 4–10, 7n3, 12, 15–16, 15n1, 15n2, 20, 23, 25, 26n6, 27– 30, 38, 41–42, 52–54, 53n3, 57–64, 67, 70–71, 74, 77, 77n4, 80–82, 119–120, 122, 124–125, 130, 132– 134, 138, 142, 145, 145n17, 147, 149, 153, 159–160, 166, 169–170, 172–74, 183–184, 189, 191–192, 194–195, 197–199, 205 factor, 4 of genes, 4 Trscr, see Transcription, 38 Truncated a-epenthesis (a-Ep), 120– 121, 123, 128, 134, 170–171 Truncation juncture, 52–53 T&R (Transcription and Repair), 41, 43, 98, 103 Underapplication, 6, 9, 22, 41 Vocabulary choice, 33, 150n18 Vocabulary item, 33–37, 33n1, 88, 145. See also Lexical item Vowel hiatus, 78, 100, 105n6, 153, Well-formed syllable structure, 12, 82, 92, 98, 112–115, 119–120, 121n3, 123–124, 142–144, 146, 198 WFSS. See Well-formed syllable structure

Linquistic Inquiry Monographs Samuel Jay Keyser, general editor 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38.

Word Formation in Generative Grammar, Mark Aronoff Syntax: A Study of Phrase Structure, Ray Jackendoff Recent Transformational Studies in European Languages, S. Jay Keyser, editor Studies in Abstract Phonology, Edmund Gussmann An Encyclopedia of AUX: A Study of Cross-Linguistic Equivalence, Susan Steele Some Concepts and Consequences of the Theory of Government and Binding, Noam Chomsky The Syntax of Words, Elisabeth O. Selkirk Syllable Structure and Stress in Spanish: A Nonlinear Analysis, James W. Harris CV Phonology: A Generative Theory of the Syllable, George N. Clements and Samuel Jay Keyser On the Nature of Grammatical Relations, Alec P. Marantz A Grammar of Anaphora, Joseph Aoun Logical Form: Its Structure and Derivation, Robert May Barriers, Noam Chomsky On the Denition of Word, Anna-Maria Di Sciullo and Edwin Williams Japanese Tone Structure, Janet Pierrehumbert and Mary E. Beckman Relativized Minimality, Luigi Rizzi Types of A-Dependencies, Guglielmo Cinque Argument Structure, Jane Grimshaw Locality: A Theory and Some of Its Empirical Consequences, Maria Rita Manzini Indenites, Molly Diesing Syntax of Scope, Joseph Aoun and Yen-hui Audrey Li Morphology by Itself: Stems and Inectional Classes, Mark Aronoff Thematic Structure in Syntax, Edwin Williams Indices and Identity, Robert Fiengo and Robert May The Antisymmetry of Syntax, Richard S. Kayne Unaccusativity: At the Syntax-Lexical Semantics Interface, Beth Levin and Malka Rappaport Hovav Lexico-Logical Form: A Radically Minimalist Theory, Michael Brody The Architecture of the Language Faculty, Ray Jackendoff Local Economy, Chris Collins Surface Structure and Interpretation, Mark Steedman Elementary Operations and Optimal Derivations, Hisatsugu Kitahara  The Syntax of Nonnite Complementation: An Economy Approach, Zeljko Boskovi´c Prosody, Focus, and Word Order, Maria Luisa Zubizarreta The Dependencies of Objects, Esther Torrego Economy and Semantic Interpretation, Danny Fox What Counts: Focus and Quantication, Elena Herburger Phrasal Movement and Its Kin, David Pesetsky Dynamic Antisymmetry, Andrea Moro

39. 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50. 51. 52.

Prolegomenon to a Theory of Argument Structure, Ken Hale and Samuel Jay Keyser Essays on the Representational and Derivational Nature of Grammar: The Diversity of Wh-Constructions, Joseph Aoun and Yen-hui Audrey Li Japanese Morphophonemics: Markedness and Word Structure, Junko Ito and Armin Mester Restriction and Saturation, Sandra Chung and William A. Ladusaw Linearization of Chains and Sideward Movement, Jairo Nunes The Syntax of (In)dependence, Ken Sar Interface Strategies: Optimal and Costly Computations, Tanya Reinhart Asymmetry in Morphology, Anna Maria Di Sciullo Relators and Linkers: The Syntax of Predication, Predicate Inversion, and Copulas, Marcel den Dikken On the Syntactic Composition of Manner and Motion, Maria Luisa Zubizarreta and Eunjeong Oh Introducing Arguments, Liina Pylkk¨anen Where Does Binding Theory Apply?, David Lebeaux Locality in Minimalist Syntax, Thomas S. Stroik Distributed Reduplication, John Frampton