The Nature of the Word: Studies in Honor of Paul Kiparsky (Current Studies in Linguistics)

Uif!Obuvsf!pg! !uif!Xpse STUDIES IN HONOR OF PAUL KIPARSKY

EDITED BY

Kristin Hanson and Sharon Inkelas

The Nature of the Word

Current Studies in Linguistics Samuel Jay Keyser, general editor A complete list of books published in the Current Studies in Linguistics series appears at the back of this book.

The Nature of the Word Studies in Honor of Paul Kiparsky

edited by Kristin Hanson and Sharon Inkelas

The MIT Press Cambridge, Massachusetts London, England

6 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. For information about special quantity discounts, please e-mail [email protected] This book was set in Times New Roman and Syntax on 3B2 by Asco Typesetters, Hong Kong. Printed and bound in the United States of America. Library of Congress Cataloging-in-Publication Data The nature of the word : studies in honor of Paul Kiparsky / edited by Kristin Hanson and Sharon Inkelas. p. cm. — (Current studies in linguistics) Includes bibliographical references and index. ISBN 978-0-262-08379-9 (hardcover : alk. paper) — ISBN 978-0-262-58280-3 (pbk. : alk. paper) 1. Grammar, Comparative and general. 2. Lexicology. 3. Word (Linguistics) 4. Kiparsky, Paul. I. Kiparsky, Paul. II. Hanson, Kristin. III. Inkelas, Sharon. P201.N375 2009 415—dc22 2007041980 10

9 8 7 6 5

4 3 2 1

To see a World in a Grain of Sand And a Heaven in a Wild Flower Hold Infinity in the palm of your hand And Eternity in an hour —William Blake, ‘‘Auguries of Innocence,’’ 1–4

Paul Kiparsky. Photo by Cleo Condoravdi.

Contents

Preface xi Biography of Paul Kiparsky

xv

I

Metrics

1

1

On ‘‘The Phonological Basis of Sound Change’’ (after Hopkins)

3

Andrew Garrett 2

On Stress and Meter and on English Iambics in Particular

5

Morris Halle 3

Nonlexical Word Stress in the English Iambic Pentameter: A Study of John Donne 21

Kristin Hanson 4

The Prosodic Word as a Unit in Poetic Meter

63

Draga Zec 5

The Word in Tiberian Hebrew

95

Bezalel Elan Dresher 6

Faithfulness and Componentiality in Metrics

113

Bruce Hayes II

Phonology and Morphology

7

The Phonology of Perceptibility Effects: The P-Map and Its Consequences for Constraint Organization 151

Donca Steriade

149

viii

8

Contents

Vowel Duration, Syllable Quantity, and Stress in Dutch

181

Carlos Gussenhoven 9

Sympathy Meets Argentinian Spanish

199

Ellen M. Kaisse 10

Vowel Length, Cyclicity, and Output-Output Correspondence

215

Cemil Orhan Orgun 11

Level Ordering in Nuuchahnulth

225

John Stonham 12

Inside Access: The Prosodic Role of Internal Morphological Constituency

241

Patricia A. Shaw 13

Morphosyntactic Correspondence in Bantu Reduplication

273

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda 14

Patterns of Reduplication in Yoruba

311

Douglas Pulleyblank 15

Multiple Tenses in the Malayalam Verb

359

Tara Mohanan and K. P. Mohanan On Pa¯nini 2.4.81 (a¯mah) ˙ ˙

387

III

The Lexicon and Change

395

17

Lexical Irregularity and the Typology of Contrast

16

S. D. Joshi and J. A. F. Roodbergen

397

Rene´ Kager 18

Derived Environment Effects in Colloquial Helsinki Finnish

433

Arto Anttila 19

A Historical Perspective on Nonderived Environment Blocking: The Case of Korean Palatalization 461

Young-mee Yu Cho 20

Lexical Storage and Phonological Change

Geert Booij

487

Contents

21

ix

The Dental Preterites in the History of English

507

Aditi Lahiri 22

Analogical Morphophonology

527

Andrew Garrett and Juliette Blevins IV

Syntax and Semantics

547

23

Second-Position Clitics in Tagalog

549

Stephen R. Anderson 24

‘‘Elsewhere’’ in Gender Resolution

567

Stephen Wechsler 25

The Force of Lexical Case: German and Icelandic Compared

Dieter Wunderlich 26

Constraints on Source/Goal Co-occurrence in Carrier

William J. Poser 27

Punctual Until as a Scalar NPI

631

Cleo Condoravdi 28

The Existential Tense in Hungarian

Christopher Pin˜o´n V

Poetics

675

29

The Art of Fusion

677

Haj Ross References 697 Contributors 743 Language Index 745 Subject Index 747

655

621

587

Preface

This book honors Paul Kiparsky, whose contributions as a scholar and teacher have transformed virtually every subfield of contemporary linguistics. In taking the form of a collection of studies of the word, it reflects the distinctive focus of his own attention, and the consequent shape of his influence. The word has traditionally been recognized as a fundamental entity of language. Ordinary speakers often identify knowing a language with knowing its words, children attend to and acquire words before any other structural units in a language, and when one language influences another it is most commonly its words that are borrowed. From a more scholarly perspective too, the word has for centuries occupied a central place as the repository of basic phonological patterns, morphological structures, syntactic capabilities, semantic content, correspondences allowing the reconstruction of historical relationships among languages, and poetic possibilities. Kiparsky’s work belongs to this tradition, but in a way unique to him. As T. S. Eliot writes of poets in ‘‘Tradition and the Individual Talent,’’ If the only form of tradition, of handing down, consisted in following the ways of the immediate generation before us in a blind or timid adherence to its successes, ‘‘tradition’’ should positively be discouraged. We have seen many such simple currents soon lost in the sand; and novelty is better than repetition. Tradition is a matter of much wider significance. It cannot be inherited, and if you want it you must obtain it by great labour. It involves, in the first place, the historical sense . . . ; and the historical sense involves a perception, not only of the pastness of the past, but of its presence; the historical sense compels a man to write not merely with his own generation in his bones, but with a feeling that the whole of the literature of Europe from Homer and within it the whole literature of his own country has a simultaneous existence and composes a simultaneous order. This historical sense, which is a sense of the timeless as well as of the temporal and of the timeless and of the temporal together, is what makes a writer traditional. And it is at the same time what makes a writer most acutely conscious of his place in time, of his contemporaneity.

Kiparsky’s ‘‘place in time’’ is in the first group of PhD students graduated in 1965 from the new program in linguistics at MIT, a program directed by Morris Halle and committed to developing Noam Chomsky’s revolutionary definition of language

xii

Preface

as a generative system, and the hypothesis that crucial aspects of its formal structure are universal because they are innately determined. Kiparsky’s ‘‘historical sense’’, it seems to us, has compelled him to express within that revolutionary paradigm ‘‘a feeling’’ that the whole of language and within it the whole of each language ‘‘has a simultaneous existence and composes a simultaneous order.’’ It is labor to obtain Kiparsky’s tradition within the revolution that has defined his particular intellectual influence that is represented here. At the outset of the early generative period, theoretical interest in productive rules in both syntax and phonology drew attention away from words, associated with the lexicon and memorized forms; but within only a few years, representing systematicity within the lexicon itself and relating the lexicon to other subsystems of grammar emerged as essential to generative theory. Di‰culties in defining the word vexatious to everyone from Scrabble players to dictionary editors to poets turned out to reflect real complexities in the structure of grammar that had to be taken into account: Is a compound word like blackbird one word or two? Why is the ‘‘g’’ of longer pronounced like that of a simple word like finger rather than that of a complex word like longing? Why do poets not treat the stress of a preposition like between quite the same way that they treat that of a verb like believe? How can we know that a word that expresses such a dazzling set of complex syntactic and semantic relationships as the Finnish adverb voimisteluttelemasta, ‘from having on and o¤ caused to do gymnastics’, really is just one word, and not the nine that English needs to convey the same meaning? Kiparsky’s serious attention to questions like these has shown the word to be as central to grammar as tradition had always suggested. His work on formalizing its role, from his theory of Lexical Morphology and Phonology to his theories of morphosyntax and of the role of analogy in sound change, has been fundamental to the development of lexicalist approaches to every subsystem of grammar within the generative tradition. The range of Kiparsky’s influence derives not only from his rigorous focus on such an intricate and central entity as the word, but also from another intellectual commitment, equally represented in this book’s title. Chomsky’s definition of language famously cast linguistics as a science, and its object of study as a phenomenon of nature. In Kiparsky this idea must have fallen on especially fertile soil, as it were; for, as fine a gardener as he is a linguist, Kiparsky approaches words rather as a botanist approaches plants, fascinated equally by their beauty, by their structure, and by their evolution, and aware that only by considering them from these multiple perspectives can one begin to understand their true nature. The contributions to this volume represent these multiple perspectives on the word that Kiparsky has explored and inspired, and are organized around the several complementary senses of the word nature that they seem to reflect. Part I, ‘‘Metrics,’’ recognizes the nature of the word not only as a source of beauty, but also as something

Preface

xiii

that is not artifice, a recognition whose profound significance for literary theory Kiparsky’s own work has drawn out. Part II, ‘‘Phonology and Morphology,’’ explores the nature of the word as a matter of its formal composition, precisely the sense in which morphology refers equally to language and to organisms. Part III, ‘‘The Lexicon and Change,’’ addresses the nature of the word simultaneously in the sense of having an idiosyncratic character, and in the sense of having a genetic inheritance, in something of the same way that nineteenth-century historical linguists’ conceptions of the ‘‘genius’’ of an individual language combined historical accidents with formal necessity. Part IV, ‘‘Syntax and Semantics,’’ captures the nature of the word in the sense of its having certain aptitudes, the way an organism’s nature determines what it can do in the world. Finally, part V, ‘‘Poetics,’’ acknowledges that the nature of the word, like nature in general, is not, finally, objectively describable, but only inferrable, partially and imperfectly, from observations that will always be di¤erent under di¤erent conditions. Poetry takes this as a central fact, and for Kiparsky, with a humbleness that paradoxically makes his work especially powerful, scientific inquiry into language is, in this regard, not fundamentally di¤erent from poetry. Representation of this range of perspectives on language has partially determined the contributors to this book, but of course much more remains to be said about the inevitably di‰cult matter of their selection. The scope of Kiparsky’s influence vastly exceeds what any single book could contain, and we have chosen to deal with this by simply acknowledging the limitations of our own perspectives, and focusing on those scholars whose close intellectual associations with Kiparsky we were especially aware of in our own work as students entering the graduate linguistics program at Stanford in 1984, the same year that he came there from MIT. These include Kiparsky’s own teachers, his previous students, our fellow students, and colleagues whose relationship to him, with the remarkable indi¤erence to institutional limitations so characteristic of him, derived purely from shared intellectual curiosities. The book largely excludes the many students and colleagues who have been important to him since that time, because we were aware when we began this project that that set would have grown and would continue to grow in ways we were simply not in a position to keep up with. It also excludes many whose own expertise overlapped insu‰ciently with our own at that time for Kiparsky’s relationship to them to have been salient to us. And it undoubtedly excludes still others simply through errors for which we can only hope we will be forgiven. In this raggedness, at least, it represents something of Kiparsky’s own helpful, honest awareness of how knowledge gets advanced: a bit here, a bit there, with gaps that hopefully can somehow, sometime be filled in. Finally, it must be added that Kiparsky’s influence does not, of course, derive from intellectual accomplishment alone. The same sense that any single word or flower, if studied closely enough in both its specialness and its generalness, can yield secrets of the universe, Kiparsky brings to his interactions with students, colleagues, and

xiv

Preface

friends—finding, cultivating, and delighting in each one’s individual talent and potential relationship to tradition, not just within linguistics but within human experience more broadly. Shared exploration of language is for Kiparsky thus inseparable from shared enjoyment of life, and hence from friendship; which is, of course, in the end the deepest motivation for this book. In this spirit we are therefore especially grateful for the patience, cooperation, expertise, and kindness of the many people who have helped bring the book into being. Dikran Karagueuzian helped us conceive, plan, and set in motion the entire project. Ann Banfield, Jim Blevins, Ed Flemming, Andreas Kathol, Paul Kay, and especially Gary Holland made helpful suggestions about individual papers, as did in fact virtually all the contributors, reading each others’ papers as well as writing their own. Jeremy Ecke helped transform twenty-nine idiosyncratic papers into a single, coherent manuscript. At The MIT Press, Tom Stone enthusiastically took on the project, and Sandra Minkkinen e‰ciently and elegantly saw it through to completion. Everyone working with her contributed stellar editorial and production work much appreciated by all the authors, who remained heroically patient during a long wait to see their papers appear. We were especially grateful for the involvement of Anne Mark—in the words of the book’s own honoree, ‘‘a legend.’’ Renowned for miracles of a di¤erent kind, Samuel Jay Keyser made a breeze blow again at a time when everything seemed becalmed. And at the Library of the Royal Botanic Gardens, Kew, with all the discretion of a royal retainer, Samantha Cox beautifully answered our peculiar request for botanical drawings of Finnish orchids.

Biography of Paul Kiparsky

Born in Helsinki on January 28, 1941, Rene´ Paul Victor Kiparsky is a scion of a long line of scholars, not least his parents Dagmar (ne´e Jaatinen) and Valentin Kiparsky. His great-grandfather Valentin Ludwig Kiparsky wrote a dissertation on plastic surgery in 1864, at a time of academic renaissance at Tartu University and the emergence of Estonian nationalism. Paul’s grandfather, Rene´ Valentinovich Kiparsky (1867–1938), was a distinguished gynecologist in St. Petersburg who developed several innovative surgeries and wrote a standard text with Dmitri Ott, head of the research institute that now bears his name (Ott 1914).1 Paul’s father, Valentin Julius Alexander Kiparsky (1904–1983), is best known among linguists for his book on Russian accent and his three-volume historical grammar of Russian (V. Kiparsky 1962, 1963–1975). Equally important, in a long career devoted to Russian and the Baltic linguistic area, are his many detailed studies of linguistic contact and di¤usion, including a dissertation on prehistoric Germanic-Slavic contact and monographs on topics ranging from the loan vocabulary of Baltic German to the history of terms for the walrus (V. Kiparsky 1934, 1936, 1952). Valentin Kiparsky is indeed still cited for his early statement, in 1938, that a language’s receptiveness to borrowing depends as much on social factors as it does on facts about linguistic structure. It is relevant to add that when he joined the University of Helsinki faculty in the same year, he introduced modern phonology (Prague School structuralism) to Finland. Paul Kiparsky’s career continues the family tradition. After studying at Alabama College and the Universities of Helsinki and Minnesota, Kiparsky received his PhD in 1965 from MIT. For two decades he taught at MIT, and since 1984 he has taught at Stanford University, where he is Bass Professor in the School of Humanities and Sciences. Among other honors he has served as Collitz Professor of the Linguistic Society of America (1973), been awarded an honorary doctorates by Go¨teborg University (1985) and the University of Konstanz (2008), and received the Alexander von Humboldt Prize (1993). He has directed over thirty dissertations in fields as diverse as poetics, language change, semantic and syntactic theory, and every area of phonology and morphology.

xvi

Biography of Paul Kiparsky

Kiparsky’s work in linguistics has been wide-ranging and foundational. He has redefined no fewer than four fields of linguistics (so far) and made major contributions to several others, notably semantics and Paninian studies. His dissertation and a series of papers mainly devoted to Greek phonology (Kiparsky 1965, 1967a, 1967c, 1967d), together with several other articles published in his second book (Kiparsky 1982a) and elsewhere, established him as the central figure in a new field of generative historical linguistics—the analysis of sound change and analogy in the grammatical terms of generative linguistics. More recent papers such as his survey of phonological change and several radical reanalyses of morphological changes in Gothic and Latin (Kiparsky 1995, 1998, 2000a, 2000b) show that he continues to be as much a leader in this field as ever. His work throughout the 1960s on phonological and morphological change led Kiparsky to assess the relationship between phonology and morphology more broadly, and the result of his and others’ research in this area was the field of lexical phonology as it arose in the 1970s and 1980s. This field remains at the center of linguistic theory, and the questions it poses are urgent for linguists of all theoretical persuasions; still among its foundational texts are a series of papers by Kiparsky (1982b, 1982c, 1982d, 1984, 1985). At the same time, starting with a classic treatment of Indo-European modality (Kiparsky 1968), Kiparsky has set a new standard in the field of historical syntax. He continues this line of research unabated, in the last decade through a dazzling set of studies tracing the evolution of Indo-European word order structures in Germanic and English and in the history of Greek (Kiparsky 1994, 1996c, 1997c; Condoravdi and Kiparsky 2001). Finally, Kiparsky created and still defines the modern linguistically informed analysis of poetics and poetic meter. His earliest work in this vein was a study of Finnish meter, to which he has recently returned (Kiparsky 1967d; Hanson and Kiparsky 1996); other major contributions appear in classic volumes on oral poetry and on the linguistics of meter (Kiparsky 1976, 1989). Kiparsky’s students and friends will agree that in accomplishment and breadth of interest he is the truest successor to his teacher Roman Jakobson, whose paraphrase of Terence equally defines Paul Kiparsky’s own career: Linguista est; linguistici nihil a se alienum putat. Andrew Garrett University of California, Berkeley Note 1. See Ol’shanets’kyi 1950, 103–104. I am grateful to Boris Maslov for archival research in St. Petersburg and to Lisa Conathan, Cleo Condoravdi, and Leslie Kurke for discussion and other information.

The Nature of the Word

I

Metrics

1

On ‘‘The Phonological Basis of Sound Change’’ (after Hopkins)

Andrew Garrett

Laulelevi virsia¨nsa¨, laulelevi, taitelevi —Kalevala 3.5–6 I taught last April spring’s companion, Ling 230’s thriller, thickly theory-thought Chapter, in its writing On di¤usion level-ordered short-a tensing claims, and gliding By there, how it does for all the facts an answer bring In analogy! then o¤, at once it’s a-swing Onto sound change brought forth in perception: structural tiding Preserved in such shifts. My heart in hiding Stirs for a word,—the achieve of, the mastery of the thing! Brute beauty and thinking react, the pattern of sound’s here Lexical! and it’s grammar guiding speech when, if alien Forms seem optimal, more simplified, they will appear! No wonder of it: she´er plo´d may scan a zillion Lines of Finnish song, but in that mere Paul trawls alone, and finds gold in prillion. [Falling paeonic rhythm, sprung and outriding]

2

On Stress and Meter and on English Iambics in Particular

Morris Halle

2.1

Introduction

An important di¤erence between poetry and prose is that all poetry is composed of lines, whereas prose has sentences, clauses, and phrases but no lines. In metrical poetry, the length of the lines is governed by a set of principles and rules in a manner detailed below; in free verse—as far as is known—no such principles or rules determine line length, yet poems in free verse are invariably made up of lines. The line must thus serve essential poetic purposes in all kinds of verse, even though in free verse the principles that govern lineation—and motivate its being printed without justification at the right margin—remain to be discovered (see Smith 1968; Steele 1990 for some discussion). In what follows we shall be concerned with metrical verse exclusively. A central proposition of this chapter is that every known type of meter is based on an iterative footing rule of the kind encountered in the assignment of word stress in many languages. As in Kiparsky (1977, 190), it is assumed here that feet are computed ‘‘by a kind of paraphonology that modifies the phonological derivations of the language and produces as output one or (more commonly) several alternative representations that may di¤er from phonetic representations.’’ In Kiparsky 1977 the metrically relevant properties of the line were characterized by means of the prosodic theory of Liberman 1975. The main innovation of the present study is that it characterizes the metrically relevant properties of the line—that is, it scans the lines—with the help of the prosodic theory of Idsardi 1992. My aim is to show that this innovation provides important insights into metrical phenomena of all kinds. 2.2

On Word Stress

An important advance in the understanding of the prosody of words was Liberman’s (1975) study of the English intonational system. Liberman’s key insight was that unlike [back], or [round], or [continuant], stress is not a phonetic feature. Instead, he

6

Morris Halle

argued, stress is a reflex of the tendency of languages to group syllables—more precisely, stressable phonemes—into feet. Stress, on this view, is the phonetic reflex of the foot structure of words. This implies that the computation of stress involves two separate steps. As illustrated below, we first compute the foot structure of the word and use it to assign stress—that is, high tone or some other phonetic mark—to phonemes that occupy special (head) positions in the feet so constructed. There have been a number of attempts to implement Liberman’s insight formally, for example, Liberman and Prince 1977, Hayes 1981, Prince 1983, Halle and Vergnaud 1987, and Idsardi 1992 (see also Halle and Idsardi 1995). It is the latter formalization that is adopted below. An obvious fact about stress is that not all phonemes are capable of bearing stress. In most languages all and only vowels are stressable; other phonemes are unstressable. This, however, is not true in all languages. For example, in Indonesian the schwa vowel may never bear stress (see, e.g., Halle and Idsardi 1994 and references therein). On the other hand, in Lithuanian, not only vowels but also glides, nasals, and liquids that are part of the syllable nucleus are stressable (Halle and Vergnaud 1987). A minimal requirement for an adequate stress notation is therefore that it include a means for indicating which phonemes in a sequence can bear stress. This requirement is implemented here formally by projecting stressable phonemes on a separate autosegmental plane, and it is sequences of the projected— that is, stressable—phonemes represented by asterisks in (1) and elsewhere that are grouped into feet. Since on the account adopted here stress is a reflex of foot structure, we need a formal device for grouping the stressable phonemes into feet. As illustrated in (1), the grouping of stressable phonemes into feet is accomplished by foot boundary markers or junctures, represented here by ordinary parentheses. A left parenthesis (foot boundary) foots the stressable elements on its right, whereas a right parenthesis foots those on its left; elements that are neither to the left of a right parenthesis nor to the right of a left parenthesis are unfooted. Thus, in the first and third example in (1b) the last asterisk (¼stressable phoneme) is not footed, but in the examples in (1a) and (1c) all asterisks are footed. The crucial di¤erence between this formalization and earlier ones (such as those of Liberman 1975; Hayes 1981; or of Halle and Vergnaud 1987) is that in the earlier notations, feet, like syntactic constituents, had two ends and therefore were delimited by a pair of matched parentheses, whereas in the present notation a single, unmatched parenthesis defines a foot. (1) a. Maranungku (Hayes 1981, 51) (* * (* * (* *( (* * (* * (* w e´ l e p e` n e m a` n t a l a´ n g k a r a` t e t `ı ‘kind of duck’ ‘prawn’

On Stress and Meter and on English Iambics in Particular

b. Pintupi (Hayes 1995, 62) )* *) * )* *) * *) j m a´ L a w a` n a t u´ T a y a ‘many’ ‘through from behind’ c. Weri (Hayes 1981, 53) )* *)* *) *) * *) * *) u l u` a m ´ı t a` k u n e` t e p a´ l ‘mist’ ‘times’

7

)* *) * *) * p u´ L i N k a` l a t j u ‘we (sat) on the hill’

It has been remarked above that feet are defined by parentheses, but to this point nothing has been said about where these parentheses come from. It is assumed here that speakers of a language have the ability to break a sequence of stressable phonemes into subsequences or feet. We picture this formally (literally) as the ability to insert parentheses into strings of asterisks, as illustrated in (1). The feet in the examples in (1b,c) are defined by right parentheses, those in (1a), by left parentheses. The parentheses are inserted by rules of Iterative Footing, that is, by rules that work their way across the stressable phonemes (asterisks) of a word starting at one edge and ending at the opposite edge. The iterative rules illustrated in (1) insert parentheses next to every other asterisk that they encounter, that is, they skip over two unfooted asterisks in each step. In (1a,b) the insertion begins at the left edge and proceeds rightward; in (1c) the insertion proceeds in the reverse direction, from right to left. In addition to binary footing rules like those illustrated in (1) there are also ternary footing rules, where three asterisks are skipped in each iteration. Stress systems with ternary feet, though considerably rarer than those with binary feet, are encountered, for example, in Cayuvava and Alutiq. (See Idsardi 1992 for some discussion.)1 The present theory provides only two kinds of iterative feet—binary and ternary—feet of greater length, such as quaternary, are specifically excluded as being beyond the innate capabilities of normal speakers. The machinery available for stress assignment—and for footing lines of metrical verse—is thus limited to counting by pairs and/or by triplets. Since the notation makes available both left and right parentheses, the type of parenthesis to be inserted must be specified in each case. As seen by comparing the rightmost examples in (1a) and (1b), the choice of di¤erent parentheses may result in scanning the same number of asterisks (syllables, stressable phonemes) into di¤erent numbers of feet. Thus, the three left parentheses in the five-asterisk sequence in (1a) generate three feet, whereas the three right parentheses in (1b) generate only two feet. Because a foot may be composed of more than one element, foot structure alone cannot define the placement of stress. To this end we have recourse to the fact that every foot has exactly one head and that stress is placed on the head of

8

Morris Halle

a foot. It is, moreover, the case that the head of a foot is almost either its left-most or its rightmost element, and the choice between the two is determined by a special, language-particular Head rule. In Maranungku and Pintupi (shown in (1a,b)) feet are left-headed, whereas in Weri (1c) feet are right-headed. Iterative Footing rules are not the only source of parentheses: parentheses may also be inserted by Accent rules. The Accent rules supply parentheses to specific syllables (more precisely, to asterisks projected from specific syllables). For example, many languages stress syllables with heavy rimes, or syllables that have special morphological characteristics. In the theory adopted here this is done by an Accent rule that inserts a parenthesis next to (asterisks projecting from) such syllables. Parenthesis insertion of this type is the main stress mechanism in languages such as Koya, Selkup, Khalkha Mongolian, and Komi. (For some discussion, see Idsardi 1992 and, from a di¤erent theoretical perspective, Hayes 1995:296¤.) In the Indo-European languages with free stress, such as Russian, Serbo-Croatian, Sanskrit, and Lithuanian, parentheses are supplied to some morphemes in their underlying representation. (For additional discussion, see Halle 1997 and Kiparsky and Halle 1977.) Of particular interest for present purposes are languages that make use both of Accent rules and of Iterative Footing. In such languages, Iterative Footing respects the partial footing imposed by the Accent rules. This is illustrated by the Finnish examples in (2), where right parentheses are inserted by Accent rules and left parentheses are inserted by Iterative Footing. (2) Finnish (Hanson and Kiparsky )* *) * )* *) (* o´ m e n a o´ m e n a` t ‘apple’ ‘the apples’ )* *) * (* l u´ m o s i v a` t ‘they enchanted’

1996, 301) )* *) (* *) l u´ m o t t u` i h i n ‘enchanted’

)* *) (* *) (* k o´ s k e m a` t t o m a` n ‘untouched’

)* *) * (* *) * (* *) v o´ i m i s t e l u` t t e l e m a` s t a ‘from having on and o¤ caused to do gym’

)* *) * (* *) o´ p p e t t a m a` s t a ‘from teaching’ )* *) * o´ m e n a t ‘the apples’

)* *) * *) l u´ m o s `ı v a t ‘they enchanted’

)* *) (* *) * k o´ s k e m a` t t o m a n ‘untouched’ In Finnish, according to Hanson and Kiparsky (1996, 301), ‘‘[s]econdary stress is subject to phonological and morphological conditioning (with variability in some contexts).’’ The formal reflex of this conditioning here is the Accent rule (3a), which inserts a left parenthesis before the syllables in question. (See the underlined paren-

On Stress and Meter and on English Iambics in Particular

9

theses in (2)). Binary feet are then constructed by a rule of Iterative Footing (3b), which inserts right parentheses left to right at binary intervals. Stresses are assigned by positing that feet in Finnish are left-headed, that is, trochaic, and that stress (high tone) is assigned to the heads. As noted, Iterative Footing respects the rudimentary foot structure assigned to the word by the Accent rule. Formally, this is implemented by ordering the Accent rule before Iterative Footing. It is to be noted that Iterative Footing applies only to sequences of unfooted syllables; it does not foot a pair of asterisks separated by a (previously inserted) foot boundary. Thus, in o´ppettama`sta and voimistelutttelemasta in (2), Iterative Footing inserts a right parenthesis after the fifth—and not after the fourth—syllable. It is the latter fact that is reflected in the rather complex part (italicized below) of Hanson and Kiparsky’s formulation of the distribution of secondary stresses: ‘‘Secondary stress falls on every second syllable after the initial one, skipping an additional light syllable if the syllable after that is heavy’’ (p. 301). There are two exceptions to the preceding. (i) According to Hanson and Kiparsky, ‘‘[f ]inal syllables not preceded by stressed syllables are optionally stressed.’’ In the formalism adopted here, this means that in word-final position the Finnish Accent rule (3a) is optional. Such alternative stressings without a left parenthesis before the word-final syllable are exemplified in the last line of (2). (ii) The second, that is, peninitial, syllable of a word is never stressed, even if heavy. This fact is taken into account in the formulation of the Accent rule in (3a).2 The basic footing rules for Finnish stress are given in (3). (3) a. Accent rule In postpeninitial position, insert left parentheses before heavy syllables and before specially marked syllables. (This rule applies optionally before wordfinal syllables.) b. Iterative Footing rule Insert right parentheses iteratively from left to right starting at the left edge of the string and skipping over two consecutive syllables at each step. Heads: left Since in Finnish the main stress of the word is distinguished from its secondary stresses, additional machinery is needed that will allow us to distinguish among stressed syllables—that is, among heads of feet. In the Idsardi notation each head of a foot is projected onto the next higher line in the metrical grid. This property of the notation also provides a straightforward way of distinguishing among heads of feet. We need only posit that the sequence of heads of feet projected onto line 1 is footed by an Accent rule that inserts a left parenthesis at the left edge of line 1 and that line 1 feet are left-headed. This is illustrated in (4), where the left parenthesis inserted by the Accent rule on line 1 is represented by a square bracket rather than a parenthesis.

10

Morris Halle

(4) * [* )* *) * o´ m e n a

* [* * )* *) (* o´ m e n a` t

* [* * )* *) * (* l u´ m o s i v a` t

* [* * * )* *) (* *) (* k o´ s k e m a` t t o m a` n

* [* * )* *) * (* *) o´ p p e t t a m a` s t a 2.3

* [* * )* *) (* *) l u´ m o t t u` i h i n

line 2 line 1 line 0

line 2 line 1 line 0

* [* * * )* *) * (* *) * (* *) v o´ i m i s t e l u` t t e l e m a` s t a

line 2 line 1 line 0

The Strict Meters of English Verse

The formal machinery sketched in the preceding section accounts also for the di¤erent types of metrical verse encountered in the languages of the world. It is not the case that the stress rules of a language are the same as those that account for its metrical verse. For example, English word stress is trochaic (see Halle 1998 for some discussion), but English metrical poetry is, of course, not restricted to trochaic meters. Though not identical, the rules governing word stress and the rules of meter are of the same kind. Underlying both English word stress and the main meters of English is a rule of Iterative Footing (cf. (3b), (9a), and (15a)). Iterative Footing is supplemented by an Accent rule both in the assignment of stress to Finnish words discussed (cf. sec. 2) and in the computation of the loose meters of English discussed in section 2.4 (cf. (3a) and (15b)). When examined in detail, the rule of Iterative Footing involves the setting of the four binary parameters in (5). (5) a. b. c. d.

The type of parenthesis inserted: left or right The direction of insertion: (left to right) L>R or (right to left) R>L The number consecutive asterisks skipped after each insertion: two or three Head placement: left- or rightmost

There are thus, in principle, 2 4 ¼ 16 distinct ways of footing a string of asterisks. Not all of these 16 footings appear to be used equally in metrical verse. In English poetry from Chaucer to the end of the nineteenth century, the overwhelming majority of lines are composed in one of the four meters (feet) in (6). In these meters, which are referred to here as the strict meters of English, head placement (5d) determines both the type of parenthesis inserted (5a) and the direction of insertion (5b):

On Stress and Meter and on English Iambics in Particular

11

when left-headed feet are constructed, left parentheses are inserted from left to right; when right-headed feet are constructed right parentheses are inserted from right to left. (6) a. Trochees Insert (, left to right, b. Dactyls Insert (, left to right, c. Iambs Insert ), right to left, d. Anapests Insert ), right to left,

heads: left, binary heads: left, ternary heads: right, binary heads: right, ternary

Examples of these four meters are given in (7). (7) a. Trochees (6a) Life is real! Life is earnest! (* * (** (* *(* *( And the grave is not its goal: (* * (* * (* * (* (Longfellow, ‘‘A Psalm of Life’’) b. Dactyls (6b) Half a league, half a league, (* * * (* * *( Half a league onward, (* * * (* * (Tennyson, ‘‘The Charge of the Light Brigade’’) c. Iambs (6c) Let me not to the marriage of true minds )* *) * *) * *) * *) * *) Admit impediments. Love is not love )* *) * *) * *) * *) * *) (Shakespeare, Sonnet 116) d. Anapests (6d) ‘Twas moonset at starting; but, while we drew near * *) * * *) * * *) * * *) L o k e r e n, t h e c o c k s c r e w, a n d t w i l i g h t d a w n e d c l e a r. *) * * *) * * *) * * *) (Browning, ‘‘How They Brought the Good News from Ghent to Aix’’) In (7) like in (4) the consecutive syllables of each line are represented by asterisks underneath the words, and parentheses are inserted into this sequence of asterisks by

12

Morris Halle

the rule of Iterative Footing given in bold type at the right. For example, in (7d) right parentheses are inserted from right to left skipping three syllables after each insertion. Since (7d) is an instance of a strict meter, head placement determines both the direction of insertion and the kind of parenthesis inserted. As shown in (7d), the result of this procedure is to construct four anapests on both lines. Footings other than those in (7) can be imposed on the lines. For example, in (8) I have footed one of the iambic lines in (7c) into anapests by means of rule (6d) and one of the anapestic lines in (7d) into trochees by rule (6a). (8) a. By rule (6d) Admit impediments. Love is not love *) * * *) * * *) * * *) b. By rule (6a) ‘Twas moonset at starting; but, while we drew near (* * (* * (* * (* * (* * (* The availability of the alternative footings in (8) raises the question as to how these patently wrong footings are eliminated in favor of the correct ones in (7). The obvious answer is that in the English verses under discussion here, the choice of the correct footings is determined by the position of the stressed syllables in the line. While traditional approaches have treated all stressed syllables as equal in determining the well-formedness of a line, it is assumed here—following Halle and Keyser 1999— that only the main stresses of polysyllabic words determine whether a particular footing is a valid instantiation of a given meter. We call such stressed syllables Maxima and require that Maxima be placed only in head positions of feet. This requirement is violated by the footings in (8), where the respective Maxima of Admit in (8a) and of moonset in (8b) occur in nonhead positions of the feet, and it is these violations that rule out the footing in (8). By contrast, none of the footings in (7) violates the requirement just proposed. The fact that stressed monosyllables play no role in determining the well-formedness of lines is illustrated by the clause Love is not love in (7c). As shown there, the first word Love is in a nonhead position, even though this word has greater stress than either of the two syllables adjoining it. Being in a monosyllabic word, however, this stress is not a Maximum and does not violate the requirement above. We have now reviewed all the main ingredients involved in the scansion of the lines in (7). The rules and conditions that define the strict meters of English are stated in (9). These consist of a definition of Maximum, a rule of Iterative Footing (9a), and two conditions (9b). (9) Definition A Maximum is the syllable bearing main stress in a polysyllabic word. (To be modified below cf. (15).)

On Stress and Meter and on English Iambics in Particular

13

a. Iterative Footing rule Select one of the four parameter settings in (6). b. Conditions i. Maxima can be placed only in head positions of feet. ii. Incomplete feet are allowed to terminate the iterative footing procedure. (Incomplete feet may therefore appear at the beginning of iambic and anapestic lines and at the end of trochaic and dactylic lines.) Condition (9bii) accounts for the variation in line length found in the examples in (7). For example, in (7a) the first line is eight syllables long, whereas the second line has only seven syllables. The di¤erence in line length is a consequence of the manner in which foot boundaries are inserted by the rule of Iterative Footing. As stated in (6a), trochees are generated by a rule of Iterative Footing that inserts left parentheses L(eft)>R(ight), skipping two consecutive asterisks after each insertion. This rule will obviously foot a sequence of eight syllables into four feet. However, since a foot may consist of a single asterisk, this rule will also foot a sequence of seven syllables into four feet. Since in (7a) Iterative Footing proceeds L>R, the incomplete foot appears at the end of the line. Where Iterative Footing applies R(ight)>L(eft) the incomplete foot appears at the beginning of the line. We see an example of this in (7d), where the first line begins with a disyllabic foot and the second line with a monosyllabic foot. Since in anapests Iterative Footing skips three syllables after each insertion, it may generate incomplete feet that are either monosyllabic or disyllabic. 2.3.1

Elision

As illustrated in the even-numbered lines of (10), extra syllables may appear in some lines. (10) a. O, n o, i t i s a n e v e r f i x e` d m a r k, )* *) * *) * *) * *) * *) That looks on tempests and is never shaken; )* *) * *) * *) * *) * *)  (Shakespeare, Sonnet 116) b. All the buds and bells of May, (* * (* * (* * (* From dewy sward or thorny spray;  (* * (* * (* * (* (Keats, ‘‘Fancy’’) Traditionally the second line in (10a) is said to end with a ‘‘feminine’’ rime, and the extra syllable is not counted for purposes of determining the meter. This same procedure is followed here and such metrically ‘‘omitted’’ syllables are signalled by

14

Morris Halle

representing them not with an asterisk but with a bullet (), as shown in (10). In (10a) the ‘‘omitted’’ syllable is located at the end of the line, that is, at the starting point of Iterative Footing (cf. (6c)). The symmetrical situation obtains in the second line in (10b), where the Iterative Footing rule applies L>R. In this case, the metrically omitted syllable begins the line. The status of the omitted syllable resembles that of incomplete feet (see (9bii)). The two cases di¤er in that incomplete feet are admitted as the last iteration of a rule of Iterative Footing, whereas the metrically omitted syllables under discussion here occur at the beginning of Iterative Footing. Metrically omitted syllables occur also line medially, but their occurrence there is more highly marked and therefore used only sparingly. A few examples are given in (11). (11) a. H e s t a r e d a t t h e P a c i f i c—a n d a l l h i s m e n * *) * *) )* *) * *) * *)  (Keats, ‘‘On First Looking into Chapman’s Homer’’) b. This fortification, gentlemen, shall we see it? )* *)  * *)* *)  * *) * *)  (Shakespeare, Othello III, 2, 5) c. But as I past, I worshipt: if those you seek. )* *) * *) * *)  * *) * *) (Milton, Comus, 302) d. Yet dearly I love you and would be love`d faine, )* *)  * *)  * *) * *) * *) (Donne, Holy Sonnet XIV) Such metrically omitted syllables are especially common in Donne’s poetry, where they occur most frequently in hiatus, that is, word finally before a vowel-initial word.3 Like Donne’s line (11d), the line from Shakespeare (11b) includes two lineinternal omitted syllables. (This and two additional lines are cited by Hanson and Kiparsky 1996, 298.) Both types of omission have traditionally been referred to as ‘‘elision,’’ which Merriam-Webster’s unabridged dictionary (1961) defines as a ‘‘deliberate syllablereducing suppression. . .of a. . .vowel in poetry for the sake of the meter.’’ The present theory treats all such vowels by not projecting them onto line 0 of the metrical grid. Special note is to be taken of the fact that metrical omission says nothing about the pronunciation of these syllables. While the meter and the pronunciation of a line are related, neither determines the other completely. 2.3.2

Ancipitia

A well-known peculiarity of English iambic verse is illustrated in (12), where every line begins with a stressed syllable of a polysyllabic word.

On Stress and Meter and on English Iambics in Particular

15

(12) a. Mighty and dreadful for thou art not so: )* *)* *) * *) * *) * *) (Donne, Holy Sonnet X) b. Kissing with golden face the meadows green )* *) * *)* *) * *) * *) (Shakespeare, Sonnet 33) c. Mindless of its just honours; with this key )* *) * *) * *) * *) * *) Shakespeare unlock’d his heart; the melody )* *) * *) * *) * *)* *) (Wordsworth, Sonnet) d. S i l e n t u p o n a p e a k i n D a r i e n. )* *) * *) * *) * *) **) (Keats, ‘‘On First Looking into Chapman’s Homer’’) Treating the line-initial stressed syllable in (12) as a Maximum results in all examples in a violation of condition (9bi). Since lines of this kind are readily found in the works of the best poets, it is necessary to modify the definition of the Maximum so as to exclude line-initial stresses. The restated definition is given in (13). (13) A Maximum is a syllable bearing main stress in a polysyllabic word provided that it is preceded and followed by a syllable with less stress. The e¤ect of (13) is to limit further the contexts where the stressed syllable of a polysyllabic word is counted as a Maximum. Since line-initial stresses are not preceded by a syllable without stress, they are also not Maxima, according to (13). The lines in (12), therefore, do not violate condition (9biii). The insight captured by this redefinition is that of anceps in classical metrics, which, according to Marouzeau 1943, 30 is ‘‘a syllable counted freely as short or as long (Lat. anceps ¼ ‘‘ambiguous’’)’’. For the role of ancipitia in Latin verse, see Embick and Halle in prep. The special treatment of line-initial stress is a peculiarity of English verse, which is not shared by the syllabotonic verse of Russian and German. This has been noted for Russian by Zˇirmunskij 1925/Eng. translation 1966, p. 53, and Bjorklund (1989, 156) for German. The definition of Maximum in Russian and German verse must therefore di¤er from that in English. Whether this di¤erence is connected with the di¤erence in phonetic stress pattern of the words in these two languages and how this connection is to be expressed formally is at present an open question. (For some discussion, see Hanson and Kiparsky 1996.) In view of the redefinition of Maximum in (13), we expect to find examples where the stressed syllable in a polysyllabic word is not counted as a Maximum in versemedial position. Such examples are given in (14).

16

Morris Halle

(14) a. A maid of grace and complete majesty (Shakespeare, Love’s Labour’s Lost I, 1, 137) b. His means of death, his obscure funeral (Shakespeare, Hamlet IV, 5, 213) c. Sir, in good sooth, in sincere verity. (Shakespeare, King Lear II, 2, 111) d. The divine property of her first being (Milton, Comus, 469) e. In profuse strains of unpremeditated art. (Shelley, ‘‘To a Skylark,’’ 6) The examples in (14), which have been copied from Bridges 1921, are frequently cited in discussions of English meter. Bridges attributed these to ‘‘recession of accent. . .not now heard’’ (p. 67). As the main (sole?) evidence for the accent recession is the scansion of the lines, Bridges’s proposal is not compelling, especially since accent movement has not been adduced by Bridges or anybody else as an explanation for the similar line-initial examples in (12); for example, nobody has suggested that Shakespeare in (12c) is subject to accent advance. In view of this there is reason to question accent recession as an explanation for the facts in (14). This is yet another instance of the indirect relationship between metrical pattern and its instantiation in verse. The definition of Maximum in (13) accounts for both classes of exceptions: those in (12) as well as those in (14).4 2.4

Loose Meters in English

In addition to the ‘‘strict’’ iambics discussed above, English poets, as well as poets writing in German and Russian, have employed what Robert Frost has called ‘‘loose’’ iambics. In addition to Frost, whose practice has been studied in Halle and Keyser 1999, loose iambics were employed by many other English poets, for example, Blake, Tennyson, Yeats, and Auden.5 The main di¤erence between strict and loose iambics involves the relationship between the footing and placement of Maxima. In strict iambics, placement of the Maxima is determined by the footing of the line: because of condition (9bi), Maxima can be placed only in head positions of the feet—that is, in even-numbered positions in iambic lines. In loose iambics it is the other way around—that is, the Maxima determine the location of the heads of the feet. Formally, we capture this di¤erence by replacing (9) with (15). This change includes the Accent rule (15b) which places a right parenthesis after each Maximum. Once the Maxima have been marked, the rule of Iterative Footing (15c) applies and inserts left parentheses from right to left, skipping two consecutive asterisks after each insertion. Since Iterative Footing is

On Stress and Meter and on English Iambics in Particular

17

ordered after the Accent rule, Iterative Footing respects the parentheses placed by Maximum marking. In sum, in loose iambics—and in loose meters, in general— footing of the line is governed by the placement of the Maxima, whereas in strict iambics, the placement of Heads is determined by Iterative Footing and the words must be chosen so that their Maxima coincide with the heads of the feet.6 The definitions and rules for English loose iambics are given in (15). (15) a. Definition A Maximum is i. the syllable bearing main stress in a polysyllabic word ii. a stressed monosyllabic word when followed by two syllables with less stress iii. a stressed monosyllabic word at the end of a major syntactic constituent b. Accent rule Insert a right parenthesis after a Maximum. c. Iterative Footing rule Insert left parentheses from right to left starting at the right edge of the line and skipping two asterisks in each iteration. Heads: right d. Condition Unfooted syllables and nonmaximal feet are admitted in all positions in the line. It is to be noted that the definition of Maximum for loose iambs in (15a) di¤ers from that of the Maxima in the strict meters in (13). The inclusion among Maxima of stressed monosyllabic words followed by two unstressed syllables expresses formally the fact noted, for example, by Tarlinskaja (1993, 57) that two consecutive stressed syllables in the line are separated by no more than two syllables. An example of loose iambs is Blake’s ‘‘The Sick Rose’’ in (16). (16) O Rose, thou art sick. (* *) * (* *( The invisible worm *( * *) * ( * *( That flies in the night (* *) * (* *( In the howling storm * (* *) (* *( Has found out thy bed (* *) * (* *( Of crimson joy, (* *) (* *(

1 2 3 4 5 6

18

Morris Halle

And his dark secret love * (* *) * (* *( Does thy life destroy. * (* * (* *) (

7 8

In (16) the left parentheses are inserted by the Accent rule (15b), and the right parentheses are inserted by the Iterative Footing rule (15c). In lines 1, 3, and 5 of (16), monosyllabic words are treated as Maxima, for they are followed by two syllables with less stress. The lines vary in length from four to six syllables, but in terms of feet there is no variation: each line has exactly two feet. The footing of the line explains the variations in syllable number on the assumption that the poem is subject to restrictions beyond those in (15). In particular, since the poem has only masculine rimes, unfooted syllables are not authorized in line-final position, nor are unary feet admitted anywhere in the poem. This admits five- and six-syllable lines into the poem in addition to four-syllable lines. Metrically the most interesting is line 7, for that line would have three feet if the lines were scanned as having no stress Maximum. The metric irregularity is removed by treating dark as a stress Maximum. This scansion is supported by the reading of dark secret as a compound adjective, like bright yellow (screen) or fair feathered (friend). (For some discussion of the metrical treatment of such structures, see Kiparsky 1975b, section 2.2, and Kiparsky 1977, section 6.) 2.5

In Conclusion

In the preceding I have attempted to show that a single grouping mechanism underlies both the assignment of stresses to words in di¤erent languages and the computations of meters in poetry. In section 2.2 the stress systems of a number of languages (Pintupi, Maranungku, Weri, and Finnish) was examined, and it was shown that the stress contours of the words in these languages were accounted for by grouping the syllables into pairs or triplets. In sections 2.3 and 2.4 I showed that the same kind of grouping of syllables into pairs or triplets accounts also for the main meters of English poetry. Both in stressing words and in computing meters, the grouping of the syllables is formally implemented by inserting diacritic markers—parentheses—among the syllables in the sequence. A left parenthesis groups the syllables on its right, a right parenthesis groups the syllables on its left. The grouping is carried out by a version of the rule of Iterative Footing (cf. (3b) and (9a)). The grouping operation, moreover, designates as head the left-, resp. right-most syllable in each of the groups. In the strict meters, the grouping of the syllables (footing) is determined exclusively by the Iterative Footing rule. In loose meters, grouping is determined by the interac-

On Stress and Meter and on English Iambics in Particular

19

tion of Iterative Footing with an Accent rule such as (15b). We encounter a parallel di¤erence in stress systems: in Maranungku and the other languages illustrated in (1), stress distribution is determined entirely by a rule of Iterative Footing; in Finnish, by contrast, stress assignment involves the Accent rule (3a) in addition to the Iterative Footing rule (3b). An important di¤erence between stress systems and meters is that in stress systems a phonetic mark—often a special pitch—is assigned to the heads of the feet, but no phonetic mark is supplied to the heads in meters; the heads of the di¤erent feet are rather subject to special conditions, such as (9bi), requiring maxima to appear in head position exclusively. Both stress and meter require information about head placement, which can be computed only by grouping the syllables in the sequence. It is therefore the grouping of syllables into feet (and into higher order groups) that underlies the parallels between stress and meter first noted by Liberman 1975. In the seven years that have elapsed since the original version of this paper was sent to the editors, Nigel Fabb and I have studied meters in many languages including, among others, French, classical Greek, Spanish, Italian, Arabic, Vietnamese, and Sanskrit. These studies have shown that with the notable exception of the metrical poetry of the Old Testament, all meters of all languages involve grouping of syllables into pairs or triplets of the same kinds as that met with in the English examples in this chapter. For details interested readers are referred to Nigel Fabb and Morris Halle (to appear in 2008) Meter in Poetry: A New Theory, Cambridge University Press. Notes I am grateful to Sylvain Bromberger, David Embick, John Frampton, Kristin Hanson, Jay Keyser, Michael Kenstowicz, Bert Vaux, as well as to an anonymous reviewer for helpful comments. 1. Hayes 1995, chapter 8 discusses additional instances of ternary stress patterns and o¤ers an alternative account of these as special cases of binary footing. I am not altogether persuaded by Hayes’s proposals, but because of the complexity of the issues involved, a discussion of these proposals must be deferred to another occasion. 2. I am indebted to Michael Kenstowicz for drawing my attention to this fact. 3. The word love`d in (11d) is counted as disyllabic for purposes of the meter. This is another deviation from the syllabification of normal speech that is common in English poetry (see Major 1901 for some discussion). Like the omission of word-medial syllables, this type of syllable epenthesis, is employed only sparingly. We take formal account of these facts by adding appropriate stipulations to the list of conditions (9b), noting in particular their marked status. 4. The treatment discussed here is along the lines of Halle and Keyser 1971.

20

Morris Halle

5. Extensive use of loose meters is found in German poetry. Among poems in these meters are such ‘‘war-horses’’ as Goethe’s Der Erlko¨nig and Heine’s Die Grenadiere and Lorelei. In an unpublished lecture (Harvard, February, 1997), I have shown that loose meters of all kinds— trochees, dactyls, and anapests in addition to iambs—are widely employed in Russian poetry, where they are referred to by the term dol’nik. 6. A parallel di¤erence is found in stress systems. As shown in section 2.2, Finnish word stress parallels the loose meters, whereas the stress systems of Maranungku, Pintupi, and Weri parallel the strict meters.

3

Nonlexical Word Stress in the English Iambic Pentameter: A Study of John Donne

Kristin Hanson

3.1

Introduction

In their study of Chaucer that inaugurated generative metrics, Halle and Keyser (1966, 201–202) suggest that the stress borne by certain function words might be irrelevant to the fundamental question of whether a line is metrical or not. As metrical theory has changed with the vicissitudes of phonological theory, that idea has been disputed (Wimsatt 1970, 777–782; Magnuson and Ryder 1970, 791–795; 1971, 205– 206), clarified and defended (Halle and Keyser 1971b, 156–157, 163–164), explicitly extended to the modern English tradition and justified (Kiparsky 1975b, 249–250), abandoned (Kiparsky 1977, 219–221), and neglected (Hayes 1989b). My purpose here is to restore it, in a revised form, as a claim that in the modern English iambic pentameter, as in other meters (Hanson 1992; Hanson and Kiparsky 1996, 321–325), whether metrical constraints apply to lexical or postlexical prosodic structure is a significant parameter of variation across poets. My argument comes from the metrical treatment by three poets of disyllabic function words. Drawing on Liberman and Prince’s (1977) model of linguistic rhythm or prosody as fundamentally relational, Kiparsky’s (1977) study of the English iambic pentameter shows that syllables that are prosodically strong relative to other syllables in the same word (1a) are always highly constrained in that meter, while those that are strong only relative to other syllables in di¤erent words (1c) are treated more laxly: (1) a. amo´k hu´nter

b. amo´ng u´nder

c. a mo´nk hu´nt for

But what counts as a word with respect to such constraints? While function words (1b) might intuitively be expected to pattern with the content words of (1a), and are predicted by Kiparsky’s (1977) analysis to do so, I will argue here that this is correct only for some poets, such as, apparently, Shelley (Hayes 1989b). For other poets function words pattern instead with the phrases in (1c). This is subtly so for

22

Kristin Hanson

Shakespeare, but strikingly so for Donne, for whom the prosodically strong syllables of (1b,c) are not constrained at all, while those of (1a) are. In fact, the possibility of a crucial distinction between (1a) and (1b,c) alongside the more intuitive one between (1a,b) and (1c) reveals systematicity in the metrical practice of Donne, the English poet most renowned for the putative irregularity of his iambic pentameter. The theory of Lexical Phonology (Kiparsky 1982d) provides a principled explanation for both possible patternings of function words. On that theory, the phonology of a language is organized into distinct components according to its morphology. The broadest distinction is between the lexical phonology, which determines the phonological structure of words in the special sense of being the output of the wordformation component of a language’s grammar, called lexical words; and the postlexical phonology, which determines the phonological structure of syntactic configurations more broadly, including that of words in the more general sense of being syntactic terminals. In English, nouns, verbs, adjectives, and morphologically complex adverbs are lexical words, and their prosodic structure is determined by the lexical phonology. Function words—which include conjunctions, prepositions, pronouns, copular and auxiliary verbs, and adverbs of the type traditionally referred to as particles (see below)—are nonlexical words and therefore have their prosodic structure determined by the postlexical phonology (Inkelas and Zec 1993). The prosodic structure of phrases must also be determined by the postlexical phonology, by definition. Hence, a metrical constraint on prosodic strength within words will apply to (1a,b) if it takes postlexical structure into account, but to (1a) alone if it respects only lexical structure. By the same token, characteristic treatments of postlexical prosodic structure may be manifest either in (1c) alone, or jointly in (1b,c). The chapter is organized as follows. Section 3.2 summarizes its basic assumptions about the English iambic pentameter and the relevant prosodic properties of the language. Section 3.3 summarizes Kiparsky’s (1977) description of Shakespeare’s treatment of the three types of structures in (1), and proposes a reanalysis of his treatment of the disyllabic nonlexical words of (1b) based on the lexical/postlexical distinction. Section 3.4, the core of the chapter, shows that this approach permits a systematic account of Donne’s metrical practice. Finally, section 3.5 briefly contrasts Donne’s practice with Hayes’s (1989b) description of Shelley’s. 3.2 3.2.1

Metrical and Prosodic Assumptions Basic Structure of the English Iambic Pentameter

Following Halle and Keyser 1966, 1971b, a meter is assumed here to be defined by an abstract underlying structure, and a set of constraints on how language can be mapped into that structure. The phonological structure of the language itself is taken as a given, although it should be noted that it may di¤er slightly from the phonology

Nonlexical Word Stress in the English Iambic Pentameter

23

of the ordinary language in ways that will not concern us here (Kiparsky 1977, 190, 235–244). A scansion is a mapping of a line into the underlying structure that satisfies the constraints, and a line that has a scansion is metrical. The underlying structure of the English iambic pentameter consists of five metrical feet (Kiparsky 1977), each composed of a weak (W) metrical position followed by a strong (S) metrical position (Halle and Keyser 1971b). The feet are in turn grouped into cola, and the cola into a line. These constitutents too are binary with a rising rhythm, but the odd number of feet in the pentameter line poses complications. Both synchronic and diachronic metrical evidence shows that the basic underlying structure of the English iambic pentameter has an initial colon as in (2a), but an alternative (2b) in which the third foot also belongs to the intial colon is available as well (Kiparsky 1977; Piera 1981).1 (2) a.

b.

The ambiguity made available by (2) functions in scansion just like ambiguity made available by phonological options a language may allow: so long as there exists some mapping of some legitimate phonological structure of a line into some legitimate underlying metrical structure of the meter in question, the line is metrical. One constraint defining the English iambic pentameter limits the amount of linguistic material that can be mapped into any single metrical position of (2). The norm is a single syllable, apart from standard elisions and the possibility of phonologically weak extrametrical syllables (‘‘hi’’) at the ends of lines and major phrases (Kiparsky 1975b; 1977, 230–236). All three poets considered here, however, also allow two unstressed syllables in W, as in (3), or a stressed light syllable followed by an unstressed syllable in the same word together in S, as in (4): (3) a. I’ th’ name of fame and honor, which dies i’ th’ search w s w s w s hi w s w s (Shakespeare, Cymbeline 3.3.51) b. To feed on that, which to’disus’d tasts seemes tough. w s w s w sw s w s (Donne, ‘‘Elegie: His Picture’’ 20)

24

Kristin Hanson

c. Of the riven soul, sent its foul dreams to sweep w sw s w s w s w s (Shelley, ‘‘The Revolt of Islam’’ 1300) (4) a. And spends his prodigal wits in bootless rhyme w s w s w s w s w s (Shakespeare, Love’s Labour’s Lost 5.2.64) b. Then the wise politique horse would heretofore, w s w s w s w s ws (Donne, ‘‘Satyre I’’ 80) c. I looked, and lo! one stood forth eloquently, ws w s w s w s w s (Shelley, ‘‘The Revolt of Islam’’ 649) Formally this means that their iambic pentameter is best analyzed as limiting each position to a moraic trochee (Hanson and Kiparsky 1996). Although the distinction between that and the more familiar limiting of each position to a syllable will not figure directly in our analysis of the distinctions in (1) apart from constraining the scansions on which it rests, it is worth noting that the analysis actually lends further support to the argument at hand.2 The other constraint defining the English iambic pentameter, which is the one of central concern here, specifies how the rhythmic contrast between W and S is manifest. As Halle and Keyser (1966) observe, S is free to contain any kind of syllable (e.g., the unstressed its of (3c), the of (4b), or -ly of (4c)), but W is not. As mentioned above, Kiparsky (1975b, 1977) argues that across the tradition, syllables that are prosodically strong relative to other syllables within the same word are generally prohibited in W. Thus in (3) and (4), for example, the stressed syllables of the polysyllabic honor, disus’d (cf. 27b), riven, prodigal, politique, and eloquently are all in S, while stressed monosyllables can occur either in S or in W, the latter case appearing in seemes (3b), foul (3c), or wise (4b). 3.2.2

Lexical Prosodic Structure

Kiparsky’s formalization of the class of syllables prohibited in W is intended to account for not only the simple distinction between polysyllables and monosyllables illustrated above but also an entire gamut of subtle distinctions poets’ practices seem sensitive to. The correct formalization has been a matter of debate ever since Halle and Keyser’s first proposals, partly because of the complexity of the class, but also because available options depend on the theory of prosodic phonology. That theory has developed rapidly since that time, including undergoing significant changes since Kiparsky’s (1977) description of the patterns under discussion here. The most substantive of these changes has been the positing of the phonological foot (‘‘f’’), as part of a universal hierarchy of prosodic constituents into which lan-

Nonlexical Word Stress in the English Iambic Pentameter

25

guage is held to be organized, which also includes the syllable (‘‘s’’) and the phonological word (‘‘o’’). In this organization each of these constituents is assumed to consist of a group of constituents of the type below it in the hierarchy, among which exactly one member is the head of the superordinate constituent. An exception is constituents created by adjunction, in which a constituent that is not in a position to be grouped into a well-formed constituent of the superordinate type joins an adjacent one as a weak sister, forming a new constituent of the same type it joins, indicated here by the superscript ‘‘ 0 ’’. In the distinction drawn by Liberman and Prince (1977), which plays a crucial role in Kiparsky’s (1977) description, constituents that also contain a nonhead are branching; their heads are strong and their nonheads weak, indicated here with the subscripts ‘‘s ’’ and ‘‘w ’’, respectively. A terminal element dominated only by heads within a constituent is that constituent’s designated terminal element (DTE). All surface prosodic phenomena are assumed to be reflections of this structure: most notably, a stressed syllable is one that heads a foot, and a primary stressed syllable is the DTE of a phonological word. For the English words to be considered here, it will be assumed that syllables are grouped into left-headed feet, specifically moraic trochees, although ones whose weight patterns are sometimes disturbed by resolution and resyllabification. It will also be assumed that any syllable adjunction is leftward if possible, otherwise rightward. These assumptions are consistent with certain major generalizations about English prosodic phonology—for example, that vowels may be reduced only in unstressed syllables, and that expletives may be inserted within words only between feet. For various aspects of all these assumptions, see Liberman and Prince 1977; Hayes 1980, 1995; McCarthy 1982; McCarthy and Prince 1986; Selkirk 1984; Kager 1989; Hanson and Kiparsky 1996; Hanson 2001. On these assumptions, the following sample words whose metrical distribution in Shakespeare is described by Kiparsky (1977) would have the structures in (5):3 (5) Lexical stress in English

26

Kristin Hanson

The syllables Kiparsky (1977) found to be generally prohibited in W are those underlined. They have in common that they are the DTE of a branching consituent within o. This property thus formalizes what it means for a syllable to be strong relative to another syllable in the same word (Hanson and Kiparsky 1996). Note that this class of syllables is only a small subset of those that are stressed:4 wise in (5a), mainin (5d), -tile in (5e), -fies in (5g), and fan- in (5i) are all stressed, but not strong, and occur in W. Thus, the constraint that Kiparsky (1977, 194–195) takes as basic to the English iambic pentameter can be updated as in (6): (6) Basic prominence constraint of English iambic pentameter A strong syllable is prohibited in W if its weak sister is within the same o. (cf. Kiparsky’s (1977) (9b)) 3.2.3

Postlexical Prosodic Structure

The metrical treatment of words is to be compared here with that of phrases; but the theory of prosodic phonology has also developed in ways that a¤ect what structure

Nonlexical Word Stress in the English Iambic Pentameter

27

phrases can be assumed to have. Kiparsky’s (1977) description assumes that for phrases the constituent structure to which meter is sensitive is syntactic structure, but Hayes (1989b) has shown that it is more plausibly the constituent structure given by the prosodic hierarchy, derived from syntactic structure but not identical to it (Selkirk 1984; Nespor and Vogel 1986; Hayes 1989b). Various versions of that hierarchy have been proposed for English; the specific one assumed here will be that of Inkelas and Zec 1993, because it gives an explicit account of the third point of comparison at issue here, the prosodic structure of nonlexical words. On Inkelas and Zec’s model, postlexical prosodic structure is based on a syntactic representation in which the individual words (the terminal elements X 0 in (10) below) have exactly the structure given by the Lexical Phonology. Thus, lexical words have foot and phonological word structure as shown in (5), but nonlexical words do not. Both types of words are grouped together into phonological phrases (‘‘j’’) according to the algorithm in (7), giving sample phrasings such as those in (8). Phonological phrases themselves are further grouped into intonational phrases (‘‘i’’), again by pairing those most closely related in the syntax (Hayes 1989b): (7) Phonological phrasing in English (Inkelas and Zec 1993, 222) a. Branching clause From the bottom up in a syntactic tree, build an obligatory branching phonological phrase over adjacent phonological words that correspond to syntactic sisters. b. Adjunction clause Phrase a terminal element that lacks phonological word structure together with its immediate sister. c. Default clause Map any terminal element left unphrased by the other clauses into a phrase of its own. (8) a.

b.

28

Kristin Hanson

c.

It is within the context of the phonological phrase that the special status of nonlexical words emerges. Phrasal stress in English normally falls on the final stressed word of a phrase (Chomsky and Halle 1968; Liberman and Prince 1977), both at the phonological phrase level, and again at the intonational phrase level. Thus in (8b), for example, Jim bears more stress than bought within the phonological phrase they share, as well as more stress than chair within the intonational phrase. But nonlexical words do not bear phrasal stress in final position: in (8c), otherwise parallel to (8b), phrasal stress falls on fought, not for; and analogously in (9) it falls on stepped and talking, not over or about: (9) a. that chair he ste´pped over b. that chair they were ta´lking about At the same time, nonlexical words in final position do not lack stress. Of course, they may sometimes have stress for special rhetorical purposes, such as contrastive stress. But even under ordinary stress patterns, a reduced vowel is not possible for most function words in phrase-final position—in (8c), for example, a reduced vowel in for, common in other positions as in a chair for Jim, is impossible—and inability of vowels to reduce is normally attributed to their syllable being stressed.5 More strikingly, disyllabic nonlexical words as in (9) obviously have fixed stress regardless of their position within a phrase: only o´ver and abo´ut are correct, never *ove´r or *a´bout. Given that it is a general assumption of prosodic phonology that only syllables that already have stress at one level can manifest stress at a higher level—it must be talk- that manifests the word talking’s phrasal stress in (9b), not -ing—the hypothesis that only lexical prosodic structure is present in the representation to which phrasal stress is assigned is what Inkelas and Zec (1993) claim accounts for the lack of phrasal stress on nonlexical words. Specifically, they propose that nonlexical words are assigned stress only later, in conjunction with a default mapping of all syntactic terminals into phonological words (shown in all representations below). Within those postlexical phonological words, stress is assigned to disyllabic words according to the

Nonlexical Word Stress in the English Iambic Pentameter

29

principle that the final syllable will be stressed if and only if it is heavy, and to monosyllabic words if they are phrase-final, as defined by (7). These rules are summarized in (10): (10) Nonlexical stress assignment in English (Inkelas and Zec 1993, 222) a. Default phonological word mapping: X 0 ! [X 0 ]o b. Within a stressless phonological word, build an obligatorily binary foot with the right node strong if and only if it branches. c. Build a foot on a phrase-final phonological word that lacks metrical structure. On this approach, then, the fact that the stress of nonlexical words as in (1b) is phonologically distinct from that of lexical words as in (1a) is formalized as its being assigned only postlexically, like that in (1c), with the result that it is irrelevant to certain phenomena in the grammar of rhythm, such as assignment of phrasal stress. We are now in a position to see that this state of a¤airs is precisely what the metrical practice of English poets reflects. 3.3 3.3.1

Shakespeare Lexical Words

Kiparsky’s (1975b, 240–241) description of Shakespeare’s metrical practice takes the prohibition in (6) against W containing syllables that are strong within words to be almost absolute. One of only two systematic sets of exceptions he notes involves a very general phenomenon whereby certain metrical constraints are relaxed on initial positions of major domains; thus all English poets allow exceptions to (6) in the first position of a line, as in (11a), and many, including Shakespeare, also do so lineinternally for the first syllable of phonological or higher-level phrases as in (11b,c) (Kiparsky 1975b, 1977, 217; Hayes 1983, 1989b, 247–249): (11) a. Angels and ministers of grace defend us! w s w s ws w s ws hi (Hamlet 1.4.39) b. Brief let me be. Sleeping within mine orchard, w s w s w s w s w s hi (Hamlet 1.5.59)

30

Kristin Hanson

c.

Lines like those in (12), in contrast, in which W contains a syllable that is strong within a word and also phrase-internal, Kiparsky claims do not occur: (12)

Nonlexical Word Stress in the English Iambic Pentameter

31

In fact this description may be rather too severe, because exceptions have periodically been noted (Youmans 1983; Golston 1998). I will return to this point briefly in section 3.4 after having explored the consequences of Kiparsky’s description more fully. 3.3.2

Phrases

The constraint in (6) bans from W syllables that are strong relative to others within the same word, but Kiparsky (1977) observes that syllables that are strong relative to an adjacent syllable in a di¤erent word in the same phrase are also sometimes restricted. Most relevant to the issue here is the case when the adjacent syllable is a nonlexical word, as in collocations like a monk or hunt for in (1c).6 For these cases, scansions with the strong syllable in S as in of grace in (11a) are clearly the norm, with exceptions allowed only under certain conditions. Kiparsky’s (1977) characterization of these conditions forms part of an argument that metrical well-formedness is sensitive not only to syllable count and prosodic prominence but also to alignment of linguistic and metrical constituency. Specifically, he notes that strong syllables of phrases may occur in W if the beginning of the constituent within which the mismatched syllable’s strength is defined coincides with the beginning of a metrical foot, as in (13) (Kiparsky 1977, 206–209):7

32

Kristin Hanson

(13)

a.

b. If I dream not, thou art Aemilia. ws w s w s w sws (The Comedy of Errors 5.1.347) c. Thou dost love her because thou know’st I love her, w s w s ws w s ws hi (Sonnets 42.6) d. Or how haps it I seek not to advance w s w s ws w s w s (1 Henry VI 3.1.31) They may also occur in W if that entire constituent is itself subordinated to a still stronger syllable, as in (14) (Kiparsky 1977, 206–209) or even (15) (Kiparsky (1975b, 243): (14)

a.

b. If I lose thee, my loss is my love’s gain, ws w s ws w s w s (Sonnets 42.9) c. I understand thy kisses, and thou mine, ws w s w s w s w s (1 Henry IV 3.1.202)

Nonlexical Word Stress in the English Iambic Pentameter

33

d.

. . . Hang there like fruit, my soul, Till the tree die! w s w s (Cymbeline 5.5.263–4) e. Of the world’s ransom, blessed Mary’s Son; w s w s w s w s w s (Richard II 2.1.56) (15)

a.

b. I do suspect I have done some o¤ense w s w s w s w s w s (Richard III 3.7.111) c. When I come where he calls, then he is gone. w s w s w s w sw s (A Midsummer Night’s Dream 3.2.414) d. Shall we hear from you, Catesby, ere we sleep? w s w s w s ws w s (Richard III 3.1.188) Lines in which neither condition holds, as in (16), are claimed not to occur, or at least to be so rare as to require inclusion of some restriction on them as part of the meter’s description (Kiparsky 1977, 207; Youmans 1983): (16)

a.

34

Kristin Hanson

b.

Syllables that are strong relative to an immediately preceding weak syllable in a di¤erent lexical word are restricted in the same way. Thus Kiparsky (1977, 207) claims that lines like Wyatt’s in (17) also do not occur in Shakespeare: (17)

Strong syllables of phrases are fine in W, however, if the preceding syllable is a lexical word in its own right (Kiparsky 1977, 208):

Nonlexical Word Stress in the English Iambic Pentameter

35

(18)

Kiparsky’s (1977, 205–206) formal description of this pattern captures so many nice distinctions that it is risky to tamper with it, but some modification is required in light of the more elaborate prosodic structures entailed by the assumption of the phonological foot described in section 3.2.2, and the assumption described in section 3.2.3 that it is the prosodic hierarchy and not the syntax directly that defines metrically relevant constituent relationships. Many relationships that are immediate under Kiparsky’s assumptions become more distant once these new assumptions are taken into account.8 For example, Kiparsky’s description refers to the relationship of command defined as in (19), but in (17), the syllable blode does not command -ent as his formalization intends it to, only the phonological word does. Similarly in (14a), the phrase her cheek is not a weak sister of melts in the same way it would be if their prosodic relationship depended directly on the syntax, because within the prosodic hierarchy its strength relative to which intervenes. Moreover, Kiparsky’s own (1977, 209) description of the subordination that allows (14) and (15) is consciously vague: he notes that the domain within which subordination is metrically relevant is limited in a way that is di‰cult to define. Hayes (1989b) argues that such limitations refer to the prosodic hierarchy: on the specific assumptions about phrasing outlined in section 3.2.3, the relevant subordination is often within the same phonological phrase (Hayes 1989b), but always within the same intonational phrase as in (14a).9 Hayes (1983, 384) also observes that the relevant subordination tends to be to the syllable in the S immediately following the mismatched strong one as in (14), but while such adjacency plausibly plays a metrical role, it does not seem to be a requirement here, since it does not hold for the lines in (15). Rather, what these lines seem to suggest is that the relevant limitations may also refer to the metrical hierarchy shown in (2), and in particular to colon boundaries, whose signficance has been somewhat neglected (Youmans 1983). The lines in (15) have in common that a rising triple rhythm is repeated twice, concluding with a strong syllable in the strong position of one colon, the latter property being shared with those in (14). Clearly any attempt to improve on Kiparsky’s description would be a major undertaking in its own right, but on the basis of his cited examples we can hazard a guess that both limitations might be relevant, a point to which we return briefly in section 3.4.

36

Kristin Hanson

Taking these considerations into account, then, we can use Kiparsky’s (1977, 205) definition of command in (19), and update his description (p. 206) as follows: (19) P commands Q if and only if there is a node R that dominates Q and immediately dominates P. (20) Additional constraint on the English iambic pentameter A strong syllable s1 is prohibited in W if a. The immediately preceding syllable s2 is not the DTE of a lexical o (¼ Kiparsky’s (1977) (46i)) and b. The head of the constituent k1 of which s1 is the DTE commands s2 (¼ Kiparsky’s (1977) (46ii)) and c. k1 is not itself commanded by a strong constituent k2 whose DTE s3 is within the same i as s1 and within the same colon, and is itself in S (¼ Kiparsky’s (1977) (46iii)) Stated less formally, (20)’s prohibition under a conjunction of three conditions is equivalent to allowance under any of their opposites. Thus (20a) allows (18) by confining the prohibition against strong syllables in W to cases where the syllable immediately preceding the mismatched prominent one is nonlexical, unstressed, or only secondarily stressed, and so to sequences that have the kinds of stress patterns words themselves might have, a property we might somewhat fancifully call lexisimilitude. (20b) allows the lines in (13) because their mismatched strong syllables are not grouped with the syllable to their left, but instead begin new constituents aligned with the beginnings of metrical feet, a property we will call left alignment. Finally, (20c) allows the lines in (14) and (15) because their mismatched strong syllables are subordinated to correctly matched stronger syllables, a property we will call right subordination. Thus (20) in e¤ect identifies various mappings that do or do not a‰rm aspects of the underlying metrical pattern (Magnuson and Ryder 1971), and states that while those that do not are not unmetrical in and of themselves, neither can they be multiplied without limit, requiring instead some compensatory a‰rmation from ones that do. For Shakespeare, (6) constrains the distribution of syllables that are strong within words, but the less absolute set of conditions in (20) constrains syllables that are strong only within phrases. 3.3.3

Nonlexical Words

The phonological theory within which Kiparsky (1977) frames these generalizations distinguishes prosodic strength within words from that within phrases by the respective absence or presence of a word boundary between the strong syllable and its weak

Nonlexical Word Stress in the English Iambic Pentameter

37

sister, predicting that nonlexical words should pattern just like lexical words with respect to (6), never occurring with their strong syllables in W. But they do not. Rather, they constitute the second systematic class of exceptions to (6): as Kiparsky himself observes (1977, 220; 1975b, 249–250), although prepositions are most frequently positioned with their strong syllables in S as in (21), they also sometimes occur with them in W as in (22) and (23): (21) a. And mock you with me after I am gone. w s w s ws w s w s (Sonnets 71.14) b. And make me travel forth without my cloak, w s w s w s w s w s (Sonnets 34.2) (22) a. To glean the broken ears after the man w s w s w s w s w s (As You Like It 3.5.102) b. Like a white hind under the gripe’s sharp claws, w s w s w s w s w s (The Rape of Lucrece 543) c. And with declining head into his bosom, w s w s w s w s w s hi (The Taming of the Shrew Ind.1.119) (23) a. Henceforth be never numb’red among men! w s w s w s w s w s (A Midsummer Night’s Dream 3.2.67) b. Words before blows; is it so, countrymen? w sw s ws w s w s (Julius Caesar 5.1.27) c. Weigh’d between loathness and obedience, at w s w s w s ws w s (The Tempest 2.1.131) d. And I will comment upon that o¤ense; w s w s w s w s w s (Sonnets 89.2) Kiparsky’s explanation assimilates the exceptional behavior of these words to exceptional behavior requiring explanation in any case for lexical words. First, as already noted in (11), Shakespeare allows strong syllables of words in W phraseinitially. When the strong syllable of a preposition is initial in its word as in (22), because of the syntactic distribution of prepositions it is also normally initial in its phrase as in (24), and so can occur in W:

38

Kristin Hanson

(24)

In support of this position, Kiparsky (1977, 221) notes the absence of lines like the following in which a mismatched preposition is preceded by a proclitic and so, as Hayes (1989) observes, would not be initial in its phrase: (25) *Unlock this casket, if after three nights w s w s w s w s w s (construct) Second, apparent exceptions to the constraint on strong syllables of words in W arise for lexical words from the operation of the Rhythm Rule, which repairs the arrhythmicality of adjacent strong constituents as in (26a) by adjusting them to the form in (26b), subject to the constraint that the first W, if terminal, must be stressed (Kiparsky 1977, 218). The Rhythm Rule thus shifts, among other things, primary stress from a syllable at the end of a word to a secondarily stressed syllable at the beginning of a word when the word is subordinated in its phonological phrase to one beginning with an even more strongly stressed syllable, making the stress pattern of a word like thı`rte´en di¤erent in isolation from what it is in a phrase like thı´rte`en me´n (Liberman and Prince 1977; Inkelas and Zec 1993; Nespor and Vogel 1986; Hayes 1984, 1989b, and others). Shakespeare’s metrical practice reflects the operation of the Rhythm Rule in the language itself; thus words like forlorn are systematically positioned with their final syllable in S as in (27a) except when the conditions for the Rhythm Rule are met as in phrases like forlorn world in (27b), suggesting that the actual stress pattern of the word is di¤erent in the two lines (Kiparsky 1975b, 247; 1977, 218): (26) a.

b.

Nonlexical Word Stress in the English Iambic Pentameter

39

(27)

Kiparsky (1977, 220) claims that words like divine that do not have any stress on their initial syllable and so remain una¤ected by the Rhythm Rule are, in contrast to (27b), positioned with their strong syllable in W even when subordinated to a more strongly stressed syllable within their phrase as in (28): (28)

The prepositions with final stress in (23) are typically subordinated to a following stronger stress, just as forlorn is in (27b):

40

Kristin Hanson

(29)

They do not, however, have the stress on their initial syllable that forlorn does and that the operation of the Rhythm Rule requires, being in that respect instead like divine in (28). Kiparsky (1977, 220) addresses this similarity and this di¤erence by suggesting that the Rhythm Rule could nonetheless still be the explanation for the positioning of the words in (23), if it is assumed to have applied more freely to function words, suspending the usual requirement that the syllable onto which stress is shifted bear secondary stress. However, given the theoretical distinction between the prosodic structure of lexical words and that of nonlexical words since shown to be available independently in the phonology, such an otherwise empirically unsubstantiated and theoretically problematic supposition of a rhythmic adjustment is unnecessary. The metrical distribution of the strong syllables of the nonlexical words of (22) and (23) is simply like that of the strong syllables of the phrases in (13) and (14)–(15). If they are initial in their word as in (22) they are allowed in W because (20b) does not obtain; that is, they are left-aligned. And if they are final as in (23) they are allowed because (20c) does not obtain; that is, they are right-subordinated. For strong syllables to usually be in S according to (6) but occasionally in W according to the set of conditions in (20) is thus the characteristic distribution for Shakespeare of syllables that are strong in postlexical prosodic structure, a¤ecting nonlexical words and phrases alike. Considering Shakespeare alone, such a reanalysis accompanies its theoretical gain with only scanty empirical improvement. Certain additional problems that might arise in treating the Rhythm Rule as applicable to a line like (23d) that has no adjacent stressed syllables (in spite of having the same tree geometry at a higher level as the unseen good old man (Kiparsky 1977, 219)) are avoided. On the other hand, the explanation for the absence of lines like (25) evaporates; all that can be said is that since the number of prepositions with their strong syllables in W is already quite a small proportion of the total (Magnuson and Ryder 1971), it is not clear that the absence of such lines should be endowed with much significance. Nonetheless, the reanalysis does e¤ect a significant improvement in the account of the tradition as a

Nonlexical Word Stress in the English Iambic Pentameter

41

whole, because it extends naturally to explain the distribution of these words in Donne, to whom we now turn. 3.4

Donne

Donne’s metrical practice is famous for being among the laxest within the entire tradition of the English iambic pentameter. Noting that Donne allows strong syllables of words in W where no other poet in his study does, Kiparsky (1975b, 258–259) takes this reputation for liberty as confirmation of his analysis of the centrality of word stress in the rest of the tradition. I propose to turn the argument around and show that Donne’s practice is actually systematic in a way that confirms Kiparsky’s analysis of the more subtle stress relations within phrases. This accords with Stein’s (1944, 1962) defense of Donne’s systematicity, and also with the findings of Tarlinskaja and Teterina’s (1974) statistical study of Donne’s ‘‘Satyres,’’ which places the ‘‘Satyres’’ beyond what their study takes to be the threshold of metricality for English iambic pentameter with respect to the number of mismatches between metrical and prosodic prominence in them, but not with respect to the distribution of those mismatches. 3.4.1

Lexical Words

In Donne’s iambic pentameter, as in Shakespeare’s, most lexical words have their strong syllables in S, as does, for example, wander in (30a). There are also exceptions of the same kind allowed by Shakespeare at the beginnings of phonological and higher-level phrases, as in (11): (30)

a.

b. Which, as the soule quickens head, feet and heart, w s ws w s w s w s (‘‘Elegie: The Bracelet’’ 37)

42

Kristin Hanson

c. Dost woe my soule, for hers o¤ring all thine: w s w s w s w s w s (‘‘Holy Sonnet: ‘Since she whome I lovd’ ’’ 10) d. Immensity cloysterd in thy deare wombe. w s ws w s w s w s (‘‘La Corona’’ 28) But there are also quite a few exceptions of other kinds, which are strikingly similar to those described by the set of constraints in (20). Strong syllables of lexical words also occur in W under left alignment, parallel to Shakespeare’s phrases in (13) and nonlexical words in (22) allowed by (20b): (31)

a.

b. Doth with doubtfull melodious murmuring, w s w s ws w s ws (‘‘Elegie: ‘Oh, let mee not serve so’ ’’ 22) c. That would have need to be pittied by thee? w s w s w s ws w s (‘‘La Corona’’ 40) d. By sicknesse, deaths herald, and champion; w s w s ws w s ws (‘‘Holy Sonnet: ‘Oh my blacke Soule!’ ’’ 2) e. And then as worst surfets, of best meates bee, w s w s w s w s w s (‘‘The Crosse’’ 39) f. Make sinnes, else equall, in mee, more heinous? w s w s w s w s w s (‘‘Holy Sonnet: ‘If poysonous mineralls’ ’’ 6) g. Bu¤et, and sco¤e, scourge, and crucifie mee, w s w s w s wsw s (‘‘Holy Sonnet: ‘Spit in my face’ ’’ 2) h. Hee keepes, and gives to me his deaths conquest. w s w s w s w s w s (‘‘Holy Sonnet: ‘Father, part of his double interest’ ’’ 4)

Nonlexical Word Stress in the English Iambic Pentameter

43

This practice is not unprecedented in the tradition at Donne’s time; it is also found in Wyatt (as well as later in the tradition in other poets, most notably Milton) (Kiparsky 1977, 202): (32) a. There is written her faier neck rounde abowte: w s w s w s w s ws (VII: ‘‘Who so list to hounte’’ 12) b. Ffyre that purgith allthing that is vnclene w s w s w s w s w s (LXXVI: ‘‘Venemus thornes’’ 5) c. Thancked be fortune, it hath ben othrewise w s w s w s w s w s (XXXVII: ‘‘They fle from me’’ 8) But unlike Wyatt, Donne also sometimes allows strong syllables of lexical words in W where this condition is not met. The lines in (33a), (34b), (37c), and (38b) are cited by Kiparsky (1975b, 258–259) as evidence of Donne’s anomalousness with respect to the English iambic tradition, but additional lines of the same type given here show that they are not anomalous with respect to Donne’s own practice. Nor are they random violations of (6). Rather, all involve some form of right subordination. Those in (33) are directly parallel to Shakespeare’s phrases in (14) and nonlexical words in (23a,b) allowed by (20c): (33)

a.

b. I, like an usurpt towne, to’another due, w s w s w s w s w s (‘‘Holy Sonnet: ‘Batter my heart’ ’’ 5)

44

Kristin Hanson

c. O perverse sexe, where none is true but shee, w s w s w s w s w s (‘‘Twicknam garden’’ 26) d. Or is thy devout Muse retyr’d to sing w s w s w s ws w s (‘‘To Mr. R.W.: ‘Zealously my Muse’ ’’ 9) e. And begin soone, lest my griefe grieve thee too, w s w s w s w s w s (‘‘To Mr. B.B.’’ 9) f. Remembers what of old, shee esteem’d most, w s w s w s w s w s (‘‘The Lamentations of Jeremy’’ 27) g. As China, when the Sunne at Brasill dines. w s w s w s w s w s (‘‘To the Countesse of Bedford: ‘You have refin’d mee’ ’’ 18) h. Of refin’d manners, yet ceremoniall man, w sw s hi w s w s w s (‘‘Satyre I ’’ 28) Those in (34) di¤er from Kiparsky’s examples from Shakespeare in that the correctly matched syllable, while still in the same colon, is not in its DTE: (34) a. I said to all my profane mistresses, ws ws w s w s w s (‘‘Holy Sonnet: ‘What if this present’ ’’ 10) b. If faithfull soules be alike glorifi’d w s w s w sw s ws (‘‘Holy Sonnet: ‘If faithfull soules’ ’’ 1) Note that the scansions of (33) and (34) cannot be consistently explained as e¤ects of the Rhythm Rule, even though that rule certainly a¤ects stress in Donne in the same way it does in Shakespeare, giving pairs of lines like (35) parallel to (27): (35) a. No Crosse is so extreme, as to have none; w s w s w s w s w s (‘‘The Crosse’’ 14) b. But extreme sense hath made them desperate; w s w s w s w s ws (‘‘Elegie: Death’’ 8) The initial syllables of many of the misaligned words of (33) and (34) do not bear secondary stress; behold God in (33a), for example, is prosodically parallel to divine soul in (28). Moreover, there is nothing systematic about the scansion of the subordi-

Nonlexical Word Stress in the English Iambic Pentameter

45

nated words in (33) and (34); they sometimes occur in prosodically similar configurations but with their strong syllables in S, parallel to (28) (Combs and Suilens 1940):10 (36) a. So my devout fitts come and go away w s ws w s w s w s (‘‘Holy Sonnet: ‘Oh, to vex me, contraryes meet’ ’’ 12) b. As my prophane Love, and as soone forgott: w s w s w s w s w s (‘‘Holy Sonnet: ‘Oh, to vex me, contraryes meet’ ’’ 6) And in any case, the correctly matched syllable does not necessarily follow immediately, but in a few cases only more distantly, as in (37), parallel to Shakespeare’s (15) and (23d):11 (37)

a.

b. Open to’all searchers, unpriz’d, if unknowne. w s w s w s w s w s (‘‘Elegie: Change’’ 6) c. In vaine this sea shall enlarge, or enrough w s w s w s w s w s (‘‘Metempsychosis’’ 52) Thus what (33)–(34) and (37) seem to represent, especially when considered together with (31), is a replacement of (6) by an extension to lexical prosodic structure of some version of (20), in a significant but systematic innovation in the tradition.

46

Kristin Hanson

The result of the extension of (20c) in particular is a signature cadence of Donne, with a misaligned prominent syllable followed by a more prominent correct alignment often having a characteristic dramatic e¤ect of inducing, then resolving, tension. Some lines pose certain complexities in relation to this analysis. Those in (38) suggest that closer attention may be required as to how the rules for phrasing in (7) apply to coordinate structures: (38) a. Corrosivenesse, or intense cold or heat. w s w s w s w s w s (‘‘To Sr. Edward Herbert. At Julyers’’ 26) b. All demands, fees and duties; gamsters, anon w s w s w shi w s ws (‘‘Satyre V’’ 39 ) Intuitively, both are like those in (37) insofar as the entire coordinate structure is a single intonational phrase with its strongest syllable in the final S of its colon. But while (38a) also seems metrical by virtue of the more direct subordination of intense to cold, like the lines in (34), and even perhaps a potential e¤ect of the Rhythm Rule, (38b) di¤ers from the lines in (37) in lacking their regular triple rhythm, instead having a stressed syllable, fees, adjacent to the mismatched one. And those in (39) are metrical by virtue of (20) only if there is also assumed to be some special rhetorical stress on thee, an assumption that is plausible in both cases: (39) a. This kinde of beast, my thought shall except thee; w s w s w s w s w s (‘‘Elegie: The Expostulation’’ 24) b. Fear frownes? And, my Mistresse Truth, betray thee w s w s w s w s w s (‘‘Satyre IV’’ 163) But only three lines seem to fall outside the scope of (20) entirely:12 (40) a. Why should’st thou forget us eternally? w s w s *w s ws w s (‘‘The Lamentations of Jeremy’’ 385) b. Is not this excuse for mere contraries, w s w s *w s w s ws (‘‘Satyre III’’ 98) (41) How fresh our love was in the beginning; w s w s w s w s*w s (‘‘Elegie: His parting from her’’ 77)

Nonlexical Word Stress in the English Iambic Pentameter

47

(40a) comes close to being allowed by (20c); the strong syllable of eternally to which the mismatched one of forget is subordinated is within the same intonational phrase, and creates triple rhythms similar to those in (37). But it cannot be regarded as in the same colon without assuming the underlying structure to be that in (2a), counterintuitively introducing still other mismatches, since the colon boundary would then not match even a word boundary, and its strongest position would contain no stress. In (40b) the subordination of excuse is not within the same intonational phrase on the assumptions sketched in section 3.2.3. Both lines seem like self-conscious transgressions: (40a) in particular is a wrenching climax to the poem’s description of alienation from God just prior to its final plea, Restore us Lord, to thee; and that in (40b) is explicitly about being contrary. Alternatively, of course, they might represent the most extreme version possible of a constraint like (20c), allowing subordination that is limited only by the line. The line in (41), which is formally di¤erent, is also textually uncertain; preceded in its clause only by ‘‘the Spring,’’ it is syntactically problematic as well. We have up to this point avoided tampering with Kiparsky’s (1977) formalization as much as possible, casting the net in (20c) wide rather than struggling to articulate constraints he himself did not identify for Shakespeare, but at this point the issue of complexity that the range of examples from Donne raises bears mention. Intuitively, the lines in (33) that have the correctly aligned stress in the strong position of a colon seem simpler than those in (34) that do not. Those in (33) that have it following immediately also seem simpler than those in (37) that do not. (How those in (37) compare with those in (34) is not clear, suggesting perhaps di¤erent stylistic choices among metrical possibilities of equivalent complexity.) Even among those in (33), those in (33a–g) in which the correctly aligned syllable concludes a constituent at the colon boundary seem simpler than that in (33h) where it does not; similarly, a line like (16b), even if unmetrical, seems simpler than (16a). Hayes (1989b) even suggests that for constructions like those described by (20c), subordination itself might not be strictly necessary so much as just nonfinality of a mismatched syllable (though see also note 15), in which case the assumption of stress on thee in (39) might make those lines simpler, but not be necessary to their metricality. All these considerations suggest that (20c) is a conflation of conditions belonging to a larger set whose members’ identities and interrelationships ought properly to be teased apart and explored.13 As Youmans (1983, 91) notes, identifying such conditions may be more fundamental to metrical description than identifying any single inviolable rule, since they reveal the ‘‘plausible metrical relationships’’ among relaxations poets allow. Even if (20) is imperfectly formulated, however, and the di‰culty of making absolute statements is compounded by the textual, linguistic, and other metrical uncertainties involved in assessing Donne’s meter, (20) seems quite clearly to describe the trajectory along which his relaxation of (6) takes place. Donne’s practice may be

48

Kristin Hanson

extreme within the English tradition, but is not outside it: in it strong syllables of lexical words come to be governed by the same basic kinds of conditions that for Shakespeare govern those of postlexical prosodic structures. 3.4.2

Nonlexical Words

For nonlexical words, however, it is not at all clear that the same conditions obtain. Nonlexical words with initial stress cannot help but escape condition (20b), and so be parallel to the left-aligned lexical words in (31): (42)

f ss sw a. Marke, if she marke whether her woman came. w s w s w s w s w s (‘‘To Mrs. M.H.’’ 44) b. Or, as a Ship after much paine and care, w s w s w s w s w s (‘‘To Mr. Tilman after he had taken orders’’ 9) c. And shee’s unto her selfe a bitternesse. w s w s w s w s w s (‘‘The Lamentations of Jeremy’’ 16) d. So wee, If wee into our selves will turne, w s w s w s w s w s (‘‘To Mr. Rowland Woodward: ‘Like one who’in her third widdowhood’ ’’ 22) e. Why shoulds’t thou that dost not onely approve, w s w s w s w s w s (‘‘Satyre I’’ 37)

But those with final stress not only are positioned at the furthest reaches of what (20c) allows, but also violate it with frequency out of all proportion to the overall number of such words in the language. Prepositions appear with their strong syllable in W whether subordinated to an immediately following stronger stress as in (43a,b), or to a more distant one as in (43c,d), or followed only by an unstressed syllable to which they are not subordinated at all, as in (42e,f ); in the latter neither the meter nor the meaning seems to urge stress on the final pronouns in quite the same way the lines in (39) do: (43) a. Here upon earth, we’are Kings, and none but wee w s w s w s w s w s (‘‘The Anniversarie’’ 23)

Nonlexical Word Stress in the English Iambic Pentameter

49

b. Which still pursues them, without strength they go. w s w s w s w s w s (‘‘The Lamentations of Jeremy’’ 24) c. And against mee all day, his hand doth fight. w s w s w s w s w s (‘‘The Lamentations of Jeremy’’ 180) d. And without such advantage kill me then. w s w s w s w s w s (‘‘The Dampe’’ 16) e. And betweene us all sweetnesse may be had; w s w s w s w s w s (‘‘Sapho to Phil¾nis’’ 43)

f.

Conjunctions as in (44) and pronouns as in (45) also occur with their strong syllables in W whether (20c) is (distantly) met as in (44a,b) and (45a), or not, as in (44c) and (45b): (44) a. Seekes her at Rome, there, because hee doth know w s w s w s w s w s (‘‘Satyre III’’ 45) b. That wilt consort none, untill thou have knowne w s w s w s w s w s (‘‘Satyre I’’ 33) c. Is’t because thou thy self art blind, that wee w s w s w s w s w s (‘‘Elegie: His parting from her’’ 15) (45) a. Which himselfe on the Doctors did bestow; w s w s w s w s w s (‘‘La Corona’’ 46)

50

Kristin Hanson

b. Durst looke for themselves, and themselves retrive, w s w s w s w s w s (‘‘To Sr. Henry Wotton: ‘Sir, more then kisses’ ’’ 44) And most strikingly, certain adverbs occur with their strong syllable in W, again whether admitted under some interpretation of (20c) as in (46a–c), or not, as in (46d): (46) a. That wak’t from both, I againe risen may w s w s ws w sw s (‘‘La Corona’’ 83) b. A better Sun rose before thee to day, w s w s w sw s w s (‘‘Resurrection, imperfect’’ 4) c. What before pleas’d them all, takes but one sense, w sw s w s w s w s (‘‘Farewell to love’’ 18)

d.

Adverbs are commonly classified as lexical in English, but those in (46) are all of the kind traditionally referred to as particles. Syntactically, like all nonlexical words other than prepositions they do not head phrasal projections. Phonologically, in contemporary English at least, their behavior is somewhat inconsistent in its patterning with respect to the lexical/nonlexical distinction: enough in (47b) rejects phrasal stress just as the indisputably nonlexical words in (9) do, while others such as before in (48b) do not, patterning instead with the lexical adverb previously: (47) a. Make sure the suspect is su‰ciently we´ak. b. Make sure the suspect is we´ak enough. (48) a. That’s the excuse she used pre´viously. b. That’s the excuse she used befo´re.

Nonlexical Word Stress in the English Iambic Pentameter

51

But even if this suggests that the distinction between lexical and nonlexical words is gray in some cases, or its consequences not fully understood, these adverbs’ syntactic closed-class status, their relation to clearly nonlexical forms, and most important, their lack of participation in any morphologically complex structures, are consistent with the assumption that nonlexical status is the basis of their special metrical treatment; they are the exceptions that prove the rule. In sum, then, while nonlexical words may conform to (20) or even (6) more often than not, there is little evidence that (20) serves as the genuine constraint on them that it seems to be on lexical words. Even allowing for the possibility that a few lexical words as in (40) may be exceptions to (20), to treat the nonconforming nonlexical words as exceptions of the same type would leave the frequency with which exceptions to (20) involve nonlexical words as simply an unexplained coincidence. And because even among the nonlexical words conforming to (20c) so many involve the most distant placement of the corrective to the mismatch that it permits, this frequency cannot be attributed to the rhythmic configuration these words fall in most often, such as having a following stronger stress as in (43a,b). Rather, they just seem to be freer. 3.4.3

Phrases

This freedom that Donne accords nonlexical words is the same freedom that Kiparsky (1977, 206) notes Donne accords phrases. One of the most important sources of the commonly noted roughness of Donne’s iambic pentameter is that syllables that are strong within phrases fall freely in W, whether the conditions of (20) are met only in some attenuated form as in (49a) or (50a), or not at all as in (49b–e) and (50b): (49) a. Both the yeares, and the dayes deep midnight is. w s w s w s w s w s (‘‘A nocturnall upon S. Lucies day’’ 45)

b.

c. Whither, why, when, or with whom thou wouldst go. w s w s w s w s w s (‘‘Satyre I’’ 64)

52

Kristin Hanson

d. Sooner may one guesse, who shall beare away w s w s w s w s w s (‘‘Satyre I’’ 57) e. If our Soules have stain’d their first white, yet wee ws w s w s w s w s (‘‘To Mr. Rowland Woodward ’’ 13) (50) a. Perfect French, and Italian; I replyed, w s w s ws w s w s (‘‘Satyre I’’ 103) b. Fishes glide, leaving no print where they passe, w s w s w s w s w s (‘‘To Sr. Henry Wotton: ‘Sir, more then kisses’ ’’ 56) And they may do so whether the preceding syllable is in a nonlexical word as in (49), or a di¤erent lexical word as in (50), parallel to (16) and to (17) respectively, which are all unmetrical for Shakespeare. Thus for Donne as for Shakespeare, strong syllables of lexical words are treated one way, while those of nonlexical words and phrases are treated another way. This pattern follows naturally from the distinction between lexical and postlexical prosodic structure. The di¤erence between the two poets is that for Shakespeare, postlexical prosodic structure is subject to a more relaxed constraint than lexical stress is, while for Donne it is not subject to any. 3.4.4

Shakespeare’s Exceptional Lines Reconsidered

Before leaving Donne, one consequence of this analysis bears mention. Starting as it does from the broad description of Shakespeare’s practice in Kiparsky 1975b, 1977, and the poet-by-poet approach to a map of the English metrical tradition advanced there, these generalizations have been presented as whole-cloth analyses of individual poets’ metrical practice. But as briefly mentioned in section 3.3.1, exceptions to Kiparsky’s generalizations have periodically been cited. Some are spurious, but others are genuine, even involving some of the same words discussed above, like enough and divine (Youmans 1983).14 Most interestingly for the pattern under consideration here, lines like that in (51) become frequent in some of the late tragedies: (51) Upon my secure hour thy uncle stole, w s w s w s ws w s (Hamlet 1.5.61) Here the phrase my secure hour has exactly the same prosodic structure of my divine soul in (28), but is positioned metrically in just the way Kiparsky (1977) claims to

Nonlexical Word Stress in the English Iambic Pentameter

53

be impossible (Golston 1998). Across Shakespeare’s practice, the number of such lines may be small enough to be easily overlooked (though see Youmans 1983), but there is no question that within some plays they become a structure to be reckoned with (Hanson 2006). However, prosodically similar prepositions, to take one example of nonlexical words, are scanned as in (23) throughout Shakespeare’s practice, not only in certain periods or plays (Bartlett 1894). Lines such as (51) therefore do not obviate the need for metrical acknowledgment of the distinction between nonlexical and lexical prosodic structure advanced here. What they have consequences for is rather its history and its aesthetic significance. Donne’s lines in (33) may not be so innovative as they have been made out to be, and the patterns discussed here may reflect style and genre as much as author in ways that merit deeper exploration. 3.5

Shelley

If strong syllables of nonlexical words pattern with those of phrases in the practice of Shakespeare and Donne, however, they do not seem to do so in the practice of all poets. In Shelley’s iambic pentameter, certain observations point to the possibility that words may be treated the same way, whether lexical or nonlexical. 3.5.1

Lexical Words

Hayes (1989b, 251–252) observes that Shelley allows lines like those in (52), in which W contains strong syllables of lexical words that are phrased with a following word: (52)

a.

b. The distinct valley and the vacant woods w s w s w s w s w s (‘‘Alastor’’ 195)

54

Kristin Hanson

c. Was raised by intense pensiveness, . . . two eyes, w s ws w s w s ws (‘‘Alastor’’ 489) d. Its stony jaws, the abrupt mountain breaks, w s ws ws w s w s (‘‘Alastor’’ 551) e. She replied earnestly:—‘It shall be mine, ws w s w s w s w s (‘‘The Revolt of Islam’’ 1000) f. And though my cheek might become pale the while, w s w s w s w s w s (‘‘The Revolt of Islam’’ 1011) g. I remained still for tears—sudden she woke ws w s w s w s w s (‘‘The Revolt of Islam’’ 1087) The configurations are similar to those of Donne in (33) and (34), suggesting that on the approach adopted here, Shelley’s constraint on strong syllables of lexical words must include some condition like the right subordination of (20c), although the apparent absence of lines like (37) and (38) suggests there are also certain di¤erences. The syllable to which the mismatched one is subordinated seems always to follow immediately, and to be in the same phonological phrase (Hayes 1989b).15 The colon also seems to be less salient stylistically. Whereas in Donne’s lines the syllable to which the mismatched one is subordinated seems most often to be a monosyllable in the strongest position of the colon as in (33a–g), in Shelley’s the syllable to which the mismatched syllable is subordinated more often initiates a polysyllabic word as in (48a–e). A significant aesthetic di¤erence follows: whereas Donne’s practice foregrounds the underlying structure at the colon boundary after obscuring it at the foot, creating the sense of resolution of tension noted in connection with lines (33) and even (37)–(38), Shelley’s continuation of words across the colon’s boundary keeps its underlying structure in the background and creates more of a sense of insistence on the natural rhythms of speech. Shelley also allows lexical words to be positioned as in (53), analogous to Donne’s practice in (31):

Nonlexical Word Stress in the English Iambic Pentameter

55

(53)

Thus, as for Donne, for Shelley strong syllables of lexical words seem to be constrained by some version of (20), suitably revised to reflect the di¤erences discussed above. 3.5.2

Phrases

Shelley is also like Donne in allowing syllables that are strong in phrases to occur freely in W (Kiparsky 1977, 210; Hayes 1989b, 251): (54)

a.

b. Like a child, half in tenderness and mirth— w s w s w s w s w s (‘‘The Question’’ 14) c. Then was heard—‘‘He who judged let him be brought w s w s ws w s w s (‘‘The Revolt of Islam’’ 1999) d. Thou darest to speak—senseless are the mountains: w s w s w s w s w s hi (‘‘Hellas’’ 475) 3.5.3

Nonlexical Words

But nonlexical words do not seem to show the latter liberty. Those with final stress listed in Ellis 1892 appear with their stressed syllable in W only in configurations like

56

Kristin Hanson

those of the lexical words in (52).16 In most, the mismatched strong syllable is subordinated to an immediately following strong syllable with which the mismatched one is phrased: (55)

a.

b. And, lo! with gentle motion, between banks w s w s w s w s w s (‘‘Alastor’’ 399) c. Orb above orb, with starry shapes between, w s w s w s w s w s (‘‘The Revolt of Islam’’ 592) d. His reign and dwelling beneath nether skies, w s w s w s w s w s (‘‘The Revolt of Islam’’ 395) e. Rose from their thrones, built beyond mortal thought, w s w s w s w s w s (‘‘Adonais’’ 398) f. How without fear of evil or disguise w s w s w s ws w s (‘‘The Revolt of Islam’’ 950) g. For, before Cythna loved it, had my song w sw s ws w s w s (‘‘The Revolt of Islam’’ 928) h. Horsed upon stumbling jades, carted with dung, w s w s w s w s w s (Charles I I.170) i. The sun will rise not until noon. Apollo ws w s w s w s w s hi (‘‘Prometheus Unbound’’ 2.5.10)

Nonlexical Word Stress in the English Iambic Pentameter

57

Only (56a) shows any clear relaxation of the constraint, while (56b) requires the again-not-implausible assumption of stress on thee if it is to conform to it: (56) a. That among such as these my youth should seek its mate. w s w s w s w s w s w s (‘‘The Revolt of Islam’’ 693) b. But beside thee, where still my heart has ever been. w s w s w s w s w s w s (‘‘The Revolt of Islam’’ 18) Apart from these, there are no lines comparable to Donne’s (43e–f ), (44c), (45b), or (46d), in spite of the exact comparability of Shelley’s phrases in (54) to Donne’s in (49b–e). Thus, the word for Shelley appears to encompass both its lexical and nonlexical forms. 3.6

Conclusion

The distinction between lexical and postlexical prosodic structure thus seems to provide a parameter of variation that is independent of, yet interacts systematically with, the variation already known to exist in how poets treat words as opposed to phrases, even when the phrases are composed of sequences of unstressed and stressed syllables of the type that might be found in words. The overall resultant pattern for the tradition for the poets discussed here, setting aside the historical complexities raised in section 3.4.4, and the need for more careful analysis of (20c) acknowledged in section 3.4.1, is thus as in (57). (57) Distribution of strong syllables in W (9 ¼ prohibited except under initial relaxation; 9 ¼ prohibited under conditions like those in (20); r ¼ free) Lexical

Postlexical

Words

Phrases

Shakespeare Donne Shelley The pattern amply supports Halle and Keyser’s (1966) original observation that the stress of function words can be metrically irrelevant. Accompanying as it did their call for a generative reconceptualization of meter itself, Halle and Keyser’s observation swiftly became the locus of a theoretical reaction to their larger project so impassioned that to this day many literary critics and linguists view themselves as separated from each other by an unbridgable chasm. For

58

Kristin Hanson

example, Magnuson and Ryder (1970, 1971), questioning the assumption that any systematic description of a meter could successfully characterize all lines in a verse corpus, noted that Halle and Keyser’s claim about function words left unexplained Magnuson and Ryder’s judgment that Donne’s (58a) is patently more metrical than their construct in (58b): (58) a.

Behind, before, above, between, below. (Donne, ‘‘Elegie: Going to Bed ’’ 26) b. *Inbetween, before, beneath and beyond

On the analysis presented here, this judgment is reflected in the facts that (58b) is characterized as unmetrical for either Shakespeare or Shelley, and that in consequence, although it is characterized as metrical for Donne, it is also characterized as at the limit of his practice as defined by the relationship of his innovations to the tradition he belongs to. Perhaps more profoundly, Wimsatt (1970, 785), rejecting the assumption that iambic pentameter could lie elsewhere than in an iambic rhythm in the language of a poem, insisted that function words alter their stress under pressure from the meter when the two conflict as in (59): (59) Ther was also a Reve, and a Millere, (Chaucer, The Canterbury Tales, General Prologue 542) Without accepting Wimsatt’s claim that pronunciation actually needs to be changed, his sense of there being less fixity to the stress of such function words can be understood here as a sense of the obligatoriness of the meter’s respect for lexical as opposed to nonlexical stress. But what is more important, perhaps, is that attributing a role in metricality to such an abstract organizational principle of grammar as the distinction between lexical and postlexical components of the phonology need not be taken to cast the experience of meter as an abstract intellectual one rather than as the sensory one critics have traditionally emphasized (Wimsatt and Beardsley 1959). Metrical generalizations drawing on generative linguistics in this way are embedded in a broader assumption that the cognitive faculty that apprehends linguistic structure is as instinctive in its operation as sight, hearing, taste, or touch (Fodor 1983). In this light, the di¤erence that the debate over stress in function words seemed to crystallize can be seen to be spurious, with a sense of the immediacy of metrical experience in fact being shared by literary critics and generative linguists alike. Notes This chapter originated in a course on metrics that Paul Kiparsky taught in my third year as a graduate student, before which I had no intention of pursuing anything remotely related to

Nonlexical Word Stress in the English Iambic Pentameter

59

metrics, and since which it has been at the heart of my working life. I would therefore like to take this opportunity to thank him for that as well as for his comments on earlier versions of the chapter itself. I am also grateful for the suggestions and help of Katy Breen, Tom Cable, Draga Zec, two anonymous reviewers, and especially Sharon Inkelas. 1. Further details of the structures in (2) reflect the assumption that a foot that is not part of a binary grouping is adjoined to one in such a way as to preserve the generalization that strong constituents are heavy, details justified for Romance decasyllabic meters (Piera 1981), but whose correctness for English remains open. In this respect, these structures di¤er somewhat from those proposed by Kiparsky (1977, 230), but nothing in his description of the phenomena at stake here depends on his assumptions about the internal organization of the two cola. The issue is of special interest, however, in light of the importance of colon boundaries suggested below. 2. Specifically, when coupled with the claim advanced below that Donne’s meter treats postlexical prosodic structure as potentially irrelevant, the claim that the limit on the amount of language in any one metrical position is a moraic trochee has the consequence that multiple nonlexical words are freely allowed in single positions, in particular without regard to their weight, a generalization that appears correct for Donne (Hanson 1996). 3. These structures are admittedly rather hairy. Other forms of representation of prosodic structure such as grids are simpler in ways that have been argued to be not only welcome but correct, eliminating representation of levels of constituency and degrees of prominence claimed to have no linguistic relevance, most notably those arising from adjunction. However, the metrical generalizations of Kiparsky (1977) discussed here require reference to some of these levels of constituency (Hammond 1991). Where these have been argued not to be necessary for meter any more than they are for language, the argument has entailed an additional argument that the correct metrical generalizations themselves are in fact rather di¤erent from what Kiparsky (1977) proposes (Hayes 1983, 1989b), an issue I return to briefly in section 3.4. (See also notes 6 and 15.) Note also that while representations using bracketed grids (Halle and Vergnaud 1987) can resolve many of these di¤erences, issues involving prominence in meters other than the English iambic pentameter remain to be worked out. (See Hanson and Kiparsky 1996.) 4. Note that this class of syllables also excludes the strongest syllables of many compounds, because compounds consist of two separate phonological words (Kiparsky 1977, 191–193, though see also Youmans 1983). It should be noted, however, that if compounds were to be analyzed as phonological words themselves, some revision to (6) might be necessary. 5. An exception is object pronouns immediately following their verb, which Inkelas and Zec (1993) therefore classify as true clitics, and assume to be adjoined within their host’s phonological word, as in (13c). 6. In many of these cases Hayes’s (1989) clitic group, a constituent not used in Inkelas and Zec’s (1993) model, would be the constituent within which the relevant strength would be defined. Several issues would need to be resolved in order for the clitic group to be used in restating Kiparsky’s (1977) generalization, however. An explicit account of the distribution of stress on function words themselves would be required, addressing Inkelas and Zec’s (1993) distinction between function words in general and true clitics mentioned in note 5. More important, the question analogous to the one raised in note 3 would need to be answered, namely, whether there is any internal bracketing within clitic groups, as Kiparsky’s approach requires.

60

Kristin Hanson

7. An additional condition not discussed here must also be met: the two syllables preceding the mismatched strong one must themselves be either function words or part of the same word as the mismatched one (Kiparsky 1977; Hayes 1983). 8. The remarks in note 3 regarding grids are thus relevant here as well. See also note 15 regarding the alternative approach suggested by Hayes (1989b). 9. On Hayes’s (1989b) slightly di¤erent assumptions about phrasing, the structures of specific lines would be a little di¤erent, but not the overall pattern. 10. Another example might be the following, but it has an alternative scansion:



(

(i) Else, being alike pure, wee should neither see, w s ws w s w s w s w sw s w s w s w s (‘‘To the Countesse of Huntingdon: ‘Man to Gods image’ ’’ 33) 11. The internal prosodic constituency of words like surrounded is debatable, given that respecting morphological structure as shown here competes with respecting the English preference for left-headed feet (Liberman and Prince 1977); for discussion of parallel cases in Wyatt see Kiparsky 1977. It should also be noted that (37a) opens a poem famous for its games of rhythmic ambiguity, because it mixes tetrameter and pentameter lines. 12. Whether there are any others depends on what assumptions are made about certain other contentious issues. For example, the following line can be scanned on the assumption of intial catalexis: (i) j Making them confesse not only mortall w s w s w s w s w s hi (‘‘Satyre IV’’ 201) 13. Hayes’s (1989) exploration of the metrical consequences of the hierarchical structure of the prosodic hierarchy has this character. So, of course, do current theoretical claims that grammars are rankings of discrete constraints (Prince and Smolensky 1993). 14. For example, Golston’s (1998, 759) list of counterexamples to Kiparsky’s generalizations neglects Kiparsky’s own discussion of many of the exact same lines as involving changes in pronounciation (Kiparsky 1975b, 246), or the operation of the Rhythm Rule (Kiparsky 1975b, 247), or even the historical process by which a poet enters a tradition in which certain practices prevail but winds up excising them from his own (Kiparsky 1975b, 241). 15. On Hayes’s (1989b) own description, Shelley disallows a peak within a word at the right edge of a phrase. This rule, along with relevant observations about Shakespeare’s pattern discussed in section 3.3.2, is formulated within a proposed typology of metrical rules that reflects the general principle that beginnings may be lax while endings are strict, and therefore includes rules that impose special strictness at right edges, with variation among poets arising from the level of the prosodic hierarchy at which such strictness is imposed. The approach di¤ers from that of Kiparsky (1977) in so many interesting and interacting ways that it would require a separate paper to do the comparison justice. As far as the core point of this chapter goes, however, in spite of there being some genuine (though marginal) empirical di¤erences between the two approaches, as well as some di¤erence in aesthetic perspective, the di¤erences do not seem to a¤ect the need described here for distinct behavior of lexical and nonlexical words to be accounted for, nor it is obvious that there would be any impediment to incorporating an

Nonlexical Word Stress in the English Iambic Pentameter

61

account like that o¤ered here within Hayes’s approach. A brief discussion of the issue was excised from this chapter out of considerations of length. 16. Only one di‰cult example presents itself, but the problem involves excess syllables as much as the stress of around; possibly Pile a- occupies, anomalously, a single position. See Gerber 2001 for discussion of such structures in Stevens. 

(i) Pile around it, ice and rock; broad vales between w s w s w s w s w s (‘‘Mont Blanc’’ 63)

4

The Prosodic Word as a Unit in Poetic Meter

Draga Zec

4.1

Introduction

The epic decasyllable is a traditional South Slavic folk metrical form. The poems created in this rich oral tradition are of a highly formulaic nature, and possess a discernible metrical organization. My focus will be on the so-called Serbo-Croatian poems within this tradition (see Maretic´ 1907, 1935; Jakobson 1933, 1952), which have been documented in numerous volumes of verse. Among the most notable are the five volumes of Serbian epic poetry collected and published by Vuk Karadzˇic´ between 1814 and 1847, which will be the principal basis of this study.1 Although extensively documented, this verse form has eluded clear characterization. Each line of the epic decasyllable has ten metrical positions, with a caesura after the fourth, which divides it into two cola. A metrical position corresponds, strictly, to a syllable, while colon boundaries coincide with word breaks.2 (1) Mje``se¯c kaˆra¯ k zvije`zdu Da`nicu sssskssssss ‘The moon admonishes the morning star’

Vk, II, 98, 1

These are the obligatory, and just about the only uncontroversial, aspects of the structural organization of this verse. The establishment of further structural characteristics of a decasyllabic line has been fairly intractable. Claims that this is a periodic meter have not been supported by a straightforward prosodic characterization of the division of a line into metrical feet. Yet, any questioning of the periodic nature of the epic decasyllable has left unaccounted for the highly suggestive signs of its trochaic organization. My claim will be that the epic decasyllable is indeed a periodic, more specifically, a trochaic meter. The goal of this chapter is to identify the prosodic indicators of its trochaic organization, which, as will be shown, are of a somewhat unorthodox nature, and crucially include the prosodic word. The theoretical frame of reference will be Hanson and Kiparsky’s (1996) paramaterized account of possible metrical forms,

64

Draga Zec

and the basic premise therein about the dependence of poetic meter on the sound system of language, which sets the range of what is possible, or to be expected, in verse (Jakobson 1960; Kiparsky 1973g; Hanson 1992). Finally, I will rely on Optimality Theory (McCarthy and Prince 1993b; Prince and Smolensky 1993) in presenting the interactions among metrical properties, formalized as constraints, that characterize this metrical form. The organization of the chapter is as follows: In section 4.1, I present aspects of the Serbian prosodic system relevant for understanding this verse. In section 4.2, I establish the structural properties of the epic decasyllable, and in section 4.3, propose a formal account of its metrical organization. In section 4.4 I discuss earlier treatments of this verse. Sections 4.5 and 4.6 address the place of function words in this verse, and section 4.7 revisits the organization of higher order metrical constituency. Concluding remarks are given in section 4.8. 4.1.1

The Prosodic System

This section provides an outline of aspects of the Serbian prosodic system whose understanding is crucial for providing proper characterization of the epic decasyllable. I present the relevant prosodic characteristics of two major dialectal areas, Old and Neo-Sˇtokavian. The prosodic distinctions between these two dialect groups will shed crucial light on the prosodic basis of this verse form, as detailed in section 4.2. I focus first on the distribution of pitch accent in the Old and Neo-Sˇtokavian dialects, then on its prosodic nature, and the minimal prosodic unit within which it is realized. The Neo-Sˇtokavian dialects are traditionally characterized as having four pitch accents, two falling and two rising (Ivic´ 1958; Lehiste and Ivic´ 1986). As illustrated in (2), the falling accents may occur on the only syllable of a monosyllabic word, and on the initial syllable of a polysyllabic one. The rising accents, which are not found in monosyllables, may appear on any syllable of a polysyllabic word, with the exception of the final one. (2) Neo-Sˇtokavian pitch accents Short falling `` n a. Monosyllabic sa ‘dream’ b. Disyllabic me``se¯c ‘moon’ `` buka c. Trisyllabic ja ‘apple’

Long falling daˆn ‘day’ suˆnce ‘sun’ naˆmera ‘intention’

Short rising None

Long rising None

ju`na¯k ‘hero’ da`nica ‘morning star’ planı`na ‘mountain’

lju´ba ‘wife’ prı´lika ‘opportunity’ juna´ci ‘hero, plural’

The Prosodic Word as a Unit in Poetic Meter

65

The prosodic properties of the Old Sˇtokavian dialects are more straightforward: these are the dialects with only the falling accents, which may appear on any syllable within the word, including the final one, as in (3): (3) Old Sˇtokavian pitch accent Short falling `` n a. Monosyllabic sa ‘dream’ b. Disyllabic me``se¯c ‘moon’ `` lju¯ba ‘wife’ `` buka c. Trisyllabic ja ‘apple’ ` danı`ca ‘morning star’

Long falling daˆn ‘day’ suˆnce ‘sun’ junaˆk ‘hero’ naˆmera ‘intention’

The traditional diacritics utilized in (2) and (3) cover a complex of prosodic properties. Each symbol marks the presence of stress, as well as a characteristic pitch, either falling or rising. What the traditional diacritics do not mark is the disyllabic nature of the Neo-Sˇtokavian rising accents, with a high pitch characterizing both the accented syllable and its immediate successor. Moreover, in both dialect groups, the long accents appear on syllables with long vowels, while the short accents appear on syllables whose vowel is short. Vowel length, which is an independent phonological property, is also found on syllables unassociated with pitch accent—for example, on the second syllables of me``se¯c ‘moon’ and ju`na¯k ‘hero’. The Sˇtokavian system of pitch accents is of a complex nature: each pitch accent is manifested as both tone and stress. Although distinct, these two components of pitch accent are integrated through the system of feet, which includes both the standard trochaic set, in (4a), and the set characterized by tonal prominence, in (4b), as argued in detail in Zec 1999. (4) Foot inventory a. Without tone i. [sm m ]F ii. [sm sm ]F

b. With tone i. [sm m ]F H ii. [sm ]F H

Pitch accent is realized within the prosodic word. Each prosodic word obligatorily contains a tonal foot that serves as its head and is the bearer of stress, as shown by the forms parsed into feet in (5) and (6). This foot (designated by underlining) is the

66

Draga Zec

leftmost one in the prosodic word in the Neo-Sˇtokavian, but not necessarily in the Old Sˇtokavian.3 (5) Neo-Sˇtokavian `` n)]po a. [(sa H d. [(lju´ )(ba)]po H H

b. [(daˆn)]po H e. [(da` )(ni)ca]po H H

(6) Old Sˇtokavian `` n)]po a. [(sa H `` )]po d. [(lju¯)(ba H

b. [(daˆn)]po H e. [da(nı``)ca]po H

c. [(me`` )(se¯c)]po H f. [( ju` )(na¯k)]po H H c. [(me`` )(se¯c)]po H f. [ ju(naˆk)]po H

Because the stressed syllable has to be associated with tone, ‘‘standard’’ trochaic feet in (4a) may never serve as heads of prosodic words. However, they may appear in nonhead position: (4ai) appears in nonhead position in (5c) and (6c–d). Nonhead tonal feet are relevant only for the Neo-Sˇtokavian, and are exemplified in (5d–f ). Both ‘‘standard’’ trochaic feet and nonhead tonal feet are active in prosodic morphology, as argued in Zec 1999. The head of the prosodic word is associated with two types of prominence: stress, which is relational, and tonal salience, which is not. The head foot bears the highest degree of prominence by virtue of stress, which is phonetically realized as increase in duration; in (5d–f ) stress alone makes the head foot more prominent than the immediately following nonhead tonal foot. The head foot is also characterized by tonal salience, which distinguishes between feet associated or not associated with a High tone. Tonal salience, as already noted, is the nonrelational sort of prominence, its phonological status being predicated solely on the presence or absence of a High tone. As already noted, pitch accent is realized within the prosodic word. Given the division of the lexicon into lexical and functional elements, only elements in the former class invariably correspond to prosodic words. As argued in Zec 2005, the prosodic properties of functional elements are more varied, their prosodic behavior depending on whether they belong to the class of clitics—that is, bound function words—or free, nonclitic function words. As will be shown in section 4.6, this distinction plays an important role in the organization of the epic decasyllable. While lexical words (WL ) are unexceptionally associated with pitch accent and prosodic word status, as stated in (7), the prosodic status of free, nonclitic function words (WF ) depends on their size. Those that are minimally disyllabic possess a pitch accent and receive the status of a prosodic word. Those, however, that correspond to

The Prosodic Word as a Unit in Poetic Meter

67

a single syllable are accented and endowed with the prosodic word status only if associated with sentential focus, as stated in (8). (7) If WL , then [WL ]po . (8) If WF , then [WF ]po i¤ WF is minimally disyllabic, or associated with focus. In sum, monosyllabic free function words lack the prosodic status and, as such, disrupt the prosodic hierarchy (as defined in Selkirk 1980, Nespor and Vogel 1986, and Hayes 1989b). As argued in Zec 2005, such monosyllabic elements are incorporated directly into the prosodic phrase. The organization of the epic decasyllable is highly sensitive to the prosodic properties of monosyllabic function words: their distribution within a metrical line is considerably restricted in those cases when they lack prosodic salience. Bound function words—those designated as clitics—cannot form a prosodic word on their own under any circumstances. Rather, they subcategorize for a prosodic word, either as proclitics, as in (9a), or as enclitics, as in (9b) (Inkelas 1989); the clitic forms a nested prosodic structure with its host. (9) a. [clitic [host]po ]po ] b. [[host]po clitic]po Both lexical and functional elements may serve as clitic hosts. Moreover, monosyllabic free function words, which are also possible clitic hosts, form with the clitic a disyllabic unit with nested prosodic word structure, which then makes them eligible for pitch accent (Zec 2005). The clitic, in this case, endows its host with the prosodic word status (see Halpern 1992 for other cases of this type). A monosyllabic free function word thus corresponds to a prosodic word in yet another case: not only when the monosyllabic function word is focused, as stated in (8), but also when it is associated with a clitic. At most one pitch accent is associated with a prosodic word, regardless of whether it is simplex in structure, as in (10a), or possesses a nested structure, as in (10b–c): (10) a. [daˆn]po ‘Day’ b. [na [daˆn]po ]po ‘On (the) day’ c. [[daˆn]po je]po ‘(The) day is’ The nested structure created by the clitic and its host is ambiguous in one respect: phonological processes whose domain is the prosodic word may select either the larger or the smaller scansion of the nested structure and, as argued in Zec 1993,

68

Draga Zec

individual dialects, or local idioms, select either the larger or the smaller scansion. However, the larger scansion is relevant for the purposes of prosodic parsing, since it alone yields an exhaustively parsed prosodic constituency. In the epic decasyllable, the larger scansion is selected in the general case, as demonstrated in section 4.6. I take this to be a fact about the linguistic, rather than the metrical, makeup of this verse. 4.2

Prosodic Basis of Metrical Organization

The arguments, both for and against, the periodic organization of the epic decasyllable have relied on the distribution of pitch accent. 4.2.1

Distribution of Pitch Accent

As observed in the literature (Zima 1909a, 1909b; Maretic´ 1907; Jakobson 1933, 1952), the occurrence of pitch accents in this verse appears to be fairly free, with one notable regularity: colon and line-final syllables, the fourth and the tenth, may not bear pitch accent. These distributional properties have been stated with the Neo-Sˇtokavian prosodic system in view, and are illustrated in (11); the accent marks follow the Neo-Sˇtokavian accentuation. ` lo`vi k cı`jel ` daˆn po go`ri, (11) a. Dmı`tar `ˇta ulo`viti: b. i ne` mozˇe k nı`s `` nese c. naˆmjera ga k pred ve``cˇe na d. na ze`leno¯ k u go`ri je``zero, e. u je``zeru k `u`tva zlato`krila; ` k sıˆvo¯ga so `` kola `` sti Dmı`tar f. pu g da u`vati k `u`tvu zlato`krilu, Vk, II, 97, 64–70

‘Dmitar was hunting, all day, in the mountain, but could not catch anything. A purpose brought him, toward the evening, to a green mountain lake. In the lake, there was a gold-winged duck. Dmitar released his gray falcon, to seize the gold-winged duck.’

While the fourth and tenth positions consistently remain unaccented, all other positions in a line, both odd and even, may bear accent: the first syllable is accented in lines (a), (c), and (f ); the second in (b), (d), (e), and (g); the third in (a) and (f ); the fifth in all lines other than (c) and (d); the sixth in lines (c) and (d); the seventh only in line (a); the eighth in all lines other than (a); and the ninth in line (a).4 In light of this, any assumption that the epic decasyllable is a periodic meter opens the issue of the headedness of metrical feet. The problem is illustrated in (12): while in (12a) (repeated from (11a)) each accent is associated with an odd position in the line, in (12d) (repeated from (11d)), each is associated with an even position. (Parentheses

The Prosodic Word as a Unit in Poetic Meter

69

in (12) designate the division into metrical feet; accented syllables are highlighted by underlining.) ` ` (12) a. (Dmı`tar) (lo`vi) k (cı`jel) (daˆn po) (go`ri), b. (na ze` )(leno¯) k (u go` )(ri je`` )(zero),

Cf. (11a) Cf. (11d)

The issue of whether the distribution of pitch accents is metrically significant for the organization of this verse led in fact to divergent views on the nature of this meter. Under one view held, most notably, by Jakobson (1933, 1952), pitch accent exhibits a trochaic tendency, manifested as an overall statistical bias toward accentedness in the odd positions of the meter, and its absence in the even positions.5 In the data he presents, based on two poems (783 lines) by the Montenegrin singer T. Vucˇic´, given in (13), the percentage of accented syllables is generally lower in the even than in the odd positions of the line.6 (13) The distribution of pitch accents in Vucˇic´’s poems (783 lines) (Jakobson 1952, 420) 1

2

3

4

5

6

7

8

9

10

62

31

53

—

75

22

34

35

42

—

Thus, according to Jakobson (1952, 420), ‘‘The even syllables of the line show a marked tendency to be unaccented and the odd syllables a less marked tendency to carry word accents: even syllables are less frequently accented than the following odd syllables.’’ But, with the exception of the fourth and tenth positions, the degree of accentedness in the even positions makes the trochaic tendency fairly weak. In (13), the eighth position has a slight advantage over the preceding seventh position, which according to Jakobson does not lessen the trochaic tendency. But the data from Vuk’s collection show that in certain poems, the eighth position exceeds not only the seventh, but also the ninth position in the line.7 In sum, if the epic decasyllable is indeed a trochaic meter, one would expect much more robust prosodic indicators than is the case with pitch accent. Under the alternative view, this meter is seen as syllabic in nature, and any apparent trochaic tendency is attributed to the phonological properties of the language rather than to the pressures of the meter (Zima 1909a, 1909b; Maretic´ 1907; Vaillant 1932; Matic´ 1964). This view, however, is overly pessimistic in assuming that no prosodic property other than syllable count is subject to metrical patterning. The one robust prosodic regularity that emerges is the absence of pitch accent  stands for from the fourth and tenth positions in the line, as stated in (14) (where s a syllable bearing any of the four pitch accents): kssssss (14) a. *s s s s  b. *s s s s k s s s s s s

70

Draga Zec

As we will see, the regularity in (14) does reflect the trochaic organization of the epic decasyllable, once the prosodic indicators of its metrical organization are properly identified. 4.2.2

Accents of Mono- and Polysyllables

A clearer metrical characterization of this verse emerges if we narrow the focus on the distribution of the accents of monosyllables, which, as will be argued, play a central role in the trochaic patterning of this verse. Relevant here is the distribution of monosyllables that constitute prosodic words and, as such, bear pitch accent.8 It is important to note that within the bounds of the Neo-Sˇtokavian accentuation, the restriction in (14) is metrically relevant for monosyllabic, but not for polysyllabic prosodic words. Polysyllabic words, whose final syllable may not bear pitch accent in this dialect, as noted in section 4.1, vacuously meet this restriction by virtue of their prosodic makeup. Thus, only the absence of monosyllables from the colon or linefinal syllable has to be viewed as metrically motivated since, by occupying these positions, monosyllables would have endowed them with pitch accent. But what role, if any, is to be ascribed to the accents of polysyllables? While the Neo-Sˇtokavian dialect, whose polysyllables may not bear final accent, provides no useful clues, the Old Sˇtokavian does since, in this dialect, the accent of a polysyllabic word may occur on any syllable, including the final one. Crucially, in the Old Sˇtokavian rendition of the epic decasyllable, accents of polysyllabic words do occur in colon and line-final metrical positions. Evidence for this comes from three Old Sˇtokavian sources of this verse: the Montenegran cycle in the fourth volume of Vuk’s collection, Njegosˇ’s Ogledalo srpsko, and the poems collected in Parry and Lord 1953.9 In (15)–(17), accents in the fourth position are given in the (a) examples, and accents in the tenth position, in the (b) examples. Accents on the fourth syllable occur in about 8 percent of lines, and on the line-final syllable, in about 4 percent.10 (15) Vuk’s collection `` n k ne obra `` c´a gla¯veˆ a. Za toˆ Iva ‘To that, Ivan did not turn his head’ `` k k svojoˆj go¯spı`` kaˆzˇe¯ Onıˆ sa¯na ‘She told that dream to her lady’ `` m rukoˆm k `u`fati oruˆzˇe Jedno ‘With one arm, he seized the gun’ `` du bje¯zˇeˆ b. Svi `o`vcˇa¯ri k uz liva ‘All shepherds ran up along the meadow’ `` nj da¯jeˆ Pa¯ `o`bjema k zˇ ˆıvı¯ oga ‘And with both, he produced live fire’ `` n k ne obra`` c´a gla¯veˆ Za toˆ Iva ‘To that, Ivan did not turn his head’

Vk, IV, 2, 91 Vk, IV, 12, 10 Vk, IV, 12, 55 Vk, IV, 12, 57 Vk, IV, 2, 119 Vk, IV, 2, 91

The Prosodic Word as a Unit in Poetic Meter

(16) Njegosˇ’s Ogledalo srpsko `ı je oraˆj k polomı`o ` krıˆla a. N`t ‘Nor did the eagle break his wing’ `` gle¯ kliku `` je¯ Ali junaˆk k iz ma ‘But the brave man called from the mist’ `` tı¯n dru¯gaˆh male``n indaˆt k `o`sam sto ‘A small squadron, of eight hundred companions’ `` c´e do `` bra¯ k ispod se``be djo¯ga `` b. da ‘He will give the good white horse under him’ `` tı¯n dru¯gaˆh male``n indaˆt k `o`sam sto ‘A small squadron, of eight hundred companions’ `` c´e za me¯ k se``dam dzˇeferdaraˆh da ‘He will give seven rifles for me’ (17) Parry and Lord’s collection11 `` rı¯ serda´ru a. Paˆ ovakoˆ k govo ‘And spoke thus to the sirdar’ Evo serdaˆr k do druzˇineˆ prı´dje ‘Then the sirdar approached the band’ `ˇta ne govo `` rı¯ `` kı¯ sˇu¯tıˆ k nı`s Sva ‘Each was silent and said no word’ ` ¯ da vo¯jskeˆ Spreˆmı¯ sultaˆn k stoˆ hı`lja ‘The sultan sent an army of a hundred thousand men’ `` zbolova¯ k se``dam go `` din’ da¯naˆ b. Ra ‘He was ill for seven years’ `` nom k devedeseˆt dru¯gaˆ Za Osma ‘Behind Osman were the ninety comrades’ ` ¯ da vo¯jskeˆ Spreˆmı¯ sultaˆn k stoˆ hı`lja ‘The sultan sent an army of a hundred thousand men’

71

Nj, 2, 7 Nj, 54, 50 Nj, 54, 86 Nj, 2, 30 Nj, 54, 86 Nj, 2, 26

P&L, 31, 86 P&L, 31, 56 P&L, 1, 43 P&L, 1, 14 P&L, 26, 2 P&L, 31, 17 P&L, 1, 14

Under the plausible assumption that the epic poems composed in the Old and NeoSˇtokavian dialects not only belong to the same poetic tradition, but are also in the same poetic meter, we conclude that the absence of the accents of polysyllables from colon-final positions in the Neo-Sˇtokavian dialect, while being a fact about its phonology, is not a fact about this verse form (contra Matic´ 1964). The accents of monosyllables, however, may not appear in colon- and line-final position in the Old-Sˇtokavian rendition of the epic decasyllable, just as they may not appear in its Neo-Sˇtokavian version. I will take this to be a strong indicator of trochaic patterning, and focus, in the following section, on the distribution of monosyllabic prosodic words in this verse.

72

4.2.3

Draga Zec

Distribution of Monosyllables

As noted in Maretic´ 1907, 53–54, the fourth and tenth positions in the line are not the only metrical positions from which monosyllables are excluded. Monosyllables are also prohibited in the eighth position. However, this metrical position is relatively frequently associated with the accents of polysyllables in both the Old and NeoSˇtokavian renditions of the epic decasyllable. Thus, while the Neo-Sˇtokavian, but not Old Sˇtokavian, may be seen as correlating the absence of the accents of monoand polysyllables in the fourth and tenth positions, the eighth metrical position does not exhibit any such correlation in either dialect group. The exclusion of monosyllabic prosodic words from the fourth, eighth, and tenth positions emerges as a central trait of this meter. Monosyllables thus may occupy a highly restricted set of metrical positions, as shown in (18)–(23). In addition to lexical monosyllables, which include numerals (the most frequent monosyllabic prosodic words in this meter), examples also include focused functional monosyllables. As is generally the case in this highly formulaic verse, specific monosyllables are often restricted to fixed metrical positions, or recur in fixed syntactic collocations.12 In the first colon, monosyllables may occur in the first syllable, as in (18), or the second one, as in (19). For the sake of explicitness, the entire colon is parsed into prosodic words. The prosodic word in the relevant metrical position, as well as its translation, is underlined. (18) First colon: First position [s]po s s s k s s s s s s a. [Grad]po [gradili]po k tri godine dana, ‘They were building the city for three years’ b. [konj]po [do [konja]po ]po k junak do junaka, ‘Horse next to horse, knight next to knight’ c. [tri]po [tovara]po k blaga zadobisˇe ‘Three loads of treasure, they acquired’ d. [svi]po mi [[da]po se]po k u so prometnemo, ‘If we all turned to salt’ (19) First colon: Second position s [s]po s s k s s s s s s a. Vec´ [daj]po [mene]po k tvoju vjeru tvrdu ‘But give me your firm word of honor’ b. Sˇto [car]po [reko]po k to i ucˇinio ‘What the czar said, he made good’ c. Jer [tri]po [fata]po k u visinu skacˇe ‘As he leaps three yards in height’

Vk, II, 25, 7 Vk, II, 49, 22iv Vk, III, 3, 97 Vk, II, 49, 10iv

Vk, II, 63, 45 Vk, III, 12, 167 Vk, III, 60, 26

The Prosodic Word as a Unit in Poetic Meter

73

Under a limited set of circumstances, monosyllabic prosodic words may also occupy the third position: when immediately followed by a functional monosyllable, which has no prosodic status (as detailed in section 4.1): (20) First colon: Third position s s [s]po s k s s s s s s a. [[da]po se]po [ja]po sad k na vojsku opremam ‘That I am now making myself ready for the army’ b. [i [tu]]po [ja]po bih k u boga zlocˇesta ‘There as well I was evil before God’ c. [[Da]po c´esˇ ]po [ti]po nas k kadgodj izbaviti ‘That you will one day rescue us’

Vk, II, 90, 21 Vk, II, 4, 85 Vk, III, 51, 23

In the second colon, a monosyllable may occupy the first syllable, as in (21); the second, as in (22); and the third, as in (23); the entire colon is, again, parsed into prosodic words. (21) Second colon: Fifth position s s s s k [s]po s s s s s a. Ja njoj dado k [struk]po [rumene]po [ruzˇe]po ‘I gave to her a stem of red rose’ b. posˇljite mu k [list]po [knjige]po [bijele]po ‘Send to him a leaf of white book’ c. Grad gradili k [tri]po [godine]po [dana]po ‘They were building the city for three years’ d. pa po svili k [sav]po [izvezen]po [zlatom]po ‘And on silk, embroidered all in gold’ (22) Second colon: Sixth position s s s s k s [s]po s s s s a. ‘‘Sjedi, care k pak [pij]po [rujno]po [vino]po ’’ ‘Sit here, czar, and drink of red wine’ b. i stolove k dje [car]po [sedijasˇe]po ‘And the tables where the czar had sat’ c. bolje nam je k [sva]po [tri]po [poginuti]po ‘It would be better if we all three died’ (23) Second colon: Seventh syllable s s s s k s s [s]po s s s a. Lov lovio k [letnji]po [dan]po [do [podne]po ]po ‘He was a-hunting on a summer day, until noon’ b. ja sam noc´as k [zao]po [san]po [usnio]po ‘Last night, I dreamt an evil dream’

Vk, II, 29, 284 Vk, II, 28, 121 Vk, II, 25, 7 Vk, III, 60, 4

Vk, II, 28, 479 Vk, II, 94, 108 Vk, II, 38, 55

Vk, II, 14, 20 Vk, II, 9, 77

74

Draga Zec

c. kovala mu k [sablju]po [dva]po [kovacˇa]po ‘His sabre was forged by two blacksmiths’ d. Neka sudi k [kao]po [ja]po sˇto [sudim]po ‘May he adjudicate, just as I adjudicate’

Vk, II, 43, 651 Vk, III, 12, l51

In contrast to the third position, however, the ninth position cannot be occupied by a monosyllabic prosodic word. This is because the prosodic configuration that allows for the occurrence of a monosyllabic prosodic word in the third metrical position, as shown in (20), may not be replicated for the ninth position. This will be attributed, in section 4.6, to a special status of the line-final foot, one of its manifestations being a prohibition against functional monosyllables in the tenth position of the meter. To conclude, monosyllabic prosodic words could in principle occupy any odd position in the line, including the third and the ninth; and, among the even positions, only the second and sixth. However, conspicuously absent are the following arrangements of monosyllabic prosodic words: (24) a. *s s s [s]po k s s s s s s [gradili]po [grad]po k tri godine dana (Constructed on the basis of Vk, II, 25, 7) b. *s s s s k s s s [s]po s s Neka sudi k [kao]po sˇto [ja]po [sudim]po (Constructed on the basis of Vk, III, 12, l51) c. *s s s s k s s s s s [s]po ja sam noc´as k [usnio]po [zao]po [san]po (Constructed on the basis of Vk, II, 9, 77) Thus, what remains to be explained is the absence of monosyllables in the fourth, eighth, and tenth positions. Their absence from the ninth position results from strong distributional restrictions associated with the final foot, to be addressed in section 4.7. It should be noted that the distribution of monosyllables in the epic decasyllable sharply contrasts with the ways monosyllables pattern in poetic traditions such as English or Russian, in which they freely occupy weak metrical positions. 4.3

Organization of the Epic Decasyllable into Metrical Feet

The relevant descriptive generalization about the epic decasyllable—that a monosyllabic prosodic word may not occupy the fourth, eighth, and tenth position in the line—clearly indicates a binary pattern of the decasyllabic verse. I will further argue that it also indicates its trochaic nature. With a line organized into five trochaic feet, two in the first and three in the second colon, a monosyllabic prosodic word is excluded from the weak position of the foot. This is represented in (25), with feet built

The Prosodic Word as a Unit in Poetic Meter

75

over syllable-sized metrical positions, and the ‘‘strong’’ (s) and ‘‘weak’’ (w) marking indicating the head/nonhead relation within a foot (Halle and Keyser 1971a; Kiparsky 1977; Prince 1989). (25) *

s w s w s w s w s w (s [s]o ) (s [s]o ) k (s [s]o ) (s [s]o ) (s [s]o )

This restriction is relaxed in colon-initial feet, the first and the third, those commonly associated with the weakening of restrictions imposed by poetic meter (Kiparsky 1977; Hanson 1992). This prohibition thus takes proper e¤ect in the second, fourth, and fifth foot. But if (25) captures the structural organization of the epic decasyllable, how is this metrical pattern prosodically realized? This will be examined by invoking the parameters of poetic organization, proposed by Hanson and Kiparsky (1996, 292). These parameters set the range of variation in periodic meters and crucially distinguish between the structure of metrical feet, in (26b), and their prosodic realization, in (26a). The latter is reflected in the prosodic characterization of metrical position, the type of prominence utilized by the meter, and the site within the foot in which prominence is constrained. (26) General parameters of periodic meters (Hanson and Kiparsky 1996) a. Structure i. Number of feet (1/2/3/4 . . .) ii. Headedness of feet (right/left) b. Realization i. Position size: maximal size of a metrical position (m/s/f/o) ii. Prominence site: where prominence is constrained (S ) sU/ S ) P/ W ) sP/ W ) U) iii. Prominence type: what prosodic category defines prominence (weight, stress, strength, pitch accent) By proposing that the epic decasyllable is a trochaic pentameter, we have already fixed the structural parameters, the number of feet and their headedness. Some of the realization parameter settings are straightforward. As already noted, the size of a metrical position clearly corresponds to a syllable—since a decasyllabic line is invariably exactly ten syllables long. Much less straightforward is the identification of prominence type. The restriction on the distribution of monosyllables in (25) strongly suggests that the prosodic entity relevant for this meter is the prosodic word, which may not exhaustively occupy a weak position in a metrical foot. Whatever prominence type may be represented by the prosodic word, it can at least be stated that prominence is excluded from the weak positions of the meter, which sets the prominence

76

Draga Zec

site to W ) sP (if weak, then not prominent). Thus, in (25), the prosodic word perniciously contributes prominence to metrically weak positions. But what is the phonological basis for the prosodic word, and in particular, for the monosyllabic prosodic word, to serve as the source of prominence in this meter? Any prominence that the prosodic word may possess is entirely due to its a‰liation with pitch accent, a bundle of properties associated with the head of the prosodic word. As noted in section 4.1, pitch accent is characterized both relationally, as signaled by stress, and nonrelationally, by virtue of tone. It is the latter property that is dominant: the entire bundle of properties inherits from tonal salience its nonrelational nature and, as a result, pitch accents in mono- and polysyllabic words will be on an equal footing. Thus, if taken as the basis of prominence, pitch accent will make mono- and polysyllabic forms equally prominent. But taking pitch accent to be the prominence type of the epic decasyllable is at odds with the fact that the accent of monosyllables, but not that of polysyllables, is subject to distributional restrictions in this meter. As already shown in section 4.2, the accents of polysyllables are in no way metrically restricted. Crucially, it is the pitch accents of monosyllables that this meter singles out as its prominence basis. The obvious question, then, is how to grant special status to the accents of monosyllables. The solution proposed is the following: while pitch accent is taken to be the prominence type selected by this meter, metrical restrictions are placed not directly on the distribution of pitch accent, but indirectly on its head domain. Since the syllable bearing pitch accent corresponds, of course, to a prosodic word’s head, the relevant domain will correspond to the prosodic word. The relevant properties regarding the realization of the epic decasyllable’s trochaic structure are summarized in (27): (27) Periodic organization of epic decasyllable a. Structure i. Number of feet (five) ii. Headedness of feet (left) b. Realization i. Position size: s (syllable) ii. Prominence site: W ) sP iii. Prominence type: pitch accent contained within its head domain Our initial empirical generalization has been that monosyllabic prosodic words play a crucial role in propelling the trochaic pattern of this meter. But although, in descriptive terms, monosyllables obviously play a significant role, there is no mention of prosodic word size in the formal account of this meter in (27). This is because the special status of monosyllables is only derivative. It results from the interaction of

The Prosodic Word as a Unit in Poetic Meter

77

two aspects of the realization of this verse: position size and prominence type. Since a metrical position corresponds exactly to a syllable, it follows, trivially, that only a pitch accent’s head domain of precisely that size exhaustively occupies a single metrical position. In sum, monosyllables have no special status among prosodic words, but appear to be playing a special role in this verse because the size of a metrical position invariably corresponds to a single syllable. A further consequence of the characterization in (27) is the asymmetry between the accents of mono- and polysyllabic prosodic words. While the former are directly targeted by the metrical restrictions on this meter, the latter are left unrestricted. Those pitch accents that are heads of polysyllabic prosodic words are free to appear in any metrical position within the line, and they indeed do, as we have seen in section 4.2. In sum, the requirement regarding the head domain of pitch accent takes e¤ect only when the head domain’s size is identical to the size of a metrical position. To conclude, the prosodic realization of the epic decasyllable’s trochaic structure is captured by prohibiting pitch accent, contained within its head domain, from occupying the weak positions of meter. This requirement is relaxed in colon-initial feet, which freely admit monosyllables in weak positions, stated in (28). (28) INITIAL LICENSE 1 In colon-initial feet (first and third), the weak position may be occupied by a pitch accent contained in its head domain—that is, by a monosyllabic prosodic word. This constraint is one of the two colon-initial relaxations of metrical conditions; the second one will be presented in section 4.6. As a result of ranking (28) above (27b), monosyllabic prosodic words are prohibited from the fourth, eighth, and tenth, but not from the second and sixth positions. In the following section I present Jakobson’s influential analysis of this meter, and compare it with my proposal, which, I argue, is superior to Jakobson’s. In sections 4.5 and 4.6, I present the behavior of function words in this meter: the distribution of monosyllabic prosodic words that serve as proclitic hosts in the former, and the peculiar patterning of monosyllabic free function words in the latter. Section 4.7 addresses the higher-level organization into cola and subcola. 4.4

Comparison with Previous Analyses

As already noted, the epic decasyllable has been characterized both as syllabic and as periodic verse. I focus here on Jakobson’s (1932, 1933, 1952) proposal, which is by far the most detailed analysis in the literature. In his studies of metrical forms, Jakobson generally distinguishes between metrical constants and metrical tendencies, and this approach constitutes the core of his analysis of the epic decasyllable.

78

Draga Zec

In his study of Slavic epic verse, Jakobson (1952, 418) grants special status to colon- and line-final positions in terms of13 ‘‘[a] ‘bridge’ or ‘zeugma’ at the end of both colons: the fourth and the tenth syllable belong to the same ‘word unit’ as the third and ninth syllable, respectively. They are consequently accentless in Sˇtokavian dialects, in which the accent has shifted from the final syllable.’’ In other words, the last two colon-final syllables are stipulated to belong to the same word. By virtue of this, monosyllables are excluded from the fourth and tenth positions. Additionally, any polysyllable satisfying the ‘‘bridge’’ will leave the colon-final position unaccented in the Neo-Sˇtokavian dialect; that this is not the case in the Old Sˇtokavian dialect is fully compatible with Jakobson’s definition of ‘‘bridge.’’ This, according to Jakobson, is a crucial metrical constant of this verse, and thus plays a markedly di¤erent role from pitch accent, which in his analysis encodes a metrical tendency toward trochaic rhythm, as mentioned in section 4.2. However, while the ‘‘bridge’’ makes correct predictions about the absence of the Neo-Sˇtokavian accent in colon- and line-final positions, it does not provide a correct empirical generalization about the distribution of monosyllables in this verse. The ‘‘bridge’’ technically excludes monosyllables from the fourth and tenth positions in a line, yet as we have seen, these metrical positions present only a subset of those that do not allow monosyllables. As demonstrated in section 4.2, monosyllables are excluded not only from the fourth and the tenth, but also from the eighth position in the line, and this is fully captured by my analysis in section 4.3. In his earlier work on the epic decasyllable, Jakobson (1932, 1933) does acknowledge restrictions on the distribution of monosyllables:14 Die einsilbigen Wortganzen fallen auf die ungeraden Verssilben. Daher ist die Konstante abgeleitet, dass sich vor der vierten und zehnten Verssilbe keine Wortgrenzen (genauer keine Grenzen zwischen autonomen Wortganzen) finden. In diesem Gesetze u¨ber die einsilbigen Wortganzen macht sich die trocha¨ische Tendenz des deseterac geltend, d.h. die Tendenz zum Alternieren ‘‘Steigung-Senkung’’: die Senkung darf hier nicht mit einem einsilbigen Wortganzen belastet sein. (Jakobson 1933, 54)

Note that only the absence of monosyllables from the fourth and tenth metrical positions is interpreted as a constant property of this verse and expressed as a constraint on the placement of word boundaries. The absence of monosyllables from the eighth position in the line is seen as simply reinforcing the general trochaic tendency of this meter. This interpretation is consistent with Jakobson’s (1952) analysis: while an explicit metrical device, notably, the ‘‘bridge,’’ is posited to exclude monosyllables from the fourth and tenth positions, no device is posited to exclude monosyllables from the eighth position. In other words, while the exclusion of monosyllables from the fourth and tenth positions is treated as a constant property of the meter, their absence from the eighth position is seen as a metrical tendency. That monosyllables are consistently excluded from the eighth position is

The Prosodic Word as a Unit in Poetic Meter

79

subsumed under the distribution of pitch accents in this and other weak metrical positions, which, according to Jakobson, tends toward trochaic patterning. Yet, as argued in section 4.2.1, the overall distribution of pitch accents in this verse is not su‰ciently systematic to reflect a trochaic tendency. Moreover, even if the accentedness of the eighth metrical position could indeed be subsumed under a trochaic tendency, this move would obscure a crucial fact about this meter: that monosyllables are absent from the eighth position, which can hardly be interpreted as a tendency. In sum, Jakobson’s analysis fails to capture the distribution of monosyllables in this meter, which is its defining property. My approach crucially di¤ers from Jakobson’s in treating the distribution of monosyllables as a defining property of the meter, which, in a robust fashion, reflects its periodic organization. This calls for a single metrical device that would capture all aspects of the distribution of monosyllables. This of course cannot be accomplished by Jakobson’s ‘‘bridge,’’ nor can it be accomplished by any other account in that same spirit. Positing two ‘‘bridge’’ points for the second half line—before the tenth and, additionally, before the eighth syllable—would incorrectly exclude lines that end with a trisyllabic word, and present a very common line type. The ‘‘bridge’’ is thus not an empirically adequate metrical device for capturing the distribution of monosyllables within a decasyllabic line. It does not cover the full range of relevant facts about this verse, nor does it provide an appropriate conceptualization of its metrical structure. As I have argued here, the distribution of monosyllables strongly suggests the organization of a line into metrical feet; in sum, it is a clear indicator of the trochaic pattern of this verse. 4.5

Monosyllabic Prosodic Words Combined with Clitics

A prosodic word that includes clitics has a nested structure, as in (29), and due to this may potentially have an ambiguous status: either the smaller scansion (i.e., poi ) or the larger (i.e., poj ) could be taken as prior: (29) [clitic [host]poi ]poj Generally, it is the larger scansion that is relevant for singling out prosodic words, both in the phonological system and in this meter. Thus, the four syllables in the first colon may be filled by a single prosodic word, which in turn may correspond to a single morphological word, as in (30a), or may have a nested structure: due to the presence of an enclitic, in (30b), or a proclitic, in (30c). (30) a. [Andje`lija]po k [mo`ja]po [vjeˆrna¯]po [ljuˆbo]po ‘Andjelija, my faithful wife’ `` nese]po b. [[naˆmjera]po ga]po k [pred [ve``cˇe]po ]po [na ‘A purpose brought him, toward the evening’

Vk, II, 97, 98 Vk, II, 97, 66

80

Draga Zec

`` tva]po [zlato`krila]po c. [u [ je``zeru]po ]po k [u ‘In the lake, there was a gold-winged duck’

Vk, II, 97, 68

In all three cases, the entire first colon is taken to be occupied by a single prosodic word—that is, by the prosodic word that corresponds to the maximal scansion. Maximal scansion is relevant also for monosyllables, which, combined with clitics, form configurations that behave like polysyllabic prosodic words. Monosyllabic prosodic words, as we have seen, are excluded from the weak positions of this meter, but when serving as proclitic hosts, they are admitted into weak metrical positions, as exemplified in (31)–(34); in fact, they are allowed in all weak positions other than the tenth—that is, the one that belongs to the final foot. A monosyllable combined with a proclitic may occur in colon-initial feet; this of course is expected, since the first and third foot are those in which metrical prohibitions against monosyllabic prosodic words are generally relaxed, due to (28). Examples for the first foot are given in (31) and for the third foot in (32); lexical monosyllables figure in the (a) and (b) examples, with a numeral in the latter, and monosyllabic function words, in the (c) examples.15 (31) First foot [s [s]po ]po s s k s s s s s s a. [u [sto]po ]po sjede k Crnojevic´ Ivo ‘At the table sat Crnojevic´ Ivo’ b. [za [tri]po ]po dana k i tri noc´i tavne ‘For three days, and three dark nights’ c. [od [njih]po ]po pusto k oduzesˇe blago ‘From them, they took away great treasures’ (32) Third foot s s s s k [s [s]po ]po s s s s a. svi mi da se k [u [so]po ]po prometnemo, ‘If we all turned to salt’ b. pa napravi k [do [tri]po ]po cˇasˇe zlatne, ‘And made three gold goblets’ c. il’ c´esˇ vatru k [sa [tim]po ]po ugasiti ‘Or will you extinguish the fire with it’

Vk, II, 88, 97 Vk, II, 52, 69 Vk, III, 5, 159

Vk, II, 49, 10iv Vk, II, 26, 29 Vk, IV, 24, 341

A monosyllable combined with a proclitic may also occupy the weak positions of the second and fourth foot, those in which metrical constraints rigorously take e¤ect. In (33), a monosyllable combined with a proclitic occurs in the weak position of the second foot: the noun lov ‘hunt’ figures in (33a), the numeral dva ‘two’ in (33b), and the function word njoj ‘it-feminine’, which, when combined with a clitic, is endowed with a prosodic word status, in (33c).

The Prosodic Word as a Unit in Poetic Meter

(33) Second foot s s [s [s]po ]po k s s s s s s a. vec´ on [u [lov]po ]po k u planinu podje ‘But he went hunting in the mountain’ b. dje se [do [dva]po ]po k druma rastavljaju ‘Where two roads fork’ c. kralj ga [u [njoj]po ]po k u svatove zove, ‘In it, the king invites him to his wedding’

81

Vk, II, 7, 53 Vk, III, 49, 178 Vk, II, 61, 14

And, in (34), a monosyllable preceded by a proclitic occupies the weak position of the fourth foot, with the noun red ‘line’ in (34a), the numeral tri ‘three’ in (34b), and the function word njoj ‘it-feminine’ in (34c). (34) s s s s k s s [s [s]po ]po s s a. Kako dodje, k s njima [u [red]po ]po stade. ‘As he came, he got in line with them’ b. te uprezˇe k konje [u [tri]po ]po reda; ‘And he harnessed horses in three rows’ c. te vidjeo k sˇto se [u [njoj]po ]po pisˇe ‘And he saw what was written in it’

Vk, II, 28, 233 Vk, II, 90, 25 Vk, III, 20, 48

This clearly shows that it is the larger scansion that is relevant—that is, the weak position is not construed as filled by a monosyllabic prosodic word. It should be noted that lines with the fourth and eighth positions filled with a monosyllable that hosts a proclitic, as in (33) and (34), are encountered much less frequently than those in which the second and the sixth positions are filled in this fashion, as in (31) and (32). Still, monosyllables hosting proclitics consistently occur in the fourth and eighth positions, albeit with a very low percentage, which is below 1 percent. Their presence is stable, with several cases emerging within any randomly picked set of one thousand lines. Thus, although lines like (33) and (34) are rare, I will consider them metrical—in other words, in full compliance with the constraints that regulate this verse. This is endorsed by the distribution of formulaic expressions across metrical positions. As already noted, the epic decasyllable is a highly formulaic verse in which certain fixed expressions recur with only minimal variation. Of interest at this point is the range of metrical positions compatible with a given formulaic expression. Thus, the phrase na um ‘to mind’, consisting of a proclitic followed by its host, appears in the fixed expression in (35), with either the order in (35a) or in (35b). (35) a. [na [um]po ]po pade b. pade [na [um]po ]po ‘Came to mind’

82

Draga Zec

The examples below show that the phrase na um is compatible not only with the first and third foot, as in (36), but also with the second and fourth foot, as in (37)— in fact, with all feet other than the final one. (36) a. [na [um]po ]po pade k begu Radul-begu ‘It occurred to beg Radul-beg’ b. Igumanu k [na [um]po ]po pade Simo, ‘The abbot recalled Simo’ (37) a. Pade [na [um]po ]po k Crnojevic´ Ivu, ‘It occurred to Crnojevic´ Ivo’ b. Ali njemu k odmah [na [um]po ]po pade, ‘But it occurred to him instantly’

Vk, II, 74, 40 Vk, II, 13, 190 Vk, II, 88, 116 Vk, II, 38, 79

Moreover, numerals often co-occur with the preposition do ‘up to’, which in this context serves merely as an empty filler. In (38), do trista ‘up to three hundred’ can only be interpreted as exactly ‘three hundred’, which suggests that do contributes no meaning of its own, and that its sole function is to ensure correct syllable count within the line. (38) a. te pokupi k do trista svatova ‘And he gathered three hundred wedding guests’ b. te nadjosˇe k do trista izvora ‘And they found three hundred springs of water’

Vk, III, 73, 42 Vk, II, 26, 26

Combinations of do with a monosyllabic numeral are found not only in the first and third foot, as in (39) and (40), but also in the second and fourth foot, as in (41) and (42): (39) First foot [do [dva]po ]po brata, dva Nedic´a mlada ‘Two brothers, two young Nedic´es’ (40) Third foot a. Vino piju k [do [dva]po ]po pobratima ‘Two bosom friends were drinking wine’ b. eto otud k [do [tri]po ]po gorske vile, ‘From there came three mountain sprites’ (41) Second foot a. kada [do [dva]po ]po k kneza pogibosˇe ‘When two princes perished’ b. dje se [do [dva]po ]po k druma rastavljaju ‘Where two roads fork’

Vk, IV, 26, 69

Vk, II, 16, 1 Vk, II, 11, 15

Vk, IV, 24, 441 Vk, III, 49, 178

The Prosodic Word as a Unit in Poetic Meter

(42) Fourth foot a. da ti kazˇem k do dve [do [tri]po ]po recˇi: ‘Let me say to you two or three words’ b. Na Balacˇku k jesu [do [tri]po ]po glave: ‘On Balacˇko, there were three heads’

83

Vk, II, 14, 95 Vk, II, 28, 608

Thus, formulaic expressions, which, in a sense, constitute the building blocks of this verse, when taking the form of a proclitic hosted by a monosyllable, are compatible with both the fourth and the eighth position within the line. However, this generalization does not extend to the final foot: a monosyllable combined with a proclitic may not occupy the tenth metrical position. While a few lines have been attested with a functional monosyllable combined with a proclitic occupying the tenth metrical position, not one case has been found with a lexical monosyllable occurring in this configuration.16 We may thus conclude that, in the final foot, it is the smaller scansion that counts, and that therefore, lines such as (43) are not within the range of metrical possibilities in this verse. (43) *s s s s k s s s s [s [s]po ]po Ali njemu k odmah pade [na [um]po ]po (Constructed on the basis of Vk, II, 38, 79) I will attribute the unmetricality of lines such as (43) to the special metrical status of the line-final foot, an important property of this verse that is addressed in section 4.7. 4.6

Monosyllabic Free Function Words

A monosyllabic free function word has no pitch accent of its own, and is thus incapable of endowing a metrical position with prominence, as noted in section 4.1. As such, it should in principle be compatible with any metrical position, and therefore also with the weak position in the foot. But while the distribution of monosyllabic function words is not constrained by any of the metrical conditions in (27), it is not free, either. In fact, it directly reflects, and is subsumed under, the set of constraints that govern the arrangement of hierarchically organized prosodic constituents (as in Nespor and Vogel 1986; Hayes 1989) within a metrical line. Both the line and the colon are subject to alignment with prosodic constituents. The right edge of the line is generally associated with syntactic boundaries, either phrasal or clausal, which strongly suggests that a metrical line is minimally aligned, at its left and right edges, with an intonational phrase boundary. The colon, on the

84

Draga Zec

other hand, is minimally aligned with the edges of a prosodic word, as stated in (44) and (45) (following McCarthy and Prince 1993a). (44) COLON LEFT The left edge of the colon coincides with the left edge of a prosodic word. (45) COLON RIGHT The right edge of the colon coincides with the right edge of a prosodic word. The e¤ect of (44) and (45) is best evidenced at the end of the first and the beginning of the second colon, and captures the already-stated structural property of this verse—that the decasyllabic line is divided into two cola by a fixed caesura, coinciding with a word boundary. To illustrate this, while colon boundaries may align with higher-level prosodic constituents—for example, the prosodic phrase, as in (46)— they minimally align with prosodic word boundaries, as in (47): (46) [[[na [Krusˇevcu]po ]po ]pf k [[gradu]po [bijelome]po ]pf ‘In Krusˇevac, the white city’

Vk, III, 10, 45

(47) [[[na [bijelu]po ]po k [gradu]po ]pf [[[Smederevu]po ]pf ‘In the white city of Smederevo’

Vk, III, 10, 53

Constraints (44) and (45) have an important impact on the distribution of free functional monosyllables. Because its morphological edges do not coincide with any prosodic edges, including the edges of a prosodic word, a monosyllabic function word that appears at the left edge of a colon violates (44), and one appearing at its right edge violates (45). In sum, a monosyllabic function word (here mwF ) prevents prosodic edge alignment both at the left edge of a colon, as in (48a), and at its right edge, as in (48b) (where k signifies colon edge): (48) a. * k mwF [po . . . ]po k b. * k [po . . . ]po mwF k A further aspect of the mapping between metrical and prosodic constituents is the general requirement that the poetic material within a metrical line be exhaustively parsed into prosodic words, as stated in (49): (49) CONTIGUITY Any two prosodic words within a colon, poi and poj , are either contiguous, or separated by another prosodic word, pok : * k . . . ]po o [po . . . k where o is a morphological word with no prosodic word status. The presence of a monosyllabic function word anywhere in the line directly conflicts with this requirement, as schematized in (50):

The Prosodic Word as a Unit in Poetic Meter

(50) * k . . . ]po mwF

po [ . . .

85

k

Monosyllabic function words are not, of course, entirely banned from a decasyllabic line, as strict adherence to Colon Left, Colon Right, and Contiguity would require. In fact, monosyllabic function words occur relatively freely at the left edges of both cola, due to another case of colon-initial relaxation of metrical conditions, Initial License 2, stated in (51); and its overriding e¤ect over Colon Left, expressed by the ranking in (52): (51) INITIAL LICENSE 2 In colon-initial feet (first and third), the weak position may be occupied by a free monosyllabic function word. (52) INITIAL LICENSE 2  COLON LEFT The occurrence of monosyllabic function words at the left edge of the first colon is illustrated in (53), and their occurrence at the left of the second colon, in (54). (53) First colon a. vec´ [isprosi]po k lijepu devojku ‘But he was accepted by a beautiful maiden’ b. on [sazida]po k kulu od kamena ‘He built a tower of stone’ (54) Second colon a. pa se svome k on [poklanja]po [kralju]po ‘And he bowed before his king’ b. ‘‘Sjedi, care k pak [pij]po [rujno]po [vino]po ’’ ‘Sit here, czar, and drink of red wine’

Vk, II, 9, 18 Vk, II, 14, 111

Vk, II, 80, 97 Vk, II, 28, 479

As the examples in (55) and (56) show, the entire colon-initial foot may be occupied solely by functional monosyllables, and thus left outside the prosodic word domain; this collocation is found much more frequently at the left edge of the first colon than of the second. (55) First colon a. vec´ on [u [lov]po ]po k u planinu podje ‘But he went hunting in the mountain’ b. al’ josˇ [sestra]po k u dusˇeku spava, ‘But the sister was still in bed asleep’ (56) Second colon a. Ne sluzˇi ga k sˇto on [blaga]po [nema]po ‘He is not serving him because he has no treasure’

Vk, II, 7, 53 Vk, II, 5, 64

Vk, II, 75, 4

86

Draga Zec

b. nek je haracˇ k ko sˇto [Murat]po [recˇe]po ‘May the taxes be as Murat had said’

Vk, IV, 24, 360

Finally, with Contiguity ranked over Initial License 2, as in (57), configuration (50) is prohibited in colon-initial feet. (57) CONTIGUITY  INITIAL LICENSE 2 To conclude, constraints and constraint rankings posited thus far have the e¤ect of excluding functional monosyllables from noninitial feet. In a narrowly defined discourse context, however, monosyllabic function words may exceptionally occur in the second and the fourth foot, as illustrated in (58)– (60). These and similar lines occur predominantly within dialogue exchanges, and are hardly ever found within the narrative portion of the poem. This strongly suggests that lines of this type are licensed by virtue of a discourse-based relaxation of both Contiguity and Colon Right in portions of the poem that call for a more discursive style: (58) Second foot: Fourth position a. [Koja [ je]po ]po tu k Roksanda djevojka ‘Which one here is maiden Roksanda’ b. [Kada [c´u]po ]po ja k i tebeka prec´i ‘When will I cross you, as well’ (59) Fourth foot: Seventh position a. Ako l’ bi se k [u [nju]po ]po kad [vratio]po ‘If I ever returned to it’ b. Al’ da vidisˇ k [cˇuda]po josˇ [vec´ega]po ‘If only you saw this yet greater miracle’ (60) Fourth foot: Eighth position a. Nit’ sam bio, k [niti [c´u]po ]po kad [biti]po ‘I never was, nor will ever be’ b. Da ogledam k [mozˇe [li]po ]po sˇto [biti]po ‘To see if anything could occur’

Vk, II, 28, 571 Vk, IV, 24, 628

Vk, IV, 28, 415 Vk, III, 38, 181

Vk, II, 49 (iii), 45 Vk, II, 66, 108

Thus, monosyllabic function words are exceptionally licensed within dialogue, possibly to render such exchanges more lively and more ‘‘natural,’’ as expressed by the following constraint: (61) DISCOURSE LICENSE Functional monosyllables are licensed in dialogue exchanges. In sum, the overall ranking of constraints that regulate the arrangement of prosodic words within metrical constituents is as in (62):

The Prosodic Word as a Unit in Poetic Meter

87

(62)

This ranking allows for the suboptimal configuration (50) to be instantiated in dialogue exchanges in colon-initial feet, and this indeed is the case, as illustrated in (63) for the first colon. (63) First colon a. [daj]po ti [nama]po k Sekulu nec´aka, ‘Give (you) to us your nephew Sekula’ b. [konj]po moj [nije]po k a djevojka tudja, ‘The horse is not mine, and the maiden belongs to another’

Vk, II, 84, 30 Vk, II, 86, 162

However, a functional monosyllable may not occur in the fifth foot, specifically, in line-final position. This distributional gap is not captured by the analysis presented thus far, as summarized in (62). One solution could be to ensure that Colon Right is enforced more rigorously in the second than in the first colon, thus capturing greater restrictiveness imposed at the end of the line. However, greater restrictiveness at the line closure has already been evidenced in section 4.5, which calls for a more general mechanism. In what follows, special restrictions associated with the line-final foot will be attributed to structural subdivisions within the line, beyond those posited thus far. 4.7

Higher-Order Metrical Units

The line-final foot is subject to greater restrictions than other positions in the line, as shown both by the distribution of monosyllabic clitic hosts (section 4.5), and of functional monosyllables (section 4.6). Rather than positing a constraint with the sole purpose of further codifying this metrical position, I will attribute these distributional restrictions to the higher-level organization of a metrical line. A notable structural property of the epic decasyllable is that a line is divided into two cola. The caesura imposes an asymmetric structure, with two feet in the first and three in the second colon, rendering the second colon weightier than the first, as represented in (64):

88

Draga Zec

(64)

Line Colon Left (s s) (s s)

k

Colon Right (s s) (s s) (s s)

As we have seen, the arrangement of prosodic words in a line crucially observes the caesura, invariably positioned after the fourth syllable; a line with a prosodic word occupying the fourth and fifth metrical positions would be unmetrical. This, however, is not the only structural division within a decasyllabic line. The heightened metrical restrictions in the line-final foot will be accounted for by positing an additional break, henceforth the subcaesura, which divides the second colon into two subcola. Unlike the caesura, the subcaesura is only partially fixed: the second subcolon corresponds to a single prosodic word and minimally includes the final metrical foot, as stated in (65): (65) The rightmost subcolon a. Subsumes the line-final foot (metrical requirement) b. Corresponds to exactly one prosodic word (prosodic requirement) Both cases of heightened metrical restrictions take e¤ect within the rightmost subcolon. Its special status follows from the structural organization of higher-level metrical constituents and their relative strength, as represented in (66): (66)

The strong/weak labeling in (66) (following Prince 1989) designates the right colon, and likewise, the right subcolon, as stronger than their left counterparts. The rightmost subcolon thus emerges as the strongest higher-level metrical subconstituent and, as such, is associated with greater metrical restrictiveness. The special status of the line-final foot follows from its obligatory inclusion into the rightmost subcolon.17 By contrast, the weak subconstituents, leftmost colon and leftmost subcolon, allow for relaxations of metrical conditions. The two Initial License constraints, stated in (28) and (51), could profitably be restated as taking e¤ect at the left edge of metrical subconstituents labeled weak. The absence of functional monosyllables follows directly from the characterization of the rightmost subcolon. Note that (65) imposes an obligatory prosodic word boundary at the end of the line. If (65) is rigorously enforced—that is, if it ranks

The Prosodic Word as a Unit in Poetic Meter

89

above Discourse License—then the final foot will not contain any functional monosyllables, which captures the heightened restriction in this metrical position, stated in section 4.6. The metrical restriction stated in section 4.5, according to which a monosyllabic proclitic host may not occupy the tenth position in a line, calls for further refinement. The rightmost subcolon is prosodically transparent: it forms a narrow window in which all prosodic structure, including prosodic word nesting, is visible to, and evaluated by, metrical constraints. As a result, any prosodic structure that it contains has to conform to the metrical requirements of this verse, including all layers of prosodic word structure. (67) STRONG SUBCOLON TRANSPARENCY Every prosodic word included in the rightmost subcolon has to comply with metrical constraints (in particular, (27b)). Thus, the clitic combination na um in the unmetrical line (30), repeated here as (68), violates Strong Subcolon Transparency because one of the prosodic words in the rightmost subcolon exhaustively occupies a weak position of the meter. The fact that, in addition to a monosyllabic prosodic word, there is also a disyllabic scansion available, does not salvage this configuration. (68)

s s s s k s s s s [s [s]po ]po *Ali njemu k odmah pade [na [um]po ]po

The special status of the second subcolon can account for another metrical peculiarity of this verse. As schematized in (69), the second subcolon—which, as already stated, corresponds to the final prosodic word in a line and subsumes the final foot— may vary considerably in size. The extreme case is (69a), in which the subcaesura coincides with the caesura, and (69b), in which the subcaesura is one syllable apart from the caesura. (69) a. b. c. d. e.

k k k k k

Colon Right Subcolon Left Subcolon Right j [s s s s s s]po s j [s s s s s]po ss j [s s s s]po sss j [s s s]po ssss j [s s]po

By far the most common are the types of lines in (69c–e), which jointly account for over 95 percent. In only one of these cases is there a further requirement imposed on the second subcolon: when it corresponds to a single foot, as in (69e). There is a strong tendency for disyllables in line-final position to possess a heavy initial syllable

90

Draga Zec

(Maretic´ 1903, 1907). This is the case in about 75 percent of the total number of disyllables in the line-final foot, compared to the situation in the third foot, in which no more than about 44 percent of disyllables have a long initial syllable. This will be interpreted as a tendency toward prominence required in the strong position when the second subcolon equals a foot, as stated in (70):18 (70) Final-foot prominence (tendency) If the rightmost subcolon corresponds to a single foot, then a. The strong position of the foot has to be prominent (S ) P). b. The prominence type is syllable quantity (P ¼ vowel length). While this is not an absolute requirement but rather a metrical tendency, this tendency is su‰ciently strong—that is, su‰ciently above chance—to deserve a place in the formal account of this verse. Moreover, the property of this verse to avoid vowel length in the seventh and eighth metrical positions, observed by Jakobson, could well be construed as a counterpart of the marked case in (70). In the unmarked case, when the rightmost subcolon’s size exceeds the size of a foot, as in (69c–d), the corresponding word unit preferably begins with a short voweled syllable.19 Cases of greater restrictiveness in line-final position, which have been captured here as metrical conditions on the rightmost subcolon, while apparently disparate in nature, do possess a unifying property: they impose greater regularity, both metrical and prosodic, within the metrically strongest constituent in the line. Constraint (70) could be motivated in this same spirit. The preferred size of the rightmost subcolon is to be greater than a metrical foot, and constraint (70) encodes what amounts to compensatory prominence when this is not the case. Thus, the unmarked size for the second subcolon is to be greater than a metrical foot, which in e¤ect maximizes the window within which a higher degree of metrical regularity is imposed. But the preference for the rightmost subcolon to be greater than a foot could be interpreted in yet another fashion. As already noted, the subdivision into cola is asymmetric, with the second colon being weightier than the first. In light of variation in the size of the rightmost subcolon in (69), the following generalization obtains: The second subcolon is weightier than the first in (69a–c), and equally weighty as the first in (69d). Only in (69e), which, as I have argued, is a marked case, is the first subcolon weightier than the second. Thus, while at the colon level, the second constituent is categorically weightier than the first, at the subcolon level, due to the mobility of the subcaesura, there is a strong preference for the second subcolon to be at least as weighty, or weightier than the first; and constraint (69) encodes what amounts to compensatory prominence when this is not the case. In other words, lines such as (69e) are marked by special prominence because they counter what appears to be the overall asymmetric organization of this verse, calling for any rightmost metrical subconstituent to be weightier than its left counterpart.

The Prosodic Word as a Unit in Poetic Meter

4.8

91

Concluding Remarks

The principal claim of this study is that the epic decasyllable is a periodic meter, corresponding to a trochaic pentameter, with the prosodic word—or rather, pitch accent contained in its head domain—serving as its crucial prominence source. Because of this, the monosyllabic prosodic word emerges as this meter’s sole prominence source, and is, as such, excluded from weak metrical positions. A further important property is the exclusion of monosyllabic free function words, those that possess no prosodic status, from all positions within the line other than colon initial. This is due to this meter’s strict adherence to the principles of prosodic parsing, much stricter in fact that in the ‘‘ordinary’’ language. The general consequence of these two principles is an overall preference in this meter for words that are minimally disyllabic. Since both lexical and free functional polysyllables invariably correspond to prosodic words, their distribution within a metrical line is fairly free, restricted only indirectly, by constraints on monosyllables. Moreover, the proposed higher-level organization of this verse into cola and, in particular, into subcola, and the metrical constraints associated with these structural constituents, further reinforce this meter’s preference for polysyllabic lexical units. The epic decasyllable thus possesses a set of metrical properties that di¤er in crucial ways from the well-known situation found, for example, in the English or Russian iambic meters, in which the distribution of monosyllables is relatively free, while that of polysyllables is heavily constrained. Notes I am grateful to Kristin Hanson and an anonymous referee for valuable comments and sugges´ upic´, former director of the Institute tions on an earlier version of this chapter, and to Drago C for the Serbian Language, for his immense help with the interpretation of the Old Sˇtokavian verse. This study would not have been possible without access to, and extensive use of, Djordje Kostic´’s Serbian Language Corpus, an invaluable resource that includes Vuk’s entire collection of epic poetry. 1. This study is based in fact on the second, third, and fourth volumes (Karadzˇic´ 1958). The first volume includes only the lyric poetry, and only some of the poems are decasyllabic; the fifth volume is not nearly as polished as the others, and is thus a less reliable source of this verse (Maretic´ 1907). Other sources include Njegosˇ’s 1845 collection (Njegosˇ 1972) and the poems collected by Milman Parry in the thirties and later published by Lord (Parry and Lord 1953). 2. In the examples, Karadzˇic´’s collection is denoted by Vk, Njegosˇ’s by Nj, and Parry and Lord’s by P&L. Specified for each example is the volume number (if relevant), the number of the poem within the volume, and that of the line within the poem. 3. The position of pitch accents in the two dialect groups diverges in a systematic fashion. Diachronically, the Old Sˇtokavian is a historical predecessor of the Neo-Sˇtokavian, which, in the fifteenth century, underwent the process of leftward accent shift whereby all noninitial falling

92

Draga Zec

accents became rising accents on the immediately preceding syllable (Belic´ 1956). While the place of the Neo-Sˇtokavian falling accents is much the same as in the Old Sˇtokavian, the Neo-Sˇtokavian rising accents correspond in the Old Sˇtokavian to the falling accents one syllable to the right. The retracted accent, and as a result, the bisyllabic status of the rising accents, are the central prosodic properties that set apart the Neo-Sˇtokavian group of dialects from the Old Sˇtokavian ones, those that had not been subject to this change (see Ivic´ 1958; Inkelas and Zec 1988; Zec 1993). 4. The type of pitch accent does not play a role in the composition of this verse, either. Again, any of the four pitch accents may appear on any of the accent-bearing positions within a line, although there are certain dispreferences. Long accents are rare in the seventh, and even more so, in the eighth position in the line. For example, in the poem from which the excerpt in (11) was taken, and that has 102 lines, we do not find a long rising accent in the seventh, nor a long falling accent in the eighth position in the line; but there are four long falling accents in the seventh, and two long rising accents in the eighth position. Jakobson (1952, 418–419) treats this as a metrical constant, but in fact it is no more than a tendency. 5. Jakobson (1933, 1952) distinguishes between metrical constants and metrical tendencies, and considers the prosodic demarcation of feet in the epic decasyllable to be of the latter type. He attributed the periodic nature of the epic decasyllable not only to the distribution of pitch accent, but also to the positioning of word boundaries. In the latter case, he identified a tendency toward the placement of word boundaries before the odd, rather than before the even verse positions. This however sheds no light on the headedness of feet—that is, on whether this is a trochaic meter. 6. According to Jakobson (1933, 53), Vucˇic´’s performance had been recorded by G. Gesemman at the Sound Division of the Prussian State Library. 7. According to Ruzˇic´ (1975, 147), the poem Marko Kraljevic´ i Arapin (II, 65), in Vuk’s second volume, which has 435 lines, exhibits the following distribution of pitch accents: 1

2

3

4

5

6

7

8

9

10

50.6

36.3

53.3

0.2

80.2

20.7

33.1

41.1

38.4

—

Note that here, the eighth position outweighs both the seventh and the ninth. High accentedness of the eighth position is characteristic of one of Vuk’s best singers, Tesˇan Podrugovic´, who preferred trisyllabic to disyllabic words in line-final position (Ruzˇic´ 1975, 147). For possible reasons, see section 4.7. 8. As detailed in section 4.1, this class includes all lexical monosyllables, as well as those functional monosyllables that possess prosodic salience by virtue of the association with focus. 9. It has been tacitly assumed that the epic decasyllable was composed in the Neo-Sˇtokavian dialect. This is mostly due to Vuk’s own e¤ort to bestow on the Neo-Sˇtokavian dialect, or rather, on one of its regional variants, the status of linguistic norm. However, while it is true that some of Vuk’s best singers are from a Neo-Sˇtokavian background, epic decasyllable has been composed by singers from more than one dialectal area, including the Old Sˇtokavian, as is clear from Vuk’s own commentary. Parry and Lord’s singers come from dialectal regions with marked Old Sˇtokavian traits, as documented by the accentuation they provide, while the Montenegrin origin of Njegosˇ’s collection associates it clearly with a specific Old Sˇtokavian idiom.

The Prosodic Word as a Unit in Poetic Meter

93

10. This is based on 506 accented lines in Vuk’s fourth volume (IV, 2, 12, 13, 17), 471 accented lines in Njegosˇ’s Ogledalo srpsko (poems 2, 50, 54), and more than 1,000 accented lines in Parry and Lord. Accentuation provided in Parry and Lord’s (1953) collection clearly reveals that the dialect of the singers is of the Old Sˇtokavian type (with some Neo-Sˇtokavian ´ upic´ for his generous help with the selection and accentuation traits). I am grateful to Drago C of poems from Vuk’s and Njegosˇ’s collections. 11. Translations of these examples are based on Parry and Lord 1954. 12. The latter property may well follow from the scarcity of grammatical roles that lexical monosyllables could perform: monosyllabic nominals and adjectives may be either in the nominative singular or, if inanimate, also in the accusative singular form. Monosyllabic verbal forms are very few, and restricted to a handful of verbs. 13. A footnote is omitted from the Jakobson quote. 14. ‘‘Monosyllabic word units are found in odd syllables of the verse. This reflects the metrical constant that before the fourth and tenth syllables in verse no word boundary (or rather, no boundary between autonomous word units) may occur. This rule regarding monosyllabic word units gives validity to the trochaic tendency of the deseterac, the tendency toward risefall alternations: the fall may not coincide with a monosyllabic word unit.’’ 15. The preposition/pronoun combinations za me ‘for me’, za te ‘for you’, za se ‘for oneself ’ are treated here as lexicalized units, rather than as sequences of a proclitic followed by its host, for the following reasons. First, the vowel of the preposition is long in this collocation but short in all other cases. Second, these expressions have ‘‘full-pronoun’’ counterparts za mene, za tebe, za sebe; the pronoun is focusable in the latter, but not in the former set. Likewise, expressions such as sa mnom ‘with me’, za mnom ‘after me’, which contain the instrumental first-person pronoun, will be treated as lexicalized, since here as well the pronoun cannot be focused. For examples, see note 16. 16. The following lines contain functional monosyllables combined with a proclitic in the tenth position: (i) da oko sˇta, k vec´e [ni [oko [sˇta]po ]po ]po ‘What about, if not about a trifle’ (ii) a ne mogu k ni da cˇuju [za [nju]po ]po ‘But they cannot even hear about her’ (iii) il ne cˇujesˇ, il’ ne hajesˇ [za [nas]po ]po ‘You either don’t hear, or don’t care about us’ (iv) hoc´e li se k nasmijati [na [nju]po ]po ‘Will he smile at her?’

Vk, II, 74, 6; Vk, II, 97, 24 Vk, II, 96, 34 Vk, III, 16, 60 Vk, III, 51, 77

These are the only lines of this type that I have found; they may well be an exhaustive list of such cases. If lines like these are to be permitted into the canon of this meter, then their privileged status should follow from a special metrical license granted to functional monosyllables that goes beyond those proposed in section 4.6. Lexicalized expressions discussed in note 15, such as za me ‘for me’, za te ‘for you’, za se ‘for oneself ’, as well as sa mnom ‘with me’, za mnom ‘after me’, do not violate any constraints on this meter when they appear in line-final foot, as in (v) Jal’ ne mare, k jal’ ne cˇuju za te? ‘Is it that they don’t care, or don’t hear about you?’

Vk, II, 96, 126

94

(vi) Stani, Andjo, ti ne idi za mnom ‘Wait, Andja, you should not follow after me’

Draga Zec

Vk, III, 54, 205

17. There are a small number of metrically infelicitous lines with lexical monosyllables in the fourth and eighth positions. Below are listed the known examples in Vuk’s three volumes: (i) Ili si [lud]po k i nisˇta ne znadesˇ, ‘You are either mad, or know nothing’ (ii) kamo ti [macˇ ]po k i megdan junacˇki ‘Where is your sword, and the heroic duel’ (iii) najprva [kc´er]po k Mihaila bana, ‘The first daughter of Mihailo ban’ (iv) pak mu sve [tri]po k otsijecˇe glave, ‘And he cut o¤ all his three heads’ (v) Meni se [zna]po k i otac i majka ‘It is known who my father and mother are’ (vi) Da se, jadna, k za zelen [bor]po vatim, ‘If, in my misery, I touch a green fir tree’ (vii) c´erasˇe ga k za puno [dva]po sata ‘They chased him for two whole hours’

Vk, II, 65, 356 Vk, III, 60, 47 Vk, II, 12, 41 Vk, II, 28, 665

Vk, III, 55, 120 Vk, II, 50, 135 Vk, IV, 28, 381

Crucially, however, no cases have been found with lexical monosyllables in line-final position, which is consistent with maximal metrical restrictiveness associated with the line-final foot. 18. This condition appears to be subject to singers’ individual strategies, which would lead to a reduced number of disyllables with a short initial syllable in line-final position (Ruzˇic´ 1975, 147). Jakobson (1952, 418–419) treats the quantitative clause as a metrical constant, on the basis of Vucˇic´’s two poems (783 lines), in which ‘‘an accented short is avoided in the penult (ninth) syllable.’’ Thus in this case penultimate syllables are reportedly long, regardless of the size of the word unit they belong to. It could be that, in Vucˇic´’s style, length in the ninth position is a condition relevant for any final foot, rather than just for those singled out by a specific shape of the second subcolon. 19. See note 4.

5

The Word in Tiberian Hebrew

Bezalel Elan Dresher

5.1

Introduction

The notion of a ‘‘word’’ is subject to various ambiguities. The text of the Hebrew Bible (Elliger and Rudolph 1977) provides us with several conflicting notions of word. The Biblical text consists of two main layers, and the written word—in the sense of letters surrounded by blank space—di¤ers in each layer. The earlier layer contains a consonantal text, devoid of almost all indications of voweling and punctuation. I will show that the word in this consonantal layer corresponds to a potential prosodic word, that is, a unit that could be an independent word for purposes of phrasing, whether or not it actually functions as such in any particular context. Since such words are not necessarily prosodic words in every context, I will call the word in this layer an orthographic word. Ordinary written Hebrew makes similar word divisions. To more precisely indicate correct pronunciation and phrasing, various diacritic marks, or ‘‘points,’’ were later added to the consonantal text, producing a ‘‘pointed’’ text. These marks include vowel signs, some allophonic consonantal distinctions, and an elaborate system of ‘‘accents’’ that indicates position of stress, division into verses, and a highly articulated prosodic parse of each verse. Orthographic words of the consonantal text (potential prosodic words) can be joined together by hyphens to create a larger unit, the (actual) prosodic word. As we shall see, these constituents are prosodic words in the sense that they count as words for purposes of phrasing. I will show that the principles for forming prosodic words—rules of cliticization—are quite complex and interact in intricate ways with other aspects of prosodic structure, such as the phonological phrase and the intonational phrase. Turning to the evidence of the phonology, I will distinguish between the prosodic word and the phonological word, which is the notion of word referred to by the phonology proper (segmental processes, syllabification, stress), as opposed to the phrasing. Though the phonological word necessarily has some relation to the prosodic word, the two concepts are not identical. Thus, a study of the word in Biblical

96

Bezalel Elan Dresher

Hebrew bears on issues of the syntax-phonology mapping in contemporary linguistic theory, as well as on the notion of levels in Lexical Morphology and Phonology (LMP). 5.2

The Consonantal Text: The Orthographic Word

In the consonantal text, all content words, such as nouns, verbs, adjectives, and numerals, are separate words separated by a space from adjacent words. Most prepositions are also written as independent words. Prepositions that consist of only a single consonant (or consonant plus schwa, depending on whether the schwa is analyzed as inserted by rule or part of the underlying form), however, are written as bound prefixes, with no space separating them from what follows. It is clear that word status is not connected to semantics in this case, because all these prefixes have variants or synonyms consisting of more phonological material, and these are invariably written as independent words. Independent word b mo (poet.), im l mo (poet.), el k mo (poet.) e e e

(1) Prepositions Prefix a. b( )b. l( )c. k( )-

Gloss ‘in, at, with, by’ ‘to, for’ ‘like, as’

e e e

Morphemes of the form C( ) do not make up a full syllable. Therefore, the above observations suggest the generalization that morphemes that consist of less than a full syllable are not written as independent words. The prepositions in (1) have variants of the form Ci- when prefixed to words that would otherwise have an initial syllable with a schwa. e

(2) Variants of prefixed prepositions a. C b Dava´r ‘in a word, matter’ (absolute) b. CibiDvar (unprefixed: d var) ‘in the matter of ’ (construct) e e

e

The @ i alternation in these prepositions is phonological, with i being the default vowel in closed syllables.1 The replacement of schwa by the full vowel i does not change the bound prefixal status of the prepositions. In this category we can also place the conjunctive w( )- ‘and’, called waw after the Hebrew letter used to represent it. This morpheme is written u- before a syllable containing schwa or a labial consonant but remains a prefix. e

e

The Word in Tiberian Hebrew

97

(3) Conjunctive waw a. Elsewhere w Dawı´D ‘and David’ b. Before schwa ux na´an ‘and Canaan’ c. Before labial umirya´m ‘and Miriam’ e

e

The u- allomorph of the conjunctive forms an exception to an otherwise regular rule that Hebrew syllables, and hence words, begin with a consonant. Thus, u- does not form a proper syllable and so also falls under the generalization that morphemes that consist of less than a full syllable are not written as independent words. Phonological subminimality is a su‰cient condition for a morpheme to be written as an a‰x, but it is not a necessary condition. The preposition min ‘from, out of, than’ frequently occurs in the form mi-, where the final n has been historically assimilated to the following consonant. Synchronically, the allomorph mi- causes gemination of a following consonant and is always written as a prefix. (4) The preposition min a. min min hae´s ‘from the tree’ b. mimidd va´sˇ ‘than honey’ c. memee´s ‘from a tree’ e

Before gutturals, which do not normally geminate, the vowel is lengthened to e. However, this lengthening does not restore the prefix to the status of independent word.2 Another morpheme with the shape CV that is always written as a prefix is the definite article ha-. Like mi-, ha- causes gemination of the following consonant; when gemination is not possible, the vowel of the article is lengthened to a.3 (5) The definite article haa. hahamme´lex ‘the king’

98

Bezalel Elan Dresher

b. hahae´veD ‘the servant’ The interrogative ma ‘what’ occurs in a number of variants: ma, ma, me. These variants occur under various segmental and prosodic conditions. This morpheme is almost always written as a separate word, though there are a few cases in which it is joined to a following word. In Isa 3:15, mlkm is to be read malla xe´m lit. ‘what to you (m.pl.)’, (i.e., ‘how dare you’); in Ex 4:2, mzh is to be read ma-zze´ ‘what is this?’ It is interesting that in both these cases the morpheme ma- is attached to a function word (a pronoun in the first case and a demonstrative in the second). Function words are more prone to be fused with other morphemes into single words.4 Thus, orthographic words are potential prosodic words. To qualify, a form must meet the minimum criterion of having at least a full syllable CV, where V is not schwa. Full vowels created by phonological processes do not count. On top of that, a morpheme must exhibit a certain syntactic-semantic independence—hence, ma is a potential prosodic word, ha- is not. 5.3

The Pointed Text: The Prosodic Word

The Biblical text was gradually stabilized and fixed in the centuries leading up to the first century of the common era (Cross and Talmon 1975; Sa´enz-Badillos 1993). At a certain point, no further changes were permitted to be made to this text. Therefore, to this day Torah scrolls that are used for public readings consist only of a consonantal text, with no indications of verse divisions, stress, or other prosodic markers. The prohibition against adding markings did not apply to texts intended for private or nonliturgical purposes, and symbols for vowels, consonant diacritics, and an elaborate system of ‘‘accents’’ to mark phrasing began to be introduced in the 6th and 7th centuries c.e., presumably to preserve the pronunciation of the traditional reading of the text (Goshen-Gottstein 1963). This activity was carried on for a number of generations by scholars known as Masoretes. A number of distinct but related schools arose; the best known was associated with a group working around the city of Tiberias and so is known as the Tiberian system (Dotan 1971; Yeivin 1980). The Tiberian system of accents represents a highly elaborated prosodic representation that, among other aspects of pronunciation, organizes the text into hierarchical groupings of verses, phonological phrases, and prosodic words (Dresher 1994). The Masoretes were not free to tamper with the consonantal text itself. While they could add diacritic marks over, under, beside, or even inside letters, they could not change or transpose letters or add or remove spaces between words. To indicate that two or more orthographic words are to be considered as a single prosodic word, the Masoretes connected the words in question by a hyphen, called maqqef.

The Word in Tiberian Hebrew

99

Whether an orthographic word in the consonantal text is cliticized to a following word or heads its own prosodic word has phonological consequences. A prosodic word has a single main word stress on the final or penultimate syllable. One rule that applies only to syllables bearing main word stress is Tone Lengthening. (6) Tone Lengthening Lengthen a vowel bearing main stress in its prosodic word. Conditions: a. The rule does not apply to the low vowel /a/ when followed by two consonants; b. The rule does not apply to verbs. (Prince 1975) This rule is exemplified by the accusative particle ey. In the majority of cases, this particle is attached by maqqef to the following word, indicating that it is cliticized to it and does not have its own word stress. In these cases the particle is pointed with the vowel e, as in (7a). When it is an independent prosodic word (7b), it is pointed with the vowel e. (7) The accusative particle a. As clitic eyey-hao´r acc-the.light (Gen 1:4) b. As independent word e´y e´y hasˇsˇama´yim acc the.heavens (Gen 1:1) 5.4

Cliticization in the Tiberian Text

Whether an orthographic word is cliticized or not depends on a complex set of prosodic, phonological, and syntactic conditions, some of which are reviewed in the following sections. It turns out that cliticization is tightly tied in with the entire Tiberian prosodic system and cannot be understood without taking into account the principles of phrasing. 5.4.1

Rudiments of the Tiberian System of Accents

For purposes of the current discussion, it is necessary to know that the Tiberian diacritics known as ‘‘accents’’ fall into two groups. A conjunctive accent on a word indicates that the word is in the same phonological phrase as the following word; a disjunctive accent indicates that its word is phrase final. Disjunctive accents, in turn, are arranged into four hierarchical classes, conventionally designated D0, D1, D2, and D3, where D0 represents the strongest disjunction (coming only at the end of a verse and at the end of the half-verse) and D3 represents the weakest (Cohen 1969).

100

Bezalel Elan Dresher

A phrase ending in an accent of level Di is divided by an accent of level Diþ1, until the D3 level. A D3 phrase is divided by another D3. Therefore, unlike much contemporary work that assumes strict layering of phonological phrases, the Tiberian prosodic representation divides each verse into nested phonological phrases.5 The principles governing the division of a verse into phrases, and hence the distribution of the accents, are extremely complex, and though some of the leading principles and rules for particular circumstances are now known, much remains to be discovered (see Arono¤ 1985; Breuer 1982; Dresher 1994; Janis 1987; Price 1990; Wickes 1887; among others). Cliticization is integral to the entire system, because phrasing is sensitive to the number of words and to the prosodic weight of words, and cliticization a¤ects both: cliticization can change two short words into one long word, for example.6 Therefore, cliticization is woven into the phrasing algorithm; it cannot be regarded as a preliminary step that takes place prior to the division into nested phrases or, conversely, as a late fix up that follows the division of words into phrases. The principles governing cliticization are therefore particularly complex, because, being situated at the interface between word and phrase, they involve general principles of phrasing as well as particular idiosyncrasies of lexical items. The most detailed discussion of cliticization in the Biblical text that I know of is that of Breuer (1982, chapter 7). Breuer proposes a series of descriptive generalizations that set out conditions under which cliticization is facilitated or blocked. These generalizations take the form of conditions akin to the ‘‘preference laws’’ of Vennemann (1988) or the constraints of much current phonological theory, notably Optimality Theory (OT; Prince and Smolensky 1993). OT proposes a theory of how conflicts among constraints can be accommodated. Generally speaking, the principal categories of cliticization are the following: 1. Small words: Some small words have an inherent tendency to be cliticized.7 2. Simplification of phrasing: Cliticization simplifies the phrasing, either by reducing the number of conjunctive accents in a phonological phrase or by reducing the number of phonological phrases. 3. Clash avoidance: Clash avoidance averts a stress clash by relieving the cliticized word of its clashing main stress. 5.4.2

Small Words

Cliticization occurs most readily to small monosyllabic words that have a short vowel in a closed syllable. Breuer divides these words into two classes: those that are generally cliticized to any word, short or long, and those that are regularly cliticized only to short words. Breuer (1982, 167) gives the following list of words of the first class.

The Word in Tiberian Hebrew

101

(8) Small function words that can be cliticized to any word ey ‘accusative particle’, al ‘on’, el ‘to’, min ‘from’, aD ‘until’, im ‘with’, im ‘if ’, al ‘not’, bal ‘not’, pen ‘lest’, af ‘also’, ma ‘what’, kol ‘all’, ben ‘son’, bay ‘daughter’, ey ‘time’ Most of these words are straightforwardly function words—the accusative particle, prepositions, negative particles, various subordinating complementizers, and quantifiers. The nouns ben ‘son’ and bay ‘daughter’ might appear to be content words; however, they are also used in contexts where their lexical meanings are attenuated or lost and take on a more functional cast. The word ben, for example, can designate a quality (ben-h a´yil lit. ‘son of valour’ ¼ ‘valiant’) or mean ‘deserving of ’ (ben mo´y lit. ‘son of death’ ¼ ‘he shall surely die’) or be part of an expression indicating age (b ne-sˇana´ lit. ‘sons of a year’ ¼ ‘of the first year’, i.e., ‘less than a year old’), and so on. Similar considerations apply to bay (e.g., e´z bay-sˇ naya´h ‘a she-goat in its first year’, bay-b liyya´al lit. ‘daughter of baseness’, ‘a worthless woman’) and less obviously also to ey ‘time’, perhaps because of its association with time (l e´y ziqnayo´ ‘in his old age’). The interrogative ma ‘what’, though it fits semantically, appears to be out of place because it has an open syllable. However, Breuer points out that it functions as a closed syllable (maC ) because it causes gemination of a following consonant. However, it is not cliticized when followed by a guttural, which does not geminate. Hence, we have the following pair. e

e

e

e

(9) Cliticization of ma when a geminating consonant follows (ma-ppisˇı´)D2 (ma´ hattayı´)D1 what-my.trespass what my.sin ‘What is my trespass? What is my sin?’ (Gen 31:36) The first instance of the word ma in this verse is cliticized to a word with a geminating consonant. In the second phrase, the word following ma is long and has an initial consonant that is a nongeminating guttural; therefore, ma acts like a word ending in an open syllable and does not cliticize in this phrase. In addition to the small words in (8), Breuer identifies another set of small words that are more restricted in their tendency to cliticize. In general, these words tend to cliticize only to short words. Breuer divides these words into two lists: the words in the first list (10a) cliticize more readily than those in the second (10b). (10) Small (mostly) content words that can be cliticized to short words a. am ‘also’, ax ‘but’, raq ‘only’, yaD ‘hand’, kaf ‘palm’, am ‘people’, dam ‘blood’, d var ‘word (const.)’, har ‘mountain’, sar ‘o‰cer’, an ‘garden’, rav ‘great’, ha' ‘holiday’, rax ‘soft’, n um ‘speech’ e

e

102

Bezalel Elan Dresher

b. af ‘anger’, mas ‘tax’, al ‘heap’, qasˇ ‘straw’, pay ‘morsel’, ay ‘winepress’, bar ‘son (aram.)’, haD ‘one (aram.)’, sˇen ‘tooth (const.)’, hoq ‘statute (const.)’, mor ‘myrrh (const.)’, tom ‘integrity (const.)’, tam ‘complete (const.)’, sˇal ‘do¤ ’, raD ‘subdue’, hay ‘live’, at ‘you (f.sg.)’, ze ‘this’, bah ‘in that (aram.)’, b aD ‘for’, n qam ‘revenge (const.)’, sˇ 'ar ‘young of an animal (const.)’, p sˇar ‘interpretation (const.) (aram.)’, l ven ‘white (const.)’, m lox ‘reign’ (const.) e e

e

e e

e

These lists consist mainly of content words, though the first three words in (10a) are function words with syntactic/semantic values comparable to those in (8). Thus, it is not clear why these words are grouped with (10) rather than (8). Some of the content words in (10), like those in (8), also have wider uses that could arguably put them into the function word class. An example is yaD ‘hand’, which combined with various prepositions can mean ‘by the side of ’, ‘next to’, ‘at the disposal of ’, and so on. Note that af ‘anger’ in (10b) is homonymous with af ‘also’ in (8). Besides the content word function word distinction, these words are also distinguishable phonologically. The final consonant in the noun af derives from an underlying geminate that surfaces in su‰xed forms, such as possessive appo´ ‘his anger’. Many of the other words in (10) likewise have underlying final geminates; thus, these words are not just semantically ‘‘heavier’’ than those in (8) but phonologically heavier, also, though the phonological distinction is neutralized at the surface in unsuffixed forms. The nouns in (10a) di¤er from those in (10b) mainly in that the former are more common. The words in (10b) consist of Hebrew nouns together with a mixed bag of other parts of speech, including the odd verb, preposition, demonstrative, and pronoun, and even some words in Aramaic. Thus, tendency to cliticize depends on a variety of factors, including phonological weight, morphological/syntactic class, semantic function, and commonness.8 Many words in (10) are construct forms. The construct raises special problems for the definition of word, but we cannot pursue this topic here. Breuer (1982, 168) points to pairs such as those in (11) as showing how cliticization of words in group (10) is sensitive to the length of the following word; cliticization of am ‘also’ applies before a short word (11a) but not before a long word (11b).9 (11) Cliticization of am depends on the length of the following word a. As clitic (w 'am-oyı´)D2 and.also-acc.me (2 Sam 2:7) b. As independent word (w 'a´m anoxı´)D2 and.also I (2 Sam 2:6) e e

The Word in Tiberian Hebrew

103

Given what I have reported on here, one might expect that the words listed in (8) and (10) are freely cliticizable—everywhere, in the case of the words in (8), and before short words for (10). This, however, is not the case. As mentioned above, cliticization interacts with other aspects of phrasing. Other constraints on the phrasing algorithm can conflict with cliticization in certain situations, and in these configurations, cliticization is systematically blocked. For example, there is a very strong constraint that the half-verse, which ends with a D0 accent, should consist of at least two phrases. In some verses, the main division is such that one of the half-verses contains only two words, one of which is a small cliticizable word. In such a case, the small word almost always remains an independent word in its own phrase, marked with a disjunctive accent. (12) Half-verse contains only two words (w e´y)D1 (b yue´l)D0 and.acc Bethuel (Gen 22:22) e

e

Another constraint that applies to D0 phrases is that a long word does not easily coexist with another word. This phenomenon can be understood as due to a slowing down of the reading in prominent positions, so that a long word in such a position counts as if it were two words, hence already enough to fill a whole phrase (Dresher and van der Hulst 1998). Thus, a small word is generally not cliticized to a long word in a D0 phrase but again is placed in its own phrase with a disjunctive accent. This phenomenon is illustrated by the following verses. (13) Cliticization of a small word to a long word a. In a D1 phrase (w ey-hairasˇ´ı)D1 and.acc-the.Girgashites (Gen 15:21) b. In a D0 phrase (w e´y)D1 (hairasˇ´ı)D0 and.acc the.Girgashites (Gen 10:16) e e

5.4.3

Simplification of Phrasing

Cliticization can also occur to reduce the number of disjunctive accents in order to create a smoother phrasing. In particular, expected phrasings of the form (14a) and (14b) below may be simplified as shown. (14) Simplification a. (w)Dnþ1 (w w)Dn ! (w w-w)Dn b. (w w)Dnþ1 (w)Dn ! (w-w w)Dn The likelihood of a word being cliticized in these contexts increases with increased shortness of the word. Of course, the small words in (8) cliticize to any following

104

Bezalel Elan Dresher

word unless blocked by the phrasing principles discussed in the previous section, so they would be cliticized in (14) without any further stipulation. The small words in (10) cliticize as a matter of course only to short words; recall (11) above. The contrast in (11) arises where the conditions of (14) do not obtain; however, such words do cliticize to a long word when Simplification is possible. (15) Cliticization of am in Simplification context a. (w ibbane´ 'am-anoxı´)D1 and.I.shall.be.built.up also-I (Gen 30:3) b. *(w ibbane´)D2 (a´m anoxı´)D1 e e

Apart from the small words discussed above, cliticization applies most commonly, according to Breuer, to the subordinating complementizers ki ‘that, for, when’, a˘sˇer ‘that’, and the negative morpheme lo. Cliticization of this type is illustrated in the following example. (16) Example of Simplification via cliticization of lo(w 'e´r lo-yone´)D1 (w lo´ yilhase´nnu)D0 and.stranger not-vex and.not oppress.him ‘Thou shalt neither vex a stranger, nor oppress him’ (Ex 22:20) e

e

The first instance of the word lo in this verse is cliticized, thereby reducing the number of words from three to two and allowing the preceding word to be phrased in the same phrase, rather than forcing it into its own phrase. This is thus an example of (14a). The second instance of this word is not cliticized, because doing so would serve no simplifying purpose, nor is there a stress clash in this phrase. This example shows that, at least in the case of this lexeme, it is preferable to have two words in a phonological phrase than for it to be cliticized, leaving just one word in the phrase, in the absence of other factors favoring cliticization. Another form of Simplification is cliticization to reduce the number of words in a phrase by replacing a conjunctive accent. Such cases can be represented schematically as in (17). (17) Reduction a. (w w w)Dn ! (w w-w)Dn b. (w w w)Dn ! (w-w w)Dn The situations in (17) are the minimal ones in which Reduction can occur. That is, a word that is not inherently cliticizable by the criteria discussed above will not cliticize to reduce a phrase from two words to one word. Cliticization of this type can occur, however, in larger phrases containing more than two words. It follows that Reduction is relevant only in phrases ending in disjunctive accents that support more than one preceding conjunctive accent, that is, that allow more than two words in a phrase. As a rule, the less prominent the phrase, the more words

The Word in Tiberian Hebrew

105

can be fit into it. Therefore, Reduction situations arise often in D3 and D2 phrases but only rarely in D1 or D0 phrases. D0 phrases, for example, normally permit a maximum of two words. This generalization is never violated before a verse-final D0 accent. However, before the D0 that ends the first half-verse (an accent called atnah ), there is a particular situation in which more than two words can occur. Breuer (1982, 156) observes that this configuration arises when the word ki is followed by a word with initial stress that is in turn followed by the D0 word (18b). When ki is followed by a word with noninitial stress in a comparable sequence, it is cliticized, resulting in an ordinary two-word D0 phrase (18a). (18) Cliticization of ki in a D0 phrase depends on the following word a. As clitic (ki-yele´x imma´nu)D0 rather-you.will.go with.us (Nu 10:32) b. As independent word (kı´ va´ vila´m)D0 that came Balaam (Nu 22:36) The pattern exemplified in (18) is the opposite of what we might have expected; since words are cliticized as a way of averting a stress clash (see the following section), the failure of ki to cliticize in just such a position is unexpected. Thus, (18b) is anomalous in two ways: the existence of a three-word phrase ending in D0 and the failure of cliticization. These two anomalies can be connected by supposing that ki in (18b) is treated by the accent system as in fact being cliticized at some abstract level; for reasons that remain unclear, the cliticization is suspended in this particular configuration. 5.4.4

Clash Avoidance

Cliticization can occur to prevent a stress clash between words in the same phonological phrase. In Tiberian Hebrew, a stress clash occurs between two words in the same phonological phrase when the first word has final stress and the second word has initial stress. If the first word ends in a superheavy syllable (a phonologically long vowel in a closed syllable), no clash is considered to occur. The cliticized word in (19a) has final stress when independent (19b). The e¤ect of cliticization is to deprive the word of its main word stress, thus averting a stress clash. Notice that a secondary stress appears on the initial syllable of the cliticized word in (19a). This secondary stress arises through the ordinary rule of secondary stress assignment, counting back two full syllables from the main stress of the entire prosodic word (Dresher 1981b).

106

Bezalel Elan Dresher

(19) Cliticization to avert a stress clash a. (va`yhi-e´rev . . .)D1 and.was-evening ‘And there was evening’ (Gen 1:8) b. (vayhı´ vae´rev)D1 and.was in.the.evening ‘When evening came’ (Gen 29:23) Another means of averting a stress clash is by stress retraction (McCarthy 1979; Rappaport 1984; Revell 1987), examples of which are shown in (20); the first word in each phrase is normally stressed on the final syllable. (20) Stress retraction to avert a stress clash a. (qa´ra la´yla)D0 he.called night (Gen 1:5) b. (to´xal le´hem)D1 you.will.eat bread (Gen 3:19) The vowel onto which stress is retracted must normally be long. For words that do not meet the conditions for stress retraction, cliticization is the only option for avoiding a stress clash, as in (19a). Where retraction can occur, it appears to be the preferred option; where retraction is not permitted, cliticization can occur. Sometimes the stress clash is left unresolved, for reasons explained in detail by Revell (1987). An example illustrating these two options side by side in the same phonological phrase is the following. (21) Stress retraction and cliticization to avert stress clashes (mo´zne se´Deq avne-se´Deq)D2 balances honest weights-honest ‘an honest balance, honest weights’ (Lev 19:36) In the word mo´zne, stress can retract onto the phonologically long vowel o; but the initial vowel of avne is short, so cliticization is the only available option (short of leaving the clash unresolved). 5.4.5

Summary

The preceding remarks on cliticization present only a partial picture of this complex phenomenon. They should su‰ce to show, however, that the principles governing the distribution of prosodic words in the text are bound up with constraints on phrasing that operate at higher levels of the prosodic hierarchy. Earlier proposals for mapping the prosodic structures indicated by the accents from the syntax assumed a derivational approach whereby prosodic structure is built up in a series of steps (Dresher 1994; Janis 1987). The above survey suggests that evaluation of candidate forms by

The Word in Tiberian Hebrew

107

means of ranked constraints, as proposed by OT, o¤ers a promising alternative. I will not, however, attempt such an analysis here (though the reader is invited to begin to construct one from the materials presented above). 5.5

The Word Level in Phonology

Up to now we have considered the notion of the word as represented orthographically in the Tiberian text and have found two types of words associated with the consonantal and pointed text, respectively. Both notions relate to the prosodic word of linguistic theory. The orthographic word of the consonantal text corresponds to potential prosodic words, that is, words that can stand as independent prosodic words in some context. The pointed text indicates which of these potential prosodic words are actually realized as such and which are cliticized. There is another notion of ‘‘word’’ that is relevant to phonological theory in the sense of a level at which certain phonological and morphological processes apply. In the theory of LMP (Kiparsky 1982d, 1985; Mohanan 1982, 1986; Pesetsky 1979), phonology and morphology apply in stages to a series of levels, such as the stem, the word, and postlexical levels. Though a prosodic word is necessarily a domain for word-level processes in LMP, the word level of LMP is not exhaustively characterized by the prosodic words indicated in the Tiberian pointed text. That is, word-level processes apply also to certain subconstituents of prosodic words. In Dresher 1983, I argued that Biblical Hebrew displays some level ordering, but the levels are not exactly what we might expect from the results of other studies. In brief, I argued that there is very little evidence for stem-level phonology, apart from some minor rules that apply to particular morphemes. I connected this fact to the nonconcatenative nature of much Semitic morphology, which does not create a suitable environment for the operation of typical phonological processes. Thus, there is no evidence that su‰xes, for example, need to be distinguished as being stem level or word level. Unlike many dialects of Arabic (Brame 1974; Broselow 1976; Kenstowicz 1981; Kiparsky 2007), for example, object su‰xes, though attached outside of subject-agreement su‰xes as in other Semitic languages, do not appear to be attached at a di¤erent level than other su‰xes. For purposes of stress, syllabification, and segmental phonology, a word like yisˇmorxa´ ‘he will guard you’, from underlying /yaþsˇmorþeþka/, can be subjected to a‰xation and word-level phonology in one pass, without any internal cycles. It follows from LMP that we should not expect the morphology to be sensitive to derived phonological properties, and this prediction is borne out; unlike English, for example, Biblical Hebrew has no wordformation rules that make reference to stress. I also argued that the word level itself must be conceived of di¤erently than in most studies of LMP. We can think of a word as existing paradigmatically in the

108

Bezalel Elan Dresher

lexicon or else as being syntagmatically placed within a phrase. In most studies of LMP, the word is thought of, perhaps correctly, as being in the lexicon. Hence, languages like English have category-changing word-level a‰xes that are best thought of as being assigned in the lexicon prior to insertion in the syntax (e.g., singaer, sadaness, nationalaize). Consistent with this is the fact that word-level phonology in English is not sensitive to the position a word has in its phrase. In Biblical Hebrew, however, these phenomena point in a di¤erent direction: here, word-level phonology applies not to words in the lexicon but to words already placed in a phrase. Evidence for this is that word-level phonology in Hebrew is sensitive to the position of a word in a phrase. This evidence comes from the so-called pausal forms, a ubiquitous feature of Tiberian Hebrew prosody. In Tiberian Hebrew, many words have one form when they are phrase internal—the contextual form—and another form—the pausal form—when they are final in a major phrase (which I take to be the intonational phrase of the contemporary prosodic hierarchy; see DeCaen 2005; Dresher 1994; Goerwitz 1993; Revell 1980, 1981). In most cases, both the contextual and pausal forms can be derived from a common underlying source by the same regular rules of the phonology. The source of the di¤erence can be located in the way rules of stress and reduction apply in pause and in context. Starting from /yaþsˇmorþeþka/, for example, the penultimate vowel is reduced when the word is in context, causing stress to appear on the final vowel and preserving the stem vowel, the result being yisˇmorxa´, as cited above. When the same word is in pause, however, the penultimate vowel is retained and stressed, resulting in the reduction of the stem vowel, yielding yisˇm re´xa (see Malone 1993; Prince 1975 for details). If word-level phonology waits until the whole word has been put together and inserted into its phrase, it follows from LMP that there should be no word-level category-changing a‰xes, and this appears to be correct. Although there exist no word-level a‰xes that apply in the lexicon, there does exist a class of word-level prefixes that create a word-level cycle in the phrase. e

(22) Word-level prefixes a. Prepositions: b( )- ‘in’, l( )- ‘to’, k( )- ‘like’, mi- ‘from’ b. Conjunctive: w( )- ‘and’ c. Definite article: ha- ‘the’ e

e

e e

The prefixes in (22) are just those discussed in section 5.2 as being noteworthy in that they are not written as independent words in the consonantal text, though their syntactic-semantic status might qualify them as being grammatical words. In the case of the prepositions (22a), we observed that they each have variants or synonyms that are independent orthographic words. From the point of view of the syntax, too, these prepositions are best viewed as being introduced into the syntax as independent morphemes with their own syntactic positions. They are then obligatorily cliticized

The Word in Tiberian Hebrew

109

but not before certain word-level phonological processes have applied. One of these is a rule changing /a/ to i in a word-initial closed syllable (Prince 1975).10 (23) A-to-I a ! i/aC

CC

This rule applies to (24a), where there is no prefix. The underlying /a/ surfaces in prefixed forms, such as (24b). However, the rule applies despite the presence of one of the prefixes in (22), as shown in (24c). (24) Examples of A-to-I a. No prefix: A-to-I applies /addel/ ! idde´l ‘he brought up’ b. Lexical prefix: A-to-I does not apply /yaþaddel/ ! y adde´l ‘he will bring up’ c. Word-level prefix: A-to-I applies /w( )aaddel/ ! w idde´l ‘and he brought up’ e

e

e

Another phenomenon that attests to the distinctive status of the word-level prefixes is spirantization. Spirantization normally applies to a (nonemphatic and nongeminate) stop that immediately follows a vowel (see Idsardi 1998 for a detailed discussion). Hence, in (25a) the /k/ and /b/ of the root /ktb/ are spirantized, but the /t/ is not. Following a word-level prefix, however, the /t/ is also spirantized (25b).11 (25) Contrast in spirantization a. Lexical prefix: Medial C of root not spirantized /laþktob/ ! lixto´v ‘to write’ b. Word-level prefix: Medial C of root spirantized /b( )aktob/ ! bixyo´v ‘when writing’ e

This spirantization attests to the presence of a vowel between the k and the t in (25b). Such a vowel would arise on the inner cycle in /ktob/ to break up the illicit initial consonant cluster, spirantizing the /t/ (26b). On the outer cycle, the prefix vowel (whether underlying or inserted to break up the CC cluster) spirantizes the /k/. The addition of the prefix also puts the derived vowel of the inner cycle in the context VC CV; this is a context in which short vowels are deleted, hence the output bixyo´v. In (26a), the prefix plus stem form a single word domain, and so there is no reason to insert a vowel between the first two root consonants (Borer 1979; Idsardi 1998; Rappaport 1984). (26) Word domains Inner cycle Input

a. Lexical prefix la-ktob (laktob)w

b. Word-level prefix ktob (ktob)w

110

Bezalel Elan Dresher

Output Outer cycle Input Output

(lixto´v)w — —

(kVyo´v)w b-(kVyo´v)w (b(kVyo´v)w )w bixyo´v

In sum, Biblical Hebrew supports the general picture of phonology-morphology interaction posited by LMP. It also shows that word-level phonology may have access to the position of a word in a phrase and that there exist postlexical levels that have many of the properties of lexical levels (see Dresher 1983 for further discussion). Of course, since phrasing itself depends on some derived phonological properties (notably, the position of stress), the nature of the iteraction between level-ordered phonology and the phrasing algorithm is not entirely clear. Putting together the results of this section and the previous one, it appears that an adequate analysis of Biblical Hebrew phonology and prosody may require a derivational component as well as parallel constraint evaluation, perhaps along the lines sketched by Kiparsky (2002).12 5.6

Conclusion

For hundreds of years, Biblical Hebrew has been at the center of important developments in linguistics. In this brief survey of some aspects of the word in Tiberian Hebrew, I hope to have shown that this position is entirely merited and that the Masoretic text continues to raise interesting and complex problems that are relevant to current issues in linguistic theory. Notes It is an honor and a pleasure to dedicate this article to Paul Kiparsky, who has contributed so much to our understanding of the word and its place in phonology and morphology. For various kinds of valuable help and illuminating discussions of Biblical Hebrew, I would like to thank Jean Balcaen, Vincent DeCaen, and Bill Idsardi. This research was supported in part by Social Sciences and Humanities Research Council of Canada research grant 410-96-0842. 1. Jou¨on (1947, section 8), against the opinion of Kautzsch (Bergstra¨sser 1962; Gesenius 1910), considers that a stem-initial schwa was actually pronounced, though in a weakened form that does not amount to a normal reduced syllable: biD var. His arguments for this assumption are first, that the schwa corresponds to a vowel that was historically present, and second, that the rule of spirantization applies to the following consonant, indicating the presence of a vowel (*biD bar). However, neither of these arguments is compelling: the historical existence of a vowel does not necessarily bear on its synchronic status; and though spirantization does point to the synchronic presence of a vowel, it does not necessarily indicate that this vowel is present at the surface—see Idsardi (1998) and section 5.5. e

2. A remark is needed concerning the transcription of vowels used here and the issue of vowel quantity. The Tiberian transcription distinguishes seven vowel signs, and the current consensus

The Word in Tiberian Hebrew

111

is that these vowels are distinguished by quality, not quantity, with values approximating to [i, e, , a, , o, u] (Bergstra¨sser 1962; Jou¨on 1947; Khan 1987). Despite the apparent seven-vowel system of the Tiberian transcription, there is a long-standing tradition (Chomsky 1952) of considering the underlying vowel system of Biblical Hebrew to comprise 10 vowels, symmetrically divided into five long and five short: /i, i, e, e, a, a, o, o, u, u/. Even while indicating vowel quantity in transcriptions, writers such as Jou¨on 1947 and Lamdin 1971 are noncommittal as to the phonetic reality of this scheme. However, there is no doubt that the quantitative interpretation makes much better sense of the phonological alternations of Biblical Hebrew than does the purely quantitative interpretation. Thus, I will refer to lengthening and long and short vowels, understanding these terms to refer to a genuine phonological reality in the grammar of Tiberian Hebrew, though not necessarily at the surface phonetic level. c

3. I omit other variants of the definite article that arise in various environments having to do with the position of stress and other peculiarities of the gutturals. 4. For example, prepositions cannot occur with independent pronoun forms, but only with su‰xal forms of the pronoun: l xa´ or ele´xa ‘to you’, never *e´l atta´, where atta´ is the independent form of the second person masculine singular pronoun. e

5. See Dresher 1994 for further discussion of the rationale behind this nesting and its connection with contemporary approaches to prosodic structure. 6. A long word has at least two full syllables before the main stress; a short word does not meet this condition. See Dresher 1981a for discussion of the theoretical basis underlying these definitions. 7. A small word is a word with only one syllable (not counting schwa). 8. Breuer (1982, 171) writes that he includes n um ‘speech’ in this list even though it has a long vowel in a closed syllable because it cliticizes frequently, particularly in certain fixed phrases. ze ‘this’ appears to be out of place here because it ends in an open syllable; however, like ma ‘what’ it causes gemination of a following consonant when it cliticizes, thereby closing its syllable. e

9. The initial /g/ is spirantized after a vowel in both examples in (11). 10. See Prince (1975, 157) for a refinement of this rule. 11. Note that the infinitive prefix laþ is distinct from the preposition l( )a. e

12. See also Dresher (in press) for further discussion of this issue.

6

Faithfulness and Componentiality in Metrics

Bruce Hayes

6.1

Introduction

The field of generative metrics attempts to characterize the tacit knowledge of fluent participants in a metrical tradition. An adequate metrical analysis will characterize the set of phonological structures constituting well-formed verse in a particular tradition and meter. Structures that meet this criterion are termed ‘‘metrical.’’ An adequate analysis will also specify di¤erences of complexity or tension among the metrical lines. Example (1) illustrates these distinctions with instances of (in order) a canonical line, a complex line, and an unmetrical line for English iambic pentameter. (1) a. The li- / on dy- / ing thrust- / eth forth / his paw (Shakespeare, Richard II, V, 1, 29) b. Let me / not to / the mar- / riage of / true minds (Shakespeare, Sonnet 116) c. Ode to / the West / Wind by / Percy / Bysshe Shelley (Halle and Keyser 1971a, 139) The goals of providing explicit accounts of metricality and complexity were laid out in the work of Halle and Keyser (1966) and have been pursued in various ways since then. From its inception, generative metrics has been constraint based: formal analyses consist of static conditions on well formedness that determine the closeness of match between a phonological representation and a rhythmic pattern. The idea that the principles of metrics are static constraints rather than derivational rules has been supported by Kiparsky (1977), who demonstrated that paradoxes arise under a view of metrics that somehow derives the phonological representation from the rhythmic one or vice versa. The idea that grammars consist of well-formedness constraints has become widespread in linguistic theory. An important approach to constraint-based grammars in current work is Optimality Theory (¼ OT, Prince and Smolensky 1993), whose basic

114

Bruce Hayes

ideas have been applied with success in several areas of linguistics. One might expect that metrics would be easier to accommodate in the OT worldview than any other area, given that metrics has been constraint based for over 35 years. Surprisingly, problems arise when one attempts to do this. To begin, OT is, at least at first blush, a derivational theory: it provides a means to derive outputs from inputs. But in metrics, the idea of inputs and outputs has no obvious role to play; rather, we want to classify lines and other structures according to their metricality and complexity. Second, there is the problem of ‘‘marked winners’’: as we will see, many existing lines or other verse structures violate Markedness constraints. Why shouldn’t these marked winners lose out to less marked alternatives? Hayes and MacEachern (1998) attempt to explain this by supposing that whenever a winning candidate violates a Markedness constraint, there are still higher-ranking Markedness constraints that are violated by all of the rival candidates. However, as we will see, this cannot be true in general. In phonology, the reason marked winners can occur is plain: they obey Faithfulness constraints that are violated by all of their less-marked rivals. But it is not immediately clear how Faithfulness can be implemented in metrics: in a patently nonderivational system, where are the underlying forms that surface candidates can be faithful to? Third, the problem of marked winners arises again when we consider metrical complexity. Intuitively, in certain cases we want to say that the Markedness violations of a winner give rise to a complexity penalty. However, as we will see, in many other cases, Markedness violations can occur without inducing any penalty at all. What distinguishes the two cases? Last, there is a problem of the ‘‘missing remedy.’’1 OT defines the output of any derivation as the most harmonic candidate, the form created by gen that wins the candidate competition. Thus, in principle, every unmetrical form ought to have a well-formed counterpart, an alternative that wins the competition that the unmetrical form loses. But this fails to correspond to the experience of poets and listeners; unmetrical forms like (1c) usually sound wrong without suggesting any specific alternative. All of these problems would have a quick and easy solution under a recent proposal made by Golston (1998); see also Golston and Riad (2000). These authors suggest that the unmetrical lines are simply those that violate high-ranked Markedness constraints, and complex lines are those that violate medium-ranked Markedness constraints. This solution is a radical one, since it claims that in metrics—unlike any other component of grammar—there are no e¤ects of constraint conflict. In other grammatical components, it is commonplace for a candidate to win (and sound perfect) even when it violates a high-ranked constraint when all rivals violate even

Faithfulness and Componentiality in Metrics

115

higher-ranked constraints. Moreover, Hayes and Kaun (1996) and Hayes and MacEachern (1998) give evidence for constraint-conflict e¤ects in metrics, so I believe that the strategy of a special version of OT just for metrics would not work in any event. My own proposal for solving the problems outlined above draws from several sources. Following the principle of the ‘‘Richness of the Base’’ (Prince and Smolensky 1993, 191; Smolensky 1996), an OT grammar can be used to delimit a set of well-formed representations, rather than derive one set of representations from another.  To derive marked winners, I adopt metrical ‘‘Faithfulness constraints,’’ which are ranked against Markedness constraints and determine which forms emerge as metrical in spite of their Markedness violations. The problem of finding the required underlying representations can be solved by fiat, simply by adopting the surface form of each metrical entity as its underlying form (Keer and Bakovic´ 1997; Bakovic´ and Keer 2001).  With Faithfulness constraints in place, the problem of metrical complexity can be addressed by using the stochastic approach to gradient well-formedness developed in Hayes and MacEachern (1998), Hayes (2000), and Boersma and Hayes (2001).  Finally, to solve the missing-remedy problem, I assume (following Kiparsky 1977) that the metrical grammar is componential and that candidate representations should be evaluated independently in each component. To be well formed, an output must win the competition for every component. This permits grammars that rule out forms absolutely, without suggesting an alternative. 

The data with which I will test my proposals involve two problems that (in my opinion) received only partial solutions in earlier work: free variation in quatrain structure (Hayes and MacEachern 1998) and the distribution of mismatched lexical stress in sung verse (Hayes and Kaun 1996). 6.2

Basics

I assume that a meter forms an abstract rhythmic pattern and that there exists for each tradition a system of principles that determine when phonological material properly embodies a pattern in verse. The verse examined here will be the sung verse of traditional Anglo-American folk songs. For many such songs, the rhythmic pattern of each line can be represented as in (2). (2) [ [ [x [x x]

x x x x ][x x [x x][x x][x

][ ][x x][x

x x x x ][x x x][x x][x x][x x

] ] ] ]

line hemistichs dipods feet

116

Bruce Hayes

This is a ‘‘bracketed grid’’ (Lerdahl and Jackendo¤ 1983; Halle and Vergnaud 1987), which embodies information about the relative prominence of its terminal positions (height of grid columns) and about grouping (constituency at various levels, labeled here at the right side of the grid). The anonymous poet/composers who collectively created the body of Anglo-American folk song sought (tacitly) to provide phonological embodiments of this and similar structures. They did so by matching the rhythmic beats (grid structure) with syllables and stress—and by matching the constituent structure with phonological phrasing. A simple example is the following.2 (3) [ [ [x [x

x x x x][x

][x x][x

It was late

x x ][ x x x ][x x ][x x x][x x][x x][x x][x x][x x

in the night when the squire came

] ] ] ]

line hemistichs dipods feet

home (Karpeles 1932, a33A)

Inspection of this line shows a good match on several grounds: the tallest grid columns are filled with stressed syllables; most of the shortest grid columns initiate no syllable at all; the syllables are fairly well matched with their natural durations; and the main prosodic break of the sentence (after night) coincides with the division of the line into two hemistichs. For discussion and exemplification of these phenomena, see Hayes and Kaun (1996). In English folk songs, it is not just lines that are metrically regulated but also higher-level structures like quatrains. Hayes and MacEachern 1998 (hereafter HM) is a study of quatrain structure, focused in particular on the sequencing of line types within quatrains. For what follows, it will be crucial to make use of HM’s typology of line types, which is reviewed below. A line type that HM call ‘‘3’’ places its final syllable on the 11th grid position, which is the third strong position of the line. The extensive empty grid structure that follows this syllable is detectable in the timing of performance. An example, with its grid structure, is given in (4). (4) [ [ [x [x As

x x x x][x

][x x][x

bright

x x ][ x x x ][x x ][x x x][x x][x x][x x][x x][x x

as the sum-

mer

sun

] ] ] ]

line hemistichs dipods feet

(Ritchie 1965, 36)

G (mnemonically ‘‘Green-O’’) has elongation of the syllable occupying position 11, with no further syllable initiated until the fourth strong position in 15:

Faithfulness and Componentiality in Metrics

(5) [ [ [x [x x]

117

x x x ][ x ][x x ][x [x x][x x][x x][x

A- mong

the

leaves

so

x x x x ][x x x][x x][x x][x x green

O

] ] ] ]

line hemistichs dipods feet

(Sharp 1916, a79)

3f (‘‘three-feminine’’3) has one weakly stressed syllable after position 11 and leaves position 15 unfilled. (6) [ [ [x [x

x x x x][x

She’s

x ] x ][ x x ] ][x x ][x x ][x x ] x][x x][x x][x x][x x][x x][x x]

gone

with the gyp-

sen

Da-

vy

line hemistichs dipods feet

(Karpeles 1932, a33A)

4 is free from any of these gaps; all of the four strong metrical positions are overtly filled and there are no elongations. (7) [ [ [x [x

x x x x][x

The

x ] x ][ x x ] ][x x ][x x ][x x ] x][x x][x x][x x][x x][x x][x x]

keep-

er

did

a

shoot-

ing

line hemistichs dipods feet

go (Sharp 1916, a79)

The distribution of these line types within quatrains is restricted. Inspecting a corpus of 1,028 Appalachian folk songs and other material, HM determined that only certain sequences of 3, G, 3f , and 4 lines can constitute a well-formed quatrain. The list of types that are well attested and assumed to be well formed appears below. For examples of these quatrain types, see HM 478–482. (8)

4444 GGGG 3 f 3f 3f 3f 3333

4G4G 43f 43f 4343 G3G3 3f 33f 3

444G 4443f 4443 GGG3 3f 3f 3f 3

GG4G 3343

G343 3f 343 3f 3G3

HM also lay out and defend an Optimality-theoretic analysis of their data, which is based on a set of 10 metrical Markedness constraints. The idea is that each quatrain type results from a song-specific ranking. The gen function is assumed to provide all of the conceivable schematic quatrain forms, each represented simply as a sequence of line types, for example, 4343. In verse composition, the poet is assumed

118

Bruce Hayes

to adopt a particular ranking of the Markedness constraints so that a single quatrain type wins the Optimality-theoretic competition. The set of possible quatrain types in (8) is modeled by assuming that the poet may freely rank the constraints for purposes of composing any particular song but adheres to that ranking for all of the song’s quatrains. Therefore, the set of quatrains that are predicted to be metrical are those that can be derived by ranking HM’s constraints. The HM analysis predicts the inventory of (8) (or something reasonably close to it) as the factorial typology (Prince and Smolensky 1993) of the constraint set.4 6.3

Problem I: Free Variation in Quatrains

The first empirical problem to be discussed here stems from an apparent inadequacy in the HM analysis, namely, its treatment of free variation. Poets do not always use the same quatrain scheme throughout a multiquatrain song. The most common pattern of variation is one in which the poet uses 3 in the even-numbered lines of each quatrain but either 4 or G for the odd-numbered lines, thus (4/G)3(4/G)3. An example of (4/G)3(4/G)3 is given below in (9), which includes four quatrains taken from the same song. The four strongest metrical beats are marked with underlining and (for silent beats) /q/. (9) 4 3 4 3

Young Edward came to Em-i-ly His gold all for to show, q That he has made all on the la´nds, All on the lowlands low. q

G 3 4 3

Young Emily in her cha´m——ber She dreamed an awful dream; q She dreamed she saw young Edward’s blo´od Go flowing like the stream. q

G 3 G 3

O father, where’s that stra´n——ger Came here last night to dwell? q His body’s in the o´ ——cean And you no tales must tell. q

4 3 G 3

Away then to some councillor To let the deeds be known. q The jury found him guı´l——ty His trial to come on. q

(Karpeles 1932, A56A)

HM note that the purpose of this variation is almost certainly to permit a wider variety of word choice on the poet’s part. The poet’s choice of 4 versus G is based

Faithfulness and Componentiality in Metrics

119

 s / and 4 for on the stress pattern of the last two syllables of the line: G for / . . . s other line endings. This dependency is illustrated by the boldface material in (9). Moreover, this pattern is the expected one, since it provides the best match of linguistic stress to rhythm grid: G provides a falling sequence to match a falling stress pattern, and 4 provides a rising sequence to match a rising one (see (5) and (7) above). The issue of how to derive the free variation in (4/G)3(4/G)3 is deferred by HM. As a stopgap, they propose ‘‘F’’ as a fifth line type, defined specifically as involving free variation between 4 and G. To this they add a Match Stress constraint, whose e¤ect is specifically to favor F. Under this arrangement, it is possible to derive quatrain types like (4/G)3(4/G)3, viewed as ‘‘F3F3,’’ simply by ranking Match Stress high enough. A more principled account would allow each of the types in {4343, G343, 43G3, G3G3} to emerge as a winner of the candidate competition under appropriate circumstances relating to the stress pattern of the line ending and hence ultimately to the poet’s choice of words. However, such a capacity is beyond the HM system, since that system evaluates only schematic representations like ‘‘4343,’’ without regard to their linguistic content. At this point, we can state the problem to be solved: to set up a grammatical system that avoids artificial constructs like F but can nevertheless derive variable quatrain types like (4/G)3(4/G)3. This will require us first to develop the formal apparatus. 6.4

Theory

To begin, it is helpful to consider what Optimality-theoretic grammars can do. 6.4.1

Defining Inventories with OT Grammars

The most familiar function of an Optimality-theoretic grammar is that of derivation; for instance, from a phonological underlying representation, we seek to derive the surface representation. In derivation, the gen function creates all conceivable surface representations, and the output is selected from among them by successively winnowing down the candidate set through a ranked set of constraints until one winner emerges. A second thing that OT grammars can do is inventory definition: the definition of a fixed (though possibly infinite) set of legal structures. The method of inventory definition described here is from Prince and Smolensky (1993) and Smolensky (1996). Let there be an additional gen, called genrb (‘‘gen of the Rich Base’’) that defines the full set (possibly infinite) of underlying representations. Submit each member of genrb to an OT grammar. When this is done, it will often be the case that distinct

120

Bruce Hayes

members of genrb will be mapped onto the same surface form. Assume further a process of collation: we remove duplicate outputs and thus collect the full set of forms that are derived as an output from at least one input. This is the inventory that the grammar defines. I will call this inventory the ‘‘output set,’’ and I will refer to an Optimality-theoretic grammar intended for defining an output set as an ‘‘inventory grammar.’’ In this view, it is not crucial for the ‘‘derivation’’ to add any new material at all. Assume in particular that genrb is su‰ciently unconstrained that it includes all possible surface representations. In this case, we can assume, as proposed by Keer and Bakovic´ 1997 and Bakovic´ and Keer 2001, that the underlying form for any candidate surface form is simply itself. We can test a form for well formedness by employing it as an input then determining whether the grammar permits it to survive into the output set.5 More precisely, to test a form F: let IF be an input form identical to F and OF be an output candidate identical to F. If OF defeats all rival candidates when IF is the underlying form, then F belongs to the output set and is legal.6 Whether OF can win the competition will depend in large degree on the ranking of the Faithfulness constraints. OF is, by definition, more Faithful to IF than any other candidate. When Faithfulness is ranked high, OF will be able defeat rival forms that perform better than OF on competing Markedness constraints. Thus, in general, inventory grammars with high-ranking Faithfulness constraints permit larger output sets (Smolensky 1996). 6.4.2

Metrics with Inventory Grammars

HM was an attempt to do metrics with an inventory grammar. However, all the constraints in their grammar were Markedness constraints, so the concepts of input forms and Faithfulness were irrelevant. To solve the problem laid out in section 6.3, we need to use inventory grammars that include Faithfulness constraints. I propose that the set of metrical quatrains, under a particular constraint ranking, should be defined as the output set for that constraint ranking. Moreover, the candidate set does not consist of schematic quatrain forms like ‘‘4343,’’ as in HM, but rather quatrains fully embodied in phonological material. To give an instance, the first quatrain in (10) can be taken to be a representative input form.7 (10)

[ [ [x [x x] Young

x x x ][ x ][x x ][x [x x][x x][x x][x x] E- mi- ly

in

her

x x x ][x [x x][x cham-

] x ] x ] x][x x] ber

line hemistichs dipods feet

Faithfulness and Componentiality in Metrics

[ [ [x [x

x x x ][ x ][x x ][x x][x x][x x][x x][x x]

She dreamed [ [ [x [x

Go

an

aw-

ful

she

saw

young

ing

like

the

x x x x ][x x [x x][x x][x x Ed-

x x x ][ x ][x x ][x x][x x][x x][x x][x x] flow-

x x x x ][x x [x x][x x][x x

] line ] hemistichs ] dipods ] feet

dream

x x x ][ x ][x x ][x x][x x][x x][x x][x x]

She dreamed [ [ [x [x

121

ward’s blood

x x x x ][x x [x x][x x][x x

stream

] line ] hemistichs ] dipods ] feet

] line ] hemistichs ] dipods ] feet

(Karpeles 1932, a56A)

Assuming that metrics is transparent, this candidate will count as well formed (i.e., metrical) if it passes the well-formedness test for inventory grammars. Specifically, if IF ¼ OF ¼ (10) and OF wins the Optimality-theoretic competition against all distinct output candidates, then (10) is predicted to be metrical. 6.4.3

Componentiality in Metrics

Before examining the candidate competition, we must add one more ingredient to the analysis: the role of components in candidate evaluation. The issue of componentiality in metrics is addressed by Kiparsky (1977), whose conception is adopted here. Kiparsky proposes that metrics is tricomponential: there is a ‘‘pattern generator,’’ which accounts for the meter; a ‘‘paraphonology,’’ which establishes the metrically relevant phonological representation; and a ‘‘comparator,’’ which evaluates the paraphonological representation against the meter to determine metricality and complexity. These three components are discussed in turn below. Pattern Generator The pattern generator for the verse described here is rather simple. In OT it can be characterized with a set of undominated constraints.

6.4.3.1

(11) a. Quatrain ¼ Couplet Couplet b. Couplet ¼ Line Line c. Line ¼ Hemistich Hemistich

122

Bruce Hayes

x d. Hemistich ¼ [x x] x e. Dipod ¼ [x x] x f. Foot ¼ [x x]

where terminals are dipod heads where terminals are foot heads where terminals are metrical positions

These constraints yield the following structure: [Quatrain [Couplet Line Line] [Couplet Line Line]], where each Line has the internal structure given above in (2). Paraphonology The paraphonology defines the phonological representations that are used in composing verse; it ‘‘constitute[s] a paralinguistic system that specifies the poetic language as a derivative of the system. . .of ordinary language’’ (Kiparsky 1977, 241). Kiparsky used a rule-based paraphonology, which included among others a paraphonological rule for John Milton’s verse that deletes stressless vowels postvocalically.

6.4.3.2

(12) Postvocalic Syncope (Kiparsky 1977, 240–241)   V !q/V stress Postvocalic Syncope can derive monosyllabic [’raIt] from underlyingly disyllabic riot /’raI. t/ and trisyllabic [v .’raI.ti] from quadrisyllabic variety /v .’raI. ti/. Because this rule applies optionally, riot can be scanned in Paradise Lost (PL) as either one or two syllables (13a), and variety as either three or four (13b). e

e

e

e

(13) a. Of riot / ascends / above / thir lof- / tiest Towrs To lux- / urie / and ri- / ot, feast / and dance b. Varie- / tie with- / out end; / but of / the Tree For Earth / hath this / vari- / ety / from Heav’n

(PL 1.499) (PL 11.715) (PL 7.542) (PL 6.640)

Examples of Postvocalic Syncope are found in Shakespeare as well. As Kiparsky points out, no examples occur in Pope’s verse, which shows that the rule is poet specific. It is quite straightforward to translate paraphonological rules into Optimalitytheoretic terms. Postvocalic Syncope reduces to the free ranking of the constraint   þsyllabic (McCarthy and Onset (Prince and Smolensky 1993) against Max stress Prince 1995). Where Onset dominates, the stressless vowel is dropped from riot in   þsyllabic order to avoid the onsetless second syllable; where Max dominates, stress the vowel is retained. A point that will be crucial below is that, at least in English, paraphonology has only modest e¤ects: schwas are lost in hiatus, nonlow vowels become glides; but

Faithfulness and Componentiality in Metrics

123

major insertions and deletions (say, of whole syllables) are not found. References on English verse paraphonology supporting this point include Bridges (1921) and Tarlinskaja (1973). I will also stipulate that paraphonology cannot alter the stress patterns of words.8 This means that when the data show a mismatch of stress against the grid, I will be assuming that this involves a Markedness violation in the comparator component, not a Faithfulness violation in the paraphonology. As Kiparsky (1977) points out, the paraphonology is independent of the mechanisms (whatever they may be) that govern the oral performance of verse. In particular, a performer is usually free not to realize paraphonological changes, even those that are crucial to metricality. The paraphonological level of the metrical grammar is, therefore, an abstract one that serves to define the phonological representations against which metricality and complexity are computed. The last part of the componential organization that Kiparsky assumes is the comparator. This is the core of the metrical system. It consists of a set of metrical filters (essentially, constraints) that examine a phonological representation from the paraphonology and a rhythmic representation from the patterngenerating component and determine whether and how they can be matched to form a unit of metrical verse. Further principles adumbrated by Kiparsky assign differences of complexity.

6.4.3.3

Comparator

Componential Organization and the Evidence Supporting It Kiparsky o¤ers empirical arguments that metrics is organized componentially. His most crucial point is that the paraphonology always provides the same representation to every metrical constraint: for example, constraints matching stress cannot regard riot as monosyllabic, while constraints matching syllable count to rhythmic positions regard it as disyllabic. In an Optimality-theoretic account of the paraphonology, there is an additional reason the system must be componential. In OT, structural changes (e.g., vowel loss) are decoupled from their phonotactic causes (e.g., the requirement that syllables have onsets). A noncomponential theory of the paraphonology would wrongly claim that structural changes could be triggered in order to obey the requirements of the metrics. A hypothetical example of this type would be as follows. Imagine we are dealing with the verse of a poet like Milton who licenses loss of stressless vowels in postvocalic position. It is necessary under any account of iambic pentameter to assume a constraint that prevents extra syllables from cropping up in random locations in the line; let us call this constraint *Ungridded s. If the system is not componential, we would expect to find lines like (14). 6.4.3.4

124

Bruce Hayes

(14) *Vivacity without end; but of the Tree

(construct)

This would follow from the constraint ranking given in (15a), with representations as in (15b) (hi marks ungridded material).   þsyllabic (15) a. *Ungridded s  Max stress [metrical constraint] [paraphonological constraint] b. [ x x ] line [ x x][ x x x ] hemistichs [x x] [x x][x x][x x][x x ] feet [vI ’væsh iti] with- out end; but of the tree e

To my knowledge, such cases do not exist. Paraphonological phenomena in metrics always have authentic phonological structural descriptions. Indeed, as Kiparsky points out, they look just like ordinary language phonology and are often grounded in the fast-speech phonology of the poet’s language. A componential organization of the metrical system implies, correctly, that paraphonological processes apply only when their phonological structural descriptions are met. With the concepts of inventory grammar and componentiality in place, we can now provide a definition of metricality. The leading idea, found in earlier work such as Inkelas and Zec (1990), is that di¤erent components simultaneously evaluate the well-formedness of the same representation.9 The components are separate, not because they apply in sequence but because they evaluate di¤erent aspects of the representation. Following this approach, we can say that a quatrain of verse is metrical when the following conditions are met.

6.4.3.5

Defining Metricality in a Componential Inventory Grammar

(16) A verse form is metrical i¤ a. The metrical pattern belongs to the output set for the pattern generator (section 6.4.3.1). b. The phonological material belongs to the output set for the paraphonology (section 6.4.3.2). c. The complete representation, with phonological material aligned to the metrical pattern, belongs to the output set for the comparator (section 6.4.3.3). Componentiality guarantees the result observed in the previous section: paraphonology cannot be conditioned metrically, because the candidate set against which forms are paraphonologically evaluated consists solely of phonological representations without regard to their metrical setting. Thus, if a schwa is lost paraphonologi-

Faithfulness and Componentiality in Metrics

125

cally, it must be lost in order to avoid a hiatus, rather than to avoid a mismatch in the scansion—if the latter holds true as well, that is a felicitous result for the poet, but the components of the metrical grammar act blindly to each other’s purposes. As will become clear below, the componential approach is also crucial to explaining metricality itself. It introduces the possibility that the rival candidates that defeat an input are sometimes themselves ill formed with respect to some other component, specifically the paraphonology. This is the crucial means by which lines may be classified as unmetrical. In the sections that follow, I demonstrate how lines are classified as metrical or unmetrical in this system. Once this is done, we can return to the problem that was stated in section 6.3, the (4/G)3(4/G)3 quatrain. 6.5

Unmetricality in 4343 Quatrains

Many English folk songs (particularly ballads) are composed in quatrains of the form 4343. The odd-numbered lines in such quatrains are consistently of the type 4, and never G. Since ballads can go on for many stanzas, we can be confident that in such cases the quatrain structure is not the (4/G)3(4/G)3 discussed in section 6.3 above. A quatrain of the type 43G3, G343, or G3G3 introduced into a strict 4343 song would count as an unlicensed deviation from the established meter, that is, as unmetrical. To illustrate the ideas above, I will construct a tricomponential, Optimality-theoretic inventory grammar that permits only 4343. 6.5.1

Constraints

The bulk of the work will be done in the comparator, and all constraints mentioned should be assumed to belong to this component unless otherwise stated. Four Markedness constraints from Hayes and MacEachern (1998) are relevant. Couplets are Salient requires that a couplet consist of an internally connected, pause-demarcated unit, roughly as follows. (17) [ ] [ x x ][ x x ] [x x ][x x ][x x ][x x ] [x x ][x x ][x x ][x x ][x x ][x x ][x x ][x x ] [ x x] [ x x] [ x x] [ x x] [ x x] [ x x] [ x x] [ x x] [ x x] [ x x] [ x x] [ x x] [ x x] [ x x] [ x x] [ x x]  ! ! su‰ciently dense spacing of gaps syllables required here required here

couplet lines hemistichs dipods feet

This constraint is obeyed fully only by the couplets 43, G3, and 3f 3. For a precise statement of the constraint and its rationale, see HM (485–486, 492–493).

126

Bruce Hayes

Lines are Salient (HM, 483–486, 493) is the analogous constraint at the line level. (18) [ x x ] [ x x ][ x x ] [x x ][x x ][x x ][x x ] [x x][x x][x x][x x][x x][x x][x x][x x] ! ! su‰ciently dense spacing of gaps syllables required here required here

line hemistichs dipods feet

This constraint is violated gradiently in a way explained in HM (p. 493) and note 15; the optimal line type defined by this constraint is 3. Fill Strong (HM, 490) requires that the four strongest positions of the line be filled by a syllable, as shown in (19). (19) [ [ [x [x

x x x ][x x][x x][x l s

x x ][ x x x ][x x ][x x x][x x][x x][x x][x x][x x l l l s s s

] ] ] ]

line hemistichs dipods feet

It is obeyed by 4 and G lines and violated by 3 and 3f . Finally, *Lapse (HM, 490) penalizes failure to place a syllable between any two of the four strongest positions, as in (20). (20) [ [ [x [x

x x x x][x

x ] ][x x ] x][x x][x x] q q q q

x [ x x [x x ][x x [x x][x x][x x][x x q q q q q

] ] ] ]

line hemistichs dipods feet

* * * *Lapse is violated by 3 and G lines but not 3f or 4. I will also assume two Faithfulness constraints, which are stated in the language of Correspondence Theory (McCarthy and Prince 1995). (21) a. MAX(s) Assess a violation for every syllable in the underlying form that is not in correspondence with a syllable in the surface form. b. DEP(s) Assess a violation for every syllable in the surface form that is not in correspondence with a syllable in the underlying form.10

Faithfulness and Componentiality in Metrics

127

It is straightforward to determine the Max(s) and Dep(s) violations for any pair of input and output when they are classified by their line type (see (4)–(7) above). For example, any line of type 4 has an extra syllable with respect to a similar line of type G and therefore incurs (at least) one Dep(s) violation when the G line occurs at its underlying representation. The full candidate set that is input to these constraints is enormous, since it comprises all phonological representations placed in correspondence with the grid (see section 6.4.2). However, for initial purposes it su‰ces to use formulae like 4343 to designate any quatrain that would be classified as 4343. Idealized in this way, the candidate set numbers 256, which is the set of logical possibilities implied by choosing from among four line types, four times per quatrain (256 ¼ 4 4 ). 6.5.2

Ranking

The ranking needed to derive 4343 is given in (22a). As tableau (22b) shows, all but 9 of the 256 candidate types are ruled out by Couplets are Salient.11 Of these 9, only 4343 maximally satisfies both Fill Strong and *Lapse. (22) a. Couplets are Salient  {Fill Strong, *Lapse}  {Max(s), Dep(s), all others} Other constraints

Dep(s)

Max(s)

*

Lapse

+

Fill Strong

/4343/

Couplets are Salient

b.

[4343]

**

**

*[G343]

**

***!

*

(*)

*[43G3]

**

***!

*

(*)

*[G3G3]

**

***!*

**

(*)

*[3f 343]

***!

**

*

(*)

*[433f 3]

***!

**

*

(*)

*[G33f 3]

***!

***

**

(*)

*[3f 33f 3]

***!*

**

**

(*)

*[3f 3G3]

***!

***

**

(*)

(*)

(*)

(*)

247 candidates

*!(*)

(*)

(*)

(*)

128

Bruce Hayes

It can be seen that under this ranking, Faithfulness plays no role in determining the output; the decision is already made once we have culled candidates with the three top-ranked Markedness constraints. 6.5.3

Ruling out Unmetrical Forms

What must now be demonstrated is that this schematic analysis is e¤ective in ruling out not just schemata but actual unmetrical quatrains under the conception of metrical grammar laid out in section 6.4. Let us suppose that the anonymous folk poet is making up a new stanza for a ballad. Assume that this ballad (like hundreds of others) is composed in strict 4343 quatrains. We must demonstrate that under the grammar with the ranking of (22a), any other quatrain type would emerge as unmetrical. In particular, consider the task of ruling out *G343. To be concrete, let us suppose that the G line of this hypothetical *G343 quatrain happens to be (10) above, repeated for convenience in (23). (23)

[ [ [x [x x] Young

x x x [x

][x x][x

E- mi- ly

x x ][ x x x ][x x ][x x x][x x][x x][x x][x x][x x in

her

cham-

] ] ] ]

line hemistichs dipods feet

ber

Although this line is thus metrical in its own context (i.e., a real song composed in (4/G)3(4/G)3), it could not metrically appear in a song composed in strict 4343; this is what we want the analysis to predict. The folk poet is assumed to have (tacitly) internalized an inventory grammar of which (22a) is a partial sketch. I assert without proof that this grammar is transparent. Under this assumption, one can show that (23) is unmetrical in the given context by using the ‘‘grammaticality test’’ laid out in section 6.4.1. Specifically, one must demonstrate that when (23) is adopted as an underlying representation, there will be a rival candidate that defeats (23) in the candidate competition. In fact, there are many such candidates, one of which is shown in (24). [ [ [x [x x] Young

x x x ][ x ][x x ][x [x x][x x][x x][x x] Em- i-

ly

in

her

x x x ][x [x x][x

[’t§eIIm

b

e

(24)

] x ] x ] x][x x]

line hemistichs dipods feet

lÔ]

The reader is asked for the moment to ignore the absurdity of the word chambeler and concentrate solely on the candidate competition. Candidate (24) is an unfaithful candidate, since it possesses a syllable [b ] where the input form (23) has a null. Specifically, (24) violates Dep(s) once. However, there is also a Markedness constraint e

Faithfulness and Componentiality in Metrics

129

that is violated by (23) but not (24), namely *Lapse. Since in grammar (22a) *Lapse dominates Dep(s), (24) will emerge as more harmonic than (23). (25)

(23) Young Emily in her cham——ber +

*Lapse

(24) Young Emily in her chambeler *(23) Young Emily in her cham——ber

Dep(s) *

*!

As (25) shows, (23) is defeated in the candidate competition, despite its obvious Faithfulness virtues. It is thus excluded from the output set of the grammar defined by this constraint ranking and therefore is unmetrical in its context. The reader will have noticed that candidate (24) is itself absurd from a di¤erent point of view: nothing in the paraphonology of English folk verse licenses the extra syllable or the inserted segmental material [ l]. Here, componentiality plays a crucial role. Candidate (24) does not directly compete with (23) with regard to its phonological content; that competition unfolds within the paraphonology—where (24) most definitely loses.12 But under the componential definition of metricality in (16), the winner must belong to the output set of each component separately and, therefore, must defeat all rivals in each component separately. In the grammar under discussion, (23) fails to defeat (24) in the comparator. The fact that its phonological material wins in the paraphonology cannot rescue (23). Consequently, for the grammar of (22a) and the underlying representation (23), there is no candidate that wins in all components. For this reason, the derivation ‘‘crashes,’’ and (23) is classified as unmetrical. One might appropriately call the ‘‘chambeler’’ line (24) a ‘‘suicide candidate.’’ In a componential inventory grammar, a suicide candidate is one that defeats the maximally faithful candidate in one component while losing to the faithful candidate in a di¤erent component. Suicide candidates usually cause derivations to crash.13 Crashing derivations provide an answer to one of the questions asked in the introduction, namely, how a grammar can predict a line to be ill formed without making any claims about which more optimal form should putatively take its place. In the case of (23), there simply is no alternative line that emerges from the grammar as the appropriate ‘‘corrected’’ version: any alternative that beats (23) metrically will be paraphonologically impossible. A folk poet dissatisfied with the unmetrical (23) must think up a di¤erent line, and the grammar does not tell her what to compose.14 e

6.6

Deriving Free Variation and Marked Winners

Having illustrated the apparatus of Faithfulness and componentiality, I will now return to the problem laid out in section 6.3, that is, of deriving the (4/G)3(4/G)3

130

Bruce Hayes

quatrain. The grammar needed for (4/G)3(4/G)3 turns out to be rather similar to that for strict 4343, except that the Faithfulness constraints Max(s) and Dep(s) are ranked higher. This permits a larger output set, which encompasses the free variants. The specific ranking needed is the one given in (26). This ranking di¤ers from the 4343 ranking in (22a) in that Max(s) and Dep(s) outrank *Lapse. (26) Couplets are Salient  Fill Strong  {Max(s), Dep(s)}  {*Lapse, remaining constraints} A tableau for the variant 43G3 is given in (27). Lines are Salient is explained in section 6.5.1 above, and Long-last is explained in HM (488–490, 493); they are included here merely to show that there are Markedness constraints in the system that would have favored di¤erent outcomes had they been ranked higher. *Lapse

Lines are Salient

Long-Last

110 15

*

**

200

*

****

20

*

*

***

110

*

**

101

***

11

*

**

101

*

**

*

***

11

*

***!*

**

*

**

2

*

(*)

(*)

(*)

(*)

(*)

(*)

[43G3]

**

*[4343]

**

*[G3G3]

**

*!

*[G343]

**

*!

*[3f 343]

***!

*

*[3f 3G3]

***!

*

*[433f 3]

***!

*

*[G33f 3]

***!

*[3f 33f 3] 247 candidates

*!(*)

Dep(s)

***

Max(s)

+

Fill Strong

/43G3/

Couplets are Salient

(27)

*!

Grammar (26) culls the rival candidates in a way similar to the 4343 grammar (22a): Couplets are Salient removes all but 9 of the 256 candidate types, and Fill Strong removes 5 more. Thus, at this point we know that the output cannot contain any quatrains other than the four targets; what is at issue is whether all four will make it into the output set.

Faithfulness and Componentiality in Metrics

131

Grammar (26) di¤ers from grammar (22a) in that whereas (22a) places the Faithfulness constraints at the bottom of the hierarchy, (26) places them just below Fill Strong, so that they are active in selecting from among the surviving four candidates. When the underlying representation is a 43G3 quatrain, any rival candidates of the types 4343, G3G3, or G343 will incur Faithfulness violations. The 43G3 form, being totally faithful to itself, thus emerges as a winner and qualifies as a member of the output set. Not surprisingly, the three other types embodied in the formula (4/G)3(4/G)3 also emerge as part of the output set of (26), since each is free of Faithfulness violations when it is selected as the underlying form. Tableaux showing how each one beats out its three best rivals are given below. *Lapse

Lines are Salient

Long-Last

200

*

*!

***

110

*

**

**!

****

20

*

**

*!

***

110

***

110

**

200

*

****

20

*

***

110

*

****

20

*

**

*[43G3]

**

*[G3G3] *[G343]

Dep(s)

b.

**

[4343]

Max(s)

+

Fill Strong

/43G3/

Couplets are Salient

(28) a.

/G343/ +

c.

[G343]

**

*[4343]

**

*[G3G3]

**

*!

*[43G3]

**

*!

*!

*

/G3G3/ +

[G3G3]

**

*[4343]

**

*!*

**

200

*

*[43G3]

**

*!

***

110

*

*[G343]

**

*!

***

110

132

6.6.1

Bruce Hayes

Distribution of 43G3

The 43G3 quatrain type has an interesting property: although there are songs composed in ‘‘strict’’ (nonvarying) 4343, strict G343, and strict G3G3 (see, HM 478– 481), to my knowledge there are no songs composed in strict 43G3. 43G3 occurs only as a free variant in songs that also allow 4343, G343, and G3G3 in other stanzas. This is predicted by the system given here. There exist rankings (see HM, 495) that permit only 4343, only G343, and only G3G3 to survive into the output set. For these rankings, the output set is culled down to a single quatrain type by Markedness constraints placed at the top of the ranking. But for 43G3 to survive into the output set, the Faithfulness constraints Dep(s) and Max(s) must be ranked relatively high. When such a ranking holds, the other three quatrain types will also be allowed into the output set, since the Faithfulness constraints will not penalize them when they are selected as the underlying representation. Note that since 43G3 cannot be selected by Markedness constraints alone, it is a kind of ‘‘marked winner’’ in the sense defined in section 6.1. The example of this section thus illustrates how Faithfulness constraints can be used in metrics to derive marked winners. This concludes the analysis of free variation in quatrain structure. The crucial idea has been that when Faithfulness is ranked higher in the grammar, a variety of underlying forms are able to defeat all their rivals and emerge as outputs of the inventory grammar. In this way, the analysis is able to derive free variation, without (as in HM) stipulating constraints that actively require it. 6.7

Lexical Inversion

As a second illustration of Faithfulness and componentiality in metrics, I will discuss a problem that was addressed but not fully solved by Hayes and Kaun (1996; hereafter HK). I define a ‘‘lexical inversion,’’ following earlier work, as a configuration in which the syllables of a simplex polysyllabic word with falling stress are placed in a metrical position that calls for rising stress. Here is an example, highlighted in bold. (29) [ [ [x [x He

x x x x][x

][x x][x

pulled

rings

x x x][x o¤

x ][ x ][x x ][x x][x x][x x][x of

his

fin-

x x x][x

] ] ] x]

line hemistichs dipods feet

gers (Karpeles 1974, 7G)

Faithfulness and Componentiality in Metrics

133

HK noticed that lexical inversions in folk verse have an asymmetrical distribution, which is strikingly di¤erent from what occurs in iambic pentameter. For folk verse, the great majority of inversions occur at the end of a line, as in (30).16 (30) a. b. c. d. e. f. g.

Who should ride by but Knight William I’ll bet you twenty pound, master I fear she will be taken by some proud young enemy There lived an old lady in the north country And two of your father’s best horses But he had more mind of the fair women Lived in the west country q

(Karpeles 1974, 27A) (Karpeles 1974, 7F) (Karpeles 1974, 45A) (Karpeles 1932, 5B) (Karpeles 1932, 5B) (Ritchie 1965, 36) (Karpeles 1974, 43E)

Most of these are of line type 4, as in (30a–f ), with a few cases of 3, as in (30g). A further asymmetry that HK note is that in quatrain types where 4 or 3 occur in free variation with G or 3f , lexical inversion is quite unusual. Thus, songs in strict 4343 include lexical inversions far more often than songs in (4/G)3(4/G)3 quatrains.17 HK propose an intuitive explanation for these facts, which I will here employ as the basis of an OT analysis. There are three points at issue: (i) why lexical inversion is disfavored in general, (ii) why it has a special privilege of occurring at the end of the line, and (iii) why this privilege should be so rarely exercised in line positions that permit (4/G) free variation. 6.7.1

Ruling out Inversion in General

For the first point, HK observe that unlike pentameter, folk verse virtually always leaves a certain number of grid positions unfilled. Therefore, when the syllables of a line are such that an inversion might arise, it is usually the case that a minor shift in the location of the syllables would make inversion unnecessary. For example, instead of producing the inversion in (31a) (mismatched stress and grid columns shown in boldface), the folk poet can sidestep the problem simply by moving William over a bit, as in (31b). (31) a. [ [ [x [x x]

x x x ][ x ][x x ][x [x x][x x][x x][x x]

x x x x ][x x [x x][x x][x x

*Wilb. [ [ [x [x x]

liam he was a x x x ][ x ][x x ][x [x x][x x][x x][x x]

noble knight (construct) x ] line x x ] hemistichs x ][x x ] dipods [ x x ] [ x x ] [ x x ] feet

Wil-

liam

he

was a

no-

ble

knight

] ] ] ]

line hemistichs dipods feet

(construct)

134

Bruce Hayes

To formalize this idea, we need to state two constraints: a Markedness constraint that bans lexical inversion and the Faithfulness constraint violated by (31b) when (31a) is the underlying representation. This is the subject of the next two sections. 6.7.1.1 MATCH STRESS In formulating a constraint to exclude lexical inversions, we are on well-explored territory. Kiparsky (1975b), Bjorklund (1978), and other scholars have shown that poets and poetic traditions often require a particularly strict match to the meter for sequences of stressed and unstressed syllables that fall within a single simplex word. Let us assume such a constraint here.18

(32) MATCH STRESS Assess a violation if    

si and sj (in either order) are linked to grid positions Gi and Gj , respectively; si has stronger stress than sj ; Gj is stronger than Gi ; and si and sj occupy the same simplex word.

In a full grammar, there would be other constraints requiring stress matching in other contexts as well; but for present purposes, (32) will su‰ce. IDENT(location) We must also formulate the Faithfulness constraint that is violated when, for example, (31b) is taken to be a candidate surface form for underlying (31a). Here, the contents of grid and phonological representation are identical, but the temporal association of syllables and grid marks is di¤erent. I will assume that this violates the Faithfulness constraint (33).

6.7.1.2

(33) IDENT(location) If si is linked to grid position G in the input and to grid position G 0 in the output candidate, assess a number of violations equal to the distance in grid positions between G and G 0 . To make (33) explicit, we need to say how violations are assessed when more than one syllable is shifted over. Various possibilities exist; since nothing matters here in how this issue is resolved, I will assume for concreteness that the violations are simply summed. Thus (31b), taken as a surface candidate for underlying (31a), incurs nine violations of Ident(location): two for Wil-, two for -liam, two for he, two for was, and one for a, as shown below.

Faithfulness and Componentiality in Metrics

(34) [ [ [x [x x]

135

x x x ][ x ][x x ][x [ :x x ] [ ::x x ] [ ::x x ] [ ::x x ]

2 ::::::::: :2:::::::::: 2 :::::::: 2::::::::::: 1:::::: ::::::: ::: :::: ::::::: ::::

*Wil-

liam Wil-

he liam

was he

x x x x ][x x [x x][x x][x x

a nowas a no-

ble ble

knight knight

] ] ] ]

line hemistichs dipods feet (31a) (31b)

6.7.1.3 Ruling out Nonfinal Lexical Inversion The goal is to construct an analysis in which (31), with a lexical inversion in nonfinal position, is unmetrical. Assume that (31a) is the underlying form and that (31b) is a rival candidate. If Match Stress outranks Ident(location), (31b) will be the winner, as (35) shows.

(35) (31a) William he was a noble knight

Match Stress

(31b) William he was a noble knight *(31a) William he was a noble knight

Ident (location) *

*!

Because it is beaten by (31b), (31a) is excluded from the output set and is unmetrical. There is independent reason to think that the ranking Match Stress  Ident(location) will prevail, because Ident(location) is generally a weak constraint and Match Stress a strong one. Here is the evidence for these two claims. The experimental data gathered by HK indicate that folk song lines are composed in a way such that the grid locations of the syllables are relatively predictable. In particular, HK’s consultants, given only text and grid, showed fair agreement among themselves as to what the proper alignment of syllables to grid should be. They could not have achieved this unless the song texts gave them clues as to where to locate the syllables. This implies that where syllables go in the grid is, to a fair degree, noncontrastive information. In Optimality-theoretic terms, noncontrastive structural information is that which is protected by low-ranking Faithfulness constraints; hence, Ident(location) must be ranked low. On the other hand, Match Stress is expected, based on our general knowledge, to be ranked rather high. There are poetic traditions (e.g., classical German and Russian verse) in which it is undominated; and even where Match Stress is violated there are usually strict limitations on where the violations may occur (Kiparsky 1977). Moreover, quite a few folk songs have no lexical inversions at all, suggesting that they are composed under a ranking in which Match Stress is undominated (see section 6.7.3 below). If rankings are relatively constrained even across di¤erent metrical forms, we expect Match Stress to be ranked relatively high in general. It

136

Bruce Hayes

would certainly be expected to outrank a characteristically feeble constraint like Ident(location). The upshot is that in the general case, candidates that match stress by ‘‘sliding’’ the syllables will be favored over candidates that mismatch stress. This provides an across-the-board pressure against lexical inversion. 6.7.2

Ruling Out Nonfinal Inversion

To explain why inversion can occur at the end of the line (in certain quatrain types), HK note that in this location, additional constraints are active. These constraints rule out all of the available ‘‘slid over’’ candidates, leaving lexical inversion behind as the best remaining option.19 The way this works can be seen if we ponder what kind of syllable sliding in principle could rescue the lexical inversion in (36). (36) [ [ [x [x

x x x x][x

Fair

][x x][x

El- li- nor

x x x][x

][ ][x x][x

she

x x x x][x

][x x][x

was a gay

la-

x x x][x

] ] ] x]

line hemistichs dipods feet

dy (Karpeles 1974, 15I)

In order to avoid a violation of Match Stress, the crucial stressed syllable la- must migrate to a stronger grid position than mismatched -dy. The migration in principle could be to the left (where gay sits in (36)) or to the right (the location of -dy in (36)). For each of these two possibilities, there are two reasonable20 possibilities for where to put -dy, making a total of four, shown in (37). The dotted arrows show where lahas been ‘‘moved’’ in each of the four possibilities.

El- li- nor

Fair Fair Fair Fair -dy . . .

ElElElEl-

lililili-

she

nor she was a nor she was a

nor nor

she she

was a gay gay was was

gay

:::: X

Fair

][x x][x la-

::::

x ][ ][x x ][x x][x x][x x][x

x x x x][x

x x x][x

lalaa a

dy gay gay

] ] ] x]

line hemistichs dipods feet

dy

original, with inversion

dy

slide la- left

: X :: X :::: ::::: ::::::::: ::::::::: ::::::::: ::::: :::: :: : : : :::: ::: X

a. b. c. d.

x x x x] [ x

::::

(37) [ [ [x [x

la- dy la-

slide la-, -dy left slide la-, -dy right slide la- right; -dy into next line

Faithfulness and Componentiality in Metrics

137

Below, we consider the four possibilities in turn and show that under appropriate constraint rankings, they are excluded. The lexical inversion setting remains as the most viable option. Sliding to G Example (37a), repeated below as (38), is a candidate in which Match Stress is obeyed by moving the penult of lady into the third strong position of the line, crowding some of the other syllables to fit it in. The syllable -dy is kept in the fourth position, so a G line results. 6.7.2.1

(38) [ [ [x [x

x x x x][x

Fair

x x ][ x ][x x ][x x x][x x][x x][x x][x

El- li- nor she was a

gay

x x x][x

][x x][x

la-

] ] ] x]

line hemistichs dipods feet

dy

This candidate will fail to defeat the lexical inversion candidate (36), provided that *Lapse, the constraint that forbids G, is ranked above Match Stress. (39) (36) Fair Ellinor she was a gay lady +

*Lapse

Match Stress

(36) Fair Ellinor she was a gay lady

Ident (location)

*

*(38) Fair Ellinor she was a gay la——dy

*!

9 *’s

As we will see in section 6.7.4 below, this ranking will necessarily hold, for independent reasons, in the quatrain types that allow lexical inversion. 6.7.2.2 Sliding to 3f Example (37b), repeated as (40), is similar to (37a): la- is again placed in the third strong position of the line, but in this case -dy is also slid over, so that a 3f line results.

(40) [ [ [x [x x] Fair

x x x ][ x ][x x ][x [x x][x x][x x][x x] El- li- nor she was a

gay

x x x [x la-

][x x][x

x x x][x

] ] ] x]

line hemistichs dipods feet

dy

If Fill Strong, which forbids 3 and 3f lines, is ranked above Match Stress, then this candidate will also fail to defeat the lexical inversion candidate (36).

138

Bruce Hayes

(41)

Fill Strong

(36) Fair Ellinor she was a gay lady +

(36) Fair Ellinor she was a gay lady

Match Stress

Ident (location)

*

*(40) Fair Ellinor she was a gay lady q

*!

11 *’s

6.7.2.3 The Fast-Syllable Candidate Example (37c), repeated as (42), is a candidate in which la- has been moved rightward to the fourth strong position of the line; -dy occupies the weak terminal position.

(42) [ [ [x [x Fair

x x x x][x

][x x][x

El- li- nor

x ][ x ][x x][x x][x she

x x x ][x x][x x][x

was

a

] x ] x ] x][x x]

gay

line hemistichs dipods feet

la- dy

Impressionistically, the e¤ect is of an uncomfortably fast rendition of lady at the end of the line, suggesting unmetricality. In fact, lines like this are quite rare in real verse. Moreover, in an experiment conducted by HK in which 670 lines of folk verse were chanted from the written text by 10 speakers of English, the consultants fairly generally avoided this kind of rendition. HK suggest that this line type is ill formed because it involves a gross mismatch of the sung duration of the final syllables versus their natural duration. (For discussion of the evidence that supports duration matching in folk verse, see HK section 6.1.) In the present case, because of the e¤ects of phrase-final lengthening (Wightman et al. 1992) and the concurrence of line and intonation phrase boundaries, the line-final syllable is normally quite long. It is therefore ill fitted to fill a single grid slot. I will therefore assume a constraint to be called Match Duration that penalizes intonational phrase-final syllables21 that are squeezed into the final grid slot of the line. I assume further that Match Duration is quite highly ranked and in particular that it outranks Match Stress. Therefore, the ‘‘fast syllable’’ candidate (42) (shown here iconically with condensed type) must lose out to the inverted stress candidate (36). (43) (36) Fair Ellinor she was a gay lady +

Match Duration

(36) Fair Ellinor she was a gay lady *(42) Fair Ellinor she was a gay lady

Match Stress

Ident (location)

* *!

6 *’s

Faithfulness and Componentiality in Metrics

139

6.7.2.4 The Overflow Candidate The fourth and last reasonable possibility for placing lady in metrically matched position was (37d), repeated below as (44).

(44) [ [ [x [x

x x x x][x

Fair [ [ [x [x

x x ][ x x ][x x ][x x ][x x x][x x][x x][x x][x x][x x][x x

El- li- nor x x x x][x

she

was

a

gay

] ] ] ]

line hemistichs dipods feet

] ] ] ]

line hemistichs dipods feet

la-

x x ][ x x ][x x ][x x ][x x x][x x][x x][x x][x x][x x][x x

dy . . . Here, the second syllable of lady spills over into the grid of the next line. This is unusual in folk verse and arguably is so because it violates general principles of alignment for phonological phrasing and metrical constituents. Evidence in support of such alignment principles is given in Kiparsky (1975b), Hayes (1989b), Hayes and MacEachern (1996), and HK section 6.2. I assume that such lines violate Alignment constraints (McCarthy and Prince 1993a), which in this case require line breaks to coincide with major phonological phrase breaks. For concreteness, we will assume that the relevant type of phrase break is the Intonational Phrase, though in a full grammar additional constraints would be needed that refer to higher and lower prosodic domains as well (Selkirk 1980; Nespor and Vogel 1986; Pierrehumbert and Beckman 1988). In McCarthy and Prince’s system, the relevant constraint is Align(Line, L, Intonational Phrase, L): ‘‘the left edge of every Line must coincide with the left edge of an Intonational Phrase.’’ Align(Line, L, Intonational Phrase, L) is a characteristically strong constraint in folk verse, and I assume it generally outranks Match Stress. Under this ranking, candidate (44) must lose out to the lexical inversion candidate. (45) (36) Fair Ellinor she was a gay lady +

Align

(36) Fair Ellinor she was a gay lady *(44) Fair Ellinor she was a gay la-]Line [dy

Match Stress

Ident (location)

* *!

7 *’s

140

Bruce Hayes

To summarize, candidates with lexical inversion will win when the inversion is in final position and lose when the inversion is in other positions, if the following rankings hold. (46) {*Lapse, Fill Strong, Match Duration, Align(Line, L, Intonational Phrase, L)}  Match Stress  Ident(location) The bottommost ranking of Ident(location) means that nonfinal inversion is prevented, because a shifted candidate will defeat any input with nonfinal inversion. Final inversion is possible because Match Stress is ranked below a group of constraints that collectively prevent the victory of any shifted candidates. As HK note, this explanation relies crucially on the notion of constraint conflict that lies at the heart of OT. 6.7.3

Songs Where Inversion Is Unmetrical

Under the ranking of (46), any line-final inversion comes out as acceptable on a faute de mieux basis. Further data, however, suggest that this pattern is not always found but rather holds true only for certain verse forms. In folk verse, many songs include no lexical inversions at all,22 and it is reasonable to suppose that such songs are composed under a ranking that classifies inversion candidates as ill formed. I posit that this ranking is Match Stress  Max(s). Under this ranking, a lexical inversion, even in final position, is defeated by a suicide candidate in which the stressless syllable is lost. One suicide candidate of this type is given in (47). (47)

[ [ [x [x

x x x x][x

*Fair

x x ][ x x ][x x ][x x ][x x x][x x][x x][x x][x x][x x][x x

El- li- nor

she

was

a

gay

] line ] hemistichs ] dipods ] feet

[leId]

The following tableau shows how the suicide candidate defeats the inversion candidate. (48)

(36) Fair Ellinor she was a gay lady +

Other constraints

Match Stress

(47) Fair Ellinor she was a gay [leI d ] *(36) Fair Ellinor she was a gay lady

Max (s) *

*!

Ident (location)

Faithfulness and Componentiality in Metrics

141

As before, the suicide candidate does not embody a metrical line of verse, since it cannot win in the paraphonology: nothing in English metrical paraphonology permits arbitrary dropping of whole syllables. In contrast, in the verse type discussed in the preceding section, where line-final inversion is allowed, Max(s) must dominate Match Stress. Under this ranking, (36) would be the winner; it would therefore belong to the output set and count as metrical.23 This result may be related to the discussion above in section 6.1. In their original account, HK assumed a naı¨ve OT approach in which the best candidate always emerges as well formed. In verse varieties that avoid lexical inversion, this assumption turns out to be wrong. The more articulated version of OT used here, incorporating Faithfulness and componentiality, is able to make the correct prediction of outright unmetricality. 6.7.4

Linking Inversion to Quatrain Type

It remains to account for one more of HK’s observations: that inversion is unusual in the odd-numbered lines of (4/G)3(4/G)3 quatrains. The argument works as follows. (1) Consider any pair of lines L4 and LG that have the same text but di¤er in that L4 is a 4 line with a final lexical inversion and LG is a G line. For example, L4 could be (36) and LG could be (38). L4 and LG di¤er in their crucial Markedness violations: L4 violates Match Stress (and LG does not); LG violates *Lapse (and L4 does not). Moreover, since L4 and LG have the same text, there will be no Faithfulness violations other than Ident(location), when L4 is taken as a candidate surface form for underlying /LG / or vice versa. In particular, although Max(s) outranks Match Stress in any verse that allows lexical inversion, it cannot a¤ect the outcome here, since both competing candidates obey it. (2) By hypothesis, the quatrain type is (4/G)3(4/G)3. Therefore, underlying G lines in the first and third lines must be able to defeat all alternative settings. For LG , this includes the rival candidate L4 . Since L4 violates only the feeble Ident(location) among the Faithfulness constraints, LG must defeat L4 on the basis of Markedness. Given the Markedness constraints that L4 and LG violate, it follows that Match Stress must dominate *Lapse. (3) Now consider what happens when L4 is the underlying form. Given what has just been said, L4 cannot survive into the output set because LG will defeat it. Specifically, the Faithfulness constraint Ident(location) is too weak to save L4 , and the Markedness constraints Match Stress and *Lapse have just been shown to be ranked in a way that causes LG to defeat L4 . The argument is summarized in tableau (49).

142

Bruce Hayes

(49) /L4 /

Max(s)

Match Stress

+ LG L4

Lapse

Ident (location)

*

*

*!

The result is that if the constraints are ranked in a way that permits G lines to occur in free variation with 4 lines, then lexical inversion candidates cannot make it into the output set and are thus unmetrical. The only exception will be when Match Stress and *Lapse are specially designated to be freely ranked. I assume that the relatively few cases where lexical inversion occurs in the odd lines of a (4/G)3(4/G)3 quatrain fall under this heading. However, even in this circumstance, there will be a complexity penalty for both inversion and G lines, for reasons discussed in the next section. 6.8

Metrical Complexity

In the final section, I will extend the Faithfulness/componentiality proposal so that it can account for metrical complexity. For earlier work on complexity, see Halle and Keyser (1966, 1971a), Kiparsky (1975b, 1977), Youmans (1989), and Golston (1998). I adopt from Youmans’s work the position that metrical complexity should be analyzed in the same terms as metricality; namely, that absolute metricality and unmetricality are only the ends of a continuum. One reason to believe this is that when one examines a variety of poets and traditions, complexity turns out to respond to the same factors that govern metricality. For instance, the coincidence of a prosodic break with the post-fourth position hemistich break in iambic pentameter is a strong normative tendency for many English poets (Oras 1960). For the verse of George Gascoigne, however, or French pentameter (the ‘‘decasyllabe’’), it is obligatory. Such cases are easily multiplied. The shared basis of metricality and complexity has a natural interpretation under OT: the traditions and poets di¤er, not in constraints that guide verse composition but only in their ranking. If this view is correct, then what is needed for analyzing gradient well-formedness is a conception under which ranking is a gradient phenomenon. For this purpose, I adopt the apparatus developed in HM, Hayes (2000), and Boersma and Hayes (2001). The version of Boersma and Hayes is quantitatively explicit and will be employed here. In this model, constraints are assigned ‘‘ranking values’’ on a continuous numerical scale. Grammars are stochastic, in that at any one application of the grammar,

Faithfulness and Componentiality in Metrics

143

the values employed for constraint strictness are determined at random. This is done by selecting a point for each constraint from a normal probability distribution, centered on its ranking value. Grammars of this type can generate a range of outcomes, with di¤erent probabilities a‰liated with each outcome depending on the ranking values of the constraints. However, such grammars can also generate outcomes that are essentially categorical; this occurs when the ranking values of the relevant constraints are extremely far apart.24 A further assumption of the model is that, at least in the crucial class of cases, gradient well-formedness can be treated in terms of probability (Boersma and Hayes 2001; Frisch and Zawaydeh 2001): forms that could be derived only under a somewhat unlikely choice of selection points are assumed to be somewhat ill formed, forms that could be derived only under a highly unlikely choice of selection points are assumed to be almost entirely ill formed, and so on. This model of gradient well-formedness has been tested by Boersma and Hayes against data on English /l/, taken from Hayes (2000), and against data involving the nasal mutation process of Tagalog by Zuraw (2000). In both cases, the model achieves a good match against scalar well-formedness ratings gathered from a panel of native speakers. In the present case, we need to adapt the stochastic apparatus to inventory grammars, which designate whether a representation is or is not in an output set. Adapting the probability-based strategy just described, I posit that the appropriate definition of complexity is as follows. (50) The ‘‘metrical complexity’’ of a line (couplet, etc.) is the probability that the constraints of a stochastic OT grammar will be ranked in a way that excludes it from the output set. On this scale, the complexity of a line or other structure varies from zero (under all possible rankings of the grammar, the line will be allowed in the output set) to one (there is no possible ranking that allows it in the output set). Zero is equivalent to perfect metricality, and one to total unmetricality. Obviously, (50) is a theory-internal definition of complexity. Complexity is also a term that is defined empirically, relating to the gradient intuitions people have about verse structures. My hypothesis is that with appropriate constraints and rankings, the probability-based theoretical values defined by (50) can be mapped onto human intuitive judgments by some monotonic function. Plainly, extensive research is needed to test this claim. However, definition (50) can already be seen to have three advantages. First, it is quantitatively explicit. Second, it is compatible with the criterion set above for an adequate theory of complexity, that is, that metricality and unmetricality should be characterized as extremes on the complexity continuum. Finally, under

144

Bruce Hayes

this approach, the same constraint inventory can be used to characterize both metricality and complexity. 6.8.1

The Complexity of Inversion

A good example of metrical complexity in folk verse is lexical inversion, analyzed in section 6.7. I think most listeners share a sense that lines with inversion are complex; certainly it has attracted the attention of scholars who have examined folk verse (Hendren 1936, 137; Karpeles 1973, 24). Experimental evidence also supports the complexity of inversion: the consultants for Hayes and Kaun (1996), asked to chant texts that in the original song included an inversion, often responded with alternative noninverted settings but seldom did the reverse.25 I will now attempt to characterize inversion quantitatively as complex. Specifically, I will develop a grammar in which lines containing lexical inversion will emerge with a complexity value no lower than .8. This value is arbitrary, being unanchored in experimental data; the point here is to show that the grammatical apparatus is capable of providing such values. I will assume that the quatrain type under examination is always of the type 4343. This quatrain type can be guaranteed by the (essentially) strict rankings given below. (51)

The units along the strictness scale are as defined in Boersma and Hayes (2001). The value 13.49 is chosen as that which results in a one-in-a-million probability against a reversed ranking for the arrows shown; hence, these rankings are essentially obligatory.26 The ranking arguments for (51) are as follows: Couplets are Salient must outrank Fill Strong if 4343 is to defeat 4G4G or any other quatrain type that fills the last strong position of a couplet; Couplets are Salient must outrank *Lapse if 4343 is to defeat 43f 43f ; Fill Strong must outrank Lines are Salient if 4343 is to defeat 3f 33f 3; and *Lapse must outrank Lines are Salient if 4343 is to defeat G3G3. A second, independent group of rankings is the following (their relative placement along the scale with respect to the constraints of (51) would not matter).

Faithfulness and Componentiality in Metrics

145

(52)

The (essentially) categorical rankings Match Duration  Match Stress, Align  Match Stress, and Match Stress  Ident(location) are defended above in sections 6.7.2.3, 6.7.2.4, and 6.7.1.3, respectively. The crucially gradient ranking is Match Stress  Max(s). When the math is done, it emerges that with the di¤erence of 2.38 in ranking values shown, there is an 80% probability that Match Stress will dominate Max(s) at any given evaluation time. As we saw in section 6.7.3, when Match Stress dominates Max(s), inverted lines are defeated in the candidate competition by suicide candidates that remove the final unstressed syllable of the line. Therefore, under the gradient ranking of (52) there is an 80% probability that a candidate with final inversion will not make it into the output set. The metrical complexity of lines with inversion (all else in the line being perfect) is thus .8, which is what we sought originally to describe. 6.8.2

Complexity in General

This analytic strategy can be extended as a treatment of metrical complexity in general. Suppose we want to characterize the complexity of a given metrical structure S in the grammar. We locate first a Markedness constraint M violated by lines (quatrains, etc.) containing S. We also locate a distinct structure S 0 , such that lines containing S 0 instead of S obey M but violate a Faithfulness constraint F and moreover are paraphonologically illegal. Under these circumstances (all else being equal), lines containing S will be unmetrical if M outranks F by a wide margin. metrical but complex if M and F have relatively close ranking values. The degree of complexity will depend on the size and direction of the di¤erence.  fully metrical if F outranks M by a wide margin.  

These rankings determine the likelihood of whether lines containing S can survive the competition with a suicide candidate that contains S 0 instead of S. In the case discussed in section 6.8.1, S was the mismatched lady in (36), S 0 was the corresponding material ([leId]) in (47), M was Match Stress, and F was Max(s).

146

6.9

Bruce Hayes

Conclusion

OT appears to have major potential advantages as an approach to metrics. It achieves explanatory force by letting the ‘‘ingredients’’ of metrical grammars be general, typologically motivated constraints, with idiosyncrasy resulting from genre- or tradition-specific rankings. Moreover, existing case studies, such as the analysis of quatrains in Hayes and MacEachern (1998) or of lexical inversion in Hayes and Kaun (1996), indicate that the data patterns seen in metrics really do reflect constraint conflict, which is the central idea of OT. The goal of this paper has been to make possible a full-fledged OT metrics by resolving the problems that were noted in the introduction. The crucial ingredients have been inventory grammars (section 6.4.1), in which surface forms are derived from identical underlying forms  faithfulness constraints (section 6.4.1)  components (section 6.4.3) that independently evaluate di¤erent aspects of the same representation  stochastic grammars and the probabilistic definition of metricality (section 6.8) 

In the data examined here, these proposals have su‰ced to explain the existence of marked outputs in metrics (section 6.6), to account for free variation (section 6.6), to permit forms to be ruled out without suggesting an alternative (sections 6.5.3, 6.7.3), and to make explicit predictions about complexity (section 6.8.1). It remains a much larger task to assess the validity of these proposals for other areas of metrics and of linguistics. Notes Thanks to Chris Golston, Patricia Keating, Gerhard Ja¨ger, Elliott Moreton, Paul Smolensky, Donca Steriade, Colin Wilson, the volume reviewers, and the participants in a winter 2000 UCLA seminar class for helpful input in the preparation of this chapter. They are absolved, in the usual way, of responsibility for defects. 1. For work on the ‘‘missing remedy’’ problem elsewhere in linguistics, see Prince and Smolensky (1993, 47–51, 175–178), Orgun and Sprouse (1999), Ra¤elsiefen (1999), and To¨rkenczy (2002). 2. Readers seeking help in interpreting the gridded examples may download chanted versions of them (in .wav format) from http://www.linguistics.ucla.edu/people/hayes/Faithfulness InMetrics/. Example (3) is rendered in musical notation in Hayes and MacEachern (1998, 475). 3. In traditional metrics, a ‘‘feminine ending’’ is one in which the penultimate syllable of the line bears stress and the final syllable is unstressed. Most 3f lines do indeed have this stress pattern in their final two syllables.

Faithfulness and Componentiality in Metrics

147

4. There is a slight complication: it is necessary to stipulate that certain constraints are undominated, so the actual set of predicted outputs is smaller than the full factorial typology. 5. For this to work, the inventory grammar must be transparent, in the sense of Kiparsky (1971b). In an opaque grammar, there are often outputs that are legal but cannot be derived from themselves. The introduction of ‘‘crashing’’ derivations below, in which some inputs yield no output, does not alter the situation with regard to testing the grammaticality of a form; it remains the case that an entity will be well formed only if it is mapped onto itself by the grammar. 6. The idea that underlying representations can be completely as rich as surface representations is proposed earlier in Inkelas (1995). 7. To give an expository simplification, phonological representations are depicted as orthography, without stress or phonological phrasing. 8. Empirically, this stipulation has massive support: if paraphonology could alter stress patterns, then words could be mismatched against the meter in all contexts; in actual fact, such mismatches are tightly constrained in a way that requires a metrical rather than a paraphonological analysis. For details, see section 6.7. 9. For other conceptions of componentiality in OT, see Pesetsky (1997, 1998), Ja¨ger and Blutner (1999), Blutner (2000), and Wilson (2001a). 10. As a reviewer points out, Max(s) and Dep(s) seem to play little role in phonology (McCarthy and Prince 1999) and might be better replaced with Max(V) and Dep(V). However, since syllables are the central elements counted in metrics I will assume that Max(s) and Dep(s) are possible metrical, if not phonological, constraints. 11. There are nine because there are only three couplets that fully obey Couplets are Salient (43, G3, and 3f 3), and each may occur in either couplet location. 12. Specifically, since*[’tSeImb l‘] includes segments [ ] and [l], which are absent in its lexical representation /’tSeImb‘/, it violates the paraphonological Faithfulness constraints Dep( ) and Dep(l). It also obeys no constraints that I can imagine that are not also obeyed by [’tSeImb‘]. Therefore, any quatrain including *[’tSeImb l‘] is always paraphonologically defeated by a candidate containing [’tSeImb‘], no matter how the constraints of the paraphonology are ranked. Substituting a real word like featherbed for chambeler does not help, since the paraphonology must construe [’fD‘ "bd], not as the word featherbed but as a candidate surface representation for underlying /’tSeImb l‘/; it plainly cannot win. e

e

e

e

e

13. The derivation does not crash when a third candidate exists that wins in all components. Such cases nevertheless cause the input to be designated as unmetrical. 14. One might speculate, however, that examining the winning candidate(s) of the comparator could give the poet a hint: ‘‘I need a line ending that sounds like . . . [’tSeIm s b‘]; perhaps featherbed?’’ 15. The rather obscure-looking numbers for constraint violations in this column implement a scheme laid out in Prince and Smolensky (1993, section 5.1.2.1), which permits a single constraint to handle di¤erent degrees of violation (there is a multivalued scale of line saliency) in multiple locations (there are four lines in a quatrain). I have also tried the alternative approach, of setting up multiple constraints each defining a cuto¤ point on the scale, and found that it also leads to a working analysis. For details of the numerical scheme, see HM 493 and http://www.linguistics.ucla.edu/people/hayes/quattabl.htm.

148

Bruce Hayes

16. A spreadsheet containing randomly collected examples of lexical inversion from Anglo-American folk song is posted at http://www.linguistics.ucla.edu/people/hayes/ FaithfulnessInMetrics/. Of the 64 inversions whose songs use the grid of (2), 58 (¼ 90.6%) are in line-final position. The exceptions mostly fall under the categories discussed in HK, 291– 294. 17. Three cases of lexical inversion in (4/G)3(4/G)3 I have noticed are Karpeles 1932, 3B; Karpeles 1974, 60G; and Karpeles 1974, 72. 18. A constraint with this name is assumed in Hayes and MacEachern (1998), but there it takes the rather artificial form ‘‘Prefer lines of type F.’’ Constraint (32) is by contrast well supported elsewhere in metrics. 19. This passage is alarmingly reminiscent of section 6.1 above in which it is argued that OT’s principle of always outputting the best candidate can be a problem in metrics. As it turns out, it is possible to eat one’s cake and have it too. Depending on the rankings, derivations can either yield winners faute de mieux or crash with no output. 20. By ‘‘reasonable’’ I mean ‘‘without gratuitous violations of other constraints.’’ 21. And, possibly, other very long syllables as well; the issue is not crucial here. 22. Some examples (all 4343) are Karpeles 1932, 31D, 42B, 97A, 186A; Karpeles 1974, 12A, 18G, 95C, 130A, 141B. 23. A further detail is that in verse types where lexical inversion is to be metrical line finally, we must also rule out suicide candidates that replace mismatched feminine endings with nonfeminine endings by inserting a syllable, as in Fair Ellinor she was a gay lady O. This will follow if Dep(s), which rules out such candidates, likewise dominates Match Stress. This assumption also holds for section 6.7.4 below. 24. I use the word essentially because the relevant type of grammar will generate certain forms with extremely low probability. If this probability is low enough, say, one in a million, then these rare outcomes could not be distinguished empirically from performance errors and thus could not sensibly be counted as wrong predictions. 25. In 74/170 cases, consultants presented with the text of lines that contained line-final inversions in the original responded with a noninverted setting. In 790 cases, consultants were presented the text of a line that had a feminine ending but was not inverted in the original; of these, they replied with an inverted rendering only 26 times. Thus the ‘‘uninversion’’ rate was 43.5 percent, whereas the ‘‘spontaneous inversion’’ rate was only 3.3 percent. 26. A tutorial on how probabilities are derived from ranking value di¤erences may be found in Zuraw (2000).

II

Phonology and Morphology

7

The Phonology of Perceptibility Effects: The P-Map and Its Consequences for Constraint Organization

Donca Steriade

7.1

Introduction

This chapter outlines a proposed revision in the structure of Optimality-Theoretic (OT) phonologies (Prince and Smolensky 1993). The proposal is to let a distinct grammatical component, which I call the P-map (P for ‘perceptibility’), project correspondence constraints and determine their ranking. The P-map is a set of statements about relative perceptibility of di¤erent contrasts, across the di¤erent contexts where they might occur. For instance, the P-map will be the repository of the speaker’s knowledge that the [p]-[b] contrast is better perceived before V’s (e.g., in [apa] versus [aba]) than before C’s (e.g., in [apta] versus [abta]). The point of departure here is the theory of correspondence set forth in McCarthy and Prince 1995, with its distinction between Max/Dep constraints, which identify the elements of two representations that stand in correspondence, and the Ident F constraints, which require a precise featural match between correspondent elements. The general rationale for the P-map proposal is that attested phonological systems display less diversity than predicted by versions of OT in which correspondence and phonotactic constraints interact freely. In particular, the range of pairings between constraint violation and ‘‘repair strategy’’ is more limited than current versions of OT would lead one to expect. An example of this need for a tighter fit between predictions and typology involves the e¤ect that constraints on obstruent voicing have on phonological systems. Consider a common constraint like (1), an underlying string like /tæb/, which violates (1), and the range of possible responses of the grammatical system to this violation, as sketched in (2). (1) A phonotactic constraint: *[þVOICE]/ ]word Voiced obstruents do not occur at the end of the word.1

152

Donca Steriade

(2) Conceivable grammatical responses to the violation of (1) in underlying representation /tæb/2 Change in underlying representation, to satisfy (1) a. Devoicing: /tæb/ ! [tæp] b. Nasalization: /tæb/ ! [tæm] c. Lenition to glide: /tæb/ ! [tæw]

Corresponding constraint ranking *[þvoice]/__]word *[þvoice]/__]word *[þvoice]/__]word [Gconsonantal] *[þvoice]/__]word *[þvoice]/__]word *[þvoice]/__]word (segments) *[þvoice]/__]word features)

d. C-Deletion: /tæb/ ! [tæ] e. V-Insertion: /tæb/ ! [tæb ] f. Segment reversal: /tæb/ ! [bæt] e

g. Feature reversal: /tæb/ ! [dæp]

 Ident [Gvoice]  Ident [Gnasal]  Ident  Max C  Dep V  Linearity  Linearity (for

The table in (2) should be read on the understanding that the correspondence constraint named in a given cell is the lowest ranked among the correspondence constraints that are in potential conflict with the phonotactic. Of the changes in (2), only the devoicing in (2a) is actually attested as a reaction to *[þvoice]/ ]word violations. This is not surprising if one consults one’s linguistic intuition, but it is unexpected in the context of OT, in its present form: if the ranking between the correspondence constraints in (2b–g) is free, one expects at least the range of fixes shown in (2). My claim is not that nasalization, C-deletion, and so forth are unattested processes, but that they are unattested as responses to the voicing problem posed by (1). This means that one does not encounter sound systems in which all the final voiced stops, and only those stops, turn to nasals, delete, or trigger epenthesis or metathesis. (3) indicates what systems of alternations would look like if some of these changes did occur.

tun,

ii. Word final:

tat,

top,

taÐ,

$

$

$

$ tim,

tek

tu,

tat,

ti,

top,

tu,

$

$

$

$

$

tud-a, tat-a, tib-a, top-a, tug-a, tek-a $

ii. Word final: b. Deletion of final voiced obstruents i. Before V:

$

$

(3) Unattested systems (lexically related forms linked by arrows) a. Nasalization of final voiced obstruents i. Morpheme shapes before V: tud-a, tat-a, tib-a, top-a, tag-a, tek-a

tek

The Phonology of Perceptibility Effects

153

tib , top,

$

$

$

$

$

tud , tat,

e

e

ii. Word final:

tug , tek e

$

c. Epenthesis after final voiced obstruents i. Before V: tud-a, tat-a, tib-a, top-a, tug-a, tek-a

The diagnosis for the problem encountered—the fact that devoicing is the only available cure for violations of (1)—starts with the observation that of all the inputoutput pairs displayed in (2), the one judged most similar is the pair [tæb]-[tæp] in (2a). (Evidence for the relevant hierarchy of similarity is reviewed in section 7.4.) The aim, in any departure from the underlying representation (UR), is to change it minimally to achieve compliance with the phonotactics. The modifications in (2b–g) are less minimal, because they result in greater input-output dissimilarity, than devoicing. This is why they are systematically avoided. If this is the correct diagnosis, then what is needed is a mechanism that relates rankings between correspondence constraints to perceived di¤erences in degree of similarity. I refer to knowledge of relative phonological similarity as the P-map. The function of the P-map I discuss here is that of guiding the speaker in search of the minimal input deformation that solves a phonotactic problem. A set of similarity rankings will also be needed in lexical access: the listener has to pick which lexical entry most closely resembles a frequently ambiguous auditory input.3 The grammatical reflex of the P-map is the projection and ranking of correspondence constraints. Thus, if the P-map identifies the pair [p]-[b] as more similar in the context V ] than [b]-[m] for the same context, then the P-map’s e¤ect on the grammar will be to rank higher the faithfulness condition corresponding to the more distinctive contrast [b][m], hence Ident [Gnasal]/V ]  Ident [Gvoice]/V ]. This idea is outlined in (4) using the same example as an illustration. (4) P-map e¤ects on the ranking of correspondence conditions P-map comparisons

More distinctive contrast (e.g., [b]-[m] in V__]

Ranking of correspondence constraints

Higher-ranked constraint (e.g., Ident [Gnasal]/V__]

vs.

Less distinctive contrast [b]-[p] in V__])



Lower-ranked constraint Ident [Gvoice]/V__])

The reader will note the parallel with Prince and Smolensky’s (1993) treatment of phonetic scales: the grammatical reflex of the physical scale is the fixed ranking between constraints referring to points on the scale. Consider now present OT. The concept of minimal modification embodied in this theory is the candidate that optimizes satisfaction of correspondence constraints, as ranked in a given grammar. No independent principle determines the ranking of

154

Donca Steriade

potentially conflicting correspondence conditions. This means, in the context of the example in (2), that either [tæp] or [tæm] will count as minimal modifications of the input /tæb/, depending only on the unconstrained ranking between Ident [Gnasal] and Ident [Gvoice]. I assume below that the phonotactic (1) induces some input modification: the question is which. (5) Devoicing as minimal modification /tæb/

Ident [Gnasal]

Ident [Gvoice]

+ tæp tæm

* *!

(6) Nasalization as minimal modification /tæb/ tæp

Ident [Gvoice]

Ident [Gnasal]

*!

+ tæm

*

The problem with this view is that for at least some phonological properties, and perhaps for all, there exists a crosslinguistically constant notion of minimal modification; that is why a violation of (1) is only resolved by (2a) and not in other ways. This study is a contribution to our understanding of this notion. The di‰culty outlined in (2)—which I call ‘‘Too-Many-Solutions’’—arises with particular clarity in Optimality Theory. This is because OT views phonology as a problem-solving system: the problem is the conflict between phonotactic constraints and lexical forms that violate them, such as the UR /tæb/ of (2). The Too-ManySolutions conundrum arises when the system of constraints and rankings predicts too many resolutions of a given phonotactic problem. But the same di‰culty comes up in any other approach to phonology in which changes in the underlying form are seen as the sound system’s responses to phonotactic violation.4 Thus Kisseberth’s (1970) insight that conspiracies arise when the sound system aims at a specific target structure via multiple means leads to equivalent questions in the context of rulebased phonology: if phonologies aim to eliminate final voiced obstruents, why don’t they employ obstruent nasalization, deletion, metathesis? 7.2

Sources of Evidence for the P-Map

The P-map hypothesis is that speakers possess judgments of relative similarity of the form in (7) and use these to determine a partial ranking of correspondence constraints.

The Phonology of Perceptibility Effects

155

(7) The pair of strings x-y is less similar than the pair w-z (abbreviated as D(x-y) > D(w-z), where D ¼ di¤erence). If some of these judgments are crosslinguistically invariant, then the ranking of certain correspondence conditions will be invariant too. The invariant correspondence ranking will give rise to the typological laws suggested earlier. We observed that if a final voiced obstruent must be avoided, then the repair is to devoice it rather than nasalize it. The P-map conjecture is that the source of this law is a similarity ranking: D(oral C-nasal C) > D(voiced C-voiceless C). This similarity ranking induces a correspondence ranking, Ident [Gvoice]  Ident [Gnasal]. If the similarity ranking is constant, the correspondence ranking is constant too—with qualifications discussed below—and this fact explains the law. Empirical tests of the P-Map hypothesis can proceed by observing invariant preferences for one repair strategy over conceivable others. Some of these observations are discussed below and in earlier work (Fleischhacker 1999, 2005; Jun 1995; Steriade 2001). What must be verified then is that each invariant preference corresponds to a constant judgment of relative similarity. Conversely, one can start from documented judgments of relative similarity and verify the P-map’s prediction that these judgments correlate with preferences for certain input modifications as against others. Both strategies are currently hampered by the scarcity of direct evidence on relative similarity. Few studies address this question and the vast majority compare di¤erent contrasts in the same position (say ba-pa versus with ba-ma5) rather than the same contrasts across positions (e.g., ba versus pa compared to ab versus ap). However, an additional hypothesis about the nature of similarity judgments expands the range of evidence bearing on the P-map. This second hypothesis is that phonological similarity is evaluated on perceptual representations (cf. Flemming 1995/20026). This means that the representations inspected in evaluating similarity are those that encode the perceptual correlates of a contrast—the auditory properties that di¤erentiate its terms. These properties vary with context. The voicing contrast, for instance, is conveyed in part by VOT di¤erences but these are unavailable when the C’s occur in final position. We can reason then that a voicing contrast like b versus p will give rise to di¤erent similarity judgments depending on context: the pair aba versus apa will register as more dissimilar than ab versus ap if one of the voicing correlates is missing word-finally. More generally, the hypothesis that phonological similarity is perceptual in nature turns evidence about cue distribution into a source of information about similarity. As a result, confusion-rate data (Miller and Nicely 1955; Wang and Bilger 1973) is potentially relevant to the P-map hypothesis: if the pair of auditory stimuli x-y give rise to a higher rate of confusion than the pair z-w, that may be because their auditory representations are less well di¤erentiated. This is at least one factor in the

156

Donca Steriade

phonological similarity judgment D(x-y) > D(z-w). Confusability and similarity are distinct matters: subjects rating similarity of auditory stimuli do not report confusion, just similarity. But if similarity is evaluated on perceptual representations, then confusability and similarity are related, because they share at least one source: some auditory representations are di¤erentiated by fewer or less salient properties than others. This makes them more similar and, in the limit, more confusable. In substantiating a claim of relative similarity one can rely on speakers’ direct judgments of similarity, or refer to confusion studies indicating that one contrast is more perceptually robust than another, or reason from the observation that in the position considered, one contrast misses an acoustic correlate and the other does not. Finally, there is the evidence of linguistic tasks that imply a judgment of identity or near-identity: rhyming and alliteration. On the former, see Zwicky 1976, Hanson 2003, as well as Steriade and Zhang 2001. On the latter, see Fleischhacker 2005 and Minkova 2002. All forms of evidence are employed below. Beyond these observations, no model of similarity computation is o¤ered here. (See Frisch, Broe, and Pierrehumbert 2004 for a model of context-free similarity evaluated on articulatory representations.) Rather, the strategy here is to observe that subjects hold certain judgments of relative similarity—whatever their source may be—and that the preference for certain repair strategies correlates with these judgments. This study focuses on only some of the predictions of the P-map hypothesis, as they a¤ect voicing neutralization, cluster resolution, epenthesis, and V-deletion. The aim is to show in each case that preferred methods of resolving phonotactic violations exist; that these preferences are not accounted for by currently available mechanisms; that each preference for a particular solution is explained by the idea that the least perceptually distinctive contrast whose modification removes the violation is always the one sacrificed; and that a solution can be obtained by ranking correspondence constraints via the P-map. 7.3

The P-Map

The P-map is a mental representation of the degree of distinctiveness of contrasts in various positions. It can be viewed as a set of statements about relative similarity between sounds or other phonological properties. The example in (8) involves a hypothetical set of similarity rankings about voicing distinctiveness in di¤erent contexts. (8) D(p-b/V

V) > D(p-b/

V) > D(p-b/V

) > D(p-b/C

C)

(8) states that p and b are less similar intervocalically than prevocalically, even more similar postvocalically, and maximally similar between C’s. The statement can be generalized to a broader class, such as all obstruent pairs that di¤er only in voicing,

The Phonology of Perceptibility Effects

157

if similarity rankings do not di¤er across class members. The P-map fragment in (8) embodies the hypothesis that voicing di¤erences are equally perceptible for all pairs of obstruents but not equally perceptible across positions, the optimal context for this being the intervocalic position (Steriade 1999a). Several properties of P-map statements like (8) are critical here. As already indicated, they must reflect the e¤ect of the syntagmatic context on perceived similarity. Voicing contrasts are not equally well perceived in all positions, and this has fundamental e¤ects on the phonology of voicing. Likewise, the distinctiveness of a contrast between any specific segment class and its absence (here Ø) is also a¤ected by context. For instance, pairs like [fIts]-[fIs]—with a t-Ø di¤erence in the V C context— register as less similar than pairs like [fIst]-[fIs]—with the same t-Ø di¤erence in the C ] context.7 A corresponding P-map statement appears in (9). (9) D(t-Ø/V

C) > D(t-Ø/C

])

Relative distinctiveness judgments can involve suprasegmental or serial position di¤erences. Thus (10) records a hypothetical similarity ranking: C1 C2 reversal under adjacency is less distinctive than the reversal of C’s separated by a V. (10) D(C1 VC2 -C2 VC1 ) > D(C1 C2 -C2 C1 ) According to (10), a pair like apsa-aspa is more similar than the pair apas-asap. The point here is not to examine the truth of or rationale for (10) but to illustrate how the terms of similarity comparisons may involve strings larger than one segment.8 While relative similarity judgments can be documented among many string pairs, one must also consider that no knowledge of relative similarity is available in other cases. Another hypothetical comparison illustrates this: Is the pair task-tass more or less similar than task-task[ ]? Suppose that the P-map fails to record a di¤erence in this case: then the P-map cannot be the source of any ranking between correspondence constraints prohibiting either type of modification (a variant of Max C in task-tass; a variant of Dep V in task-task[ ]). This does not mean that the two constraints cannot be ranked, but if crosslinguistically invariant correspondence rankings emerge only from crosslinguistically invariant similarity rankings, then the Max C versus Dep V ranking will be free to vary on a language-specific basis. The final point to emphasize about the P-map is that distinctiveness and its opposite, similarity, are properties of contrasts (Flemming 1995 [2002]): the statement ‘‘a is more perceptible than b’’ means ‘‘a is more reliably distinguished from a reference term x than b is distinguished from x.’’ It is not the sounds or the articulations a and b that are being compared for perceptibility but the contrasts a versus x and b versus x. This point is fundamental to the success of the P-map as an analytic tool and follows from the assumption that the P-map is so structured as to permit a definition of e

e

158

Donca Steriade

the concept of minimal modification. Consider what information is needed to discover the minimal modification that will render a representation like /tab/ compatible with a constraint like *[þvoice]/ ]word . Since several modifications of this input achieve compatibility with this constraint, the P-map must indicate which of these represents the minimal modification. For example, if we compare [tap] and [tab ] as potential modifications of input /tab/, the necessary comparison involves the inputoutput pairs /tab/-[tap] and /tab/-[tab ]. If there is a guide to the minimal modification, this guide must exist in the form of statements about the relative perceptibility of contrasts like these, or their generalized forms. The contrast is that between the unchanged input string and its modified output correspondent, as it occurs in the context of the modification. There is another sense of contrast and another sense of perceptibility, distinct from the one used here. Suppose that there are invariant properties that underlie sound classes—either invariant acoustic properties or articulatory gestures common to all manifestations of a given class. Then we can talk about the fact that some context allows a better recovery of these invariants. For instance, suppose that the invariant properties of [b] may be better recovered intervocalically than interconsonantally. What that means is that we can better distinguish [b] in V V from all other sounds that could have occurred there. This is the broad sense of contrast. This may be a useful notion but not for the purpose of defining the minimal modification: it does not tell us which pair—/tab/-[tap] or /tab/-[tab ]—is the most similar input-output pair. e

e

e

7.4 7.4.1

A P-Map Account of Voicing Neutralization Differences of Relative Similarity

As a preliminary to the P-map analysis of devoicing—our answer to one aspect of the Too-Many-Solutions problem—I now outline the evidence for a hierarchy of perceived similarity between the pairs in (11). Each pair corresponds to the contrast between an input string with a voiced obstruent in final position and its modified counterpart. (11) a. b. c. d. e. f.

D vs. T/V D vs. N/V D vs. G/V C vs. Ø/V Ø vs. V/C C1 VD2 ] vs.

]word ]word ]word ]word ]word D2 VC1 ]word

D ¼ voiced stop, T ¼ voiceless stop N ¼ nasal G ¼ glide or lateral

The present task is to show that, among these, the voicing contrast D vs. T/V ]word is least distinctive—that is, that its terms are perceived as more similar

The Phonology of Perceptibility Effects

159

than those of the other pairs. In fact, the voicing contrast (11a) stands out because it is the only one that lacks its primary perceptual correlate: the VOT value. This is one reason to expect the (11a) pair to be considered most similar. There are, however, no studies that compare overt similarity judgments for the relevant five pairs in (11) (ab, a-c, a-d, a-e, a-f ). This gap can be filled by combining rhyming studies, studies of foreign-accent perception, and similarity studies for CV pairs, where C quality is systematically varied. Regarding the latter, if the study of CV sequences shows that voicing pairs (e.g., [ba]-[pa]) are more similar than oral-nasal pairs (e.g., [ba]-[ma]), then we can reason that the same result will obtain a fortiori for the VC pairs, since the voicing contrast is, if anything, further attenuated in VC sequences. Voicing versus Manner Studies of imperfect rhyming provide a direct comparison between the final voicing contrast D vs. T/V ]word and the manner contrasts D vs. N/V ]word and D vs. G/V ]word . Zwicky (1976) analyzes 236 half-rhymes in which a consonantal mismatch is ignored, out of a corpus of rock-and-roll rhymed texts with a total of 700 half-rhymes. Relevant here is that next to the 18 instances of a voicing mismatch (pairs like died-light or wise-price) there are only 5 comparable cases where nasality or obstruency di¤erences are discounted (i.e., mid-sin). Hanson (2003) studies slant rhyme in the poetry of Robert Pinsky: here V’s di¤er freely in rhyming pairs, while final C’s stand under a violable requirement of identity. She notes that of the 128 imperfect slant rhymes in her corpus, 96 percent di¤er only in voicing (e.g., woes-loss). Pinsky is not an isolated case. Hanson discusses similar data in Pope and Yeats; for the latter, out of a total of 66 rhyming pairs containing a difference in the final C, 94 percent involve a voicing di¤erence. No rhymes are cited where nasality or laterality is ignored. The rhyming results are supported by the studies of similarity comparing CV sets (Walden and Montgomery 1975) or isolated C sets (van den Broecke 1976). The first of these studies identifies four dimensions of contrast: sibilant versus nonsibilant, sonorant versus obstruent, stop versus nonstop, and, to a much lesser extent, the [p]-[t]-[k] place contrast. Voicing was not a global contrast factor and the overall similarity between voicing cognates (e.g., [pa]-[ba]) emerges as much greater than that between oral/nasal or continuant/noncontinuant pairs. Van den Broecke’s study records Dutch subjects’ impressions of similarity between single isolated C’s uttered silently, and here too di¤erences based on nasality and sonority emerge as dominant. Conversely, pairs judged to possess the highest degree of similarity are pairs of similar sonority, most of them [p]-[b]-type pairs. Greenberg and Jenkins (1964) report similar results in one of their experiments, where subjects were asked to list associates of nonsense stimuli like [klæb]. For all forms that, like [klæb], could yield a lexical item through a change of the final C’s voicing, the most common responses involved voicing changes. Thus for [klæb], the most commonly mentioned forms were [klæp] 7.4.1.2

160

Donca Steriade

(23/46 responses) and hands (a clear associate of clap: 12/46 responses). Significantly, there were other potential associates that also di¤er by exactly one feature from the stimulus: for [klæb] a minority associate is [klæm] (11/46). The one feature di¤erentiating the stimulus [klæb] from the [klæm] response is nasality. Apparently, however, nasality is more significant than a di¤erence in obstruent voicing, since [klæm] was much less frequently elicited than [klæp]. This brief review indicates that voicing is, in any context, perceived as less distinctive than contrasts based on obstruency di¤erences, and moreover that this weak voicing contrast is being suppressed—by final devoicing—in one position where it is least distinctive to begin with. This supports the proposal that devoicing is preferred to nasalization, gliding, or lateralization as a means of complying with the voicing constraint (1), because devoicing induces the smallest input-output dissimilarity. Next I consider evidence on the distinctiveness of voicing as compared with the C-Ø contrast. The aim here is to suggest that dropping the C, to avoid violating (1), is a more salient departure from the input than simply devoicing it. To this end, we could note that the C-Ø contrast involves multiple dimensions of di¤erence (because C and Ø di¤er in voicing, labiality, obstruency), whereas the voicing contrast involves just one of these dimensions. This relates to theories of correspondence in which feature values stand in correspondence and their mismatch is penalized by Max and Dep constraints (Casali 1997; Lombardi 1998). In these theories, discussed below, the loss of a segment necessarily violates a superset of the constraints violated by simple featural modification. However, I claim that the perception of similarity does not reduce to counting features. We thus look for independent support for rankings like D(b-p) < D(b-Ø). Fleischhacker (1999) solicited from English speakers relative similarity judgments between a target word and a modification of it. Some modifications involved changes of final obstruent voicing, while others involved C-loss, metathesis, or V-insertion. The results relevant to us were of the type in (12). 7.4.1.3

Voicing versus the C-Ø Contrast

(12) Voicing versus C-Ø similarity di¤erences: Fleischhacker (1999) Reference term

More similar to

Than to

print

prind

prin, prit

Fleishhacker also tested possible correlations between, on the one hand, greater perceived similarity between target and modified form and, on the other hand, greater preference for one modification than another. She did this by ensuring that, in some of the sets compared, the more similar form was also phonotactically disfavored. Thus prind is judged more similar to print that prin or prit, but it is phonotactically disfavored relative to these, both because it violates (1) and because it

The Phonology of Perceptibility Effects

161

contains a complex coda. Despite the phonotactic improvement, the preference test correlates with the similarity test, suggesting that preference for a given modification is first and foremost a function of its similarity to the source, and only secondarily a matter of phonotactic well-formedness. I consider now the e¤ects of precedence, bearing in mind that the voicing constraint (1) could be satisfied by displacing the C: [tæb] ! [bæt]. Data presented in Fleishhacker 1999 allows us to indirectly compare these e¤ects to those induced by a voicing di¤erence. In the absence of a direct comparison between voicing and serial position contrasts, one can rely on the assumption that similarity is a transitive relation. Thus if C-Ø di¤erences are more distinctive than di¤erences of voicing, and if serial position di¤erences (D(C1 VC2 C2 VC1 )) are more distinctive than C-Ø di¤erences, then by transitivity, serial position is more contrastive than voicing. Fleischhacker’s (1999) study shows that precedence modifications are judged more significant than either coda or onset C-deletion. 7.4.1.4

Voicing vs. Precedence Relative to the V

(13) Fleischhacker (1999) C-Ø contrast versus precedence relative to V Reference term

More similar to

Than to

flip gulf

fip guf

filp gluf

Since C-Ø contrasts are more distinctive than voicing, we infer that contrasts involving position relative to the V are more distinctive than voicing. Feature Transfer Available similarity data does not bear on the possibility of single-feature transfer as an alternative to final devoicing—that is, /tæb/ ! [dæp] as against /tæb/ ! [tæp]. Here however it is safe to reason without data: whatever the dissimilarity degree of /tæb/ versus [tæp] might be, that of /tæb/ versus [dæp] will be greater, since two C’s modify their voicing value in the case of featural metathesis, as against only one in the case of devoicing. This case appears irrelevant to the discussion of standard correspondence theory: single-feature movement of the /tæb/ ! [dæp] sort will violate twice Ident[Gvoice], whereas mere devoicing will violate it only once. In this case the correct preference appears to be built into the existing system. However, the variant of correspondence theory that adopts Max [aF] constraints—instead of or in addition to Ident [GF] constraints (Casali 1997; Lombardi 1998)—will allow the [dæp] candidate to emerge as the minimal modification of the input /tæb/, under rankings like *[þvoice]/ ], Max [þvoice]  Linearity [voice]. 7.4.1.5

162

Donca Steriade

(14) Feature reversal in a Max [aF] theory /tæb/

*[þvoice]/__]

tæp

Max [voice]

Linearity [voice]

*!

*

+ dæp

**

The aim here is not to argue against Max [aF] constraints, but to point out that without recourse to a theory of perceived similarity their mere existence causes candidates to emerge that need further weeding out. However, since standard correspondence theory accidentally avoids this issue, I consider in what follows only full segment reversal as an option that needs to be explicitly excluded. The last case discussed is the possibility of removing violations of (1) through epenthesis. The question involves the relative distinctiveness of the [tæb]-[tæp] contrast as against [tæb]-[tæb ]. I am discussing only one choice of epenthetic V, [ ], because any other V will likely represent an even more salient departure from the original. On this, see below. We can directly compare for distinctiveness the voicing and Ø-[ ] contrasts on the basis of data reported by Magen (1998), who sought to determine which features of the Spanish accent are most noticeable to English speakers. Among the more common aspects of Spanish-accented English are schwa insertion (as in [ spik] for speak and [kloz d] for closed ), deletion of final sibilants (as in stand for stands), and the modified realization of the voicing contrast: medial [z] realized as [s] and initial voiceless stops realized without aspiration and perceived as voiced. Magen asked her English subjects to rate for native quality the original, Spanish-accented utterances as well as edited versions of these originals, in which specific manifestations of the Spanish accent had been targeted and changed, so that the utterances would acquire nativelike quality in those specific respects. In this way, one can observe how English speakers rated the V-Ø di¤erence between the original and edited version of forms with epenthesis (e.g., Spanish-accented [kloz d] versus modified [klozd]) and compare this with the rating di¤erence between the original and edited version of forms with voicing changes (e.g., Spanish-accented [ris n] reason versus edited [riz n]). The relevant results are that voicing changes did not elicit statistically significant rating di¤erences; in contrast, C-deletion and epenthetic schwa significantly altered the ratings and in fact ranked as the most noticeable di¤erences observed. We reach the same conclusion about the relative salience of voicing versus V-Ø in a di¤erent way, on the basis of Fleischhacker’s (1999) study, supplemented with results of earlier work done on English and Swedish by Wingstedt and Schulman (1988). These researchers did not directly compare devoicing and epenthesis, but V-Ø vs. Voicing

e

e

e

e

e

e

e

e

7.4.1.6

The Phonology of Perceptibility Effects

163

rather epenthesis and C-deletion. Wingstedt and Schulman’s subjects rated Cdeletion outputs as preferable to the epenthesis outputs: the relevant triplets were in this case three modifications of a base form like conduct: conduc versus condu[k t] versus condut. e

(15) Preference judgments in Wingstedt and Schulman 1988: Final C-Ø versus Ø-V Reference term

Best modification

Worse

Worst

conduct

conduc

condu[k ]t

condut

e

Fleischhacker’s results allow us to compare a di¤erent version of the same question, because she inserted the V after the last C. She also distinguished similarity and preference ratings and was thus able to show that these ratings correlated. (16) Fleischhacker (1999) similarity judgments: Final C-Ø contrast versus Ø-V Reference term

More similar to

Than to

Than to

heft

hef

heft

het

e

(17) Fleischhacker (1999) preference judgments: Final C-Ø contrast versus Ø-V Reference term

Best modification

Worse

Worst

heft

hef

heft

het

e

In this context, Wingstedt and Schulman’s preference data becomes relevant to the issue of similarity. Recall that Fleischhacker had compared the e¤ects of devoicing with those of C-deletion and had verified that forms related via C-deletion ( printprin) are perceived as more dissimilar to the base relative to forms related via voicing ( print-prind ). Reasoning again from the assumption of transitivity, it follows that devoicing will be less distinctive a departure from input than V-insertion. From this I deduce that devoicing is preferred to epenthesis because it is a less salient modification of the input. As noted above, Magen’s study leads to the same conclusion. This exhausts all the alternatives to devoicing considered in (2). 7.4.2

The Analysis

The discussion of relative similarity has yielded the dissimilarity hierarchy in (18). (18) A hierarchy of distinctiveness in contrasts D(C1 VC2 -C2 VC1 ), D([ ]-[Ø]) > D(C-Ø) > D([Gson]/

]) > D([Gvoice]/

])

e

Relevant to the discussion of final devoicing is only the fact that the word-final voicing contrast emerges as less distinctive than other contrasts considered. Next I

164

Donca Steriade

show that this fact alone resolves the Too-Many-Solutions problem as applied to final devoicing. The solution anticipated earlier is that correspondence constraints are ranked as a function of the relative distinctiveness of the contrasts they refer to. Since it is the Pmap that contains information on distinctiveness, the analysis must establish a link between correspondence constraints and P-map statements. There are two aspects of this process. First, if the P-map encodes a similarity ranking between two contrasts— D(x-y) > D(z-w)—each of the contrasts must map onto a distinct correspondence constraint, Corresp (x-y) and Corresp (z-w). Otherwise relative similarity rankings will exist that fail to be reflected in the structure of the correspondence system. I formulate this requirement below. It amounts to the claim that the dimensions and degrees of similarity di¤erentiated by the system of correspondence are projected from the P-map. (19) P-map projects correspondence constraints Let D(x-y) Ki stand for the perceptual di¤erence between members of sound classes x and y in context Ki .  If D(x-y) Ki > D(w-z)/ Kj , then there exist distinct sets of correspondence conditions, Corresp (x-y/ Ki ) and Corresp (w-z/ Kj ). 

Corresp (x-y/ Ki ) is an abbreviated reference to constraints prohibiting the correspondence of a member of class x in one representation, to a member of class y in a related representation, when both x and y occur in context K. The basic requirement on the relation between the P-map and the correspondence system is that the more distinctive contrasts are protected by higher-ranked correspondence conditions. (20) Ranking correspondence constraints by relative distinctiveness If D(x-y)/ Ki > D(w-z)/ Kj , then any correspondence constraint referring to D(x-y)/ Ki outranks any parallel constraint referring to D(w-z)/ Kj . The term parallel constraints refers to constraints that link the same pair of representations: input-to-output and varieties of output-to-output correspondence (baseto-reduplicant, una‰xed base-to-a‰xed base, and so on). Thus, if D(Gnasal)/ K > D(Gvoice)/ K, then (20) requires that Ident [Gnasal]/ K I-O  Ident [Gvoice] K I-O. However, Ident [Gnasal]/ K I-O may or may not outrank Ident [Gvoice]/ K Base-Derivative, because these two constraints do not link the same pair of representations and thus are not parallel constraints. I now clarify the reference in (20) to ‘‘any correspondence constraint referring to D(x-y)/ Ki .’’ Ident [GF] constraints refer to P-map di¤erences of the form D(x-y)/ K, where x and y are distinct values of the same feature. In contrast, Max and Dep constraints refer to P-map di¤erences of the form D(x-Ø)/ K, where x is

The Phonology of Perceptibility Effects

165

a phonological property (e.g., a segment) and K is a context where x occurs in one representation and its absence occurs in a related representation. Thus Max segment/ K could be paraphrased as: ‘‘There is no D(Ø-C) between I and O, such that I contains C in K and O contains Ø in K 0 and K corresponds to K 0 .’’ Regarding Linearity, it is possible to view the di¤erence D(AxB-ABx), where A and B are context strings and x is a segment, as the sum of the di¤erences D(x-Ø)/ A B and D(x-Ø)/AB . If so, then under the P-map hypothesis, Linearity refers to the same type of P-map di¤erences as Max and Dep, and thus Linearity reduces to conjunctions of context-sensitive Max and Dep constraints. This suggestion has great potential benefits as well as problematic aspects. The observation of phonological similarity judgments implicit in rhyming practices suggests that modifications of linear order induce much greater perceived di¤erences between strings than featural mismatches or segment loss. Rhyming pairs involving metathetic mismatches like task-fax are virtually absent from several large corpora of half-rhymes, in clear contrast to subsequence half-rhymes like desk-mess (Steriade and Zhang 2001; Steriade 2003). This observation is explained if metathesis induces a superset of the D(x-Ø) di¤erences caused by deletion or insertion: cumulative di¤erences between rhyming lines create worse, hence less frequent half-rhymes. A related benefit of this proposal is its ability to explain the extreme rarity of metathesis as a solution to phonotactic violation.9 The possible drawback is that, by the same token, if Linearity reduces to Max and Dep conjunctions and all Max and Dep constraints are context sensitive, it becomes di‰cult to characterize any systems that prefer metathesis over deletion as a solution to phonotactic violation. In what follows I continue to refer to metathetic di¤erences as D(AxB-ABx) and to Linearity as an independent class of constraints. The problem noted is left unresolved here. Returning now to the analysis of final devoicing, we observe that from the principle in (19) and the distinctiveness hierarchy in (18) it follows that each contrast distinguished by (18) gives rise to a distinct set of correspondence conditions. From (20) it follows that correspondence conditions extracted from (18) are ranked by distinctiveness as in (21). (21) Ranking of I-O correspondence constraints by the distinctiveness scale (18) Linearity (C1 VC2 vs. C2 VC1 ), Dep ( vs. Ø)/C ]  Max (C vs. Ø)/V ]  Ident [Gson]/V ]  Ident [Gvoice]/V ] e

Recall now that the correspondence constraints in (21) are the only ones whose violation could in principle satisfy the *[þvoice]/ ] constraint, for inputs that violate it. Our starting point was the observation that each of the five constraints in (21) can, in the present version of correspondence theory, be ranked lower than the others, thus predicting at least five distinct solutions to violations of (1). The constant ranking in (21)—derived from the constant similarity relations in (18)—eliminates

166

Donca Steriade

this di‰culty. Although there are in principle six di¤erent ways of ranking *[þvoice]/ ] relative to members of the correspondence hierarchy in (21), only two sets will yield distinct e¤ects: one set contains the five rankings in which *[þvoice]/ ]  Ident [Gvoice]/V ], all of which amount to final devoicing; the other set contains the ranking in which Ident [Gvoice]/V ]  *[þvoice]/ ]. Only two distinct outcomes are thus predicted: violate the phonotactic or apply final devoicing. This is the result we were aiming to derive. 7.4.3

Consider the Alternatives

The next step is to consider if alternative views on correspondence and constraint interactions, unaided by the P-map, can achieve the desired result of cutting down appropriately on the number of solutions to phonotactic constraint violation. Two of the solutions listed in (2), nasalizing the voiced stop or leniting it to an approximant, will have the e¤ect of changing not one feature but possibly two: in both cases an oral stop ([son, nasal, cont]) becomes a sonorant (either [þson, þnas, cont] or [þson, nas, þcont]). One might hope to discover a formal solution to the Too-Many-Solutions problem, by noting that a one-feature modification (i.e., violating one Ident F constraint) is better than a two-feature modification (violating two Ident F constraints). It is not clear how this idea can be implemented, because rankings of the form Ident F  Ident G, Ident H cannot be ruled out in principle. However, there is independent reason to believe that the cause of our problem does not reside in the count of features being modified. This can be shown by observing that in languages like Turkish (Inkelas and Orgun 1995) where stops—not fricatives—are subject to final devoicing, the active constraint must be (22). (22) The Turkish version of *[þVOICE]/ *[þvoice, son, cont]/ ]

]

This constraint can, in principle, be satisfied by turning voiced stops into fricatives to avoid devoicing. But Turkish reacts to violations of (22) exactly as Russian or Dutch react to violations of (1): by final devoicing. Underlying forms like /kitab/ are devoiced ([kitap]), not lenited (*[kitaB]10 or *[kitav]). The real generalization is that stricture contrasts are not being sacrificed when the phonotactic problem at hand is readily solved by voicing adjustments. As discussed above, changes in stricture induce greater perceived di¤erences than changes of voicing. The same point arises in connection with place phonotactics. Certain heterorganic obstruent clusters—among them tp, dp, tk, dk—are frequently disfavored or impermissible, as in Korean, ancient Greek, or classical Latin. Consider now ill-formed /dk/ inputs (e.g., Latin ad-kelera:re, surface [ak:elera:re]). Such inputs lend themselves to multiple fixes: [ak:elera:re] versus *[askelera:re], *[alkelerare], and so on.

The Phonology of Perceptibility Effects

167

One of these solutions is widely attested—gemination or place assimilation—while the others are simply unheard of. The mapping /adkelerare/ ! [ak:elera:re] entails the loss of two features (voicing and coronality): it is unclear what feature-counting alternative to the P-map will establish this mapping as preferable to the /adkelerare/ ! *[alkelera:re] mapping, in which obstruency is lost and laterality is added. The generalization here is that when the same phonotactic problem can be addressed by adjusting either place or stricture, the solution is to change place. Consider now the viability of C-deletion as a solution to final voicing violations. Recall that underlying forms like /tæb/ could—but never do—satisfy (1) by dropping the final voiced stop altogether. Here too one might imagine that a di¤erent modification of the theory of correspondence, one that substitutes Max [aF] for Ident F constraints (Casali 1997; Lombardi 1998), will explain the preference for devoicing. Thus the output of devoicing, [tæp], violates only Max [þvoice], while the output of C-deletion, [tæ], violates Max [þvoice], plus Max [labial], Max [cont], Max [nasal], and so on. On this view, the C-deleting candidate loses under any ranking of the Max constraints. But this cannot be the answer either. Consider the constraint against stop þ noncoronal stop sequences (tp, tk, dp, dk) active in ancient Greek. Most such inputs arise at the boundary between the verb root and the perfect ending -ka, and the constraint is satisfied in this case through [t/d] deletion —for example, ke-komid-ka ! [kekomika] ‘I have eaten’. Thus the Greek solution to the tp, tk problem is not to place-assimilate, as in Latin or Korean, but rather to drop the first stop altogether—that is, to violate Max [a voice], Max [coronal], Max [cont], and so on. If we look at this problem in terms of the number of features being sacrificed from the input, we cannot understand why the [d] of komidhad to drop, when it could well have been turned into [l], [r], [s], yielding well-formed *[kekomilka], *[kekomirka], or *[kekomiska]. Each of these alternatives contains fewer Max F violations than the solution actually adopted, which was to eliminate the [d] altogether. (23) Failed attempt at C-deletion in a system with Max F and no Ident F /ke-komid-ka/

Max [cont]

Max [coronal]

kekomika

*

*!

+ kekomilka

*

A theory that relies exclusively on Max [aF] cannot explain any pattern in which a segment is deleted in contexts where the phonotactic violation can be met by modifying a subset of its features. If we adopt, along with the Max [aF] constraints, Dep [aF] constraints, then the ranking Dep [þstrident], Dep [þnasal], Dep

168

Donca Steriade

[þcontinuant]  Max [son], [cont] can induce [t/d] deletion. This move, however, brings back the original problem: in a system where every feature value possesses its Dep constraint, devoicing violates Dep [voice]. Then what rules out Dep [voice]  Max [son], [cont]? What emerges from this discussion is that some hierarchy of features must be assumed in any approach; one must recognize that modifications of voicing, especially final voicing, matter less than modifications of obstruency. This is the first step on the way to the P-map. For the P-map analysis, the Greek [t/d] deletion process raises not the formal problem faced by Max F analyses but an empirical question: Is the contrast between unreleased t/d and Ø in prestop position judged less distinctive than that between t and s or t and n or t and l in the same position? If yes, then the we predict that, in such a context, straight deletion is more likely than fricativization or lenition to a sonorant. We lack similarity data bearing on this, but studies of assimilation and cluster simplification (Wilson 2001a; Steriade 1999a) suggest that a perceptibilitybased solution will be fruitful for this case too. 7.5 7.5.1

P-Map Effects in Cluster Simplification Size-of-Cluster Constraints

Many languages constrain agglomerations of C’s when they exceed some specified size. If the constraint responsible for size-of-cluster phenomena prohibits strings of the form C i C j / K, where K specifies a context, segmental or prosodic, then a representation violating it can achieve compliance in at least three ways: by deleting C i , by deleting C j , by modifying either of them or by adding a V, the insertion of which will yield further choices regarding site and V quality. In this section I briefly suggest that this wealth of apparent choices in dealing with size-of-cluster constraints fails to reflect phonological reality: the actual solution comes much closer to being predetermined by the composition of the string containing the violation. While the choice between V-insertion and C-deletion might remain free in resolving a size-of-cluster violation, other decisions (which C to delete; which C to modify and how; where to insert a V and which V to insert) are partly or fully predictable. They are predictable largely in terms of the relative similarity between the input and the modified output: it is the most similar input-output pair that is predominantly selected. The issue of predictability in intervocalic CC cluster simplification has been independently identified by Wilson (2001a), whose formal proposal di¤ers from mine but whose discussion raises points related to those made here. The partial predictability of epenthesis site in initial clusters is analyzed in a framework akin to the P-map by Fleishhacker (1999, 2005). Here I extend Wilson’s observations by considering briefly the choice of C’s to delete in sequences more complex than VCCV.

The Phonology of Perceptibility Effects

7.5.2

169

Similarity with Ø in Cluster Simplification

Wilson (2001a) notes that when C-deletion targets a sequence VC i C j V, the lesser perceptibility of C i leads to its loss. I consider here the choice between deletion of C i and C j in VC i C j Ck V. To simplify matters, I assume, with Wilson, that the prevocalic Ck is undeletable here. My basic empirical point is that the target of deletion is predictable from considerations of confusability, not from its prosodic position or its adjacency to the V. The P-map’s predictions for cluster simplification in VC i C j Ck V are derived from the degree of distinctiveness of two contrasts: C i vs. Ø/V C, and C j vs. Ø/C C. To be judged su‰ciently distinct from Ø in a given context, the sound must in fact be su‰ciently distinct from both of the elements adjacent to it. To see this, suppose that C j in VC i C j Ck V is confusable with C i : the percept resulting from this confusion is VC i C i Ck V or VC i :Ck V. The e¤ect of shortening-in-clusters (Haggard 1973; Klatt 1973) renders VC i :Ck V confusable, in turn, with the simplified VC i Ck V. Thus C j ’s similarity to C i leads, under C i -C j adjacency, to C j ’s similarity to Ø. The same holds if C j is similar to Ck . Likewise, postvocalic C i in VC i C j Ck V is similar to Ø, if it is too similar with either the preceding V (a confusion leading to the V:C i Ck V percept) or to the following C j . Finally, consider a sequence C j Ck V in utterance-initial position: the initial C j is confusable with Ø if it is confusable with either the absence of sound that precedes it or the Ck that follows. Mutatis mutandis, the same holds for confusion with Ø of Ck in an utterance-final VC j Ck . Similarity with Ø means then similarity with either of the adjacent elements, whether silence or sounds. From this we predict that position relative to the syllable boundary or a V will not guarantee that C i is less confusable with Ø than C j . The simplest example illustrating this is the loss of postvocalic liquids in systems where other postvocalic C’s are preserved. In such cases what identifies the element deleted is not proximity to the V or syllable position but similarity to the neighboring V. Cho (1999) presents an example of this type in his analysis of postvocalic [1] deletion in Korean VlCCV. A more complex illustration of the same point is the di¤erence between the confusability with Ø of interconsonantal stridents and stops. Consider first the case in which the sequence VC i C j Ck V contains three stops. As a stop, C i is su‰ciently distinguishable from the immediately preceding V. Moreover, since this V carries C i ’s transitional cues to place and voicing, it provides information distinguishing C i from other C’s, including C j , which might have occurred in the same position. Therefore C i is not confusable with either the V or the following C j , hence it is not confusable with Ø. The medial C j , on the other hand, is confusable with Ø. Because no V is adjacent to C j , the string VC i C j Ck V contains less information allowing the listener to di¤erentiate C j from any other stop that might have occurred in the VC i Ck V position, including from C i or Ck . If C j is confusable with either of the adjacent stops, then the string VC i C j Ck V is confusable with VC i :Ck V or VC i Ck :V and hence

170

Donca Steriade

with the shortened variant VC i Ck V. And therefore C j is more confusable with Ø, than C i . In this way we reach the conclusion that, if cluster simplification targets the C that is most confusable with Ø, then it will operate in this case at the expense of the medial C j . However, this holds only for cases in which all but position relative to the V is equal between the three C’s. Suppose that C j in our VC i C j Ck V string is a strident. Then the distinctiveness di¤erence between D(C i vs. Ø)/V C, and D(C j vs. Ø)/C C might be obliterated or reversed, because the inherent noisiness of the strident C j identifies it as distinct from any adjacent nonstridents, even in the absence of vocalic transitions. If so, the relevant correspondence constraints (Max C/V C and Max strident/C C) either remain unranked or Max strident/C C might in fact rank higher. 7.5.3

Test Cases

Several predictions follow from this. First, a language may delete interconsonantal stops but not interconsonantal stridents. This pattern occurs in Dihovo Macedonian (Groen 1977). To analyze it we need a size-of-cluster constraint, interpreted as the requirement that each C be adjacent to a V: (24) C//V Every C is adjacent to a V. The Dihovo pattern of cluster simplification corresponds to (25): (25) Stops, not stridents, are deleted between stops in VCCCV. Max [cont]/V C, Max strident/C C  C//V  Max [cont]/C

C

The clear e¤ect of the P-map in this case is the ranking Max stop/V C, Max strident/C C  Max stop/C C. The position of the phonotactic C//V relative to the correspondence constraints is left undetermined by the P-map, which allows us to predict variation in the patterns of cluster simplification. Thus the modified hierarchy in (26), where C//V has climbed higher, requires that some cluster simplification take place even in V-stop-strident-stop-V clusters. (26) All VCCCV clusters are reduced to VCCV. C//V  Max [cont]/V C, Max strident/C

C  Max [cont]/C

C

Further elaboration of (26) yields two types of simplification for VC i C j Ck V sequences where C i is a stop and C j a sibilant: either VC i C j Ck V ! VC i Ck V or VC i C j Ck V ! VC j Ck V. What is, however, invariant is that if the middle C j is a stop surrounded by obstruents, it will always be deleted. Colloquial Latin illustrates the more revealing pattern: interobstruent stops are lost, whereas interobstruent [s] is preserved at the expense of the stop preceding it.

The Phonology of Perceptibility Effects

171

(27) Two types of cluster simplification in Latin (Niedermann 1953) VC i stop Ck (V) ! VC i Ck (V)

VC i s Ck (V) ! V s Ck (V)

pa:sktus ! pa:stus nokts ! noks temptare ! tentare lampterna ! lanterna k w inktus ! k w intus

sekstus ! sestus obstendo ! ostendo apsporto ! asporto sekskenti: ! seskenti pinstus ! pi:stus

This cluster-reduction pattern suggests that the strident-Ø contrast is more distinctive, even in the absence of contextual cues, than the postvocalic stop-Ø contrast. (28) Simplified cluster-reduction hierarchy for Latin11 C//V  Max strident/C C  Max [cont]/V C  Max [cont]/C

C

(29) Cluster reduction in obstendo /obstendo/ optendo

Max strident/C__C

Max [cont]/V__C

*!

+ ostendo

*

(30) Cluster reduction in kwinktus /kw inktus/ kw iktus + kw intus

Max [cont]/V__C

Max [cont]/C__C

*! *

The Latin asymmetry between postvocalic stops and sibilants as targets of cluster simplification is encountered in several languages, among them Finnish, Catalan (Wheeler 1979), and colloquial Polish (Madejowa 1992). The alternative pattern of deletion, where every interconsonantal obstruent deletes, whether it is a stop or a sibilant, is perhaps also attested in Greek, Sanskrit (Steriade 1982), and Korean (Kim-Renaud 1974), but alternative interpretations are available for these cases. In particular the analysis of cases like Korean kaps-to ‘price-and’ ! [kapto] must take into account the fact that all fricatives are prevocalic in Korean: if [p] had deleted, the actual outcome would have to be *[katto], not *[kasto]. The tableau in (31) indicates that the Latin ranking of correspondence constraints need not be changed to derive this case.

172

Donca Steriade

(31) Cluster reduction in kapsto /kapsto/ kasto

*s/__C

Max strident/C__C

*!

Max [cont]/V__C *

katto

*

+ kapto

*

*!

The Greek and Sanskrit instances of deleted interobstruent [s] are sparsely attested and involve exclusively su‰xal [s]. It is possible then that the Latin reduction pattern represents the general case. My general claim, however, is more modest. A P-map account, as sketched here, predicts only this: insofar as a C-cluster contains one and only one C whose similarity to Ø is greater than that of the other cluster members, cluster reduction will target this one C. Similarity to Ø means similarity to an adjacent element. We have seen that an interobstruent stop—or a stop flanked by a nasal and an obstruent—can be identified as the most confusable with Ø among all components of its C-cluster. This corresponds to the observation that stops in such contexts are the systematic, invariant targets in cluster simplification. It may also turn out that the inherent salience of stridents renders the D(strident-Ø)/C C contrast more distinctive than D(stop-Ø)/V C. If so, a stronger prediction is made: the postvocalic stop will always be deleted, unless morphological factors intervene, in V-stop-s-C sequences. 7.6

Insertion and the Ranking of DEP Constraints

The P-map account of the choice of epenthetic segments derives from the hypothesis of a context-dependent hierarchy of similarity between individual segments and Ø. If a phonotactic constraint requires insertion of a segment in some context K, then the segment most confusable with Ø in K is predicted to be the choice of insertion. I outline now how this prediction follows from proposals made thus far. The class of correspondence constraints violated by insertion take the form in (32). (32) DEP (I-O) schema There is no D(Ø-x) between I and O, such that I contains Ø in K and O contains x in K 0 and K corresponds to K 0 . Like all correspondence constraints, the Dep constraints are projected from the Pmap. This means that if D(Ø-x)/ Ki > D(Ø-y)/ Kj , then corresponding to this similarity ranking there exists a ranking of correspondence constraints so that the more confusable contrast with Ø corresponds to the lower-ranked Dep constraint (cf. (20)). It follows that the outcome of phonotactically motivated insertion is to a large extent

The Phonology of Perceptibility Effects

173

predetermined. This prediction is mitigated by the e¤ect of conflicting phonotactics and thus the e¤ect of markedness on epenthesis has to be considered as well. Next I outline only one aspect of the evidence bearing on this point: the selection of epenthetic segment quality. 7.6.1

Epenthetic Glottals

The typology of epenthetic C’s has been usefully outlined by Lombardi (2002), who identifies a general pattern, insertion of [], and minor deviations from it, due either to structure preservation (in the form of constraints forbidding []), morphological constraints, or the dispreference for rhyme []. The central question is what accounts for the preference for inserted []. Lombardi assumes that the relevant factor is markedness: [] is the least marked among C’s. But what fact other than its propensity to get inserted reflects []’s extreme unmarkedness? This is a harder question: the standard evidence for markedness, the implicational universals, suggest otherwise: []’s presence in an inventory is not asymmetrically implied by the presence of all other C’s, or indeed by the presence of all other members of its stricture class. Under the P-map hypothesis, [] has, with [h], a uniquely favorable property for an epenthetic C: it does not possess an oral constriction and thus it will fail to induce coarticulatory changes on neighboring segments, unlike the orally articulated C’s. If we compare input-output pairs of the form V(input)-CV(output), the most similar ones will be V-V or V-hV or V-GV, where G is homorganic to V. Both epenthesis of [h] and epenthesis of homorganic glides represent in fact the only widely attested epenthesis patterns, along with the more common case of [] insertion. Thus, if the lack of coarticulatory V modification translates into the similarity hierarchy in (33), then the P-map hypothesis predicts the preference for [] as epenthetic segment, regardless of how it rates in markedness. (33) D(Ø-t)//V; D(Ø-k)//V; D(Ø-p)//V > D(Ø-)//V The view that markedness determines the choice of epenthetic segments runs into independent di‰culties. Several of the languages Lombardi cites, where [] occurs as an exclusively epenthetic C, must in fact be assumed to rate the markedness of [] as higher than that of all their other C’s. Thus to explain the fact that German allows only epenthetic [] we have two options. We can assume that a constraint *[V  *[]  Max C while at the same time assuming Max C  *[p], *[k], and *[t]. But this contradicts the universal markedness ranking Lombardi assumes: *[p], *[k]  *[t]  *[]. The alternative is to rely on the P-map-based ranking: Max/Dep [p],[t],[k]/ V  Max/Dep []/ V. The German choice of epenthetic [] follows when this correspondence ranking is embedded in the complete analysis: Max [V, Max/Dep [p],[t],[k]/ V  *C  *[V  Max/Dep []/ V. On this view, it is correspondence, not markedness, that di¤erentiates C types. The same conclusion follows

174

Donca Steriade

from Lombardi’s analysis of Asheninca, one of the rare languages where something other than a laryngeal or a glide is inserted in hiatus. Lombardi argues that [t] is inserted in Asheninca because *[] is undominated. If we grant this, it follows that *[]  *[t], since *[t], but not *[], is outranked by Max C. This too is incompatible with the claim of unmarked status for []. I conclude that there is either no constant context-free, all-purpose preference for glottal as against other stops, or, if there is a preference, it is the opposite from the one needed to predict the proper choice of epenthetic C. The choice is predicted by the P-map. 7.6.2

Epenthetic Schwa

I document next the status of [ ] as the inserted V of choice. This too follows from a hierarchy of similarity with Ø. What defines for our purposes the class of schwa vowels is not their midcentral quality (Romanian and English [ ], for instance, are not epenthetic, while Romanian [ ] is) but rather the fact that the schwalike V is significantly shorter and more variable in quality than all other V’s in an inventory. This characterization allows for some diversity in the actual quality of a language’s neutral V, while permitting us to make specific predictions about what di¤erentiates it from other V’s. A systematic di¤erence of duration between schwa and other V’s of Dutch is documented in Koopmans-van Beinum 1994, and known informally to obtain for English and French schwa. Further, Dutch schwa is also more variable in its F2 values than other Dutch V’s (Koopmans-van Beinum 1994; van Bergem 1995). Assuming then that the defining properties of schwa are shortness and variability, the preference for schwa as an epenthetic element follows form the fact that it is, in both duration and relative absence of invariant articulatory properties, the closest thing in a V system to no segment at all (i.e., to Ø). Note that this is not the same as saying that schwa has no properties. First, it is a vowel. When it does occur, speakers count an extra syllable. This is invariant. Further, schwa in Dutch is less durationally variable than other V’s (Koopmans-van Beinum 1994): it is least subject to contextual or context-free lengthening. In many languages where schwa is unstressable, as in Dutch, Indonesian, and French, this can be attributed to the fact that schwa cannot be lengthened. In that respect then it does have a second invariant property: short duration. Thus we cannot explain why schwa is preferentially inserted by assuming that it possesses no properties, or a subset of the invariant properties of other vowels. Schwa epenthesis is preferred because the P-map identifies it as the closest thing to no epenthesis at all. The preference for schwa insertion may manifest itself in a language independently of the composition of the V inventory. However, when structure preservation does not constrain its occurrence—that is, in languages possessing contrastive or nonalternating schwas—this preference for inserting schwa is absolute. The statement in (34) holds of all relevant cases I have encountered, some of which are listed below. e

v

The Phonology of Perceptibility Effects

175

(34) If a language contrasts schwa and zero in some context, or if it contains nonalternating forms with schwa, and if it resolves clusters through epenthesis, then the choice of productive epenthetic vowel is limited to schwa. (Indonesian (Adisamito 1993), Romanian (Avram 1990 and below), German (Giegerich 1987), Damascene Arabic (Bohas 1986), French (Dell 1978 and below), Meitei (Chelliah 1997), Miya (Schuh 1996), Welsh, English, Dutch (Booij 1995; Kuijpers, van Donselaar, and Cutler 1997), Berber (Kossman 1995; MacBride 1990)) The comments about the markedness of [] apply here too: schwa insertion does not stem from a context-free preference for this segment. It most clearly does not in languages where schwa is permissible only where a V is otherwise needed. In Berber, for instance, schwa—but no other V—must be prevented from occurring in open syllables (MacBride 1999); in Miya, it cannot occur after a sonorant (Schuh 1996). Phenomena of this sort require *[ ] constraints, whose high ranking is not paralleled by other *V conditions: this precludes a claim of unmarked status for schwa. What explains the V-epenthesis generalization is the existence of a hierarchy of Dep V constraints containing, at its bottom, Dep [ ]. Here too I speculate that the source of this Dep V hierarchy are speakers’ judgments of relative similarity between individual V’s and Ø. e

e

7.7

All-Purpose Segments

I turn next to a di¤erent respect in which the P-map proposal tightens the theory of correspondence. The suggestion here will be that the segments most likely to be inserted are also most likely to be deleted. The behavior of [ ] will illustrate this, but reports about specific segments being both preferentially inserted and deleted go beyond this case.12 This phenomenon is not predicted by the classic theory of correspondence, but it follows from the same principle (21) that solved the Too-ManySolutions puzzle involving final devoicing. (21) predicts that if D(x-Ø)/ K > D(y-Ø)/ K then not only will Dep (x)/ K  Dep (y)/ K but also Max (x)/ K  Max (y)/ K. Then y has priority for both insertion and deletion over x. More concretely, this means that if some y-Ø contrast is identified as more confusable than other segment-Ø contrasts, the insertion and deletion of y will be the preferred response to all phonotactics for which insertion or deletion of segments from y’s class represent a potential solution. The general class of situations described is this: a language avoids hiatus, hence it must delete a V when adjacent to others or insert a C between them. The same language avoids CCC clusters, hence it must insert some V in such clusters, or else delete a C. Because it appears impossible to predict the preference between C-insertion e

176

Donca Steriade

and V-deletion, or between C-deletion or V-insertion, these choices are settled on a language-specific basis. The language we are interested in eliminates clusters by Vinsertion and hiatus by V-deletion. Under these conditions, the P-map hypothesis predicts that the V deleted in hiatus is the same as the V inserted as a clusterresolution strategy. That is because the criterion that selects a V for deletion is the same as the one selecting it for insertion. This criterion is the greater similarity of the contrast between that V and Ø relative to all other V’s. I summarize this below, using [ ] as the V judged to be most confusable with Ø. e

(35) If, for any choice of V0 , D(V-Ø) > D( -Ø) Then Max V, Dep V  Max ( ), Dep ( ) e e

e

e

Note that, aside from the P-map, nothing guarantees that the Max and Dep constraints corresponding to di¤erent V’s will be ranked as pairs. Nothing excludes rankings like (36), which predict that schwa is deleted but that [a] is inserted. (36) Schwa deleted and [a] inserted in system lacking P-map, where MAX/DEP constraints exist for individual segments Max (a), Dep ( )  Phono-constraints  Max ( ), Dep (a) e

e

Systems exist in which hiatus is resolved at the expense of certain V’s only (Pulleyblank 1988; Casali 1997; below). As Casali argues, these systems reflect a hierarchy of distinct Max constraints, independent of the markedness scale: deletion targets the V associated with the lowest ranked Max V. This dispenses with alternatives to the P-map analysis that assume that targets of deletion/insertion are determined by markedness conditions alone, interacting with a monolithic Max V, Dep V. One can document systems where both V-insertion takes place and specific V’s are deleted, either to avoid hiatus or to shorten the word. In all such cases, it is the prediction of the P-map analysis, (35), that is upheld. The pattern emerges more clearly if we restrict our attention to productive, lexically unrestricted insertion and deletion. Thus in French, it is schwa that deletes in hiatus, regardless of its location relative to the other V. (37) French schwa deletion a. Optionally deleted in VC CV contexts; no other V deletes la pelouse [lapluz] ‘the lawn’ cf. phrase-initial pelouse [p luz] pas de role [pad‚ol] ‘no role’ cf. phrase-initial de role [d ‚ol] b. Obligatorily deleted in hiatus; no other V deletes t’entendre ‘to hear you’ cf. te remercier [t ‚m‚sje] ‘to thank you’ vivre ailleurs [viv‚ aj˛‚] ‘live elsewhere’ cf. vivre la` [viv‚ la] ‘live there’ e e

e

e

Compare ni entendre [ni þ tþ d‚ ] ‘neither to hear’ and vivra ailleurs [viv‚a aj˛‚] ‘will live elsewhere’, both of which surface with hiatus, in the absence of a deletable vowel. Schwa is also inserted, optionally, to avoid clusters of obstruents. e

The Phonology of Perceptibility Effects

177

Romanian [ ] gives rise to identical patterns: (38) Romanian schwa a. [i] optionally deleted next to a V; no other V deletes vine ‘ndat ‘comes immediately’ cf. indat ‘immediately’ vine ‘nainte ‘comes before’ cf. inainte ‘before’ b. Nondeleting V’s vine odat ‘comes once’; vine la ‘comes that one-masc.’ vine aja ‘comes that one-fem.’ c. [i] optionally inserted in obstruent clusters CCC(C); no other V is inserted opt-spre-zece [optsprezetSe] @ [opt sprezetSe] ‘18’ (‘eight-to-ten’) Comparable patterns appear in Dutch (Booij 1995; Kuijpers, van Donselaar, and Cutler 1997), Meithei (Chelliah 1997), and English (Hooper 1978). 7.8

Conclusion

I note in closing potential points of dispute relating to the P-map. At the most basic level one can dispute the premise this account shares with most modern phonology, namely that phonology is a problem-solving system, or—as Goldsmith (1993) puts it—‘‘an intelligent system.’’ If the phonotactic in (1) is not viewed as a problem to be solved, or as a standard of well-formedness independent of the lexicon’s contents, but rather as a generalization over the words that happen to be attested in one’s language, then no Too-Many-Solutions problem arises: learners, on this view, do not seek to find a solution to (1) but to learn whatever patterns happen to be instantiated by their lexicon. Similarly, one may question whether the Too-Many-Solutions problem arises in the initiation of sound change. The view presented here is that innovators may aim to improve a sound system and that they do so in the safe regions of confusability identified by the P-map. We assume, for instance, that speakers who initiate final devoicing have a choice of methods to satisfy (1)—or a choice of spontaneously occurring speech variants to promote (Lindblom et al. 1995)—and choose final devoicing because it involves the smallest departure from established speech norms. But it may be possible to look at the initiation of sound change in di¤erent terms if most naturally occurring variants to an established lexical form represent its common misperceptions. In that case, innovators have the more passive role of simply favoring the more commonly noted deviations from the norm, without reflecting on their phonotactic virtues or on their similarity to canonical forms. This possibility has been discounted here on the strength of evidence that speakers know not only what the more common deviations from the norm are but also which deviations are more similar to the norm. We have seen that knowledge of similarity is displayed in rhyming

178

Donca Steriade

practices and experiments seeking overt similarity judgments. It remains to be established, however, that the available knowledge of similarity is exploited by speakers in constructing grammars and in favoring one change over others. (See also Steriade 2001.) A di¤erent class of possible objections to the P-map involves the fact that there is, at least at first sight, a considerably greater variety of alternations than a theory of perceived sound similarity may predict. If the case for predictability of C-deletion or V-insertion appears overstated, a way of testing the P-map proposals is to focus on fully productive, not yet lexically entrenched processes. For the moment, it seems necessary only to acknowledge the existence of parochial constraints governing alternations, in addition to phonotactics and P-map generated correspondence constraints. Finally, I have focused here on aspects of perceived similarity that correspond to broad crosslinguistic generalizations, and for this reason it may appear that a claim of universality is made regarding the contents of the P-map. This is not the intention. If the perception of similarity is governed, in part, by ‘‘the contents of the universe of discourse’’ (Tversky, cited in Frisch, Broe, and Pierrehumbert 1997), then the same pairs of sounds will rate di¤erently for similarity, when embedded in di¤erent systems. The existence of such e¤ects is not denied; the development of a firstapproximation version of the P-map will hopefully make it possible to identify them. Notes I am grateful to the editors, Kristin Hanson and Sharon Inkelas, and to Bert Vaux for detailed and helpful comments on a first draft. Questions from audiences at Berkeley, Stanford, UCLA, and the 1st North American Phonology Conferences at Concordia University, Montre´al, have helped clarify some of the issues discussed here. Most of the writing was done in fall 1999, while I was in residence at the Miller Institute for Basic Science, at UC Berkeley. Many thanks to Andrew Garrett, Larry Hyman, and Sharon Inkelas for engineering my stay at the Miller. This study is dedicated to Paul Kiparsky, whose writings converted me in the late ’70s from Indo-European studies to phonology, and whose teaching and example since then have inspired my work. 1. What exactly causes final obstruents to devoice is a matter of debate; see Lombardi 1999 and Steriade 1999 for two recent proposals, among many. The precise nature of the phonotactic does not a¤ect the problem outlined in this study. 2. The correspondence constraints cited are those proposed by McCarthy and Prince 1995. The argument carries over to other views on correspondence. See section 7.5.3. 3. See Connine, Titone, Deelman, and Blasko 1997. 4. See Ito 1986, Paradis 1988, Goldsmith 1992, Calabrese 1995. 5. See Mohr and Wang 1968, Walden and Montgomery 1974, van den Broecke 1976.

The Phonology of Perceptibility Effects

179

6. Flemming’s point is that it is faithfulness that’s evaluated on perceptual representations: ‘‘The output must sound like the input.’’ The same holds for any similarity relation. 7. See Wingstedt and Schulman 1988 and Fleischhacker 1999 for relevant experimental evidence. A further indication of the same point is that partial subsequence rhymes of the form VC1 C2 -VC1 Ø (e.g., best-less) are frequent in English (Zwicky 1976) and elsewhere (Steriade 2003) while subsequence rhymes of the form VC1 C2 -VØC2 (e.g., bets-less) are extremely rare. 8. See, however, section 7.4.2. 9. See Blevins and Garrett 2004 for relevant discussion and references. 10. Readers who object that [kita”] is a non-structure-preserving change—as Turkish lacks bilabial fricatives—will recall that under the ranking Ident [Gvoice]  *B, Ident [Gcont] structure preservation should be irrelevant. Indeed, voicing adjustments can be non-structurepreserving in languages like German and Catalan, where final devoicing is incomplete and does not obliterate the contrast. Thus a formal ranking generating [kita”] exists under the standard account. The question is why it is unattested. 11. I am not claiming that C//V is undominated in Latin: there exist clusters like mbr, ltr, str, and so on. But the focus here is on the fate of medial C-obstruent-obstruent sequences in which cluster simplification did occur regularly in the spoken language. To obtain the more accurate account, one must assume Max constraints that outrank C//V. For these Max constraints, the P-map’s claim is that they involve clusters whose individual members are better distinguishable from Ø than the C’s that do in fact delete. 12. See Pulleyblank 1988 for the observation that the same vowel may be both the prevalent target of deletion and the preferred inserted element in selected languages. A particularly interesting case of epenthetic/deletable C is discussed in McCarthy 1993, namely postvocalic rinsertion and deletion in New England varieties of English. Not surprisingly, postvocalic [r] in most varieties of American English is an approximant hardly distinguishable from the end of a preceding low back vowel; it may thus be the closest thing to Ø in that context.

8

Vowel Duration, Syllable Quantity, and Stress in Dutch

Carlos Gussenhoven

8.1

Introduction

A persistent conundrum in the analysis of Dutch word prosodic structure has been the fact that closed syllables require a foot head, while long vowels apparently fail to project a foot (van der Hulst 1984; Kager 1989, 261; Zonneveld et al. 1999, 499; among others). This type of selective quantity sensitivity is highly marked: in other quantity-sensitive languages, long vowels are heavy (bimoraic), and will be in a stressed syllable (type A), while in addition such languages may require closed syllables to be heavy, and be stressed (type B). The latter option was termed weight-byposition by Hayes (1989a). In (1), this well-known typology is given. (1) a. Short vowel (V) b. Long vowel (VV) c. Closed rhyme (VC)

Type A m mm m

Type B m mm mm

?Dutch m m mm

There have been three responses to the apparent quantity-weight mismatch in Dutch. First, Lahiri and Koreman (1988) proposed that weight and quantity are represented separately: while weight is counted in moras, quantity is counted in X-slots. As a result, the light long vowels of Dutch can be represented as having one mora dominating two X-slots. Second, Kager (1989, 261) suggested that weight, which in other languages is determined by the branchingness of the syllable peak (short vowel versus long vowel, diphthong, short vowel plus C) is determined by ‘‘melodic complexity,’’ or the number of segment root nodes associated with moras after the first mora in the rhyme (light monophthong versus heavy diphthong or VC). A third response is that by van Oostendorp (1995), who assumes that so-called long vowels are not in fact represented as long, but di¤er from short vowels in lacking the vocalic feature [lax]. The specification of the duration of Dutch vowels will, in his view, be provided during the phonetic implementation.

182

Carlos Gussenhoven

It is argued that none of the above responses is tenable, and that the solution lies in a reevaluation of the phonetic facts. Unlike what is generally assumed, the long tense vowels of Dutch are only longer than short lax vowels in stressed syllables— that is, in the head of the foot. This suggests that Dutch stressed syllables are bimoraic, while unstressed syllables are monomoraic. The dependence of vowel length on stress will allow us to assume, with van Oostendorp (1995), that there are no underlying moraic representations for Dutch syllables. (For long ‘‘marginal’’ vowels, see section 8.3.) Bimoricity (and restricted occurrences of trimoricity) are the result of (i) weight-by-position (the projection of moras by coda consonants) or (ii) stress-to-weight (a bimoricity requirement on stressed syllables). Thus, in the analysis presented here, quantity plays a major part in the phonological representation of Dutch vowels, unlike van Oostendorp’s (1995, 1997) analysis, which assumes mora-less surface representations as well as mora-less underlying representations. It will be shown that moraic representations of vowels are in fact part of the lexical phonology of Dutch, and that a description of Dutch word prosodic structure is impossible if the moraic structure is left unspecified. In section 8.2, some new facts about Dutch vowel duration are given, together with a brief description of the experiment that yielded them. Section 8.3 shows that vowel quantity is partly determined in the lexical phonology, and that the specification of quantity cannot be left to the phonetic implementation in the sense of van Oostendorp. Section 8.4 describes the regular stress patterns of Dutch, drawing on the analyses in Nouveau 1994 and van Oostendorp 1997, while section 8.5 shows that the success of this description crucially depends on the correct analysis of the moraic structure of Dutch words, including the part before the main stress. A conclusion is o¤ered in section 8.6. 8.2

Duration and Distribution of the Vowels of Dutch

Table 8.1 lists the vowels of Dutch that can appear in a stressed syllable nucleus (e.g., Moulton 1962; Gussenhoven 1992; Booij 1995). There is a set of lax, short vowels, a set of tense vowels of which the [high] vowels are long, a set of tense or lax, oral or Table 8.1

The vowels of Dutch that can appear in stressed syllables Short (lax)

Long or short (tense)

Long

I y 

i y e ø

i y  ˇ  

þ



u o a

Diphthong u    þ þ 

i ˇy u

Vowel Duration, Syllable Quantity, and Stress in Dutch

183

nasal, long ‘‘marginal’’ vowels, which only occur in recent loans and onomatopoeic words, and finally, there are three diphthongs, as shown in table 8.1.1 The vowel [ ] (not listed) only appears in unstressed syllables. The heaviness of VC-rhymes is not obvious, but can be seen in trisyllabic words. Dutch has quantitity-sensitive trochees, built from right to left. The atypical ‘‘fact’’ that closed syllables are heavy and long-voweled open syllables light can be seen by comparing words of the type almanak ‘almanac’ with words like Gibraltar (the proper name). Both of these have a closed final syllable, but they di¤er in the structure of the penult. The closed penult attracts stress, leading to [xi’brþltþr], while the open penult is regularly skipped, causing main stress on the antepenult. The above interpretation of vowel quantity thus gives rise to the belief that long vowels, like [a] in [’þlma"nþk], are light, even though closed syllables are heavy. In reality, the duration of [e,ø,o,a] in unstressed positions is equivalent to that of short vowels. Using reiterant speech, Rietveld, Kerkho¤, and Gussenhoven (2004) investigated the duration of [a] and [i] in a large number of word prosodic contexts. While [i,y,u] are tense and have the same distribution as ‘‘long’’ tense vowels, their duration happens to be identical to that of short lax vowels: the vowels of [zIt] ‘sit’ and [zit] ‘see-3sg’ have equivalent durations, which may be only 50 percent of the duration of the vowel in [zat] ‘seed’ (Nooteboom 1972, 66). To identify the prosodic positions in which the duration contrast between short [i] and long [a] is made, nine word prosodic patterns were identified, between them covering all word prosodic positions. (E¤ects of codas on vowel duration were left out of account.) Word stress (as opposed to secondary stress), stress (as opposed to no stress), serial position (initial, nonperipheral, and final syllable in foot, and final foot in word) were each shown to have an e¤ect on the duration of each of the two vowels. A list of words illustrating these patterns is given in (2), where S ¼ word stress, s ¼ secondary stress, and w ¼ unstressed. e

(2) "rodo’dndr n "para’dis "pasifi’kasi "lokomo’tif "mini(")mali’zasi ’oli"fþnt ’oli"fþnt pi’rat pi’jano

swSw swS swwSw swwS swswSw Sws S w s w (inflected word) wS wSw

‘rhododendron’ ‘paradise’ ‘pacification’ ‘locomotive’ ‘minimalization’ ‘elephant’ ‘elephants’ ‘pirate’ ‘piano’

c

e

To exclude confounding segmental factors, the experiment made use of reiterant CV syllables, as produced by four speakers, with alternating occurrences of two out of the three consonants /k/, /s/, /m/ for the C-position, in all possible permutations.

184

Carlos Gussenhoven

Because the reiterant words were pronounced in a carrier sentence of the type Ik DOE (. . .) niet, ‘I do (. . .) not’, where the word doe was realized with a pitch accent, the data abstract away from accentual lengthening and utterance-final lengthening (Cambier-Langeveld 2000). Forty-eight (12 reiterant versions  4 speakers) realizations were obtained for each of the nine prosodic patterns. The results showed that there was never a significant di¤erence between the durations of the two vowels in positions outside the foot head, while occurrences of [a] in the foot head were significantly longer than [i]. The contexts in which a significant duration contrast was absent included the word-initial syllable immediately before the word stress, as well as the word-final syllable immediately after the word stress—for example, in the first and last syllables respectively of [pi’jano] ‘piano’. 8.3

Making the Moraic Structure Reflect the Phonetic Facts

Since phonological representations of vowel quantity are reflected in phonetic duration (Duanmu 1994; Hubbard 1995; Broselow, Chen, and Hu¤man 1997), it is proposed that the ‘‘long’’ vowels in the second column of table 8.1 are bimoraic in stressed position and monomoraic in unstressed position. Thus, from now on, a word like piano will be given as [pi’jano]. This suggests, first, that Dutch ranks a Stress-to-Weight Principle (SWP) (3) high,2 and second, since closed syllables attract stress, that it also ranks a Weight-to-Stress Principle (WSP) (4) high. (3) STRESS-TO-WEIGHT PRINCIPLE (SWP) Foot heads are (minimally) bimoraic. (4) WEIGHT-TO-STRESS PRINCIPLE (WSP) Bimoraic syllables are foot heads. WSP is not only relevant to closed syllables. As observed by Zonneveld (1993), the truly long vowels (see table 8.1, third column) do not tolerate being in an unstressed position: (Rio de) Janeiro [‰a’nro], *[’‰an"ro].3 Nor could the diphthongs (see table 8.1, fourth column) appear in unstressed penults: Khomeiny [ko’mini], *[’komi"ni]. These truly long vowels must thus be represented in the lexicon with two moras, while diphthongs are bimoraic by virtue of the fact that they contain two segments in the nucleus. These facts are consistent with Dutch being an unexceptional ‘‘type B’’ language. Broselow et al. (1997) assume a default markedness constraint SylMon (5), whereby syllables are monomoraic (cf. NoLongVowel in Kager, chapter 17, this volume). This constraint will make ‘‘long’’ vowels short in weak positions. (5) SYLMON Syllables are monomoraic.

Vowel Duration, Syllable Quantity, and Stress in Dutch

185

Tableau (6) illustrates how these constraints characterize [’bata] as the correct form of the brand name Bata. The trochee, assumed in (6), causes the first vowel to be bimoraic by SWP, while the weak syllable defaults to a monomoraic [a], since bimoraic [a] in weak position violates WSP. (6)

bata

WSP

SWP

+ a. (’bata)

*

b. (’bata) c. (’bata)

SylMon

*! *!

**

The bimoricity of syllables with short lax vowels is due to the following tautosyllabic consonant, which is obligatory in this context. Constraint LaxþC (cf. (7)), a reformulation of part of a constraint proposed by van Oostendorp (1995), requires a lax vowel to be monomoraic and be followed by a consonant in the same syllable.4 The moricity of the coda consonant is ensured by high-ranking Weight-by-Position (WbP), given in (8) (Hayes 1989). (7) LaxþC s m [þlax][þcons] (8) WEIGHT-BY-POSITION (WbP) A consonant in the coda projects a mora. As shown by van der Hulst (1985), a consonant after a short lax vowel is ambisyllabic in Dutch if it is required to be in the onset of the next syllable by Onset (9). For instance, a word like Hetty (a proper name) has an ambisyllabic [t], transcribed [t.t]. (9) ONSET A syllable has an onset. In (10), SWP  SylMon, together with Onset, causes the [t] to be ambisyllabic, and the first syllable to be bimoraic as a result. Lengthening of [], as in candidate (10c), is correctly prevented by LaxþC.

186

Carlos Gussenhoven

(10)

hti

Onset

WSP

SWP

LaxþC

SylMon

+ a. (’ht.ti)

*

b. (’h.ti)

*!

*

c. (’h.ti) d. (’ht.i)

*!

*

*!

*

e. (’ht.ti)

*!

**

Our account does not so far explain why words like [lat] ‘late’ have long vowels: the bimoricity of the syllable would already be guaranteed by the coda [t]. It is suggested that the long tense vowel in these forms results from the e¤ect of a constraint that maximizes the sonority of the syllable peak. The bimoricity of tense vowels before a tautosyllabic consonant is thus unrelated to the prosodic status of the syllable. The relevant constraint, given as SonPeak (11), can be seen as part of the family of HNuc (Prince and Smolensky 1993, 134). SonPeak must be ranked below LaxþC, to prevent short lax vowels from lengthening. (11) SONPEAK m [cons] Constraints Onset, WSP, and SWP are irrelevant in tableau (12), which shows the crucial ranking LaxþC  SonPeak (12) a.

lþt

Onset

WSP/SWP

LaxþC

+ i. lþt ii. lþt b.

SonPeak

SylMon

*

*

*!

**

lat i. lat + ii. lat

*!

* **

Finally, the diphthongs (column 4 of table 8.1) and the truly long (marginal) vowels will behave like long vowels: both elements of the diphthong are [cons]. If these long vowels are lax, they escape shortening by LaxþC because of high-ranking FaithMora (13), which preserves the lexical mora structure.

Vowel Duration, Syllable Quantity, and Stress in Dutch

187

(13) FAITHMORA Preserve mora structure. The rankings FaithMora  LaxþC, and LaxþC  SonPeak  SylMon, together with Onset and WSP/SWP, account for the moraic structure of Dutch vowels, with the exception of short tense [i,y,u], which are short even in stressed position. In the next section we will see how the quantity of these vowels must be accounted for in the lexical phonology, which provides us with an argument for rejecting the (implicit) assumption of van Oostendorp (1995) that the duration of Dutch vowels is purely a result of the phonetic implementation.5 8.3.1

Short Tense Vowels

The quantity of Dutch [i,y,u] is determined by the segmental context: they are long when appearing before [r] in the same foot (Gussenhoven 1993). In (14), words in which [i] appears before [r] in the same foot are compared with words in which this vowel appears in prosodically identical words, but in which it is not followed by [r]. wier Olivier giro pierement fakir

[Vir] [’oli"fir] [’xiro] ["pir ’mnt] [’fakir] e

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

‘algae’ ‘Oliver’ ‘giro’ ‘barrel organ’ ‘fakir’

wiek kolibri kilo lineaal kieviet

[Vik] [’koli"bri] [’kilo] ["lini’al] [’kivit]

‘wing’ ‘kolibri’ ‘kilo’ ‘ruler’ ‘peewee’

When [r] is in the next foot over, the high vowels are short, as in piraat [pi’rat] ‘pirate’, corduroy [’k rdy"r j] ‘corduroy’, admiraal ["þtmi’ral] ‘admiral’. The constraint HighV-m (15) reflects the widespread tendency for high vowels to be shorter than nonhigh vowels. The relatively short distance between the tongue body and the roof of the mouth explains this e¤ect (see Laver 1994, 435). c

c

(15) HIGHV-m High vowels are monomoraic. To ensure that [i,y,u] are long before [r] in the same foot, we postulate Pre-r-mm (16). The articulatory motivation for this constraint is probably to be found in the conflict between a vocalic tongue posture (a convex dorsum and a tongue blade curling down into the lower jaw) and the tongue posture for a coronal [r], for which the front is held in a concave shape behind a tongue tip that curls up. The articulatory transition from a vocalic posture to that required for [r] will thus be more elaborate than a transition to the position for postvocalic [t,s,n,l], for which the front of the tongue may, but need not be concave. Evidently, Pre-r-mm  HighV-m, for otherwise high vowels could never be bimoraic in Dutch. (16) PRE-r-mm Tense vowels are bimoraic before [r] in the same foot.

188

Carlos Gussenhoven

Tableau (17) shows how this works for short and long occurrences of high tense vowels. Candidate aiii, in which the intervocalic onset consonant is ambisyllabic, is ruled out by SonPeak, which for this reason must be ranked above WSP/SWP. That is, we cannot satisfy SWP by filling a second mora with the following [t]. (17) a.

rita +

Pre-r-mm

HighV-m

i. ’ri.ta

WSP/SWP

SylMon

*

ii. ’ri.ta

*!

iii. ’rit.ta b.

SonPeak

* *!

*

xiro +

i. ’xi.ro

*

ii. ’xi.ro

*!

iii. ’xir.ro

*!

* * *

*

I give the constraint rankings established so far, with the generalizations they account for: WSP/SWP  SylMon: Stressed syllables are bimoraic, unstressed syllables monomoraic. Pre-r-mm  HighV-m: High tense vowels are long before [r] in the same foot, but short otherwise. LaxþC  SonPeak: Lax vowels are short, despite their occurrence in bimoraic syllables, and thus must be followed by C in the rhyme (or: do not maximize a lax vocalic peak). SonPeak  SylMon: Tense long vowels are long even before a coda C (or: do not minimize moraic structure at the expense of a vocalic peak). SonPeak  SWP: Do not make consonants ambisyllabic merely to have bimoricity in stressed syllables. Faithmora  LaxþC: Do not shorten long lax (marginal) vowels. 8.3.2

Why Vowel Duration Is Represented in the Phonology

At this point it might be objected that the durational phenomena we have dealt with could be accounted for in the phonetic implementation. Specifically, the question arises of whether we couldn’t have a single articulatory instruction, to be carried out with reference to a phonological representation without quantity distinctions, which says: ‘‘make nonlax vowels long when stressed or followed by a tautosyllabic conso-

Vowel Duration, Syllable Quantity, and Stress in Dutch

189

nant, but not when they are high, except when [r] appears in the same foot.’’ This course of action is unavailable, however, because the quantity of high tense vowels is subject to a lexical process. Crucially, this means that moraic structure must be represented in the lexical phonology. There are five irregular past-tense forms that have short [i] before [r] in the same foot. In (18), an example is given. (18) wierp

[Virp]

‘threw’

(cf. vier-t [firt] ‘celebrateþ3sg-pres’)

As Booij (1995, 94) points out, these forms end in a cluster of [r] and a labial obstruent, while no other words do. A phonological account that exploits this observation might assume, first, that all coda consonants project a mora, except postconsonantal [t,s], and second, that there is a constraint *mmmm (19), undominated in Dutch, disallowing tetramoraic syllables. The first assumption is widely supported in work on Dutch phonology (see Booij 1995, 26). The second is evidently supported by the typology of quantity, trimoraic syllables already being rare. (19) *mmmm Syllables are maximally three moras long. As shown in (20), to represent wierp with a long vowel would mean violating undominated *mmmm. The form [firt] escapes shortening, because coronal [t] fails to project a mora. The correctness of this solution is underscored by examples like Ataturk [’ata"tyrk] ‘Ataturk’ and kirsch [kirS] ‘Kirsch’ (Paul Boersma, personal communication), which have short pre-[r] vowels, as expected under this analysis. (20) a. *

mmmm u i

rp

b.

mmm f i rt

While these facts already look pretty phonological, the clinching reason why it is not possible to translate the e¤ect of *mmmm into a phonetic implementation rule is that the short [i] of the past-tense forms survives an inflectional a‰xation process that removes the labial obstruent from the coda. The structure that arises can be compared with phonologically underived forms, as in (21), to show that long [i] appears in phonologically comparable contexts. [’Vir.p n] [’kir.k "xart] e e

(21) a. wierpen b. Kierkegaard

‘threwþplur’ ‘proper name’

This means that vowel quantity di¤erences are involved in lexical representations. The facts of (21) are consistent with a lexical phonology version of Optimality Theory (Kiparsky, forthcoming): wierp is subjected to the constraint grammar at the stem level, and its output is evaluated as the input of the constraint hierarchy at the word level, where the moraic structure specified at the stem level is preserved.

190

8.4

Carlos Gussenhoven

Dutch Stress

In this section, I consider the implications of our representation for a description of regular Dutch stress. The description essentially follows Nouveau 1994 and van Oostendorp 1997, but di¤ers from these in that it describes the foot structure of the whole word, not just of the foot with main stress. I show that regular main stress cannot be satisfactorily described if the prosodic status of the section of the word before the main stress is ignored. Crucially, the assumption must be made that, in Dutch, quantity sensitivity is restricted to the right-hand part of the word, and that before the main stress Weight-by-Position is not in e¤ect. 8.4.1

The Regular Stress Pattern

The basic facts about Dutch stress are summarized in (22) and (23) (van der Hulst 1984; Kager 1989; Nouveau 1994; van Oostendorp 1997; among others). The examples in (22) show the simplest case: main stress falls on the penult. (22) a. a.’xa.ta b. a.’mþn.da c. ’a.r n

‘Agatha’ ‘Amanda’ ‘Aaron’6

c

Penultimate stress systematically fails to appear in three situations. First, it does not appear when the word is monosyllabic, as in (23a). Second, it is systematically omitted when the word is minimally trisyllabic, has an open penult and a final closed syllable, and word stress is on the antepenult, as illustrated in (23b). Third, as shown in (23c), superheavy syllables, which can appear in word-final position only, attract main stress. (23) a. ’la ’kþt b. ’ma.ra."t n c. "þt.mi.’ral "le.di.’kþnt

‘drawer’ ‘cat’ ‘marathon’ (but: pa.’lm.bþÐ ‘Palembang’) ‘admiral’ ‘bed’

c

A treatment in OT was presented by Nouveau (1994, 184¤.) and modified by van Oostendorp (1997). The present analysis di¤ers from these earlier ones in that the representation of long vowels is bimoraic in stressed syllables, rather than monomoraic, but otherwise it essentially follows the older analysis in the characterization of regular main stress. All constraints are from Prince and Smolensky 1993 as well as McCarthy and Prince 1995 unless indicated otherwise. In section 8.3, we already took the e¤ect of RhythmTrochee (24) for granted. NonFin (25) is interpreted to ban main stress on the final syllable, as in Nouveau 1994, while Fs Right (26) will see to it that a foot with main stress is rightmost in the word.7

Vowel Duration, Syllable Quantity, and Stress in Dutch

191

(24) RHYTHMTROCHEE Feet are left-dominant. (25) NONFIN Main stress is not on the word-final syllable. (26) Fs RIGHT, or Align(Pw,Rt,Fs ,Rt) The right edge of the word coincides with the right edge of a strong foot. NoClash (27) forbids adjacent stressed syllables (i.e., adjacent foot heads). Highranking NoClash ensures that monosyllabic feet can only exist word-finally (see Gussenhoven 1993). (27) NOCLASH Foot heads are not adjacent. Finally, FootBin requires that feet are binary, either at the moraic level or at the syllable level. Violations are incurred by trisyllabic feet and monomoraic feet. In e¤ect, because high-ranking WSP/SWP already weeds out all monomoraic foot heads, FootBin’s only role in our analysis is to ban ternary feet. (28) FOOTBIN Feet are neither monomoraic nor trisyllabic. Cases like Agatha, with three open syllables, are derived straightforwardly: the winning candidate manages to obey all five relevant constraints. Candidate (29b) is ruled out by NoClash, and forces the prestress initial syllable to be unfooted—that is, directly attached to the Pword node. This is in accordance with the finding in Rietveld et al. (2004) that no quantity contrast exists in such syllables. Candidates (29c–f ) all founder on one of the other three constraints, as shown in tableau (29). (29)

axata

FootBin

NoClash

NonFin

WSP/SWP

Fs Right

+ a. a’(xa.ta) b. (a)’(xa.ta)

*!

c. ’(a.xa)ta

*!

d. ’(a.xa)(ta)

*!

e. (a.xa)’(ta) f. ’(a.xa.ta)

*! *!

The same result obtains if the penult is closed, as in [a.’mþn.da], the only di¤erence being that the equivalents of candidates (29c,d) are taken out by WSP as well as Fs Right. The ranking of WSP/SWP becomes critical in words with final closed

192

Carlos Gussenhoven

syllables. To stay with trisyllabic words, type (23b) (e.g., marathon) provides evidence that Dutch is quantity-sensitive, as shown in tableau (30). The winning candidate (30ai) violates Fs Right, which ranks below WSP/SWP: it alone incurs no WSP/ SWP violation, because both foot heads are bimoraic and the only weak syllable is monomoraic. Words like [pa.’lm.bþÐ] ‘Palembang’, which in addition have a closed penultimate syllable, incur WSP/SWP violations regardless of whether the main stress is on the penult or the antepenult (see candidates (30bi) and (30bii) in particular), and avoiding a violation in the penultimate syllable, as in candidate (30bi), is therefore pointless. The decision falls to Fs Right. Candidates (30aiv), (30biv), and (30bv) are excluded by NoClash. FtBin is omitted, but will reappear in section 8.5. (30) a.

maratn +

NoClash

WSP/SWP

i. ’(ma.ra)(tn)

Fs Right *

ii. ma ’(ra.tn)

*!

iii. ’(ma.ra)tn

*!

iv. (ma) ’(ra.tn) b.

NonFin

*!

!

palmbþÐ i. ’(pa.lm)(bþÐ)

*

+ ii. pa. ’(lm.bþÐ)

*!

*

iii. ’(pa.lm)bþÐ

*!*

iv. (pa) ’(lm.bþÐ)

*!

v. (pa) ’(lm)(bþÐ)

*!*

* *

Penultimate stress in disyllables with a closed final syllable is due to NoClash and NonFin, both of which rank above WSP/SWP. In tableau (31), [’a.r n] shows that the language prefers incurring a WSP violation to violating NonFin (see candidates (31a,b)). c

(31)

arn

NoClash

NonFin

+ a. ’(a.rn)

WSP/SWP

Fs Right

*

b. a.’(rn)

*!

c. ’(a).(rn)

*!

d. (a).’(rn)

*!

* *

Vowel Duration, Syllable Quantity, and Stress in Dutch

193

To continue with the examples in (23), monosyllabic words (see 23a) have stress because of the constraint Lex A Pwd (not given as a numbered item), which requires every morphological word to be footed. Superheavy syllables (see 23c) only occur word-finally. The relevant contextual constraint allowing this (‘‘No trimoraic syllable unless at right word edge’’) is here taken for granted. Superheavy syllables attract the word stress because trimoraic syllables are disfavored in weak positions, in the foot as well as in the word. This constraint, the Superheavy-to-Stress-Principle (32), can best be seen as belonging to the same family as WSP—that is, as a stricter version of the latter (see Prince 1990). Nouveau (1994) and van Oostendorp (1997) achieve the e¤ect of SHSP by postulating a degenerate or abstract final syllable for the final consonant, which makes these words escape the censures of NonFin; the effect is that of final consonant extrametricality. (32) SUPERHEAVY-TO-STRESS PRINCIPLE (SHSP) Trimoraic syllables are strong foot heads. Ranking SHSP above NonFin will have the desired e¤ect, as shown in tableau (33). (33)

kapital

NoClash

SHSP

+ a. (ka.pi).’(tal)

NonFin

WSP/SWP

Fs Right

*

b. ’(ka.pi).(tal)

*!

c. ka’(pi.tal)

*!

* **

In this section, the following generalizations were achieved: Lex A Pwd undominated: SHSP  NonFin: NonFin  WSP/SWP: NonFin  Fs Right: WSP/SWP  Fs Right:

8.5

Monosyllables have word stress. Final superheavy syllables have word stress. No word stress on a final closed syllable in disyllables. No word stress on a final closed syllable in trisyllables. Trisyllables with final closed syllable and open penult have antepenultimate stress (i.e., do not make a disyllabic final foot with heavy weak syllable).

Whole-Word Foot Structure

The description of Dutch stress presented in the previous section would appear to be seriously challenged once words with closed initial prestress syllables like armada [þr’mada] ‘armada’ are considered. If the attested candidate (34a) is disregarded for the moment, high-ranking NoClash would incorrectly characterize *[’þrma"da],

194

Carlos Gussenhoven

candidate (34e), as the optimal form. Candidate (34b) violates NoClash, and candidate (34c) violates NonFin. The choice between candidates (34d) and (34e) would be decided by WSP/SWP, which will not tolerate an unfooted bimoraic syllable. Importantly, to characterize the attested candidate (34a) as optimal, we must remove the foot from the initial syllable, so as to satisfy NoClash, and declare it monomoraic, so as to satisfy WSP/SWP.8 (34)

þrmada

NoClash

NonFin

WSP/SWP

Fs Right

m + a. þr.’(ma.da) b. (þr).’(ma.da) c. (þr.ma).’(da) mm d. þr.’(ma.da) e. ’(þr.ma)(da)

*! *! *! !*

The inevitable conclusion that closed word-initial prestress syllables are monomoraic in Dutch is supported by three independent arguments. The first is based on duration measurements. The best comparison we can make is with the first syllable of a word-initial weak foot. If, for example, the duration of [kþn] in cantorij ["kþnto’ri] ‘church choir’9 were to be shorter than that of [kþn] in kantoren [kþn’tor n], this would be strong evidence that kantoren has an unfooted first syllable, for if it were footed, it would either have to be equal in duration to the first syllable of cantorij, or, in view of the fact that there are widespread tendencies to shorten the foot head as more syllables occur in the foot, longer. In a production experiment with four speakers, it was consistently the case that the prestress syllable was shorter than the segmentally identical head of a disyllabic foot before the main stress (Elise Hofhuis, unpublished research). This is strong evidence that closed prestress syllables are not footed. The second argument was presented in Gussenhoven 1993 and concerns the fact that the Dutch intonational pattern known as the ‘‘chanted call,’’ which potentially produces a new pitch level on every postnuclear foot, systematically fails to produce such a pitch level on word-initial prestress syllables, whether closed or open, but may produce one on (binary) feet before the word stress. The di¤erence would be brought out if cantorij and kantoren were to be used as second members in compounds, as in nepcantorij and nepkantoren, where nep ‘fake’ would be accented. Whereas can- could begin a second pitch level, kan- could not. The third argument is distributional. Monomoricity of the word-initial prestress syllable predicts that one type of segment should be systematically excluded there, namely, the ‘‘mare

Vowel Duration, Syllable Quantity, and Stress in Dutch

195

ginal’’ long vowels (see table 8.1, third column). Rhymes consisting of strings of different segments can be accommodated, however marked multiple association of segments with a single mora may be, as shown by pneumatisch [pnœy.’ma.tis] ‘pneumatic’, which has a monomoraic diphthong in its first syllable. However, the combination of a single mora and a long vowel amounts to a contradiction in terms. Interestingly, while words like creˆmerie ["kr.m .’ri] ‘creamery’ can exist, in which [] occurs in a foot head, words like *[.’tþp.p ] are impossible (cf. e´tappe [e.’tþp.p ] ‘leg (sport)’). In conclusion, not only open, but also closed word-initial prestress syllables are monomoraic and unfooted, as argued in Gussenhoven 1993, which explains the fact that candidate (34a) is optimal. e

e

e

8.5.1

Accounting for Monomoraic Closed Syllables

If closed word-initial prestress syllables are monomoraic, Weight-by-Position (WbP)—the constraint that requires coda consonants to project a mora—must not be operative in that syllable. In fact, to the left of the main stress there is little or no evidence for the working of WbP at all (see Booij 1995, 106; Zonneveld et al. 1999, 504). Indeed, van der Hulst and Kooij (1992) proposed that main stress in Dutch results from quantity-sensitive footing from the right, but that the rest of the word is subsequently footed quantity-insensitively from the left.10 The weightlessness of syllables before the word stress can be observed in words that contain a string of three syllables prior to the main stress, of which the first is open and the second closed. If the second syllable attracted stress, the three-syllable stretch would be realized as an unfooted syllable followed by a binary foot. The words in (35) belie that expectation: the secondary stress is on the first syllable.11 (35) "aristo’cratisch ‘aristocratic’, "decompo’sitie ‘decomposition’, "enunci’atie ‘enunciation’, "evange’list ‘evangelist’, "emanci’patie ‘emancipation’, "identi’teit ‘identity’, "paterna’listisch ‘paternalistic’, "potenti’eel ‘potential’, "protestan’tisme ‘protestantism’, "tubercu’losis ‘tuberculosis’ Evidently, quantity sensitivity only obtains in the stretch from the main stress to the word end. Constraint WbP 0 (36), a version of WbP which is confined to the stretch beginning with the main stress, expresses this. (36) WbP 0 From the main stress onward, a coda consonant projects a mora. The pattern illustrated in (35) suggests that Dutch ranks high All-Ft-Left (37), a constraint that imposes a violation for every syllable by which the left edge of any foot fails to coincide with the left edge of the word. With Parse-s (38), which requires that syllables be parsed into feet, ranked above All-Ft-Left, exhaustive footing is achieved (Prince and Smolensky 1993). Tableau (39) shows this for enunciatie. The tableau dispenses with NoClash, Fs Right, SHSP, and NonFin, which all

196

Carlos Gussenhoven

relevant candidates satisfy: the point is that the low ranking of generic WbP allows the second syllable to escape the censure of WSP/SWP, which, had it been bimoraic, would have had to be a foot head, causing candidate (39d) to be optimal. As it is, the competition is decided by All-Ft-Left, which the winning candidate (39a) best satisfies; the fact that the syllable [nyn] is in a weak position and therefore violates WbP is no longer relevant to the outcome. (37) ALL-FT-LEFT, or Align(Ft,Lt;Pwd,Lt) The left edge of every foot coincides with the left edge of the Pword. (38) PARSE-s Syllables are parsed into feet. (39) en nsiasi

Foot Bin

+ a. (e.n n).si.’(a.si) b. (e.n n.si).’(a.si)

8.6

WbP 0

WSP/ SWP

Parse-s

All-FtLeft

WbP

*

***

*

***

* *

*!

c. e.n n.si.’(a.si)

**!*

***

d. e(n n.si).’(a.si)

*

***!*

Conclusion

The present account of the duration of the vowels of Dutch, the moraic structure of its syllables, and its foot structure shows that Dutch is an unexceptional ‘‘type B’’ language: truly long vowels and diphthongs (see columns 3 and 4 in table 8.1) attract stress in the same way that closed syllables do. The reason the language has been characterized as atypically requiring long vowels to be light and closed syllables to be heavy is that earlier researchers failed to accommodate the fact that the long vowels concerned only acquire bimoricity (and thus length) as a result of their being in a stressed location, as determined by the regular foot structure of the language. By attributing their length to the working of Stress-to-Weight, a phonetically realistic representation of vowel quantity has become possible. Moreover, there has been no need to represent [i,y,u], which are short even in stressed positions, as bimoraic, either underlyingly or on the surface, as earlier analyses were forced to do. They are only long, and bimoraic, when occurring before [r] in the same foot. Solutions to the special status of quantity sensitivity in Dutch provided by Lahiri and Koreman (1988) (separate representation of weight and length) and Kager (1989) (counting segments in the rhyme rather than moras) are thus rendered unnec-

Vowel Duration, Syllable Quantity, and Stress in Dutch

197

essary. My solution confirms the representation proposed by van Oostendorp (1995): underlyingly, the di¤erence between [þ] and [a], for instance, is captured by including [lax] in the representation of the former vowel, and quantity is not present in underlying representations. However, it disagrees with van Oostendorp’s analysis in requiring moraic structure to be present in the surface representation, a position enforced by the fact that quantity di¤erences are in part morphologically determined. Exceptionally long lax vowels (see column 3 in table 8.1) are accounted for by lexical specification of bimoricity, which is respected due to the high ranking of FaithMora. Interestingly, as pointed out by Sharon Inkelas, this analysis correctly predicts that there are no exceptionally short tense vowels (other than [i,y,u]): marking [a] as monomoraic will not prevent this vowel from being long in stressed positions, due to SWP/WSP. The relevance of moraic structure was further underscored by the foot structure of the word before the main stress foot, where crucially Weight-by-Position must be suspended. Failing to do so predicts incorrect main stress in words like ar’mada, which contain a closed word-initial prestress syllable. My analysis demonstrates that it may be unwise to present analyses of part of the data representing some phenomenon, like the data for main stress, and ignore other parts, like the data for secondary stress and vowel duration. Notes I thank the audiences at the universities of Groningen, Constance, Leiden, Nijmegen, and Amsterdam (Free University), where earlier versions of this chapter were presented in 1998 and 1999, for their useful comments, in particular Geert Booij, Mirco Ghini y, Haike Jacobs, Wouter Jansen, Aditi Lahiri, and Dominique Nouveau. Rene´ Kager, Aditi Lahiri, and Marc van Oostendorp commented on earlier drafts, thus allowing me to make various improvements. This chapter represents a revised version of the first half of a paper ‘‘Duration and Quantity in the Dutch Word’’ presented at the Conference on the Phonological Word in Berlin (ZAS), held October 24–26, 1997. The second half of the paper, on exceptional stress, will appear elsewhere. 1. The marginal nasal vowels (third column) are generally given without the length mark, but as Ton Broeders pointed out to me, they are in fact long. 2. Pace Prince (1990), who argues that SWP does not exist, and that the tendency is for the heads of trochees to shorten, as in Hayes 1987. Prince admits that what looks like the e¤ect of SWP may be seen in languages with dynamic stress. In fact, Dutch also has Trochaic Shortening, though it is variable and confined to disyllabic feet to the left of the main stress, as in ["po()li’tik] or substandardly ["p l ’tik] ‘politics’. Rene´ Kager points out that an e¤ect similar to SWP is obtained by HNuc, as shown in a handout to his Metrical Phonology class at the University of Utrecht of November 29, 1996: ‘‘The best main stress is the heaviest syllable.’’ As in van Oostendorp 1997 he assumes that length is coded by an underlying tenseness feature, while HNuc is held responsible for the bimoricity of main stressed vowels. His restriction to main stress equates the categorical occurrence of short vowels in unfooted syllables and in weak ec

198

Carlos Gussenhoven

syllables with syllables with secondary stress, leaving their (frequent) bimoricity unexplained. Also, no account of the quantity of [i,y,u] is given. 3. Trommelen and Zonneveld (1989) assume that [] is the surface form of underlying [i] before [r], observing that the latter combination does not occur and that all occurrences of [] before a full-voweled syllable precede [r]. However, there are counterexamples in both directions: theta [’tta] ‘theta’ and Teixeira (de Mattos) [tk’sira] (proper name). 4. Van Oostendorp also makes the complementary assumption that vowels that are not lax— that is, our ‘‘long’’ tense vowels—have no coda, in spite of the fact that they apparently do (e.g., laat ‘late’). This assumption is not made here. 5. Zonneveld et al. (1999, 500) make the notion of such a phonetic implementation rule explicit, but they reject the analysis and go on to defend the solution presented in Lahiri and Koreman 1988. 6. The biblical name is often pronounced [a’ar n], as pointed out to me by Marieke Polinder. c

7. Both Nouveau and van Oostendorp have a constraint that requires the main stress to be on the last syllable of the word, while the present analysis assumes a constraint aligning the main stressed foot with the right word edge. No crucial di¤erences in coverage follow from this. 8. Unlike what is the case in English, where a closed word-initial prestress syllable is footed if it does not represent a Latinate prefix (Chomsky and Halle 1968), in Dutch all prestress initial syllables, whether closed or open and whether or not they are prefixes, are unfooted (Gussenhoven 1993). 9. This word has a su‰xal [i], which attracts main stress; compare simplex [’sld "ri] ‘celery’, which has the regular pattern of words with final heavy and penultimate light syllables (see section 8.3). e

10. The notion of di¤erent directions for main stress assignment and secondary stress assignment already occurs in Booij 1983 with reference to Kenneth Pike’s work on Auca (though without di¤erentiation for quantity sensitivity). 11. There are words like this that, in addition to the pattern exemplified in (35), may alternatively be pronounced with secondary stress on the second, closed syllable karakteri’stiek ‘characteristic’, gerontolo’gie ‘gerontologie’, appendi’citis ‘appendicitis’, electrici’teit ‘electricity’, amonti’llado ‘amontillado’. A case for quantity sensitivity cannot easily be made on the basis of these words, all the more so since, as pointed out by Booij (1995, 106), there are words like piraterij [pi"rat ’ri], which show that a light second syllable may have secondary stress. That example also shows that the situation is more complex, since the reason for the location of the secondary stress on the second syllable in this word would seem to be that [a] is opener, more sonorous, than [i]; similarly, as pointed out by Haike Jacobs, the word caleidoscoop [ka"lid ’skop] ‘kaleidoscope’ illustrates that diphthongs can attract the foot in competition with [a] (cf. variable amonti’llado). e

e

9

Sympathy Meets Argentinian Spanish

Ellen M. Kaisse

9.1

Introduction

The fact that phonological processes are often opaque—that is, that they may appear to have applied when they should not have, or to have failed to apply when they should have—has occupied the attention of phonologists ever since Kiparsky’s (1971b, 1973e) landmark treatments of the phenomenon. As new theories of phonology emerge, they must be tested against opaque surface forms. A theory that cannot generate such forms must be judged inadequate. Opacity is a well-known problem for two-level, nonderivational theories such as standard Optimality Theory (Prince and Smolensky 1993). While a number of solutions have been proposed that attack this problem piecemeal, McCarthy (1999a) stands out as the first comprehensive attempt to treat virtually all opaque phenomena with the same single enrichment to standard OT: ‘‘sympathy.’’ Sympathy allows us to identify a form that is not underlying and use it as the basis for evaluating the faithfulness of potential outputs. McCarthy’s goal is to supplant all other mechanisms accounting for opacity, many of which are poorly understood and controversial—output-output constraints (Benua 1997), locally conjoined constraints (Smolensky 1995),1 stratal OT (Kiparsky 1998; McCarthy and Prince 1993; Orgun 1996; Rubach 2000), two-level constraints (Koskenniemi 1983; McCarthy 1996), and so forth. The strategy, which we must applaud, is to fully investigate and understand the characteristics of this one novel mechanism and push it to its full capacities while relying as much as possible on ‘‘familiar and indispensable OT constraints.’’ In addition to providing a careful explication of the characteristics of sympathy, McCarthy o¤ers a comparison between its predictions and those of standard derivational treatments of opacity using rule ordering. While the overlap in coverage, and indeed in concept, between the two theories is large, McCarthy is able to isolate one significant di¤erence—a class of cases that are ruled out by sympathy and thus predicted not to occur in natural languages, but that are easily expressed in a

200

Ellen M. Kaisse

derivational framework. He demonstrates that within sympathy theory, the following prediction holds: (1) If two notionally distinct processes . . . violate exactly the same faithfulness constraints, then they must always act together in rendering a third process opaque. (McCarthy 1999, 344) But is McCarthy correct in predicting that natural languages do not contain similar but distinct processes that respectively do and do not opacify a third process? I will argue that he is not. In this chapter, I will discuss data from Rı´o Negro Argentinian Spanish involving a sometimes-opaque interaction between a coda-based process and resyllabification into onset. I first treat the phenomenon in a multistratal derivational framework (Lexical Phonology), following the analysis in Kaisse 1999 and Harris and Kaisse 1999. Translated into monostratal OT using sympathy theory, this phenomenon is shown to meet the description of two processes violating the same faithfulness constraint, yet not to comply with McCarthy’s prediction. I will conclude that sympathy cannot, indeed, handle all cases of opacity. Because this example is emblematic of cases where intraword and interword phenomena diverge in their behavior, the best solution, I will argue, is to return to the derivational account based on Lexical Phonology’s division between lexical and postlexical strata, or perhaps to adopt a derivational OT model (Kiparsky 1999; McCarthy and Prince 1993; Rubach 2000) that allows us to maintain a division between lexical and postlexical evaluation. In general, cases violating the prediction in (1) may not be terribly hard to find (Levi 2000, for instance, contains another example, of a quite di¤erent character) and throw doubt on the ability of sympathy to adequately treat all cases of opacity within Optimality Theory. This chapter thus joins Odden’s (2000) paper in arguing that sympathy, despite its power, cannot be the sole solution to problems of opacity in OT. 9.2

S-Aspiration in Rı´o Negro Argentinian Spanish

Like many dialects of Spanish, Argentinian varieties realize /s/ as [h] in the coda of a syllable.2 The phenomenon, often referred to as ‘‘s-aspiration,’’ is illustrated below within morphemes and between a stem and su‰x, environments where virtually all dialects with aspiration show the same behavior. (2) ka.s-a gu.sa.n-o pa.s-ar dje.ses

‘house’ ‘worm’ ‘to pass’ ‘tens’

kah.ka.r-a guh.t-o pah.tel djeh.-mo

‘rind’ ‘pleasure’ ‘cake’ ‘tithe’

The rule for dialects like that of Rı´o Negro is given in (3), and is shown more formally in (4) to be a debuccalization after nonconsonantal segments (glides and

Sympathy Meets Argentinian Spanish

201

vowels), expressed as the spreading of [consonantal] to rhymal ‘‘/s/.’’ The loss of place can be seen as the automatic consequence of the loss of consonantality. (3) s ! h/ [cons] /

]s

(4)

The data presented in (2) is valid for the vast majority of dialects that exhibit the s-aspiration phenomenon. Preconsonantal /s/ is aspirated within a morpheme (kahkara) or before a consonant-initial su‰x (djeh-mo); within a morpheme or between a stem and a vowel-initial su‰x, prevocalic /s/ is not aspirated, since it is not in coda (gusan-o, kas-a). Things get more interesting when we look at other morphological contexts, where dialects diverge in their behavior. In this chapter, I concentrate on a little-studied dialect spoken in the northern Patagonian provinces of Neuque´n and Rı´o Negro, about 800 miles to the southwest of the capital, Buenos Aires. I will refer to this dialect as Rı´o Negro Argentinian, or RN for short, since the majority of my informants are from that province or right across the river in the city and province of Neuque´n. However, I suspect that this is the dialect of most of Patagonia. It has long been noted that s-aspiration in Spanish interacts in various ways with other phenomena in the phonology, among them resyllabification. Guitart (1979) and Harris (1983, 1993) have pointed out, for instance, that in some Caribbean dialects, word-final /s/ is realized as [h], even if it is not in a coda at all on the surface, but forms the onset of a following vowel-initial word: (5) tje.ne. heh.pa.sjo

tienes espacio

‘you have space’

The same is true for RN: word-final /s/ is realized as [h] not only before pause3 and preconsonantally, where its appearance is transparently due to its position in coda, but also before a word-initial vowel, where its appearance is opaque: (6) bwe.noh. di.ah loh. la.bjoh

‘good days’ ‘the lips’

bwe.no. haj.reh lo.h o.xoh

‘Buenos Aires’ ‘the eyes’

The prevocalic and prepausal aspiration is the source of much merriment among porten˜os—speakers of the prestigious dialect of Buenos Aires—among whom RN is commonly mocked as the loh ojoh (i.e., [lohoxoh]) dialect, a reference to the way

202

Ellen M. Kaisse

speakers from Rı´o Negro pronounce the phrase meaning ‘the eyes’. Porten˜os have transparent aspiration—they do not aspirate a prevocalic /s/ that ends up in an onset—and they do not aspirate before pause, so ‘the eyes’ emerges as [lo.so.xos] when uttered in isolation. In undergoing aspiration, prevocalic word-final /s/ in RN thus exemplifies the type of opacity in which a process appears to have occurred when its structural description is not present. The environment for the emergence of the first [h] in [bwe.no.haj.reh] or [lo.ho.xoh] is not present on the surface. Thus, aspiration is nonsurface-apparent, or overapplies, in McCarthy’s (1999) terminology. Within a derivational framework, such cases are handled via counterbleeding orders. Aspiration occurs at the word level, then resyllabification between words applies, removing [h] from its position in coda and providing the following syllable with an onset. (7) a. Word level

b. Phrase level

Initial syllabification Resyllabification within words Aspiration Resyllabification between words

/los oxos/ los. o.xos ——— loh. o.xoh lo. ho.xoh

The transparent Buenos Aires derivation, where aspiration is postlexical and does not occur before a pause, would proceed as in (8). (8) a. Word level b. Phrase level

Initial syllabification Resyllabification Resyllabification between words Aspiration

/los oxos/ los. o.xos ——— lo.so.xos ———

The important additional wrinkle added by RN (and not by other, better-known dialects) is this: between a prefix and the base4 to which it attaches, aspiration applies transparently. That is, if the base begins with a consonant, the s-final prefix des is realized as [h], but before a vowel, it is realized as [s]. (9) dehkargar desarmar dehkremada desinteresada

‘discharge’ ‘disarm’ ‘defatted, skim’ ‘disinterested’

Ample evidence that prefixes are syllabified on a separate cycle from their bases, then (speaking derivationally) resyllabified, is presented in Harris and Kaisse 1999. As described there, Argentinian Spanish has a process of glide consonantalization that realizes a high front vowel as [ˆ] syllable-initially. In the (a) forms below, there is no prefix, and the Cþglide sequence, whether underlying or derived from diphthongiza-

Sympathy Meets Argentinian Spanish

203

tion, is syllabified together at the beginning of the second syllable. The glide is not syllable-initial, and therefore does not consonantalize. However, in the (b) forms, the base is syllabified on its own. Its initial glide (again be it from diphthongization or underlying) emerges as a fricative and a preceding /s/ aspirates, while a preceding nasal agrees in place of articulation. (10) a. /deserto/ /boniato/ b. /des-elo/ /kon-iuxe/

! ! ! !

[de.sjer.to] [bo.nja.to] [deh.ˆe.lo] [ko.ˆu.xe]

‘desert’ ‘sweet potato’ ‘defrost’ ‘spouse’

The transparency of word-internal aspiration versus the aspiration between words is easily accounted for in a derivational theory using rule ordering and strata: (11) a. Word level

b. Phrase level

Syllabification Resyllabification between prefix and base Consonantalization Aspiration Resyllabification between words

A typical derivation would proceed as in (12): (12) a. Word level

b. Phrase level

Initial syllabification Resyllabification within words Aspiration Resyllabification between words

/desþarmar/ des.ar.mar de.sar.mar ——— ———

/los oxos/ los. o.xos ——— loh.o.xoh lo.ho.xoh

Before closing this section, we should address the question of whether /s/ is the underlying form for all or some [h] in RN, or whether the aspiration rule has been lexicalized, yielding /h/ in underlying representations. There are several reasons to opt for the analysis with all [h] coming from underlying /s/. First, unlike aspirating dialects spoken in the Caribbean, Argentinian Spanish does not have a phone [h] in any contexts other than those of s-aspiration. (The segment spelled ‘‘j’’ is pronounced [x], a voiceless velar fricative.) Thus, /h/ would be an ill-distributed phoneme, occurring in underlying representations only in preconsonantal position and as the last segment in a few morphemes. Second, most [h]’s alternate with [s] in Rı´o Negro. Prefixal ones appear as [s] before vowels; word-final ones are subject to variable realization as [s] before pause, as discussed in note 3. Root-final /s/, it is true, is usually realized exclusively as [s], since most su‰xes begin with vowels, and most roots require su‰xes, but the occasional consonant-initial su‰x also induces alternation.5 Furthermore, roots that do not require su‰xes when they appear as singular nouns, such as /lus/ ‘light’ and /rais/ ‘root’, alternate between a word-final [h] and [s] before a plural or

204

Ellen M. Kaisse

other vowel-initial a‰x ([luh] ‘light (n)’, [luseh] ‘lights’ (n.pl), [lusir] ‘to light up’). Even a mass noun like /ras/ ‘level’, which virtually never occurs with a su‰x and thus might appear to be a decent candidate for a root with an underlying /h/ that is realized consistently as [h], has a verbal derivative /a-ras-ar/ that shows a surface [s] throughout its paradigms. I conclude that the simplest analysis does not include /h/ in the inventory of any Argentinian dialect. I should also reassure the reader that the word los in the example los ojos is representative of all s-final words in RN, whether they are monosyllabic function words like los, or polysyllabic content words like the plural americanos ‘‘Americans’’ or the singular, s-final place name Caracas. All aspirate their final /s/ even if the next word begins with a vowel. 9.3

Treatment within Sympathy Theory

We have seen how the apparent overapplication of aspiration in RN is handled in a derivational framework with strata. Essentially, we have a ‘‘sandwich,’’ where wordinternal resyllabification precedes word-internal aspiration (aspiration being a lexical rule in RN), but where interword resyllabification follows and opacifies the lexical rule. The solution is unremarkable and entirely workable. Now we must ask how a phenomenon like this would be handled in Optimality Theory. We already know that opacity is generally problematic in OT, and that sympathy theory has been introduced as a method—hopefully the method—with which opacity will be treated. In this section, I show how the transparent portion of the syllabification-aspiration interaction would be handled in OT, why the opaque portion is problematic, and why the architecture of this example makes an insightful solution within sympathy theory unavailable as well. I begin by following McCarthy’s (1999) general method for recasting a phonological process in OT. The existence of a phonologically motivated process whereby an underlying form is not realized perfectly faithfully on the surface is captured in OT by ranking a markedness constraint above a faithfulness constraint. We have three processes to recast: resyllabification between prefix and base, resyllabification between words, and aspiration. Resyllabification processes are typically handled by ranking anchor constraints (a type of faithfulness constraint) lower than constraints favoring unmarked syllable structure. Anchor constraints enforce a matchup between the edges of morphological constituents, present in underlying representations, and the edges of phonological/ prosodic constituents in the output. Resyllabification blurs the boundaries between morphemes or words by placing a consonant belonging to the end of one morpheme at the beginning of a syllable containing the next morpheme so that the syllable and morpheme boundaries do not line up:

Sympathy Meets Argentinian Spanish

(13) a. CV.C]þ[V b. CV.C]a[V

205

Resyllabification between prefix and base Resyllabification between word and word

To start out, I will make the simplifying assumption that the morphological category that a prefix attaches to is always a root, though that root may, in addition, initiate a stem and/or a word. (I sketch a slightly less simplified morphological analysis at the end of this section.) One possible faithfulness constraint that is violated when a prefix-final consonant is resyllabified is (14): (14) ANCHOR (ROOT, s, LEFT) The beginning of a root corresponds with the beginning of a syllable. The markedness constraint that outranks (14) and forces its violation is Onset: (15) ONSET Syllables must have onsets. We do not need to add any more machinery to also translate our second process, resyllabification between words. It also results from mismatching syllable and morpheme boundaries so that a root (this time, a word-initial root) does not begin its own syllable. (Again, I temporarily make the simplifying assumption that all words, like all bases of prefixation, begin with roots.) From the fact that consonants do not always end up syllabified with all and only the material from their own morpheme, I induce the ranking in (16): (16) Onset  Anchor Aspiration is roughly translated as follows:6 (17) *s]coda [s] may not appear in the coda of a syllable. This markedness constraint dominates a faithfulness constraint requiring that the place of an input consonant (coronal, in this case) be realized faithfully on the surface. (18) IDENT (PLACE) The place of an output consonant is identical to that of the corresponding input consonant. Our debuccalization process violates Ident (Place). Since it occurs nonetheless, we conclude (19) *s]coda  Ident (Place) We are now able to construct the tableau needed to account for the transparent case, [desarmar] ‘to disarm’.

206

Ellen M. Kaisse

(20) Generation of transparent form /desþarmar/

Onset

*s]coda

*!

*

des.ar.mar

Anchor (Root, s, Left)

+ de.sar.mar

Ident (Place)

*

deh.ar.mar

*!

*

de.har.mar

*

*!

However, the opaque, interword result cannot be generated. In (21), following McCarthy 1999, the + symbol calls out a problematic candidate that actually wins the evaluation, while + marks the attested output we are trying to derive. In a successful analysis, the + and + candidates coincide and we only need the + symbol. I have simplified the tableau to only look at Ident violations for the first s, in los: (21) Attempt to generate opaque form without sympathy /los oxos/

Onset

*s]coda

los. o.xoh

*!

*

+ lo.so.xoh loh.o.xoh + lo.ho.xoh

Anchor (Root, s, Left)

Ident (Place)

* *!

* *

*!

The reason the opaque form [lo.ho.xoh] cannot win, as McCarthy explains clearly, is that it has a superset of the violations of the transparent [lo.so.xoh]. Both resyllabifying the /s/ and turning in into [h] yields a gratuitous faithfulness violation. To deal with opacity involving overapplication within sympathy theory, we find a member of the candidate set in which aspiration, the opaque, nonsurface apparent process, has applied but where it has not overapplied. That is, we look for a candidate where the environment for aspiration is present. This is the form [loh.o.xoh], which has the [h] from /s/ in coda position. This so-called sympathetic or flower candidate7 is often but not always identical to the form found at the intermediate stage of the serial derivation. If only faithfulness constraints can serve as selectors (McCarthy 1999), then the sympathetic candidate must more closely resemble the input form than does the actual, opaque output. It has failed to undergo the last process in the serial

Sympathy Meets Argentinian Spanish

207

derivation—the one that resulted in opacity. In the case at hand, the sympathetic candidate, [loh.o.xoh], is more faithful than the output because it does not violate the anchor constraint—it did not undergo phrasal resyllabification,8 and so the initial o of oxoh begins both the root and the syllable. This is not the case in the actual output, [lo.ho.xoh]. The faithfulness constraint that picks out the sympathy candidate and eliminates the opaque output temporarily is called the selector, and is marked in tableaux with the symbol ¶. The sympathy candidate is the most harmonic candidate that does not violate the selector. In the case at hand, the selector is Anchor (Root, s, Left). Now how do we choose the actual, opaque output over the transparent candidate? This is done via a high-ranking sympathy constraint enforcing partial identity with the sympathetic candidate. The sympathy constraint and sympathetic candidate are marked in tableaux with the flower symbol {. The sympathy constraint in our case is {Ident (Place), a high-ranked clone of the Ident (Place) that enforced resemblance to the input. {Ident (Place) requires that the output form have [h] by resemblance to the sympathetic candidate, rather than [s] by resemblance to the input. We are now ready to construct the tableau. A check mark calls out those candidates that obey the selector and may thus be the sympathy candidate. (22) Tableau with sympathy: Opaque form correctly chosen as optimal *s]coda

{Ident (Place)

¶Anchor (Root, s, Left)

Faithfulness

los. o.xoh

*

*

B

Transparent form

lo.so.xoh

Sympathy

{ loh.o.xoh

Opaque form

+ lo.ho.xoh

*!

*!

*

Ident (Place)

Onset *!

/los oxos/

B

*

*

*

However, we now cannot derive the transparent prefixþbase form we had no trouble with before the introduction of the selector, ¶Anchor (Root, s, Left).

208

Ellen M. Kaisse

(23) Tableau with sympathy: Transparent form now cannot be chosen *s]coda

{Ident (Place)

¶Anchor (Root, s, Left)

Faithfulness

des. ar.mar

*

*

B

Transparent form

+ de.sar.mar

Sympathy

{ deh.ar.mar

Opaque form

+ de.har.mar

*!

*!

*

Ident (Place)

Onset *!

/desþarmar/

B

*

*

*

We now see why McCarthy’s prediction in (1) holds true. If two processes violate the same faithfulness constraint, and that constraint is the selector for the sympathetic candidate, forms violating the selector can never be chosen as the sympathetic candidate themselves, and thus are forced to resemble the sympathy candidate in whatever way the sympathy constraint requires. Here two notionally distinct processes are intraword and interword resyllabification; the faithfulness constraint they both violate is Anchor (Root, s, Left); and the third process they act together in rendering opaque is aspiration. Both resyllabification of prefixes and resyllabification between words violate Anchor (Root, s, Left). Both are eliminated from the set of sympathetic candidates and both must therefore resemble the sympathy candidate, which has [h] in coda. This is correct for [lo.ho.xoh] but not for *[de.har.mar]. The tableau in (23) thus illustrates an unfortunate property of analyses using sympathy, dubbed ‘‘chaotic interaction’’ by Idsardi (1997). The addition of a sympathy constraint and a selector constraint to account for opaque forms can have unexpected results downstream, making transparent forms that were once easy to generate become nonoptimal because they are not sympathetic to the flower candidate. Of course, in this case, the result is not unexpected at all. McCarthy explicitly predicted that a situation like this would not be treatable within sympathy theory. Before moving on to consider alternative analyses of the Rı´o Negro case within OT, let’s partially remedy the oversimplification of Spanish morphology perpetrated at the beginning of this section. We said that the category to which des- attaches is a root, and also that all words begin with roots. That gave us an easy way to unify the anchor constraint that both forms of resyllabification violate. However, des- actually

Sympathy Meets Argentinian Spanish

209

attaches only to independently occurring words, so it is not obvious that the category to its right, its base, should be called a root. It should more properly be called a word. A form like [de.sar.mar] is in violation of Anchor (Word, s, Left) then, armar being a word. And turning to the interword cases, we should recognize that not every word begins with a root. Many words begin with prefixes: des-armar ‘disarm’, intro-duccio´n ‘introduction’, en-chilada ‘stu¤ed tortilla seasoned with chilies’. Now if a word begins with a vowel-initial prefix, interword resyllabification will not violate Anchor-Root. The phrase las enchiladas, pronounced [la.hen.tSi.la.dah] in Rı´o Negro, has a mismatch between the beginning of the word enchilada and the beginning of a syllable, but that mismatch involves the anchoring of the word, not of its root, chil-. Thus to be accurate, the faithfulness constraint that both intraword and interword resyllabification violate is Anchor (Word, s, Left).9 But let us not get too involved in the details of how best to formulate the anchor constraints. As the next section will show, there are many versions of these constraints and of the sympathy tableaux that I could have constructed—indeed there are far too many. 9.4

Ways Out

The most obvious way to save the sympathy analysis is to split the relevant faithfulness condition into two. Then our two processes, resyllabification within words and resyllabification between words, will not violate the same faithfulness constraint and will escape the prediction in (1). To achieve the sought-after result, let us refer to the right edge of the morpheme containing the /s/ rather than to the left edge of the following morpheme: (24) a. ANCHOR (PREFIX, s, RIGHT) The right edge of a prefix coincides with the right edge of a syllable. b. ANCHOR (WORD, s, RIGHT) The right edge of a word corresponds with the right edge of a syllable. The tableaux in (25) show how the correct forms can now be generated choosing Anchor-Word (24b) as the selector. The correct form, [lo.ho.xoh], has a word/ syllable edge mismatch and thus is eliminated as the sympathy candidate by the selector; it therefore must match the actual sympathy candidate, [loh.o.xoh], in having the apparently gratuitous [h]. However, [de.sar.mar] has no violations of the selector. It is itself, then, the most harmonic candidate that does not violate the selector, and emerges as both the sympathy candidate and the optimal output. (I mark it as both the sympathy and the transparent candidate—it is both the optimal form and, vacuously, the sympathy candidate.) It matches itself with respect to the sympathy constraint, {Ident (Place), and thus emerges with [s].

210

Ellen M. Kaisse

(25) a. Anchor (Word, s, Right) as selector: Interword {Ident (Place)

¶Anchor (Word, s, Right)

los. o.xoh

*

*

B

Transparent form

lo.so.xoh

Sympathy

{ loh.o.xoh

Opaque form

+ lo.ho.xoh

*!

*

*!

Ident (Place)

*s]coda

Faithfulness

Anchor (Prefix, s, Right)

Onset *!

/los oxos/

B

*

*

*

b. Anchor (Word, s, Right) as selector: Prefix þ stem

Transparent form

+ de.sar.mar

Sympathy

{ deh.ar.mar

Opaque form

de.har.mar

B B

*!

Ident (Place)

*

Anchor (Prefix, s, Right)

des.ar.mar

¶Anchor (Word, s, Right)

*s]coda

Faithfulness

{Ident (Place)

Onset *!

/desþarmar/

*

B

*!

B

* * *

*

Mechanically, the solution works. But the underlying problem with this line of attack is that it attempts to replicate the insights of a lexical/postlexical distinction in the absence of an elaborated theory of phonology-morphology interactions in OT. What kind of anchor constraints can there be? We do not know. One can imagine other anchor constraints I might have devised, but again, we do not know whether they would accurately reflect true generalizations about the interaction of morphology and phonology. Perhaps worse, as Bill Idsardi has pointed out to me, there is

Sympathy Meets Argentinian Spanish

211

no way to limit which anchor constraints can be chosen as selectors. In the case at hand, what if we chose Anchor-Prefix instead of Anchor-Word as the selector? This would allow us to generate a language in which word-internal aspiration (or the process of our choice) is opaque but aspiration between words is transparent—a situation that does not occur in natural languages, to my knowledge. The tableau in (26) shows how easy it would be to generate the unnatural result—happily nonreplicable in Lexical Phonology—where [de.har.mar] and [lo.so.xoh] emerge from the evaluation. (26) Anchor (Prefix, s, Right) as selector

Transparent form

+ lo.so.xoh

Sympathy

{ loh.o.xoh

Opaque form

B *

*!

*

lo.ho.xoh

Ident (Place)

*

¶Anchor (Prefix, s, Right)

los.o.xoh

Anchor (Word, s, Right)

*s]coda

Faithfulness

{Ident (Place)

Onset *!

/los oxos/

*!

*

B B

*

B

*

/desþarmar/ Faithfulness

des.ar.mar

Transparent form

de.sar.mar

Sympathy

{ deh.ar.mar

Opaque form

+ de.har.mar

*!

*!

*

*

B

*!

* B

*

*

*

We must conclude that sympathy cannot insightfully handle the case of Rı´o Negro aspiration. What of other mechanisms in the armamentarium of OT? One could handle this case with an output-output condition, requiring that a word ending in /s/ show [h] before a vowel so as to match the [h] that the word has in isolation. However, McCarthy (1999) rightly points out the output-output conditions only work when

212

Ellen M. Kaisse

there is a base form on which to model the opaque form. There happens to be one in this case, but often there is not. Output-output conditions therefore cannot function as the sole means of capturing opaque phenomena in OT, and so are no better than sympathy if our goal is to unify the treatment of opacity with a single, wellunderstood addition to the theory. In general, output-output conditions are not really well understood, because, like earlier theories dealing with analogy, they trade in the ill-understood question of what can serve as a base and under what circumstances bases influence other forms. We might stop at this point and insist that the Harris and Kaisse (1999) treatment of aspiration in RN is the one to stick with. It has the virtue of being explicit, and of meshing correctly with a wealth of other facts about the phonology of Argentinian Spanish, such as glide-consonant alternations, stress assignment, and syllabification. However, there is one variant on OT that can handle this particular case with similar insight to the derivational treatment given by Harris and Kaisse. That is Derivational Optimality Theory (Rubach 2000) and the philosophically similar Stratal OT (Orgun 1996; Kiparsky 1999). Both these theories reject sympathy as a method for dealing with opacity. Instead they select output candidates via a series of evaluations, each evaluation corresponding, more or less, to a stratum of the Lexical Phonology model.10 In the case at hand, we will first derive [de.sar.mar], [loh], and [o.xoh] as outputs of the word-level evaluation; we will then take these outputs as the new inputs to the phrase level. (27) Word-level derivational OT tableau a. /desþarmar/ des.ar.mar

Onset

*s]coda

*!

*

+ de.sar.mar deh.ar.mar

Ident (Place)

* *!

*

de.har.mar b.

Anchor (Root, s, Left)

*

*!

/los/ los

*!

+ loh

*

/oxos/ o.xos

*

+ o.xoh

*

*! *

Sympathy Meets Argentinian Spanish

213

Nothing further happens to [de.sar.mar] at the phrasal level, and it emerges as the optimal form. However, at the phrasal level, the outputs [loh] and [o.xoh] are strung together and it becomes possible to fulfill the onset constraint by violating Anchor (Root). The opacity emerges naturally from the tableau, without sympathy or other additional devices, because the input form for los already has /h/—there is no gratuitous faithfulness violation, as there was in the parallel OT derivation. (28) Phrase-level derivational OT tableau /loh.o.xoh/ loh.o.xoh

Onset

lo.so.xoh

Anchor (Root, s, Left)

Ident (Place)

*!

+ lo.ho.xoh los.o.xoh

*s]coda

* *!

* *

*!

Successful as this translation to derivational OT is, we must not forget McCarthy’s (1999) observation: not all cases of opacity result from well-motivated stratal segregation. There are numerous cases in the derivational literature that at least appear to involve opaque interactions within a single stratum. Therefore, the nascent theory of derivational OT has a burden to bear—it must show that all genuine cases of opacity result from stratal organization. Should this prove impossible—and my guess is that it will—we are back to the conclusion that a unified treatment for opacity is not currently available within Optimality Theory. Notes An earlier version of this chapter was presented at the First North American Phonology Conference. This chapter fleshes out a brief suggestion made in Harris and Kaisse 1999 concerning the di‰culties of dealing with the Rı´o Negro dialect of Argentinian Spanish within OT in general and within sympathy theory in particular. Thanks to Sharon Hargus, Jim Harris, Bill Idsardi, Susannah Levi, and other members of the audience at NAPHCI. The following abbreviations occur throughout the text: pl ¼ place; cor ¼ coronal; n ¼ noun; n.pl ¼ plural noun. 1. Actually, McCarthy reserves locally conjoined constraints as the treatment of choice for one kind of opacity—counterfeeding on focus, as originally proposed by Kirchner (1996). This type of opacity results from ‘‘chain shifts,’’ where y becomes z, and x becomes y, but new y’s do not go on to become z’s. The reason for this exception to the general sympathy treatment is not worked out in either McCarthy 1999 or 2003. 2. This description oversimplifies the phenomenon, which is described at greater length in Kaisse 1996, 1999, as well as in Harris and Kaisse 1999 and references therein. Among other details not directly relevant here, the phenomenon is variable in many dialects, can involve

214

Ellen M. Kaisse

near or total loss of any aspiration, may require a preceding nonconsonantal segment, and so forth. 3. Realization as [h] before pause is variable in RN—sometimes word-final /s/ emerges as [s] in that context. This variation is almost certainly due to the normative e¤ect of the prestige dialect, that of Buenos Aires, where /s/ is always [s] before pause. 4. The base to which a prefix attaches may be a root, a stem, or a word. des- attaches to words. The prefix /kon-/ illustrated in (10) is attaching to a stem; iuxe does not appear in isolation. 5. For instance, the morpheme /dies/ ‘ten’ appears in [djeh.þmar] ‘decimate’, [djeh.þmo] ‘tithe’, [djeh] ‘ten’, and [dje.sþes] ‘tens’. 6. I am merely aiming for a quick, schematic view of this process. A more insightful translation of aspiration would involve a member of the family of Coda Conditions—constraints on what may appear in coda and, in particular, the inability of codas to license place on their own (Coda Condition—*Place). Indeed, as discussed in Kaisse 1996, the placeless h often takes on the place of a following velar consonant ([pax.kwal] for ‘Pascual’), indicating that a shared dorsal place node is permitted in the coda. 7. The term ‘‘sympathetic’’ candidate is confusing, in my opinion, because it refers to the flower candidate that the output, opaque form sympathizes with, not to the output form that exhibits sympathy by coming to resemble the flower candidate. One might more logically refer to the flower candidate as the pathetic form—the one that compels sympathy. 8. In cases involving anchoring, resemblance to the input is at one remove. The input does not contain syllabification per se, but it does contain discrete morphological elements. The sympathetic candidate resembles the input by better matching syllable edges to input morpheme edges. 9. Though not every prefix attaches to words, des- is virtually the only s-final prefix in Spanish, and it is a word-level prefix. (The rare and unproductive prefix dis- does not attach to any vowel-initial roots, as far as I can determine.) 10. The stratal OT material by Kiparsky available to me at this writing is more concerned with the stem/word distinction than the lexical/postlexical (word/phrase) distinction. The work of Rubach is thus a more direct source for the proposal advanced here. The ranking of constraints can vary between the strata. However, reranking does not appear to be necessary in this example.

10

Vowel Length, Cyclicity, and Output-Output Correspondence

Cemil Orhan Orgun

10.1

Introduction

For about three decades, cyclicity and level ordering enjoyed a privileged theoretical position in Generative Phonology in accounting for synchronic alternations reflecting morphological relatedness between forms (Chomsky, Halle, et al. 1956; Chomsky and Halle 1968; Kean 1974; Allen 1978; Kiparsky 1982b; Mohanan 1982; etc.). Recently, however, alternatives to cyclicity based on the notion of paradigm uniformity (the idea that morphologically related forms are grammatically constrained to be phonologically similar) have come to the forefront of phonological research. Particularly popular among these is the Output-Output Correspondence approach within Optimality Theory (Benua 1997; Burzio 1994; Kenstowicz 1995; McCarthy 1996; Steriade 1997). A number of researchers, however (Dolbey 1996, Dolbey and Orgun 1996, Booij 1996, Kiparsky, forthcoming), have argued that traditional interleaving (cyclicity or level ordering) is empirically or theoretically superior to paradigm uniformity as a means for accounting for synchronic alternations. In this chapter, I present data from English and Turkish vowel length alternations that seem to favor the traditional interleaving approach (cyclicity and level ordering) over Output-Output Correspondence. The English and Turkish data that I discuss show that interleaving e¤ects do not necessarily result from the interaction between related word forms, as paradigm uniformity posits. Rather, individual morphological constructions may use either a lexical stem or a word as the base to which the given morphological process is to be applied. Even when a given root is free (can surface as a word without further morphology), a morphological construction may use its stem form rather than its surface (word) form. Such cases crucially fail to exhibit output correspondence e¤ects.

216

10.2 10.2.1

Cemil Orhan Orgun

English Vowel Length Alternations Data

In addition to well-known morphophonemic alternations involving length as well as quality, such as Closed Syllable Shortening (wide [aI] versus width [I]) and Trisyllabic Shortening (opaque [eI] versus opacity [æ]), English vowels undergo allophonic variation in length alone. Introductory textbooks often portray this alternation in deceptively simple terms, for example, as ‘‘the general rule in English whereby a vowel is longer if the following consonant is a voiced sound’’ (Kenstowicz and Kisseberth 1979, 299). I present a fuller set of facts (based on work with several consultants, all native speakers of American English from California), which prove to have significant implications for the phonology-morphology interface. The basic facts in American English are that vowels are long before tautosyllabic voiced consonants and in final position, short otherwise. (1) a. Long high hide

[a:I] [a:I]

b. Short hybrid height

[aI] [aI]

Vowel length interacts with morphological structure in interesting ways. The relevant phenomena will be illustrated in this chapter for vowels in open syllables. However, an examination of compounds with vowel-final first members shows that the vowel in question is short in some compounds and long in others. In my work with native speaker consultants, I have found that the exact set of compounds with short vowels varies from speaker to speaker. However, all of my speakers have some compounds with long vowels and some with short. The particular examples presented in this chapter reflect the speech of one consultant. (2) a. Long dry mop pie pan lie detector

[a:I] [a:I] [a:I]

b. Short high chair high school skyscraper

[aI] [aI] [aI]

Why are long vowels found in nonfinal position in the first members of the compounds in (2a)? Alternatively, if the final vowel of the first member of a compound is in ‘‘final position’’ for the purposes of vowel length, why are short vowels found in (2b)? 10.2.2

A Lexical Phonological Account

In the framework of Lexical Phonology (Pesetsky 1979; Kiparsky 1982b,d; Mohanan 1982), an immediate solution is available. The lexicon is divided into at least two strata, corresponding to the traditional Levels 1 and 2. Vowel lengthening

Vowel Length, Cyclicity, and Output-Output Correspondence

217

applies on Level 2 but not Level 1. For the purposes of this chapter, we may assume Level 1 to be the ‘‘stem’’ level and Level 2 to be the ‘‘word’’ level, although a full account of English phonology and morphology would be much more complicated. Compounds may be formed at either level. (3) Level 1 Level 2

Phonology

Morphology

No vowel lengthening Vowel lengthening

‘‘Stem’’ compounding ‘‘Word’’ compounding

Typical derivations show how the full range of vowel length alternations receive a satisfactory account in this system: (4) Level 1 Level 2 Surface

high

high chair

dry

dry mop

high [a:I] — high [a:I] — high [a:I]

high chair [aI] — — — high chair [aI]

— — dry [a:I] — dry [a:I]

— — dry [a:I] dry mop [a:I] dry mop [a:I]

Note that we need to assume, following Inkelas 1989, a prea‰xal cycle on Level 2. Whether we have a prea‰xal (‘‘root’’) cycle on Level 1 or not is immaterial to the discussion here, as Level 1 has no vowel lengthening. I assume one for the sake of consistency. While the theoretical details of this account may or may not be acceptable to a given linguist, there is an obvious insight that this approach reflects. Words but not stems need to end in long vowels in American English. Stems as well as words may be compounded. When two stems compound, the final vowel of the first stem is not subject to lengthening because it is not a word. When two words are compounded, the final vowel of the first word is of course word final. Accordingly, it lengthens. The crucial insight here is that morphological constructions may call for a stem form or a word form. 10.2.3

Output-Output Correspondence

The Output-Output Correspondence approach to almost-allophonic alternations thwarted by morphological structure assumes that markedness constraints in the relevant dimension outrank competing faithfulness constraints (Benua 1996 for Sundanese; Kenstowicz 1995 for Italian). In English, this would mean ranking constraints that govern the distribution of long and short vowels below constraints requiring any

218

Cemil Orhan Orgun

underlying vowel length to be preserved. This way, regardless of whether vowels are underlyingly long or short, we are guaranteed a long vowel in high (word-final lengthening) and hide (lengthening before a voiced coda) and a short one in hybrid and height (neither word-final lengthening nor lengthening before a voiced coda are applicable). In order to account for the long vowel in pie pan, we posit an OutputOutput Correspondence constraint that requires vowel length to be the same in morphologically related words. This constraint outranks the markedness constraints governing the distribution of vowel length. (5) Output-Output faithfulness to vowel length a.

Base: none Input: hide

OO-Faith

hide

V-Length

IO-Faith

*!

(*)

+ hi:de b.

(*)

Base: pie, pan Input: pie pan pie pan + pie: pan

*!

(*) *

(*)

According to a common assumption in Optimality Theory called Richness of the Base (Prince and Smolensky 1993), there are no direct constraints on underlying forms. Thus, forms such as pie and hide may have underlyingly long or short vowels. Depending on which is chosen, there will be di¤erent IO-Faith violation patterns (hence the parenthesized asterisks in the tableau). However, since IO-Faith is never the ruling constraint, this does not matter. Since underlying length contrasts do not have an e¤ect on the surface forms, there will not be a vowel length contrast in English. However, OO-Faith constraints outrank well-formedness constraints. Therefore, we find unexpectedly long vowels in some morphologically derived forms when the form’s base contains a long vowel, as in (5b). But then what about compounds with short vowels? Why do we have a short vowel in high chair? Is high not a base here? Why not? Would one have to abandon the claim that there is a uniform family of Output-Output Correspondence constraints and make up special constraints for each morphological construction? Level 1 (stem) compounds would then correspond to their bases via Level 1 correspondence constraints, and Level 2 (word) compounds would correspond to their bases via Level 2 correspondence constraints. This move, of course, would amount to abandoning the enterprise: we could no longer say that cyclic phonological e¤ects are

Vowel Length, Cyclicity, and Output-Output Correspondence

219

due to general paradigm uniformity constraints. In fact, Benua (1996) has remarked that Output-Output Correspondence is superior to phonology-morphology interleaving precisely because it uses one family of general correspondence constraints rather than construction- (or stratum-) specific constraints. But if the facts make using one family of constraints untenable, that argument is lost. Thus positing constructionspecific constraints is simply a thinly veiled reconstruction of Lexical Phonology. One might as well admit that Lexical Phonology is the right approach. It is clear that Output-Output constraints do not help. Do they have to hurt though? If one could make up an even higher-ranking constraint to circumvent the undesirable e¤ects of Output-Output Correspondence, one would simply be admitting that Output-Output Correspondence does not account for English vowel length. This would not be too high a price to pay. After all, why should Output-Output Correspondence account for everything? 10.2.4

A Prosodic Account

The basis of the Lexical Phonological account was a distinction between stems and words implemented by means of level ordering. By translating this morphological account into prosodic terms, we can account for the alternations by surface wellformedness constraints alone. We assume that ‘‘stem compounds’’ consist of a single prosodic word (w), while ‘‘word compounds’’ contain two. (6) Bare forms Stem compound Word compound

[hide]w , [pie]w [high chair]w [[pie]w [pan]w ]w

With these representations, everything follows from surface prosodic structure. Interleaving and Output-Output Correspondence are simply irrelevant. However, this account in turn su¤ers from a serious defect. How are the appropriate prosodic structures assigned to these forms? In a framework that includes interleaving, this problem is not present. According to Inkelas 1989, prosodic structure assignment takes place cyclically. Thus, pie pan, pie, and pan undergo a cycle on which they are assigned word status. On the next cycle, they are compounded and the whole form is assigned (recursive) word status. Short vowel compounds such as high chair must undergo compounding at the stem level (Level 1). As a result, the individual members high and chair are not prosodic words and therefore not subject to vowel lengthening. The compound high chair is assigned word structure on Level 2 as a whole, where vowel lengthening does not apply to high since only prosodic word-final vowels are subject to lengthening. Of course, if prosodic structure assignment is done cyclically, there is little point in insisting that vowel length is determined noncyclically. Proponents of Output-Output Correspondence would therefore need to find a principled way of assigning the

220

Cemil Orhan Orgun

appropriate prosodic structures to these English forms noncyclically, perhaps by using Alignment (McCarthy and Prince 1993a). 10.3

Judgment Day

While it so far seems that the prosodic account could work if principled means of assigning the right structure could be found and the argument against Output-Output Correspondence could vanish, there are two kinds of evidence in favor of the interleaving account. The first is morphological and language internal; the second is phonological and crosslinguistic. 10.3.1

Stem versus Word Compounds

It has long been known that there is a compounding construction in English that uses irregular plural nouns (Marchand 1969). For nouns with regular plurals, the ‘‘singular’’ (that is, bare stem) form is used instead. (7) Singular mouse rat

Plural mice rats

Compound mice-infested rat-infested

There are also compounds using singular forms of regular or irregular nouns (e.g., mouse trap); these are irrelevant to our purposes. Less widely acknowledged is the fact that some compounds do use regular plural forms. (8) Singular form in compound sport jacket pants pocket

Regular plural form in compound sports pages pant press1

What is the di¤erence between these two kinds of compounds? A natural answer is that the kind in (7) is a stem-compounding construction, while the kind in (8) is word compounding. Irregular plurals are lexically listed stems, while regular plurals are inflected words (with a word-level su‰x). The theoretical tools used in this account are identical to those used in the Lexical Phonology account of vowel length alternations. Thus, Lexical Phonology unifies two apparently separate phenomena: regular versus irregular plurals in compounds, on the one hand, and long versus short vowels in su‰xed forms and compounds, on the other. To the extent that such unification of apparently disparate phenomena is an indicator of theoretical success, the Lexical Phonology account is supported. 10.3.2

Turkish Vowel Length

As Inkelas and Orgun (1995) show, Turkish has vowel length alternations that are quite similar to those that I have presented for English, although the environment

Vowel Length, Cyclicity, and Output-Output Correspondence

221

for vowel lengthening is more restricted. Due to a minimal size condition of two moras, vowels of CV roots lengthen. Long vowels are found in una‰xed as well as a‰xed forms of these roots: (9) Turkish orthography do doyu

IPA (with morpheme breaks) do: do:-ju

Gloss ‘musical note C’ ‘C-accusative’

However, these roots surface with short vowels in some compounds: (10) do bemol do mino¨r

do-bemol j do-minør

‘C flat’ ‘C sharp’

Within Lexical Phonology, these facts receive a treatment parallel to that of their English counterpart. Two levels are needed; we may call them stem and word. Shortvowel compounds are formed at the stem level.2 Vowels lengthen at the word level, where a‰xes are also added. As in English, we need to assume a prea‰xal cycle. (11) Surface (orthography in parentheses) Underlying Stem level Word level

do: (do)

do:ju (doyu)

dobemol j (do bemol )

do do do:

do do do:-ju

do, bemol j do-bemol j —

Although Turkish seems to be similar to English, there turns out to be a crucial twist. It can easily be shown that the prosodic approach that could come to the rescue of Output-Output Correspondence in English is not available in Turkish. Let us first see what a prosodic analysis of Turkish vowel lengthening would look like. On a parallel analysis, we could assume that short-vowel compounds contain one prosodic word, while a‰xed forms contain an embedded prosodic word: (12) [do:]w [[do:]w ju]w [dobemol j ]w Given the generalization that vowel lengthening applies in order to bring prosodic words to the required minimum size of two moras, it is crucial that the accusative su‰x attach to prosodic words. If we found evidence that the base of a‰xation is not a prosodic word, the motivation for vowel lengthening would be lost, and we would have to conclude that the prosodic analysis does not work. Since the prosodic analysis is crucial to saving Output-Output Correspondence, we would have to conclude that Output-Output Correspondence faces an insurmountable challenge. Lexical Phonology would emerge as the clear winner.

222

Cemil Orhan Orgun

In Turkish, coda clusters must obey Sonority Sequencing. Underlying clusters that do not obey Sonority Sequencing are repaired by vowel epenthesis. If a vowel-initial su‰x follows a root ending in such a consonant cluster, such as /resm/ ‘picture’, epenthesis does not apply, as the second consonant in the cluster is able to syllabify as an onset. (13) resim resmi

resim resm-i

‘picture’ ‘picture-accusative’

These forms show that the root cannot be a prosodic word in su‰xed forms: (14) *[[resm]w i]w Since [resm] is not a possible prosodic word, the embedded prosodic word representation is ill formed. The only escape would seem to be to claim that prosodic words are redefined according to syllable boundaries: (15) [[res]w mi]w Introducing a slight phonology-morphology mismatch of this kind would not in itself be an unreasonable theoretical move. Unfortunately, however, it would have the undesirable consequence of predicting vowel length where none is found. In particular, what happens when a su‰x is added to a CVC root, such as /kin/ ‘hatred’? The prosodic word account predicts vowel lengthening in this case. Since we concluded that prosodic words are redefined according to syllable boundaries, we must have the structure in (16): (16) [[ki]w ni]w In fact, no vowel lengthening takes place in CVC roots followed by vowel-initial a‰xes: (17) kin kini

kin kin-i

‘hatred’ ‘hatred-accusative’

However, the representation in (16), corresponding to the data in (17), is ill formed! If vowel lengthening takes place in (12) because of prosodic minimality, it must take place in (16). The prosodic word in (16) is clearly monomoraic. In addition, the representations in (16) and (17) make the unfortunate prediction that the first syllable in these forms will bear stress, as every prosodic word bears stress in Turkish. In fact, stress is final in these forms. Prosodic restructuring would present a similar problem in English.3 If syllables are redefined in a compound like rice paper, a complex onset [sp] will be formed, and the [p] of paper should therefore be unaspirated. This, however, is not the case; the [p] is fully aspirated.

Vowel Length, Cyclicity, and Output-Output Correspondence

223

The inevitable conclusion is that the domain for vowel lengthening is a morphological stem, not a prosodic constituent of any sort. Therefore, the proper account of the vowel lengthening data must be based on interleaved phonology-morphology interaction in the style of Lexical Phonology. 10.4

Conclusions

One of the major observations in phonology-morphology interaction is that bound roots are never subject to cyclic phonological e¤ects. Output-Output Correspondence captures this by observing that bound roots are by definition not pronounced as independent words and therefore cannot stand in correspondence. The flip side of this is that free morphemes are possible independent words and therefore stand in correspondence with related forms and give rise to cyclic phonological e¤ects. In this chapter, I have shown that being a free morpheme does not guarantee cyclic e¤ects. The grammar is free to choose the root form or the word form for such a morpheme. Even within a language, this choice might vary from construction to construction. Thus, English Level 1 a‰xation and compounding call for the root high [haI], while Level 2 a‰xation and compounding call for the word high [ha:I]. Output-Output Correspondence has no additional insight to o¤er here; the best we can do is try (somewhat unsuccessfully) to circumvent problems caused by the framework by using other mechanisms, such as ad hoc assignment of prosodic structure. Lexical Phonology not only handles the facts with ease, it also relates them to independent morphological facts having to do with presence or absence of inflectional morphology in compounds. Notes Many thanks to Bob Ladd, who first brought the English vowel length phenomenon to my attention. Without his observation of the short vowel in high school, this chapter would have never come into existence. I also thank Sharon Inkelas, Carlos Gussenhoven, Andrew Dolbey, and the attendees of a UC Berkeley colloquium and a Stanford phonology workshop for valuable comments. 1. Attested in a mail-order catalog. 2. Inkelas and Orgun 1995 in fact shows the need for at least four, while Inkelas and Orgun 1998 argues for five. 3. I thank Carlos Gussenhoven for bringing this fact to my attention.

11

Level Ordering in Nuuchahnulth

John Stonham

11.1

Introduction

One of the primary contributions of Lexical Phonology and Morphology (henceforth LPM) to linguistic theory is the use of stratal separation to explain behavior shared by only a subset of the elements of the lexicon within a language. The ordering of strata allows linguists to apply phonological rules to the result of combining one group of morphemes with another while excluding other nonparticipating groups of morphemes. Perhaps the best-known example is that of the application of the English stress rules at only the first stratum, discussed, inter alios, by Kiparsky (1982b, 1982d, 1985) and Mohanan (1982, 1986), but for every language one might expect to find particular phenomena indicative of such stratal membership. For any given language, the number of strata will depend on factors such as the complexity and interaction of the phonology and morphology, but a guiding principle should be that the minimum number of levels necessary for explaining the facts is to be preferred.1 Assuming this principle and the fundamental tenets of LPM, we will now proceed to the structure of the word in the Nuuchahnulth language of Vancouver Island, Canada, examining various phenomena that provide clear evidence of the need to separate morphemes into strata in the grammar. There are a number of areas of Nuuchahnulth grammar where there appear to be distinct di¤erences in the application of rules. Some of these include       

Glottalization e¤ects Delabialization t´ ! 2  deletion Stress assignment Bound root/free form distinctions Su‰x combining forms

We will consider each of these cases in turn, paying particular attention to the possible arguments for di¤erent domains of application for phonological and

226

John Stonham

morphological rules in Nuuchahnulth and, in the end, arguing for a two-strata model of the morphology. 11.2

Derivation/Aspect versus Inflection

There are strong arguments for dividing Nuuchahnulth morphology into two sets, as first suggested in Sapir and Swadesh (1939) (henceforth S&S): (i) basically derivational and aspectual a‰xes, labeled ‘‘formative’’; and (ii) ‘‘incremental’’ a‰xes, as described by S&S.3 In fact, no great weight should be placed on the significance of these basically descriptive terms, but the segregation of morphemes into di¤erent groups is the crucial issue, as will be demonstrated in what follows. It should also be noted that, while in large part the analysis of morphemes that is presented here will be the same as that of S&S, there will also be some di¤erences based on more recent reanalysis of the data. The distinction between formative and incremental morphemes is relatively robust in Nuuchahnulth. The distribution is marked most clearly in the 500 or more su‰xes of the language, where typically the order of derivational su‰xes is relatively free, being based on the semantics of the construction, but su‰xes occurring after the derivational and aspectual ones are rigidly ordered, following the order presented in table 11.1.4 Additional criteria for the division include the category-changing nature and lexicality of derivational su‰xes and the paradigmaticity of incremental morphemes. As Table 11.1

Incremental su‰xes Nonparadigmatic

Paradigmatic

1

2

3

4

5

6

7

8

9

10

11

m’in¡ pl

is dim

q¡ mw

’aqt´ intent

’ap caus

’at´ now

’at pass

mit past

cnd

qt

infer

’at inal(ienable) uk/ak poss(essive) ’aa¡ irr(ealis)

ind(icative) rel(ative) sub(ordinate) dub(itative) rel(ative) dub(itative) abs(olutive) inter(rogative) purp(osive) indef(inite) rel(ative) pres(ent) imp(erative) fut(ure) imp(erative)

Level Ordering in Nuuchahnulth

227

for linear order, derivational su‰xes invariably appear closer to the root. As an example of the productivity of a‰xation in Nuuchahnulth, note the following.5 (1) aaat´qim´¡timji´m’in¡aqt´itsu dup-dup-at´-qim´-¡ta-ma´-’i´ k -m’in¡-aqt´-m‡itsu rep-suf-two-units-on foot[r]-moving-in house-pl-intent-2p.ind ‘You carry two dollars on your feet.’ As can be seen from this example, a‰xation is very productive in Nuuchahnulth. Reduplication, which may perform such tasks as marking certain types of lexical suffixes and indicating distributivity and repetitive aspect, is also common, as demonstrated by this example. The innermost copy is a concomitant of the su‰x marked [r] (reduplication), and the second indicates the repetitive aspect of the form. There is no prefixation or compounding (but see Stonham 1999a). One complication arises with respect to aspectual markers, which typically appear between derivation and inflection, as demonstrated by the following examples. (2) a. tSutSuk w ainmasnakSiat´ma tSutSuk w ainma-’as-na‡k w -Sit´ k -at´-ma‡ b. hask w a´Sit´itwin hask w a´-Sit´ k -mit-w‡in

‘they had people go about the village inviting’ invite-outside-have-mom-now-3s.ind ‘it was nearly . . . ’ nearly-mom-past-3s.qt

However, further evidence demonstrates that it is possible for such aspect markers to occur intertwined with the derivational su‰xes, although never with inflection, as in the examples in (3). (3) a. ¡isSit´mapt ¡is-Sit´-mapt b. ´utSnakSit´maiqstuat´a¡ ´utSna‡k w -Sit´-maiqstut´ k -’at´ -ma‡¡

‘species of plant’ bleed-mom- . . . plant ‘I want to get married’ marry-mom-want to (mom)-now1s.ind

This evidence supports S&S’s decision to combine derivation and aspect into a single class of ‘‘formative’’ morphemes and will be used here to support the decision to combine these elements on a single stratum. A further issue is the status of certain su‰xes presented in table 11.1. Su‰xes marking plural nouns, -m’in¡, and diminutives, -is, may arguably be considered derivational, although we will see that they fit better into the category of stratum 2 a‰xes in any case. This highlights the common observation that a distinction based solely upon the derivation/inflection division often runs into di‰culties that are not encountered by a simple stratal distinction (e.g., Hargus 1989; Inkelas and Orgun 1995).

228

11.3

John Stonham

Arguments for Stratal Separation

In what follows, various phenomena will be presented that serve to clarify the issue of stratal segregation. Some of these cases are phonological in nature, while others may more accurately be described as morphological. Some indicate a need for a separate level of (quasi)inflectional morphology, while others provide evidence for a combined level of derivation and aspect. All go to support the hypothesis that Nuuchahnulth morphology must be divided into two distinct strata. 11.3.1

Glottalization

Glottalization involves the interaction of a specific subset of su‰xes with a preceding segment. The outcome will vary depending on the segment involved but typically results in the change of stops to glottalized stops, fricatives to laryngealized glides, and sonorants to their laryngealized counterparts. The chart in (4) illustrates the various changes that may occur.6 (4) Plain p t ts t´ tS k k w q qw s´S ´w xw ¯w ¡w mnwj x¯¡

!

Glottalized p’ t’ ts’ t´’ tS’ k’ k’ w  j’ w’ m’ n’ w’ j’ —

Basically, the glottalization is ‘‘triggered’’ by some characteristic of the su‰x that S&S represent by a glottalizing mark /’/ at the beginning of the appropriate su‰x. Such su‰xes are reasonably common in the language and instances of the process are frequent. They all begin with a vowel, although not all vowel-initial su‰xes trigger glottalization. How one should appropriately distinguish those su‰xes that do from those that do not cause this process is perhaps an issue for further debate elsewhere. However one does it, it will obviously be necessary to make some distinction between the two categories of su‰xes, and for the purposes of this chapter the convention used in S&S will be employed. Examples of su‰xes that trigger glottalization (5a) and those that do not (5b) are provided below. (5) a. -’um -’it´ -’at´ b. -im -it´ -at´

‘ . . . on rocks’ ‘invite . . . ; go for . . . ’ now ‘ . . . thing’ mom caus ‘ . . . receptacle (in Deer’s speech)’

Level Ordering in Nuuchahnulth

229

Examples of how these su‰xes combine with preceding elements are provided below. The su‰xes in (6a) are all glottalizing su‰xes, while those in (6b) are not. (6) a.

i. wiSk’um wiSk-’um ii. hita¡t’it´Siat´ hita-¡t-’it´-Sit´ k -’at´ iii. qa¡nak’at´7 qa¡-na‡k w k -’at´ b. i. titinkum8 dup-ti-n’uk w -im ii. ajimkit´qas aja-mik w -it´ k -qa‡s iii. k w an’ux w at´ k w an’ux w -at´

‘angry on the rocks’ angry- . . . on rocks ‘they started out of the woods’ loc-exit woods-go for[l]-mom-now ‘he was dead now’ dead-have . . . -now ‘handwiper’ suf-wipe-in hand[r]- . . . thing ‘may I be a getter of many . . . ’ many-getter of . . . -mom caus-1s.sub ‘chamber-pot (in Deer’s speech)’ urinate (female)-receptacle

As can be seen from these examples, the result of the combination of a glottalizing su‰x with a preceding stop/a¤ricate-final base is the creation of a glottalized version of the consonant. Where there is no available appropriate version, as in the case of fricatives, which have no glottalized counterparts in Nuuchahnulth, the nearest possible consonant is provided, namely, for labialized fricatives, a laryngealized labial glide, /w’/, and for coronal fricatives, a laryngealized coronal glide, /j’/. The data follow the pattern below. (7) a. t´iw’inap t´ix w -’in k -’ap b. atuj’isap’anitin a:tuS-’i‡s k -’ap-’at-mit-min c. mat´itsuaj’in mat´-wits-jua´-’in d. Saw’aat´ Sax w -’aa k -’at´

‘he laughed at him’ laugh-sound of . . . -caus ‘they gave us deer meat to eat’ deer-eat-caus-pass-past-1p.ind ‘had been seen wearing it around his head’ tied-around head[l]-perceive-treated as ‘they fled’ flee-do for . . . sake-now

When the base ends in a consonant that is not a possible candidate for glottalization, such as /x ¯ ¡/, then no change occurs to the fricative and // is inserted. (8) a. cici¡aqt´mapt dup-ci¡-’aqt´-mapt b. tSijanuxat¡ tSijanux-’at¡

‘crab-apple wood’ pl-sour-inside- . . . plant ‘the Chiyaanuh Tribe’ Chiyaanuh- . . . tribe

Where the base ends in a vowel, the result is similar to that above, with the insertion of a glottal stop between the two vowels preventing coalescence.

230

John Stonham

(9) a. ink’ w aap ink w -’a‡a k -’ap b. taktSiinim tak-tSi k -’i‡nim

‘he made a fire’ fire-in hearth-caus ‘let us each do . . . ’ each-along with . . . -1p>3obj.fut

The most interesting aspect of this process with respect to stratal separation arises with regard to the fricatives. When a fricative constitutes part of an incremental morpheme, the concomitant glottalization does not occur. At this later level of morphology, there will be no change to produce laryngealized sonorants. S&S observe that ‘‘while glottalizing formative su‰xes . . . change voiceless fricatives to glottalized semivowels, glottalizing incremental su‰xes a¤ect only non-glottalized stops’’ (p. 236). Thus, the rule applies in the same fashion to stops and a¤ricates at either level, but with fricatives there is a distinct di¤erence. Rather than posit systematic exceptions to the application of the rule, we may capture the regularities of the process by the introduction of rule di¤erences associated with stratal membership: in short, the phonological rule of glottalization occurs at both strata with somewhat di¤erent environments (see Kiparsky 1985). As a demonstration of this distinction, examine the results of the application of the su‰x -’at´ ‘now’, a clear example of an incremental su‰x (position 6 in table 11.1), compared with the su‰x -’a‡a ‘on the rocks’, an obviously derivational morpheme. (10) a. i. hiniasat´ hina-ias k -’at´ ii. ujua´at´ u-jua´ k -’at´ b. i. nawaj’aa nawa‡s-’a‡a ii. hij’aaq¡aqt´itq hi´-’a‡a k -q¡-’aqt´-i‡tq

‘leaving the house’ loc-go outside-now ‘she noticed . . . ’ ref-perceive-now ‘while sitting idly chatting on the rocks’ sit idly chatting-on rocks ‘to the rocky point where they were to . . . ’ loc-on rocks-being-intent-3s.rel

From this evidence we can clearly see the distinction between the treatment of fricatives within the so-called formative component and in the incremental component. This suggests the need for two strata within the morphology of the language, with two alternative formulations of the environment for the rule of glottalization, depending on the stratum where the rule takes place. 11.3.2

Delabialization

One of the conditioning factors for the loss of labialization of rounded velars and uvulars is the boundary between the two strata, at which point delabialization occurs. This suggests a type of ‘‘exit feature’’ that applies at the end of stratum 1.

Level Ordering in Nuuchahnulth

231

This would be possible only if there were a stratal distinction in order to provide the means to apply the rule. In Nuuchahnulth there is an underlying distinction between round and unround velar and uvular obstruents as shown below.9 (11) ka ki´ qi t´’a¯ ak

‘measure’ ‘lift up canoe’ ‘shift position’ ‘vertically flat’ ‘dog-salmon’

k w a k w i´ q wi t´ax w ak w

‘move backward’ ‘blow spray’ ‘that which’ ‘carved’ ‘whittle’

When rounded consonants appear at the end of the first stratum, they will be delabialized, necessitating a distinction between levels in the grammar. This is illustrated by the examples in (12). (12) a. qa¡nak’at´ qa¡-na‡k w k -’at´ b. hisik’atwin his-i‡k w k -’at-w‡in

‘he was dead now’ dead-have . . . -now ‘they passed by’ loc-go along . . . -pass-3s.qt

Compare the examples above with those in (13) in which the labialization remains intact. (13) a. ¡ajutSink’ w aa ¡aju-ink w -’a‡a b. upa´nak w inj’ap’at upa´-na‡k w -inj’u k -’ap-’at c. jasik’ w as jas-i‡k w -’as

‘Ten-together-on-Rocks (man’s name)’ ten-together-on rocks ‘after giving me my club’ club-have . . . -left behind-caus-pass ‘someone going along . . . ’ there-go along . . . -outside

In all of the cases with labialization, the immediately following su‰x is derivational (13), while in those in (12) it is incremental, indicating that the rule applies at the end of stratum 1. Clearly, delabialization may be used as a test of stratal separation, providing further evidence of the bistratal nature of Nuuchahnulth morphology. 11.3.3

t´ ? 

The application of this rule, which merges a lateral a¤ricate with a following glottalizing su‰x resulting in a glottal stop rather than the anticipated glottalized lateral a¤ricate /t´’/, is again confined to the boundary of the two strata, further supporting a di¤erence between the two levels of Nuuchahnulth morphology. The /t´/ in these cases is typically part of a morpheme associated with the momentaneous, inceptive, or graduative aspects and is therefore attached prior to the

232

John Stonham

attachment of the triggering morphemes /-’ap/, /-’at´/ and /-’at/, and /-’a¡/, which are found in positions 5, 6, and 7 in table 11.1. The examples in (14) demonstrate the application of this rule. (14) a. ¡at’uk w iap ¡a-at’ut´-Sit´ k -’ap b. naj’aqnakSiat´ naj’aqnak-Sit´ k -’at´ c. ´u¡tqapiatqa ´u¡-tqa-’ipit´ k -’at-qa‡ d. ak’uatwin ak’ut´ k -’at-w‡in e. ¡isi´itSia¡at´quwin ¡i-si´a-itSit´ k -’a¡-’at´-qu-w‡in

‘making it (burn) bright’ complete-on fire-mom-caus ‘she gave birth’ give birth-mom-now ‘it was leaking in the house’ flow-underneath-in house-pass-sub ‘it was borrowed’ borrow (mom)-pass-3s.qt ‘they could not get through’ unable to-do . . . -inc-irr-now-cnd3s.qt

Furthermore, the rule only applies in the environment of these glottalizing su‰xes, not anywhere else within the level of inflection, as demonstrated in (15). (15) a. tS’itasSit´m’in¡at´ tS’itas-Sit´ k -m’in¡-’at´ b. wiktaqSit´isqu wik-taqSit´ k -is-qu c. haukSit´aqt´’at´min hauk-Sit´ k -aqt´-’at´-min

‘they felt cold’ feel cold-mom-pl-now ‘as if it had never been’ not-before . . . ing-dim-cnd ‘we shall eat’ eat-mom-intent-now-1p.ind

If the su‰x is another glottalizing inflectional su‰x, then the result will be a glottalized lateral a¤ricate, [t´], as in the following cases. (16) a. hiSimj’up’at´’itS hiSimj’awup k -’at´-’i‡tS b. ts’iqSiat´’i ts’iq-Sit´ k -’at´-’i‡ c. hawiat´’inim hawit´ k -’at´-’i‡nim

‘get together now!’ assemble-now-2p.imp ‘now sing your sacred chants’ sing secret chant-mom-now-2s.imp ‘you can now stop asking’ finish[m]-now-1p>3o.fut

It is interesting to note that this process does not apply to those su‰xes that trigger it, specifically -’at´, as demonstrated by the following examples. (17) a. waat´’at wa k -’at´-’at b. m’aak w aap’at´’at m’a-ak w a‡ k -’ap-’at´-’at

‘it was said’ say-now-pass ‘he got chewed to pieces’ bite-in pieces-caus-now-pass

Level Ordering in Nuuchahnulth

233

Since the su‰xes triggering this rule are clearly within the domain of stratum 2, occupying positions 5, 6, and 7 in table 11.1, and the a¤ected elements are found within the derivational/aspectual categories, this process occurs precisely at the boundary of stratum 1 and stratum 2, again suggesting some sort of ‘‘exit feature’’ as suggested for the rule of delabialization above. 11.3.4

 Deletion

The deletion of // occurs when a base ending with a fricative comes in contact with a su‰x beginning with a glottal stop //. Such su‰xes should not be confused with those that trigger glottalization, represented by an initial /-’/, as discussed in section 11.3.1. In such situations, there are two possible outcomes that depend on the category of su‰x involved. If the su‰x is a stratum 1 element, then the glottal stop will disappear. If, on the other hand, the su‰x occurs at stratum 2, the glottal stop will remain following the fricative. Thus, this rule too is subject to domain distinctions. Take, for instance, the case of the su‰x /-aqsu´/ ‘at the mouth’. In the following examples one can see the outcome of this derivational su‰x combining with basefinal obstruents and fricatives. In the case of stops (18a), the // remains, whereas in the case of fricatives (18b), it disappears. i. k w isitaqsu´ k w ist-it-aqsu´ ii. apaqsu´ ap-aqsu´ b. i. ts’a¡aqsu´ ts’a¡-aqsu´ ii. ¡asaqsu´ ¡as-aqsu´ iii. hi´aqsuas¡at´t´a hi´-aqsu´-’as k -q¡-’at´-t´a

(18) a.

‘other side of the mouth’ di¤erent-at side-at mouth ‘mouth’ loc-at mouth ‘Tough-mouthed (man’s name)’ tough-at mouth ‘loud voice’ loud-at mouth ‘they were outside the door again’ loc-at opening-outside-being-nowagain

Compare this with the results of an inflectional su‰x, /-aqt´/, intent, in similar environments and we can see that the results are the same in both cases: stops (19a) and fricative-final bases (19b). (19) a.

i. ts’awakaqt´qa ts’awa‡ k -ak-aqt´-qa‡ ii. hinipaqt´qa hina-ij’ip k -aqt´-qa‡ iii. mup’itaqt´in mu-p’it k -aqt´-min

‘one would . . . ’ one-nom-intent-sub ‘that they find . . . ’ loc-obtain . . . -intent-sub ‘we would sing four times’ four- . . . times-intent-1p.ind

234

John Stonham

b.

i. tsiasaqt´qa tsias k -aqt´-qa‡ ii. tS’itSm’uq w i´aqt´qa tS’itSm’u‡-i´ k -aqt´-qa‡ iii. hi´¡aqt´itq hi´ k -q¡-aqt´-i‡tq

‘he was going to talk marriage’ woo-intent-sub ‘he got ready to exact sacrifices’ scapegoat-make . . . -intent-sub ‘the place he was . . . -ing’ loc-being-intent-3s.rel

As can be seen by these data, there is no // deletion at the later level. This rule of deletion operates only at the earliest level of the morphology, further supporting the notion of stratal separation and a stratum 1 phonological rule of // deletion. 11.3.5

The Domain of Nuuchahnulth Stress

In Nuuchahnulth, the domain of primary stress is the first two syllables of the word (i.e., the first foot). Weight, in the form of vowel quantity distinctions or the presence of a nasal in the coda, is a determining factor in stress assignment, but closed syllables containing a short vowel followed by one or more nonnasal consonants count as light for the purposes of stress assignment, as demonstrated in (20e). (20) a. b. c. d. e. f.

tSu´Suk w it´ ku´¡sinqin´ap n’a´csat´ ¡ı´j’i¡ qa¡na´k’at´ haja´akSiat´

‘he begins to suspect’ ‘he always causes a hole in the side’ ‘he sees now’ ‘he is after blood’ ‘someone died now’ ‘he did not know now’

As can be seen from these data, the stress always appears on either the first or second vowel of the word, regardless of the presence of long vowels farther on in the word. It is also the leftmost of two heavy syllables that is assigned the stress in the case where there are two in the first foot (e.g., 20b, 20c, 20d). The question arises as to what happens if the word contains no long vowels. The following data present the results. (21) a. t’a´n’anak b. na´j’aqak c. witSup’at´’isak

‘have a child’ ‘baby’ ‘come and cause me now to sleep!’

Here, stress is assigned to the leftmost vowel. Note that this result obtains even in the case where there is a longer vowel later on in the word. (22) a. ta´nakmitSiat´ b. hu´atsatSit´aqt´ma c. q w a´jats’iktaqim´

‘they turned now into mosquitoes’ ‘he will come back’ ‘wolf band’

Level Ordering in Nuuchahnulth

235

Much work remains to be done in working out the intricacies of Nuuchahnulth stress assignment and the interaction of the various factors that determine it, but one interesting observation is that stress assignment appears to be limited to the realm of derivation and aspect, that is, at stratum 1. In all of the previous examples, stress has appeared on either the root or on a derivational su‰x immediately following it, so it is worth asking the question, what happens when an incremental su‰x is in competition for stress assignment? Naturally, this is not a common occurrence given the prolific nature of Nuuchahnulth a‰xation, but it occasionally arises that a free root with a short vowel may be followed directly by one of those su‰xes described in table 11.1. Examples in (23) illustrate this situation. (23) a. u´¡qa u¡ k -qa‡ b. wı´kqu wik k -qu c. u´¡ma u¡ k -ma‡

‘it was’ be-sub ‘it might not . . . ’ not-3.cnd ‘it is . . . ’ be-3.ind

These examples support the proposal that stress assignment in Nuuchahnulth occurs before the attachment of incremental a‰xes, that is, at stratum 1. Since stress is already assigned, the weight of the vowels of a stratum 2 su‰x has no bearing on the outcome of stress assignment. It should, however, be noted that a definitive treatment of Nuuchahnulth stress awaits completion and there are some seemingly contradictory examples in the original data gathered by Sapir (Stonham 1999b, 60–61; Sapir field notes). Thus, while the claim that stress assignment occurs at an early stratum holds some weight, the evidence must still be viewed with some healthy skepticism. 11.3.6

Root/Free Form Distinctions

A further distinction encountered in the data is the morphological consideration of root allomorphy. In Nuuchahnulth, there is a high degree of root allomorphy that distinguishes between stratum 1 and stratum 2 a‰xation. In the following table the first column provides the free form that occurs with stratum 2 su‰xes and on its own. The second column indicates the basic word class of the root, and the third column provides a rough gloss. The final column shows the form of the root that is found in combination with derivational and aspectual suffixes. The examples in (24d) are of borrowed words, mainly from English. (24)

Free Form a. tsapin tsinwa

Class n n

Meaning ‘sawbill duck’ ‘mussel species’

Bound Form tsapitq tsitq

236

John Stonham

tS’aq’uniS hax w inma¯ k’i´anus t´uk w ana b. mak’it q w a¡in tik w in c. pua´ t´au d. tSikinis miSin tipin pinis

n n n n v v v adj adj n n n n

‘edible berry species’ ‘wren’ ‘furseal’ ‘Wolf Ritual’ ‘play with shells, dolls’ ‘fish with prong-spear’ ‘bake in sand, ashes’ ‘sleepy’ ‘other’ ‘chicken’ ‘boarding school’ ‘table’ ‘apple’

tS’aq’utq hax w itq k’i´atq t´uk w atq mak’itq q w a¡itq tik w atq puitq t´au tSikitq miSitq tipatq pitq

The choice of bound versus free form is not a property of any particular word class or a historical remnant, and it may even be applied to borrowed words.10 The bound allomorph is found to exist only at the level of derivational/aspectual morphology and never at the later level, suggesting that here again a distinction exists between the two domains. Take, for instance, the following examples where the first in each pair involves the free form and the second the bound form. (25) a. i. tsapin tsapin ii. tsapitqin’akj’ak tsapin-i‡n’ak w -j’ak b. i. t´uk w anat´ t´uk w ana k -’at´ ii. t´uk w atquwis t´uk w ana-u´ w -’is c. i. t´atmaptuki t´atmapt k -uk-i‡ ii. t´atmaqan’u´i t´atmapt-a‡n’u´ k -i‡

‘sawbill duck’ sawbill duck ‘(I) have a sawbill dance’ sawbill duck-imitate . . . in dance-device ‘I had given a Wolf Ritual’ Wolf Ritual-now ‘Wolf Ritual Beach’ Wolf Ritual- . . . place-on beach ‘his yew tree’ yew-poss-def ‘the one of yew’ yew-along . . . [l]-def

These are not the only instances of bound roots however. Bound roots may occur in a variety of shapes, including examples such as the following, where the first instance in each pair (a) is the free form and the second (b) is the bound form. Note that in some cases the only di¤erence between the forms is vowel quantity. (26) tS ’itSm’u / tS ’itSm’uq ‘scapegoat’ a. i. tS’itSm’uaqt´’ap ‘he was forced to make sacrifices’ tS’itSm’u k -aqt´-’ap scapegoat-intent-caus

Level Ordering in Nuuchahnulth

ii. tS’itSm’ui tS’itSm’u k -i‡ b. i. tS’itSm’uqtsk w i tS’itSm’u-tsk w i‡ ii. tS’itSm’uqi´ tS’itSm’u-i´ (27) ha mut / hamut ‘bone’ a. i. hamuti hamut k -i‡ ii. hamutuk t´’unim hamut k -uk t´’unim b. i. hamuttsk w i i¡tup hamut -tsk w i‡ i¡tup ii. hamut¡tin tS’itu´i hamut -¡tin tS’itu´ -i‡ (28) ma¡t’i / ma¡t’iq ‘house’ a. i. ma¡t’imiti ma¡t’i k -mit-i‡ ii. ma¡t’im’in¡i ma¡t’i k -m’in¡-i‡ b. i. ma¡t’iqapu´isi ma¡t’i-apu´ k -is-i‡ ii. ma¡t’iqtsk w ak ma¡t’i-tsk w i‡ k -’ak

237

‘the scapegoat’ scapegoat-def ‘what is sacrificed’ scapegoat-remains of . . . ‘make a scapegoat’ scapegoat-make . . . ‘the bones’ bone-def ‘an elk’s bone’ bone-poss elk ‘remains of whale bones’ bone-remains of . . . whale ‘the warclubs were made of bone’ bone-made of . . . warclub-def ‘the former house’ house-past-def ‘the houses’ house-pl-def ‘the little pretended house’ house-imitating . . . -dim-def ‘the remains of his house’ house-remains of . . . -poss

As can be seen from the examples above, the (a) forms involve free roots combining with stratum 2 su‰xes and the (b) forms involve bound roots combining with stratum 1 derivational and aspectual su‰xes. There is a variety of forms that bound roots may take, diverging from the free forms to a greater or lesser extent, but the su‰xes found to occur with bound forms are consistent throughout the paradigm, suggesting some close a‰liation. Such su‰xes include derivational and aspectual suffixes but no members of the previously presented incremental category. What best accounts for this a‰liation is the principle of a common stratal membership, bound root allomorphs being restricted to stratum 1, while free forms occur at stratum 2. 11.3.7

Suffix Combining Forms

Akin to the bound roots of the previous section are a set of su‰x forms that occur in a special shape only in combination with derivational and aspectual su‰xes. Since

238

John Stonham

‘‘bound su‰x’’ would be somewhat incongruous, the term ‘‘combining form,’’ first introduced by S&S, will be employed to refer to this class of a‰xes. (29) -sj’aqsti / -sj’aqstiq ‘main or leading . . . ’ a. i. uksj’aqsti ‘the leaders were . . . ’ u-sj’aqsti ref-leading . . . ii. jaqsj’aqstiitqa´ ‘the one who was the head’ jaq w -sj’aqsti k -i‡tq-a´ rel-leading . . . -3s.rel-pl b. i. uksj’aqstiqistasi ‘I was chief of the crew’ u-sj’aqsti-ista k -si‡ ref-leading- . . . persons in canoe1s.abs ii. jaqsj’aqstiqij’ipitqa´ ‘the most valuable thing they had got’ jaq w -sj’aqsti-ij’ip k -i‡tq-a´ rel-leading-obtain-3.rel-pl (30) -api / -apiq ‘up in the air’ a. i. kitsapii kitsuk-api k -i‡ ii. t´’ikapim’in¡ t´’ik-api k -m’in¡ b. i. ts’isapiqSiat´ ts’is-api-Sit´ k -a’t´ ii. sutS’aqim´ma´apiqij’ip sutS’a-qim´-ma´-api-ij’ip

‘a fallen tree that was slanting upward’ log-in air[l]-def ‘they were placed stretching upward’ have hands in position-in air[l]-pl ‘send a telegram’ rope, line-in air[l]-mom-now ‘they got five (birds) in the air’ five-units-move about-in air[l]-obtain

The (a) examples above show what happens when the su‰x completes the stratum 1 stem, whether or not further a‰xation at later levels occurs, while the (b) examples demonstrate the results within stratum 1, where a ‘‘combining’’ form is necessary for further derivational su‰xation. Again, the su‰xes that concatenate with these combining forms are the same ones that occur with the bound roots discussed in the previous section. As such, it seems obvious that there is a clear distinction made in the grammar between stratum 1 combining forms of certain su‰xes and a stem-final form, which enters stratum 2 and is subject to stratum 2 a‰xation. 11.4

Conclusions

In summary, there are a number of arguments for separating Nuuchahnulth morphology into two strata. A variety of phonological and morphological processes provide evidence for making such a distinction in the language. On this view, phonological rules such as stress assignment and glottal stop deletion would be situated at stratum 1. Furthermore, stratum 1 would be the domain of both bound roots and su‰x combining forms. Only free roots and complete stems would

Level Ordering in Nuuchahnulth

239

be allowed to enter stratum 2. At the boundary between strata would be the phonological rules of delabialization and the rule of t´ ! . There would be two versions of the glottalization rule: one at stratum 1 a¤ecting all eligible consonants, and another at stratum 2 restricted to just the stops and a¤ricates. Taking this approach to Nuuchahnulth morphology allows us to capture the special properties of the morphology in a clear and straightforward fashion. Certain phenomena are associated with stratum 1, and others are associated with stratum 2. The modules of the lexical morphology of Nuuchahnulth can be said to operate along the following lines. (31) Stratum 1

Stratum 2

Morphology

Phonology

Bound roots Su‰x combining forms Derivational su‰xes A‰x-triggered reduplication11 Free roots and stems Distributive reduplication Plural reduplication

Stress assignment  Deletion Glottalization 1 Delabialization t´ !  Glottalization 2

It should be noted here that phonological rules applying at stratum 1 do not necessarily reappear at stratum 2, apparently violating Kiparsky’s (1985) hypothesis that there is only a single set of phonological rules that may apply throughout the lexicon, possibly under distinct conditions. Nuuchahnulth would appear to constitute a counterexample to this proposal, although perhaps not the first one to be brought to light (see, e.g., Buckley 1992 on Kashaya and Inkelas and Orgun 1995 on Turkish). A number of issues remain to be addressed, including the status of clitics in the overall picture. There are, in fact, a number of clitics or clitic-like objects in Nuuchahnulth that invariably occur outside of the inflectional su‰xes in (1). In addition there are a number of postlexical phonological rules, including aspiration, diphthongization, and echo-vowel placement. These issues will eventually have to be addressed in a full treatment of Nuuchahnulth morphology. For the time being, a two-stratum model appears to be capable of dealing with the various issues encountered in Nuuchahnulth morphology and phonology. Notes I would like to thank Umberto Ansaldo, Sharon Inkelas, Stephen Matthews, Joel Nevis, Winnie Yiu, and an anonymous reviewer for comments on an earlier draft of this chapter. Data for this chapter come from a project on the Nuuchahnulth textual material gathered by Edward

240

John Stonham

Sapir and others from the early part of the twentieth century. This research was supported, in part, by a Hong Kong Universities Grants Council award (RGC No. HKU 7183/97H). The abbreviations used in table 11.1 and throughout the chapter are as follows: [l] ¼ lengthened vowel; [m] ¼ momentaneous aspect; [r] ¼ reduplication; 1s ¼ first person singular; 2s ¼ second person singular; 1p ¼ first person plural; 1p>3obj ¼ first person plural acting on a third person object; 2p ¼ second person plural; 3 ¼ third person; abs ¼ absolutive mood; being ¼ existential; caus ¼ causative; cnd ¼ conditional mood; def ¼ definite; dim ¼ diminutive; dub ¼ dubitative mood; dup ¼ reduplication; fut ¼ future; imp ¼ imperative; inal ¼ inalienable possession; inc ¼ inceptive; ind ¼ indicative; indef ¼ indefinite; infer ¼ inferential; intent ¼ intentive; inter ¼ interrogative; irr ¼ irrealis; loc ¼ locative; mom ¼ momentaneous aspect; mw ¼ meanwhile; nom ¼ nominalizer; now ¼ contemporaneous; pass ¼ passive; past ¼ past tense; pl ¼ plural; poss ¼ possessive; pres ¼ present tense; purp ¼ purposive mood; qt ¼ quotative mood; ref ¼ referential base; rel ¼ relative; rep ¼ repetitive aspect; sub ¼ subordinate mood; suf ¼ su‰x; n ¼ noun; v ¼ verb; adj ¼ adjective. k indicates the boundary between derivation (before k) and inflection (after k): 1. See Stonham (1997) for discussion of this issue. 2. Examples will be represented in a phonemicized form of the IPA. The length marks // and /‡/ represent the phonological distinction of long versus variable length vowels. Morphemic analysis that is provided in certain cases employs the free form as UR and marks the stratum boundary by means of the symbol hki. 3. I will avoid use of the terms ‘‘derivational’’ versus ‘‘inflectional’’ to distinguish the strata, as the arguments would be circular, and will use the traditional terms introduced by S&S instead. 4. Henceforth the glosses of so-called incremental su‰xes will be represented in small caps, those of derivation in lower case, and those of aspect with an initial uppercase letter. It should be noted that this is a tentative division in some ways and that there are still some a‰xes whose membership remains unclear. 5. No attempt will be made here to explain the various phonological rules that interact to produce striking di¤erences between the underlying and surface forms of words. The reader is referred to Stonham (1999b) for further details. 6. Since /h/ never occurs inside the word, it does not factor into this characterization. 7. This form undergoes the further rule of delabialization discussed further on. 8. [u] results from the coalescence of /w/ with /i/. 9. It should be noted that a rule of labialization a¤ecting velars and uvulars when preceded by a round vowel obscures the facts in certain situations. 10. The exact mechanism at work here has yet to be determined, but it seems to be related to the presence of a nasal consonant somewhere near the end of the free form. 11. See Stonham (1990, 1994) for discussion of this phenomenon.

12

Inside Access: The Prosodic Role of Internal Morphological Constituency

Patricia A. Shaw

12.1

Introduction

This chapter introduces into the theoretical literature a complex variety of stress facts ,, , , from the central Coast Salish language h nq min m (Musqueam), focusing on issues fundamental to understanding the prosody-morphology interface. The main claim of the chapter, framed within the restrictive constraints of the Parallelist Hypothesis in Optimality Theory (OT; Prince and Smolensky 1993), is that constraints on prosody may access internally embedded morphological constituency. This hypothesis is crucial to the analysis of what would otherwise be opaque stress phenomena. The argument focuses on recognition at the Output level of the left edge of the innermost morphological category, the morphological root. Among the major contributions of Kiparsky’s broad-based impact on linguistic theory are many fundamental insights into the role that internal morphology can play in the course of phonological derivation. The complexities of this interface are patently at the core of the framework of Lexical Morphology and Phonology (hereafter LMP). Kiparsky’s evolving theoretical perspectives (Kiparsky 1982a, 1982d, 1982e, 1983b, 1985) on issues such as Strict Cyclicity, Subjacency, Level Ordering, lexical versus postlexical properties, and so on, have consistently focused on identifying appropriate constraints on phonological reference to internal morphological structure. These several postulates cohered into a derivational model of the phonology-morphology interface that hypothesized that once morphological derivation had progressed from an inner morphologically defined cycle or level to a subsequent stage of morphological layering, then the internal morphological structure was no longer accessible or ‘‘visible’’ to phonological processes at the later stage of derivation (‘‘Bracket Erasure’’). Subsequent theoretical perspectives—for example, OT and Declarative Phonology (Scobbie 1991)—have challenged explicitly derivational models of phonology, necessitating a reevaluation of the empirical support for those constraints on the morphology-phonology interface that were conceived from a generative derivational e

e

e

242

Patricia A. Shaw

perspective. Within OT, the diverse roles of cyclicity in a grammar have been transfigured in various ways; see for example Kager and Zonneveld 1999 as well as Hermans and van Oostendorp 1999 for overviews. Two principal constructs are of particular relevance to the issues focused on here. One is the foundational Parallelist Hypothesis (Prince and Smolensky 1993), which holds that there is no serial derivation. Rather, parallel constraint satisfaction operates on a direct pairing of Input and Output representations. Mitigating the potential e¤ect of the Parallelist Hypothesis to e¤ectively obscure all internal structure within the earlier purview of cyclic or level-ordered access is the subtheory of constraints on Generalized Alignment (McCarthy and Prince 1993a), defined as a family of constraints that align edges of morphological categories (henceforth MCat) and/or prosodic categories (henceforth PCat). Although the Parallelist Hypothesis expressly prohibits access to any level of representation akin to an intermediate stage of a derivation, Generalized Alignment counteracts its restrictiveness, allowing more liberal access to internal morphological constituency than was provided even in LMP, which adhered to the general constraints of Strict Cyclicity, Subjacency, and Bracket Erasure. Further increasing the descriptive power of OT is Output-Output correspondence, which permits morphologically related words to influence each other’s phonology (e.g., Benua 1997). The challenge from the OT perspective is how to constrain ‘‘inside access’’ in principled ways to just those domains for which there is clear empirical justification. Kiparsky (2000c, 351–352) argues that if the Parallelist Hypothesis is interpreted in terms of ‘‘the intrinsic morphological and prosodic constituency of words and phrases, as characterized by the Stem, Word, and Postlexical levels’’ of LMP, then cyclicity e¤ects can be captured in terms of Input-Output (I-O) faithfulness, rather than requiring recourse to significantly more powerful Output-Output or Paradigm Uniformity constraints. The basic tenet of Kiparsky’s OT-based LMP model holds that ‘‘the way processes interact depends [on] their functional organization in the grammar into domains, which define a hierarchy of phonological levels’’ (Kiparsky 2000c, 356). The principal focus of the present study is the issue of what kinds of access to in,, , , ternal morphological constituency are required by the stress system of h nq min m, and what kinds of constraints can be defined to delimit such access appropriately. It is argued on the basis of this prosodic system that OT, under the Parallelist Hypothesis, requires ‘‘inside access,’’ specifically, reference to an internal domain defined by the left edge of the innermost morphological constituent of the word, the morphological root (hereafter MRoot). Contributing to a growing body of recent literature (e.g., Booij 1997c; Czaykowska-Higgins 1998; Downing 2000; Hyman and Katamba 1999; Inkelas and Zoll 2000; Kiparsky 2000c; and so on) that argues for the necessity of recognizing internal morphological constituency in the analysis of various pro,, , , sodic processes, the h nq min m case provides clear empirical evidence that the e

e

e

e

e

e

Inside Access

243

MRoot must not be rendered invisible by the presence of outer layers of prefixation. Properties of stress, vowel reduction, epenthesis, infixation, and reduplicative shape variance all converge in identifying this internal morphological edge as phonologically significant. Of further theoretical import, given the apparent crosslinguistic rarity of quantity-sensitive trochaic stress systems and various controversies about their properties (Hayes 1995; Kager 1995a, 1995b), this chapter makes a substantial contribution to the detailed documentation and analysis of a moraic trochee system. Of descriptive import is the fact that the language under discussion here is critically endangered.1 ,, The chapter is organized as follows. General phonological properties of h nq , , min m relevant to the analysis are presented in section 12.2, along with detailed ,, , , argumentation for a treatment of h nq min m schwa as featureless, weightless, and nonlexical. Section 12.3 presents an overview of the general stress patterns of the language, motivating its analysis as a moraic trochee system. In section 12.4, foot construction and foot alignment are shown to be sensitive to the internal domain defined by the left edge of the MRoot. Stress patterns of morphological word (MWord) prefixes (section 12.4.1) are contrasted with those of morphological stem (MStem) prefixes (section 12.4.2); both support stress foot alignment to the left edge of the MRoot. Section 12.5 summarizes the argumentation and conclusions. -e

e

e

e

e

e

,, , , General Phonological Properties of h nq min m e

e

e

12.2

,, , , As is characteristic of Salish languages, the consonant inventory of h nq min m2 (also referred to as Musqueam, traditionally spoken by peoples living near the mouth of the north arm of the Fraser River in British Columbia) is extensive:3 e

m , m

t t’ n , n

c , ¨’ c ´ s l l’

(cˇ)

(k)

[sˇ]

x y , y

k, w kw xw w , w

q , q w

e

t’ y

y

e

(1) p , p

qw , qw ww h 

In contrast, the vowel inventory is relatively small, consisting basically of three distinct full vowels /i e a/ plus schwa [ ]. The limited occurrence of /u/ gives it marginal status as a fourth full vowel.4 Only the full vowels can carry length.5 The model in (2) represents the basic tripartite parsing of the morphological struc,, , , ture of h nq min m words into MWord, MStem, and MRoot domains:6 e

e

e

e

(2) [MWord Nonredup prefixes [MStem Redup prefixes [MRoot Lexical root . . . Analyses of Salish stress have traditionally countenanced a high degree of lexical stress specification. The present analysis argues that much of this apparent specificity

244

Patricia A. Shaw

is instead derivable from prosodic constraints interacting with morphological domain properties: only the left edges of the three domains in (2) are focused on, since stress feet are left-aligned. The question of what edge stress feet are aligned to is not transparent. The MWord domain is defined by a set of prefixes that are argued to be outside of the stress domain; all nonreduplicative prefixes belong to this outer set. The MStem domain consists of various reduplicative prefixes. Although stress may indeed surface within the MStem domain, the major claim, in section 12.4, is that it is the innermost MRoot that demarcates optimal foot alignment. Conflict with a higherranked prosodic constraint renders this alignment generalization opaque. Significantly, the salience of the internal MRoot edge is not eclipsed by prefixation within the two outer levels, contrary to what a theory assuming Bracket Erasure would predict. 12.2.1

Properties of Schwa

Any discussion of Salish phonology, including stress, requires an understanding of the behavior of schwa ([ ]). Extensive argumentation is presented in Shaw et al. ,, , , 1999 for an analysis of schwa in h nq min m as featureless, weightless, and nonlexical. These are independent hypotheses; all three hold in the epenthetic schwa behavior documented here.7 A summary of some of the principal kinds of empirical evidence supporting these hypotheses follows. Supporting the first claim, that schwa is featureless, is the fact that schwa takes on coloration from adjacent consonants ranging in realization throughout both the vertical and horizontal dimensions of the vowel space—for instance, [i] before palatovelar /x/ as in /cel x/ [clic¸] ‘hand’; [V] before tautosyllabic rounded velars /k w /, /x w / as in /x w lm x w / [x w lmVx w ] ‘person of First Nations ancestry’; [þ] before a tautosyllabic glottal stop as in /sm y y ´/ [sm y yþ´] ‘belonging to a deer’ (e.g., tracks). Further, a pervasive, low-level process of translaryngeal harmony entails full assimilation of schwa to the quality of another vowel across an intervening //, e.g. [st x w ´] @ [stx w ´] ‘children’, [t ı´] @ [ti ı´] ‘this one’. It is certainly the case that the full vowels are also phonetically variable and influenced by adjacent consonants. For example, /e/ is generally realized as [], as in [mn] ‘father’ and, often, as a high [æ:] under length, , as in /e:ny / [æ:ny ] ‘1 s. ind pro’; it raises and fronts to [e] before /y/, as in [sk w ey] ‘impossible’. To take another example, phonemic /i/ is , , characteristically lowered to a high o¤glided [e i ] by a preceding uvular, as in /qic y/ , i, [qe ciy] ‘Katzie’ (place name). However, the phonetic range of variation for full vowels is much more narrowly restricted in the vowel space than is the range for schwa. Coarticulatory e¤ects from adjacent segments add phonetic detail to, but do not override, the lexically distinctive core feature specification of each full vowel. The hypothesis that schwa has no distinctive features accounts for surrounding consonants having complete influence over its realization and for its range extending throughout the vowel space. e

e

e

e

e

e

e e

e

e

e

e

e e e

e

e

e

e

Inside Access

245

Motivating the hypothesis that schwa is weightless—that is, nonmoraic—are several aspects of its stress behavior. As will be exemplified in detail in section 12.3, ,, , , stress in h nq min m consistently di¤erentiates the full vowels /i e a (u)/ from schwa, preferentially stressing a full vowel. The hypothesis that stress is weight-sensitive, with schwa being weightless, o¤ers a principled account of these patterns. Further, under the standard moraic hypothesis that vowel length is represented by two as opposed to one mora, the complementary hypothesis that schwa is nonmoraic explains ,, , , its systematic failure to lengthen in h nq min m in contexts where other vowels (i.e., the moraic full vowels) do lengthen. The third hypothesis is that schwa is basically nonlexical. The lexical-versusepenthetic status of schwa is one of the most complex and controversial issues in Salish. The role of schwa in the present analysis is that of an unmarked default vowel fulfilling a headedness requirement for a syllabic nucleus in various contexts defined by universal markedness constraints on prosodic parsing and well-formedness. Unfortunately space limitations preclude a motivated account of constraints governing schwa epenthesis (but see Shaw 2002); consequently, Input representations will generally include schwa where it is realized in the Output and when it is not directly relevant to the specific argumentation in focus. e

e

e

e

e

e

12.2.2

ProperHeadedness

The hierarchical set of prosodic well-formedness constraints hypothesized to govern headedness (ProperHeadedness, or PropHead) within the Prosodic Hierarchy is given in (3a–c) (see Ola 1995; Ito and Mester 1992). The constraint in (3c), also known as sNuc or Nuc (Prince and Smolensky 1993), ensures that syllables have a nuclear head (3c). All potentially violable, the constraints are in an entailment relation such that the satisfaction of PropHead at each successively higher level of prosodic structure is directly dependent on the satisfaction of PropHead in one of the constituents it immediately dominates; this is expressed in (3d). Epenthesizing a schwa to satisfy sNuc violates the lower-ranked correspondence constraint DepNuc, specified in (3e). In the full range of data being considered here, all syllables satisfy Onset (3f ).8 (3) Prosodic hierarchy: ProperHeadedness a. PROPHEAD-PrWd A Prosodic Word (PrWd) is headed by a Foot. b. PROPHEAD-Ft A Foot (Ft) is headed by a Syllable. c. PROPHEAD-s A Syllable (s) is headed by a Nucleus [¼sNuc]. d. PropHead-PrWd I PropHead-Ft I PropHead-s

246

Patricia A. Shaw

Correspondence e. DEPNUC Every Nuc element in the Output has a corresponding Nuc element in the Input. Syllable structure f. ONSET Syllables must have an Onset. Given this general overview, let us proceed to the detailed investigation of the conditions governing stress. ,, , , Stress in h nq min m e

e

12.3

e

The traditional view of stress in Salish languages is of a highly lexicalized system of competing strengths between strong, weak, and variable roots and a‰xes (see especially Czaykowska-Higgins 1993 and Czaykowska-Higgins and Kinkade 1998, 15– 16). The present approach diverges significantly from these treatments, drawing instead on the interpretation in Shaw et al. 1999, where it is argued that many of the perceived lexical properties reduce to a major phonological distinction between schwa as opposed to full vowels, combined with a sensitivity of stress placement to internal morphological structure. Assuming schwa to be nonmoraic, the complexities ,, , , of h nq min m stress receive a coherent analysis as a quantity-sensitive trochaic stress system, with the left edge of the stress domain being the MRoot. The lexical residue is reduced primarily to a subset of su‰xes that are lexically endowed with certain distinctive lexical or subcategorizational properties. ,, , , The general phonological properties of word-level stress in h nq min m are outlined in section 12.3.1 and formalized in OT terms in section 12.3.2. Section 12.4 investigates the influence of morphological prefixation domains on stress behavior. e

e

e

e

e

e

12.3.1

General Properties of Stress

All lexical categories surface with stress (primary stress is marked with an acute accent over the vowel, secondary stress with a grave accent), even if monosyllabic as in (4), and regardless of whether the vowel is full—that is, [i, e, a] and marginally [u] as in (4a), or schwa as in (4b). (The s-/sˇ- prefix in these and the following data is a nominalizer prefix.)9 , (4) a. yı´ ‘big’ s-ca´tw ‘halibut’ ´cı´ws ‘tired’ ‘harbor seal’ pu´s ‘cat’ (loan) qa´ ‘water’ e´sˇx w , b.  ´y ‘good’ s-m ´k w ‘ball’ m ´l’q w ‘uvula’ , , s-p ´x w ‘tripe’ q w ´s ‘enter the water’ p ´q ‘white’ e e

e e

e

e

Inside Access

247

In polysyllabic words, stress is determined with reference to the left edge and, as in Proto-Salish (Thompson 1979, 721), is sensitive to the distinction between full vowels and schwa. The data in (5a) exemplify words where the full vowel precedes a schwa: the initial full vowel receives stress. If, as in (5b), an initial schwa is followed by a full vowel in the subsequent syllable, then the full vowel attracts stress: t’a´m n pu´l’ s , h ma´ k pu´ e e

‘to walk’ ‘sasquatch’ ‘living parent’s sibling’ ‘beaver’

e e

(5) a. ı´m x , se´sq c b. sˇ-x w mnı´k w , s-q le´w

‘wall’ ‘cats’ (loan) ‘pigeon’ ‘coat’ (loan)

e e e e

However, if both vowels are schwa, as in (6), then initial stress surfaces, even though it results in stressing a schwa. This establishes stress as basically trochaic, optimally aligned with the left edge. All data thus far are MRoot-initial or prefixed only by the nominalizer s-/sˇ-. A more precise characterization of ‘‘left edge’’ is given in section 12.4. (6) a. b. c. d. e. f.

q ´l m ´ ´m x w k w ´n-n- x w , m ´n s-n ´x w ´ w ´w s

(7)

Root , nı´w , pe´t’, y xa´k w

‘eye’ ‘(to) rain’ hold-n.tr-3su ¼ ‘manage to get, hold’ ‘child’ ‘canoe’ ‘tree frog’

e e e e e e e e e e ee

,, , , Sequences of adjacent short full vowels are virtually nonexistent in h nq min m. Where such sequences would arise through morphological concatenation, one or the other full vowel reduces to schwa,10 attesting to the active role played by grammatical constraints in e¤ecting a strong-weak rhythmic pattern. For example, the reduplicated Progressive forms in (7) show that vowel identity in Base-Reduplicant correspondence is sacrificed in order to optimize a left-headed (V ) foot rather than a (V V) foot. e

e

e

e

Progressive-Root (-T.TR) , , nı´-n w- t (*nı´-niw- t) , , , , pe´-p t’, y (*pe´-pet’, y ) xa´-x k w - t (*xa´-xak w - t) e

e

e

e

e e e

a. b. c.

Gloss ‘advise/be advising someone’ ‘sew/be sewing’ ‘bathe/bathing someone’11

In accordance with the claim that schwa is nonmoraic (see section 12.2.1), the re,, , , sult of such pervasive schwa-reduction e¤ects in h nq min m is a stress system actively favoring the ‘‘uneven’’ trochee.12 Further confirmation of a basic trochaic system is seen in patterns of secondary stress on polysyllabic forms. The words in (8) illustrate that footing is iterative: syllables throughout the word are grouped into bisyllabic trochaic feet, foot parsing being e

e

e

248

Patricia A. Shaw

indicated by parentheses. The head syllable of the rightmost foot in the word receives the greatest degree of prominence—that is, primary stress. The morphological composition of each form is shown on the left; its trochaic foot parse is on the right. (8) a. /¨ [-l-]qt¼ l’¼ew n/ (¨’ `lqt )(l’e´w n) [pl]-long¼connective¼side ¼ ‘long arms’ ( `´t )(na´y n) b. / ´t n¼ay n/ eat¼margin/edge ¼ ‘eat along the way’ (WS, 290) , , c. /cew- y-amx- s/ (ce´w )(ya´mx s) help-t.tr-1sobj-3trsu ¼ ‘she/he helps me’ d. /it t- lm n/ (ı`t )(t ´lm n) sleep-desiderative ¼ ‘to want to sleep’ e

e

e

e

e

e

e e

e

e

e

e

e

e e

e

e

e e

e

e e

Because the forms in (8) are exhaustively footed into bisyllabic trochees, they are not revealing with respect to edge-alignment constraints. That foot alignment is optimally to the left edge is established by two di¤erent types of data. First, trisyllabic forms with exclusively schwa nuclei are footed at the left edge, as shown by the forms in (9), all of which are MRoot-initial: * (n ´x w y t) e

e

( ´n x w ) y t (9) a. / n x w -y t/ stop-refl ¼ ‘stop oneself ’ , , (s ´wq ) n p b. /s wq¼ n p/ seek¼land ¼ ‘look for land’ , , (q ´m ) n ´p c. /q m n¼ ´p/ maple¼tree ¼ ‘maple’ (´ ´q l) l x w d. /´ q l-n- x w / Root-n.tr-3agr ¼ ‘know’

e

e

e e

e

, *s w (q ´n p)

e

, *q (m ´n ´p)

e e

e e

e

e

*´ (q ´ll x w ) e e

e

e

e

e

e

e e

e e

e e

e

e e

e

e

e e

e e

In the (unasterisked) foot parsings in (9), feet are aligned to the left edge, with stray unparsed syllables at the right margin. The parsings marked with an asterisk, with right-edge alignment, manifestly stress the wrong syllable. A second argument for left-edge alignment comes from the distribution of degenerate feet: monosyllabic feet occur only at the right edge. While, as seen in (9) and (10a), a rightmost odd-parity syllable headed by a schwa nucleus is not parsed into a degenerate foot, a final odd-parity syllable headed by a full vowel is footed, as illustrated in (10b). i. /tey¼ w ´/ (te´y ) w ´ race¼canoe/vessel ¼ ,‘racing canoe’ , w (k w e´c ) t s ii. /k ec- t- s/ see-t.tr-3trsu ¼ ‘she/he saw him/her’ , , iii. /k w in¼ q n/ (k w ´ı.n ) q n how.many¼containers ¼ ‘how many containers?’ e

e

e e

e

e

e e

e e

e e

(10) a.

Inside Access

249

iv. /pax l q w / (pa´.x ) l q w ‘yellow cedar’ (WS, 203) , , (ı`.t )(-te´wtx w ) b. i. /it t¼ewtx w / sleep¼building ¼ ‘hotel’ , , ii. /k w in¼ winx w / (k w `.n )(wı´:nx w ) how.many¼years ¼ ‘how many years’ , , , , (q `.q )(ya´s) iii. /q q¼ yas/ be.taut¼circular ¼ ‘barrel, washtub’ , , iv. /´qec s¼mat/ (´qe`.c s)(ma´t) five¼bundles/kinds ‘five bundles of a kind’ e e

ee

e

e

e e

e

e e

e

e

e

e

Left-edge foot alignment correctly accounts for the right-edge asymmetry in the distribution of degenerate feet. Although left-edge alignment is optimal, the sensitivity of stress to the schwa– versus–full vowel distinction militates against strict left-edge alignment when an initial schwa syllable is followed by one headed by a full vowel. (11a) illustrates a bisyllabic form, trisyllabic forms are given in (11b), and (11c) illustrates quadrisyllabic forms. The quadrisyllabic forms in particular strongly support the analysis of stress developed thus far. The first syllable, headed by schwa, is unfooted; the trochaic foot includes the second and third syllables, leaving the final syllable, headed by schwa, unfooted as well, as documented in (9). , , h (ma´) (¼ (5b)) (11) a. /h ma/ ‘pigeon’ b. i. /x w nit m/ x w (nı´.t m) ‘white person’ c (l’e´.q ´) ii. /c l’eq- ´/ fossilized.root-past ¼ ‘yesterday’ iii. /q w¼()ic n/ q w (ı´.c n) warm¼back ¼ ‘Cowichan’ ,, ,, iv. /ck w x¼el’ / ck w (xe´.l’ ) twenty¼people ¼ ‘twenty people’ n (wa´q w .c s) t n c. i. /n w¼aq w ¼c s-t n/ enter¼head¼hand-instr ¼ ‘thimble’ ii. /y sel’ ¼ q n/ y (se´.l’ ) q n two¼containers ¼ ‘two containers’ e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e e e

e

The proposed analysis models a fundamental asymmetry in the behavior of full vowels as opposed to schwa. The footing of full vowels is exhaustive;13 syllables headed by a full vowel are systematically parsed as a trochaic foot head, and a final full-vowel syllable at the right edge is parsed as a degenerate (monosyllabic) foot (10b). In contrast, the footing of schwas is much more restrictive. Aside from

250

Patricia A. Shaw

monosyllabic words compelled by culminativity (see section 12.3.2.3), a foot may be headed by schwa, but only under the following two conditions: (i) the foot is bisyllabic, and (ii) the syllable in weak metrical position is not headed by a full vowel. As a consequence, parsing of syllables headed by schwa to the foot level is not necessarily exhaustive. As seen most dramatically in (11c), more than one single syllable headed by schwa can remain unfooted in a word. However, even in a word all of whose syllables are schwa vowels, some degree of footing is obligatory; in all forms seen thus far it is the case that no two adjacent syllables may remain unparsed.14 12.3.2

An Optimality-Theoretic Interpretation

The distinction between full vowels and schwa, clearly central to an account of the ,, , , h nq min m stress system, is modeled in (12) as a moraic distinction. Each full vowel has underlying moraic structure, while schwa is entirely devoid of both moraic and featural specification. Completely unmarked, epenthetic schwa realizes a prosodically required syllable Nucleus position, abbreviated as Nuc, thus functioning as the default segment. e

e

e

(12) Nucleus

a. Full vowel Nuc

b. Schwa Nuc

c. Schwa reduction Nuc ¼ m

Moraic weight

m

Root node





([f ])

([f ])

Features

Although weight (moraicity) and syllable headedness are optimally correlated, as reflected by universal constraints such as Wgt-to-Str and PeakProminence (Prince and Smolensky 1993), the hypothesis here (following Shaw 1992, 1993, and so on; Jiang-King 1996; Ola 1995; Blake 2001; Howe 2000) is that they have independent prosodic status. Giving Nuc formal status independent of moraic weight predicts the well-known fact that weight may be nonnuclear (e.g., where a coda consonant contributes to a heavy syllable), as well as the complementary fact that a Nuc consisting of schwa may be weightless (Anderson 1982; Kager 1990; Kinkade 1998; among others). If schwa is represented, as in (12), as simply a nuclear node, its phonetically variable realization is predictable through feature spreading from its paradigmatic and syntagmatic context. As diagrammed in (12c), the familiar process of ‘‘schwa reduction,’’ whereby a ‘‘full’’ vowel in weak metrical position reduces to a light default vowel, is formalized as deletion of all content below the Nuc, consistent with the basic hypothesis that schwa is nonmoraic and lacks distinctive featural (represented as [f ]) content. Vowel deletion, which subsumes schwa deletion, entails deletion from above the Nuc node, including all content dependent therefrom.

Inside Access

251

The claim that schwa is nonmoraic plays a significant role in accounting for the stress behavior of schwa in contrast to full vowels: its weightlessness makes it less optimal as a prosodic Foot head under constraints such as Wgt-to-Str (see (13c)) that value a correlation between weight and prominence. Given these structural assumptions regarding the appropriate representation of full vowels as opposed to schwa, the general stress patterns of data sets (9) through (11) derive straightforwardly from the constraints in (13). The basic parsing into trochaic feet results from the high ranking of (13a), FootForm ¼ Trochee (FtForm). Although the data thus far have shown only that feet are built from the left edge, section 12.4 will show that it is in fact the left edge of the MRoot, rather than the MStem or the MWord, to which foot alignment is sensitive. Align-(Foot, L, MRoot, L) (Al-L(Ft, MRt)) constrains the left (L) edge of a foot (Ft) to correspond with the left edge of a lexical MRoot (MRt) (13b). The preferential stressing of a full vowel over schwa derives from the Weight-to-Stress (Wgt-to-Str) constraint in (13c), assuming the proposal in (12) that a full vowel is moraic (i.e., has weight), whereas epenthetic schwa is not. Wgtto-Str is ranked above Al-L(Ft, MRt). Hence, if the initial syllable is headed by schwa, footing does not start until the second syllable.15 However, as was seen in (6) and (9), if the second syllable does not provide a more optimal head, Al-L(Ft, MRt) is not compromised. 12.3.2.1

A Moraic Trochee System

(13) a. FOOTFORM¼Trochee (FTFORM) Feet are left-headed: ( s s) or ( s). b. ALIGN(FOOT, L, MROOT, L) (AL-L(FT, MRT)) Align the left edge of a foot with the left edge of the MRoot. c. WEIGHT-TO-STRESS (WGT-TO-STR) If a syllable nucleus has weight, then it is prominent within a foot. d. Ranking FtForm, Wgt-to-Str  Al-L(Ft, MRt) The interaction of these constraints to e¤ect the appropriate parsings of the data considered thus far is illustrated in the tableaux in (14),16 in which a left bracket demarcates the left edge of the MRoot. The crucial ranking identified in (13d) of FootForm and Wgt-to-Str above Al-L(Ft, MRt) is established by the evaluation of competing candidates in the tableau for [x w nı´t m] ‘white person’. Consider, first, candidate (14ai). The foot head is the full vowel [i], which satisfies Wgt-to-Str. Moreover, the left edge of the foot (x w nı´) aligns directly with the left edge of the morphological root, thereby satisfying Al-L(Ft, Rt). The problem with candidate (14ai) is that the foot is a right-headed iamb, not a left-headed trochee. The fact that candidate (14aiii), not (14ai), is the correct output requires FtForm to be ranked above Al-L(Ft, MRt). Candidate (14aii) conforms to FtForm and Al-L(Ft, e

e

e

252

Patricia A. Shaw

MRt), but at the expense of fatally violating Wgt-to-Str, thus motivating the crucial ranking of Wgt-to-Str above Al-L(Ft, MRt). The second tableau, in (14b), for [ ´n x w y t] ‘stop oneself ’, shows that unless the second syllable has weight—that is, a full vowel—then Al-L(Ft, MRt) will not be compromised. e

e e

(14) a. x w nı´t m ‘white person’ (¼ (11bi)) e

[x w nit m

FtForm

i. (x w nı´) t m

*!

e

e

e

e

e

ii. (x w ´ni) t m

Wgt-to-Str

Al-L(Ft, MRt)

*!

e

e

+ iii. x w (nı´t m)

*

e

e

b.  ´n x w y t ‘stop oneself ’ (cf. (9)) e

e e

[ n x w -y t e

e e

i. ( n ´x w ) y t e

Wgt-to-Str

Al-L(Ft, MRt)

*!

y t) e

e e

e

ii.  (n

´x w

FtForm

*!

e

+ iii. ( ´n x w ) y t e

e e

The Role of Binarity The generally alternating pattern of stress in su‰ciently long words (8) is governed by two constraints. Maximal parsing of syllables into feet is e¤ected by Ft-Bin-s, which requires feet to be disyllabic (15a), and by Parse-s-2, which is violated by any sequence of two unfooted syllables (15b). Degree of stress, in words with more than one foot, is handled by the constraint Rightmost (15c), which captures the generalization that the head syllable of the rightmost foot in the word gets primary stress, while the heads of all other feet surface with secondary stress. (15d) shows crucial rankings among these constraints as well as constraints in (3) and (13). 12.3.2.2

(15) a. FT-BIN-s Feet are binary at the syllabic level. b. PARSE-s-2 Adjacent syllables cannot both remain unparsed.17 c. RIGHTMOST The rightmost foot in a word has greatest prominence. d. Ranking Wgt-to-Str, Parse-s-2  Al-L(Ft, MRt)  FtBin-s PropHead (3)  FtBin-s, DepNuc (3) Given the criterial role of Wgt-to-Str in optimizing prominence on full vowels, it is important to consider how it is that words with no full vowel get footed at all. This

Inside Access

253

is shown in the tableau in (16), which shows how Parse-s-2 and Al-L(Ft, MRt) can function crucially in such cases. (16)  ´n x w y t ‘stop oneself ’ (cf. (9)) e

e e

[ n x w -y t e

Parse-s-2

e e

a.  n x w y t

*!

b. ( ´) n x w y t

*!

Al-L(Ft, MRt)

*

e

e e

e

e

e

y t)

*!

d. ( `n x w )(y ´t)

*!*

e

e

e

c.  (n

´x w

Ft-Bin-s

*

e

e e

+ e. ( ´n x w ) y t e

e e

f. ( ´n x w y t)

*!

e

e e

Candidates (16a) and (16b) both fatally violate Parse-s-2: this constraint allows a single syllable to remain stray, but requires two adjacent syllables to be parsed into a foot. (16c) and (16e) have only one stray syllable each, satisfying Parse-s-2; (16c) is less optimal because its foot does not align with the left edge of the MRoot. Interestingly, the fully footed candidate in (16d) is even worse than (16c) because its final (degenerate) foot incurs two alignment violations, since it is two syllables away from the Root edge. The left-aligned exhaustive parse in (16f ) violates Ft-Bin-s. It is thus candidate (16e), with a final stray syllable, that ends up being optimal since it satisfies both Parse-s-2 and Al-L(Ft, MRt). The tableaux in (17) illustrate cases where Ft-Bin-s makes it optimal to leave an initial schwa syllable stray.18 (17) a. x w nı´t m ‘white person’ (cf. (11bii)) e

e

[x w nit m e

e

Al-L(Ft, MRt)

i. x w (nı´t m)

*

ii. (x w `)(nı´t m)

*

e

e

e

+

Wgt-to-Str

e

, b. h ma´ ‘pigeon’ (¼ (5b), (11a)) , Wgt-to-Str [h ma , i. (h ´ma) *! , ii. (h `)(ma´) , + iii. h (ma´)

Ft-Bin-s

*!

e

e

Al-L(Ft, MRt)

Ft-Bin-s

e

*

**!

e

*

*

e

254

Patricia A. Shaw

With respect to the role of Ft-Bin-s, recall the distinction in the data of (10) where a single final syllable headed by schwa is not footed, whereas a single final syllable headed by a full vowel is footed. The tableau in (18a) shows the e¤ect of the constraint ranking in (15e) on a stress-initial odd-parity word where the final syllable is headed by schwa. Under the postulated constraint rankings, there is no optimal way to parse the final syllable into a foot, which is exactly the correct result. In contrast, however, a final odd-parity syllable headed by a full vowel is parsed into a degenerate foot. As shown in the tableau in (18b), the very same ranking accounts for cases like this. In successful candidate (18bi), the fact that the full vowel in the final syllable is footed (te´wtx w ), thus bearing stress, satisfies the highly ranked Wgt-to-Str constraint. The degenerate foot does violate lower-ranked Ft-Bin-s and Al-L(Ft, MRt), since the final degenerate foot is misaligned from the L-edge by two syllables, but it still fares better than candidate (18bi), which violates Wgt-to-Str. , (18) a. k w e´c t s ‘she saw him’ (¼ (10aii)) , Wgt-to-Str Al-L(Ft, MRt) Ft-Bin-s [k w ec- t- s , *! * i. (k w e`)(c ´t s) , w` *!* * ii. (k ec )(t ´s) , + iii. (k w e´c )t s ee

e e

ee

e e e e

b. ı`t te´wtx w ‘hotel’ (¼ (10bi)) e

[it t¼ewtx w e

i. (ı´t ) tewtx w e

+ ii. (ı`t )(te´wtx w )

Wgt-to-Str

Al-L(Ft, MRt)

Ft-Bin-s

**

*

*!

e

A direct comparison of these two tableaux shows that the right-edge asymmetry in the distribution of degenerate feet that was observed in the data of (10) falls out straightforwardly from a single ranking of the constraints Wgt-to-Str  Al-L, FtBin-s in the proposed analysis. The other constraint relevant to whether stray syllables surface is Parse-s-2. Evidence that it too is crucially ranked above Align-L(Ft, MRt) is established through consideration of even-parity words like (19a). Since schwa is weightless, the nonparsing of a syllable headed by schwa does not violate Wgt-to-Str. One unparsed schwa syllable is tolerated, as in all the examples in (11) and (17), but two adjacent unparsed schwa syllables are not: Parse-s-2 ensures that they will be parsed into a bisyllabic foot. Rightmost (15c) (not shown in the tableaux here due to space) accords it primary stress.

Inside Access

255

The tableau in (19b) illustrates the prosodic parsing of a word with an initial sequence of schwa-headed syllables. In each candidate form, the second foot consistently violates Al-L(Ft, MRt) by two syllables. Candidate (i) shows that Ft-Bin-s will rule out any stray schwa syllables if they can be parsed into a binary foot. Candidate (ii), with one foot and two adjacent stray syllables, fatally violates Parse-s-2. The successful candidate (iii) satisfies both Parse-s-2 and Ft-Bin-s. (19) a. ı`t t ´lm n ‘to want to sleep’ (¼ (8d)) e

ee

[it t¼ lm n e

Wgt-to-Str

Parse-s-2

e

e

i. (ı´t )t l.m n

Al-L(Ft, MRt)

*!

e

e e

+ ii. (ı`t )(t ´lm n)

**

e

e e

b.  `´t na´y n ‘eat along the way’ (¼ (8b)) e

e e

[ ´t n¼ay n e

Parse-s-2

e e

i. ( `´)t (na´)y n e

e

e

ii.  ´.t (na´y n)

*!

e

e

e

+ iii. ( `´t )(na´y n)

Al-L(Ft, MRt)

Ft-Bin-s

**

*!*

** **

e

e e

As seen in section 12.4, prefixed forms may violate Parse-s-2; however, all the forms evaluated thus far are MRoot-initial, and obey Parse-s-2. Tableau (20) illustrates an odd-parity five-syllable word beginning with two schwa syllables and ending with a schwa syllable. In accordance with what has been seen so far, schwa cannot head a degenerate (nonbinary) foot, and thus the final syllable remains unfooted. The established ranking of Parse-s-2  Al-L(Ft, MRt) predicts this outcome. ,, , , ,, , , (20) h `nq mı´n mq n ‘the h nq min m way of speaking’ (cf. q n ‘throat’) ,, , , h nq min m¼q n Parse-s-2 Al-L(Ft, MRt) Ft-Bin-s ,, , , **,*!*** * a. (h `nq )(mı`n m)(q ´n) , , ,, **,*!** * b. (h `nq )(mı`)(n ´mq n) , , , , *! ** c. h n.q (mı´.n m)q n ,, , , + d. (h nq )(mı´n m)q n ** e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e

e e

e

e

e

e

e

e

e e e e

Candidates (20a) and (20b) represent the two most plausible exhaustive parsings of the five-syllable string; both satisfy high-ranked Wgt-to-Str (not shown in (20)). Comparing these with the successful, nonexhaustively parsed candidate in (20d), we

256

Patricia A. Shaw

see that Al-L(Ft, MRt) prevents the final schwa syllable from being footed; the respective accumulation of additional misalignment violations renders (20a) and (20b) much less optimal than candidate (20d). In contrast, candidate (20c), the most plausible nonexhaustive parsing that respects both Wgt-to-Str and Ft-Bin-s, ties with (20d) in number of alignment violations but fatally violates Parse-s-2; treatment of the initial sequence of two schwa syllables is the critical factor favoring (20d) over (20c). 12.3.2.3 Culminativity Given the argumentation above that a single schwa syllable can remain unparsed, as in (20), (17a,b), and (16), it is important to consider what ,, , , accounts for the stressing of such syllables in monosyllabic words such as /pq/ [p ´q] ‘white’ (4b). First, note that in monosyllabic words with a full vowel, like [e´sˇx w ] ‘harbor seal’ (4a), the high-ranking (in fact inviolable) constraint Wgt-to-Str forces violation of Ft-Bin-s.

e

(21) e´sˇx w ‘harbor seal’ (¼ (4a)) [esx w a. esˇx w + b. (e´sˇx w )

Wgt-to-Str

Ft-Bin-s

*! *

A deeper generalization, however, is that all lexical forms in the language require stress on at least one syllable. This is captured by the hypothesis that the constraints on proper headedness (PropHead, in (3)) are consistently satisfied. The ranking in (22) of PropHead constraints above DepNuc (3) ensures that words without a lexical full vowel will surface with schwa as a default syllable head. Similarly, the systematic stressing (footing) of monosyllabic words, regardless of nuclear weight, is captured by ranking PropHead constraints above Ft-Bin-s, thus ensuring that each prosodic word has at least one foot, even if foot binarity cannot be satisfied. , , (22) p ´q ‘white’ (¼ (4b)) ,, [pq PropHead-PrWd, -Ft PropHead-s Ft-Bin-s DepNuc ,, a. pq *! * , , *! * b. p q , , * * + c. (p ´q) e

e

e

To review the basic claims of the stress analysis developed thus far, the reader is referred to the constraints and rankings summarized in (13) and (15). Based on these

Inside Access

257

generalizations, we turn to an investigation of the interplay of morphological domains with left-edge stress alignment. 12.4

Morphological Domains of Stress

In the following subsections, various sets of apparent exceptions to the basic generalizations governing stress are explored, and it is argued that each set receives a principled explanation under the basic hypothesis that it is the left edge of the innermost morphological domain, the MRoot domain, which defines optimal stress foot alignment. The analysis is structured with reference to the MWord, MStem, and MRoot domains, presented in (2). First, in section 12.4.1, the outer layer of MWord-level prefixes is examined, and shown to be consistently outside of the stress domain. The complex interaction of stress with MStem-level progressive and plural reduplicative prefixes, as well as a plural infix, is analyzed in section 12.4.2, with the several independent patterns each converging to the same conclusion: constraints governing stress make crucial reference to the internal MRoot-domain left edge. 12.4.1

MWord-Domain Prefixes

No MWord-domain prefix19 has a lexical full vowel: only schwa surfaces in this domain, and all such schwas are epenthetic, thus permitting syllabification of an adjacent resonant. The data in (23) illustrate the consistent lack of prosodic prominence on these prefixes. The stress pattern in the data of (23a) is expected from the analysis given in section 12.3.2. However, the stress patterns in (23b–d) are not. (23) a. t m-lı´l’ season-salmonberry ¼ ‘salmonberry time’ , x w n-a´m t early-be.at.home ¼ ‘get home early, soon’ , b. t m-w ´y¨’ season-cold ¼ ‘winter’ , w -p ´k w suddenly-billow.forth ¼ ‘suddenly billow forth’ (WS, 283) c. x w n-h ´ye20 early-leave ¼ ‘to leave early, soon’ t w-t ´k y somewhat.like-turkey (loan) ¼ ‘like a turkey’ (WS, 255) w -s-k w ´yw-y t established-s-move-refl ¼ ‘he moved’ , d. w -m ´k w - l p established-all-2p ¼ ‘you all’ (WS, 254) e

e

e

e

e

e

e e

e

e

e e

e

e

ee

e

e

e

e

258

Patricia A. Shaw

t l- ´n c 21 from-where ¼ ‘from where?’ e e e

e

In (23a) it is to be expected that Wgt-to-Str would initiate trochaic foot construction with the second-syllable full vowel, leaving the initial schwa syllable as stray. But, in (23b–d) where there is a sequence of syllables headed by schwa, stress would be expected on the basis of phonological criteria alone to fall on the leftmost schwa. Specifically, given initial stress , on words like [w ´w s] ‘tree frog’ , and similar data of (6), one would expect *[w ´p k w ] rather than the attested [w p ´k w ] in (23b). Similarly, in the trisyllabic forms in (23c), one would expect initial stress, as previously attested by the data of (9) in words like [ ´n x w y t] ‘stop oneself ’. The morphological parsings of these contrastive data sets are compared in (24), where the left MWord- and MRoot-domain edges are identified: e e

e e

e e

MRoot [w ´w , s [p ´k w [ ´n x w y t [k w ´yw-y t e e

MWord [ [w [ [w -s-

w ´w s, w p ´k w  ´n x w y t w sk w ´ywy t

e e e e e e

e e

e e e e e e e e e e

e e

a. b. c. d.

e

(24)

‘tree frog’ (16) ‘suddenly billow forth’ (23b) ‘stop oneself ’ (9) ‘he moved (himself )’ (23c)

The morphological composition of these words clearly provides a systematic explanation for the observed di¤erence in stress patterns. The unifying generalization for the data represented by (23) and (24) is that stress falls on the initial syllable of the MRoot. Data such as these provide strong empirical evidence for Al-L(Ft, MRt) (13b). Stress feet are aligned with the left edge of the internal morphological MRoot, rather than with the MWord edge. Further corroboration of MRoot alignment is provided by data like (23d), which have a sequence of four schwas. If footing originated at the leftmost MWord margin, one would expect exhaustive parsing into two feet as in (25a). The actual parse, in (25b), best satisfies the three previously motivated constraints—Parse-s-2  AlL(Ft, MRt)  Ft-Bin-s—as shown in tableau (25). MRoot [

n c

Parse-s-2

[(t `l [ ) (n ´c ) e

[t l [( ´n ) c

c.

[t l [( ´) n c

Ft-Bin-s

e

e

+ b.

Al-L(Ft, MRt) *!

e e

a.

l-

e

MWord [t

e e e

(25)

*!

*

e e

e e

e

e e

The fact that it is optimal in (25) to leave both the initial and the final syllables unparsed follows directly from the analysis and constraint ranking developed in sec-

Inside Access

259

tion 12.3.2, most specifically from the claim in (13b) that feet are aligned with respect to the internal MRoot domain. The core of the above argument is based on the di¤erential behavior of stress on words with no full vowels, only schwas. In some data—for example, (6) and (9)— stress is initial; in other data (23b,c,d), stress is pen-initial. In the absence of any full vowel to bring Wgt-to-Str into play, the criterial role of the MRoot edge emerges as the crucial factor behind the di¤erence between the two data sets. The next argument is based on data in which stress falls on the third syllable from the beginning of the word. Representative examples are given in (26a,b). In each case word-level prefixes attach to an MRoot with a schwa in the initial syllable. Note that (26c) ¼ (10biii), (26d) ¼ (8b). , , a. t ns y, e´m , b. y x w k w ı´q n

MWord MRoot , , [t n[s, ye´m , [y -x w - [k w ¼ı´q n

, , c. q `q ya´s

[

, , [q `q¼ ya´s

d.  `´t na´y n

[

[ `´t n¼a´y n

e

e

e

e

e e

e

e

e e

e

e

(26)

‘from the upper class’ (WS, 279) along-loc-[ascend¼belly ¼ ‘belly up’ (WS, 536) taut¼circular ¼ ‘barrel’ (WS, 313)22 ‘to eat along the way’

e e e e e

e

e e

What is initially striking here is the contrast between the data in (26a,b) where there is an initial sequence of two unparsed schwa syllables, and previously analyzed data, repeated as (26c,d), which motivate a high-ranking role for Parse-s-2 (15b). This contrast reduces directly to the principal claim of this section: MWord-domain prefixes are outside of the purview of stress constraints. Not only are word-level prefixes excluded from the domain of foot alignment, as was seen in (23–25), but they are also excluded from the domain of Parse-s-2. , , Consider the example in (26a), [t ns ye´m], where the only stress is on the third syllable. If foot alignment could not access the left edge of the internal MRoot, we , , would expect secondary stress on the initial syllable, *[t `ns ye´m], just as in examples , , like [q `q ya´s] (26c). However, as shown in the tableau in (27), the interplay of foot alignment to the MRoot edge with constraints on weight and binarity, correctly accounts for the stress pattern in words of this type. e e

e e

e e

e

e

e e e

+ d.

e

c.

, , yem

, , [(t `n [s )(ye´m) , , [t n [(s ´ yem) , , [t n [(s `)(ye´m) , , [t n [s (ye´m) e

b.

MRoot [s

Wgt-to-Str

e

a.

n-

e

MWord [t

e

(27)

Al-L(Ft, MRt)

Ft-Bin-s

*!*

*

*

**!

*

*

*!

260

Patricia A. Shaw

Word-initial stress, as in candidate (27a), is nonoptimal. The only constraint that outranks Al-L(Ft, MRt) here and could therefore compel foot alignment to extend leftward from the MRoot edge would be Wgt-to-Str, but it is not applicable to the MWord-domain prefixes because it lacks a full vowel. Formally, (27a) is disqualified by the double Al violation: the left edge of the first foot (t `n.s ) is misaligned from , , the MRoot by one syllable on the left, and the second foot (ye´m) is misaligned by one syllable on the right. Though the candidate in (27b) does achieve perfect foot alignment with the left edge of the MRoot, it is not optimal overall, because its failure to assign prominence to the final moraic vowel [e] violates the higher-ranked Wgt-to-Str. Parsing both syllables of the root into independent monosyllabic feet, as in (27c), avoids violating Wgt-to-Str, but incurs an extra violation of Ft-Bin-s, fatal in comparison to (27d). Note that even the optimal candidate (27d) has one infraction of Al-L(Ft, MRt). The interesting consequence of aligning feet to the MRoot is that the optimal output (27d) sustains a sequence of two syllables with unstressed schwas. This raises an important question regarding the correlation of well-formedness constraints on prosodic parsing, such as Parse-s-2, with domains defined by alignment constraints. The claim advanced here is that the domains of Parse-s-2 and Al-L(Ft, MRt) directly coincide. Consider first ranking relations, under the strong hypothesis that ranking relations in a particular OT grammar are fixed. It was demonstrated, on the basis of the tableau in (19a), that Parse-s-2 is crucially ranked above Al-L(Ft, MRt). The issue that the tableau in (28), based on (27), foregrounds is this: treating (28b) ¼ (27d) as a violation of Parse-s-2 would result in a ranking paradox, since it would result in the correct output, (28b), not being evaluated as optimal. e

e

(28) Wgt-to -Str, Parse-s-2  Al-L(Ft, MRt)  FtBin-s (cf. (15d)) , , Parse-s-2 Al-L(Ft, MRt) MWord [t n- MRoot [s yem , , ** L a. [(t `n [s )(ye´m) , , *! * b. [t n [s (ye´m) e

e

e

e

e

e

If, however, the domain of Parse-s-2 is formally recognized as coextensive with the domain of stress foot alignment, then MWord-domain prefixes like t n- simply would not count in the evaluation of violations to Parse-s-2, in the same way that they would not contribute to an Al-L(Ft, MRt) violation. This is shown in (29). e

Inside Access

, , t ns ye´m

MWord [t n" N/A N/A e

e e

Parse-s-2/MRt Al-L(Ft, MRt)

MRoot , , [s (ye´m) ‘from the upper class’ " nonviolation (because only one s) violation (by one s) e

(29)

261

A second prediction of establishing the MRoot as the domain of Al-L(Ft, MRt) and Parse-s-2 is that foot construction and stress will be insensitive to the number of of syllables preceding the MRoot domain. Consider, for example, a sequence of two schwa-headed syllables preceding the MRoot in the MWord domain: both will remain unparsed. As a consequence, there should be no stress on either syllable. The form and tableau in (30) confirm this prediction: e

e

, [w -y - [ey e e

, a. [(w `y ) [(e´y) , + b. [ w y [(e´y)

, ¼ [w y e´y] ‘. . . it is (you who will) continue along’ (WS, 520)

Parse-s-2

e e

(30) MWord MRoot , [w - y [e´y [established-along-[continue

Al-L(Ft, MRt) *!*

e e

e e

The candidate in (30a) exhaustively parses syllables into feet. The undesirable consequence, however, is that the leftmost foot incurs two Al-L violations, given its twosyllable distance from the MRoot domain edge. In contrast, the candidate in (30b) leaves both the MWord prefixes unfooted. Under the hypothesis that the highly ranked Parse-s-2 constraint has jurisdiction only in the MRoot domain, the two MWord-domain prefixes may remain unparsed without violating Parse-s-2. To summarize, then, the morphological structure outlined in (2) correlates with di¤erent prosodic domains. Prefixes in the MWord domain, even multiple prefixes, consistently do not get stressed. The formal interpretation of these facts is that prosodic constraints on footing (Parse-s-2 and Al-L(Ft, MRt)) apply with explicit reference to the MRoot domain. Recall that the initial observation in this section was that no MWord-domain prefix has a lexical full vowel: only epenthetic schwa surfaces in this domain.23 Given the argumentation throughout sections 12.3 and 12.4.1 that Wgt-to-Str is highly ranked (as in (13) and (15)), it is clear that the lexical absence of full vowels in any of the prefixes within the MWord domain is not accidental. Their systematic absence preserves the significant generalization that foot alignment functions optimally to signal the word-internal MRoot edge.

262

Patricia A. Shaw

12.4.2

MStem-Domain Prefixes

There are several di¤erent reduplicative morphemes, distinct in form and function, which may occur as a prefix to a root. These reduplicative prefixes essentially define the morphological content of the MStem domain. The initial argumentation in this section draws on the interaction of stress with examples from each of two principal reduplicative canons on verb roots: an open-syllable pattern, marking Progressive aspect, and a closed-syllable pattern, marking Plural.24 Infixation of an -l- Plural allomorph provides further support for the prosodic relevance of the MStem domain. The first pattern to be considered, building on Shaw et al. 1999, involves a reduplicative CV-prefix that, in conjunction with glottalization of subsequent resonant(s), marks the Progressive aspect of certain classes of verb stems.25 As seen in (31), if the MRoot has a lexically full vowel, the Progressive reduplicative prefix surfaces with a copy of the vowel and concomitant reduction of the MRoot vowel to schwa: Progressive , , pe´-p t’, y , ya´-y k w - t , , q w´ı-q w m , nı´-n w, t xa´-x k w - t e e e e e

Root (-TR) , pe´t’, y ya´k w - t , q w ´ım , nı´w, - t xa´k w - t

e e

e

e

e

a. b. c. d. e.

e

(31)

Gloss ‘sew’ ‘smash it, break it up’ ‘disembark, get out of ’ ‘advise him’ ‘bathe him’

Two properties of these forms need to be accounted for. First, the lexically distinctive vowel quality of the MRoot surfaces only in the reduplicant. Second, the Progressive reduplicative prefix is systematically stressed in such forms, clearly establishing that stress can extend leftward beyond the MRoot-domain edge. In the analysis proposed below, these properties are integrally related. Consider first the fact that in (31) the lexically distinctive quality of the MRoot vowel surfaces only in the reduplicant, not in the base; thus neither Base-Reduplicant (BR) faithfulness nor Input-Output (IO) faithfulness in vowel identity is satisfied. The relevant faithfulness constraint is MaxNuc, which subsumes faithfulness in both moraic and featural content. Assuming the Full Model of reduplication developed in McCarthy and Prince 1995, prefixal vowel quality in Progressive reduplication can be accounted for with a highly ranked Input-Reduplicant (IR) faithfulness constraint, MaxNucIR (32a).26 Setting aside for the moment the question of domain alignment, the tableau in (32b) shows how the previously motivated constraints Wgt-to-Str (13) and Ft-Bin-s (15) interact with MaxNucIR . The other FaithNuc constraints, ranked below Ft-Bin-s, are not shown here. (32) a. MaxNucir A nucleus in the Input has a correspondent in the Reduplicant.

Inside Access

263

vi. vii. + viii.

Ft-Bin-s *

*! *!* *!

*

*!

*

*! *!

e

v.

*!

e e

iv.

MaxNucir

e

iii.

Wgt-to-Str

e

ii.

‘be sewing’ (¼ (31a))

e

i.

MRoot , [p t’ y , [Prog-[pet’ y , , pe-(pe´t’ y ) , , (pe´-pet’ y ) , , (pe`-)(pe´t’ y ) , , p -(pe´t’ y ) , , p -(p ´t’ y ) , , (p ´-p t’ y ) , , (pe´-pt’ y ) , , (pe´-p t’ y ) e

b. MStem , [pe´-

Pairwise comparisons of candidates in this tableau establish that identity of lexically distinctive vowel quality between the Base and the Reduplicant is sacrificed in favor of an optimal Heavy-Light moraic trochee, in conformity with high-ranking prosodic constraints on Wgt-to-Str and Ft-Bin-s. Now let us turn to an account of the stress pattern exhibited by Progressive reduplication. Considered in isolation, the data in (31) might lead one to think that foot alignment extends to the left edge of the MStem domain, thereby including the reduplicative prefixes in its purview. However, two other sets of data show that this is not the case, and that stress aligns with the MRoot edge. The first set of data involves underlyingly vowelless roots (see (22)); in the Progressive stems of such roots, the reduplicative prefix takes the form C -. As expected, where syllabification requires a nuclear head, a default epenthetic schwa appears. Representative data are given in (33). In these data, schwa is not included in underlying representation (see section 12.2.1). e

Reduplicated Progressive stem t -t ´s , , , c -c ´q w , s, -s ´q, k w -k w ´´ , x w -x w ´y ´ -´ ´t’- m t’ -t’ ´p l’ e e e e

Unreduplicated MRoot t ´s , , c ´q w , s, ´q k w ´´ x w ´y ´ ´t’- m t’ ´p l’ e e e

e e

e e e e

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

MRoot /ts/ ,, /cq w / , /sq , / /k w ´/ /x w y/ /´t’/ /t’pl’/

e e e e e e e e e e e e

(33)

MRoot gloss ‘approach’ ‘get pierced’ ‘get torn, get split’ ‘spill over, tip over’ ‘wake up’ ‘rake herring’ ‘play cards’

264

Patricia A. Shaw

What is of immediate interest here is the locus of stress in the Progressive forms. The data of (31) established that the Progressive reduplicant is, at least potentially, in the stress domain. One might therefore expect a stress pattern like *[t ´-t s] (33a) or *[´ ´-´ t’- m] (33f ), parallel to the stress exhibited by MRoot-initial words beginning with a sequence of schwas—for example, [w ´w s] ‘tree frog’ (6f ) or [ ´n x w y t] ‘stop oneself ’ (9a). The di¤erence between the stress patterns in (31) and (33) shows that stress is not predictable simply on the basis of phonological properties of a word; reference to internal morphological domain structure is also essential. When, as in (33), nuclear weight is not present as a dominant factor, it is apparent that foot alignment to the MRoot is in force. The absence of lexically distinctive vowel quality in (33) means that Wgt-to-Str is not a factor, allowing the lower-ranked Al-L(Ft, MRt) (13b) to emerge. This is illustrated in the following tableaux, where Progressive reduplication of vowelless /CC/ roots in (34a) is compared with Progressive reduplication of /CVC/ roots in (34b). Recall that in the latter case, high-ranking MaxNucIR (32) ensures retention of the underlying vowel quality in the reduplicative prefix: e e

e e e

e e

e

e

[Prog- [t s

‘be approaching’ (MRoot ¼ /ts/) (¼ (33a)) Wgt-to-Str

e

i. [(t ´- [t s)

Al-L(Ft, MRt)

e

e

*

e

e

MRoot , [p t’ y , [Prog- [pet’ y , , i. [(p ´-[p t’ y ) , , ii. [pe-[(p ´t’ y ) , , + iii. [(pe´-[p t’ y )

Ft-Bin-s

*!

+ ii. [t - [(t ´s) b. MStem , [pe´-

e e

MRoot [t ´s

e

(34) a. MStem [t -

, ‘be sewing’ (MRoot ¼ /pet’ y /) (¼ (31a); cf. (32)) Wgt-to-Str

MaxNucir

Al-L(Ft, MRt)

*!

*

e

e

*! *

e

e

e

The incorporation of an MStem syllable into a stress foot in (34b) is the straightforward consequence of the previously well-established ranking of Wgt-to-Str above Al-L(Ft, MRt). The second set of data that confirms the role of the MRoot domain edge even in forms with MStem prefixes involves Plural/Distributive/Iterative (henceforth Pl) re,, , , duplication. Although an analysis of the full range of Pl allomorphy in h nq min m is beyond the scope of this chapter, two of its reduplicative realizations are directly relevant. Their stress behavior supports the morphoprosodic generalizations estabe

e

e

Inside Access

265

lished above. As shown in (35), Pl reduplication prefixes a CVC reduplicant. In stems with an underlying full vowel, high-ranked MaxNucIR preserves the full vowel in the reduplicant and results in stress on the reduplicative prefix, parallel to what was seen with Progressive reduplication in (31): Underlying MRoot , w w a. /k aq / b. /t’ax w / c. /t’ic m/

Reduplicated PL stem , , w´ w w w k aq -k q t’a´x w -t’ x w t’ı´c-t’ c m e

e e e

(35)

MRoot gloss ‘get hit’ ‘go/come down’ ‘swim’

e

Underlyingly vowelless roots exhibit a C C-reduplicative Pl prefix, as seen in (36). Parallel to the Progressive forms in (33), stress falls on the MRoot, rather than on the reduplicative prefix. e

Unreduplicated MRoot , , c ´q w l ´k w p ´n- t , n ´p- x

Reduplicated PL Stem , , , , c q w -c ´q w l k w -l ´k w p n-p ´n- t , , n p-n ´p- x e e e e

e e e e e e

a. b. c. d.

Underlying MRoot ,, /cq w / /lk w / /pn/ , /np/

e e e e e e

(36)

MRoot gloss ‘get pierced’ ‘get broken’ ‘bury -tr’ ‘eat, chew -tr’

The analysis of Progressive stress in (32) carries over directly to these Pl stems: in the absence of an underlying full vowel, Wgt-to-Str is not a factor, and Al-L(Ft, MRt) compels alignment of the foot to the MRoot. Compare the tableau in (37a), in which Wgt-to-Str stresses a full-vowel reduplicant, to that in (37b), in which the impact of the lower-ranked Al-L(Ft, MRt) emerges. (37) a. MStem , [k w a´q w -

MRoot , , [k w q w (MRoot ¼ /k w aq w / ‘get hit’) (¼ (35a)) , Wgt-to-Str Al-L(Ft, MRt) [Pl-[k w aq w , , * + i. [(k w a´q w - [k w q w ) , , ii. [k w aq w - [(k w ´q w ) *! e

e e

b. MStem , , [c q w -

MRoot , , ,, [c ´q w (MRoot ¼ /cq w / ‘get pierced’) (¼ (36a)) ,, Wgt-to-Str Al-L(Ft, MRt) Ft-Bin-s [Pl-[cq w , , , , *! i. [(c ´q w - [c q w ) , ,w , ,w * + ii. [c q - [(c ´q ) e

e e

e

e

e

In all of the data sets in (37), there is a single syllable to the left of the Root. An important question is whether Wgt-to-Str can draw Foot-headedness more than

266

Patricia A. Shaw

one syllable to the left of the Root. Interestingly, various kinds of evidence converge on the conclusion that a foot edge can be misaligned from the MRoot edge by no more than one syllable. One type of evidence leading to this conclusion derives from the interaction of the Plural -l- infix, realized as [- l-] or [-l -] depending on context,27 with CV-diminutive reduplication. Consider the following data: e

(t’ y a´t’ y - (l-a´(t’ y a´-l -)

e

h- (l- e´-

MRoot (t’ y a´m) t’ y m) t’ y m) t’ y m , (he´wt) , h wt) e

e

e

e

e

i. ‘rat’ ii. dimþpl

MStem

e

e

e

e

b.

MWord sss*s-

s-t’ya´m , s-t’ y a´-t’ y m , s-t’ y - l-a´-t’ y m , *st’ y a´-l -t’ y m , he´wt , h- l-e´-h wt

i. ‘bone’ ii. dim iii. dimþpl

e

a.

e e e

e

(38)

As seen in (38aiii,bii), the Plural infix disrupts the contiguity of the reduplicative Diminutive prefix (violating Contiguity). As a consequence, even when both Stem prefixes are present, the full vowel of the Diminutive prefix is not dislocated more than one syllable away from the MRoot edge. The ungrammaticality of the starred form in (38aiii) follows directly from the independently motivated ranking of Wgtto-Str over Al-L(Ft, MRt), as shown in (39).

Wgt-to-Str *!

e

*!

‘little bones’ (¼ (38aiii))

Al-L(Ft, MRt)

Contiguity

*

e

**!

e

e

e

e

e

+ iv.

[s- [dim-l- [t’ y am , [s [t’ y al [(t’ y ´m) , [s [t’ y la [(t’ y ´m) , [s [(t’ y a´l ) [t’ y m , [s [t’ y (la´ [t’ y m)

[st’ y la´t’ y m]

e

iii.

e

ii.

e

i.

MRoot t’ y am/ t’ y m)

e

b.

MStem dim-lt’ y -(la´-

e

(39) a. MWord ss-

*

*

All of the candidates in (39b) satisfy the high-ranked MaxNucIR (not shown) as well as a constraint, not formalized here, aligning the reduplicative Diminutive prefix with the left edge of the MStem. Candidates (39bi,bii) show that, regardless of whether Contiguity is maintained in the reduplicant or not, perfect alignment with the MRoot has the fatal consequence of violating Wgt-to-Str. Candidate (39biii) moves the foot head further left into the MStem domain, thereby maintaining Contiguity and respecting Wgt-to-Str, but at the expense of a double violation of Al-L(Ft, MRt). The candidate in (39biv) misses perfect foot alignment by one syllable, and the plural infix violates Contiguity, but it is optimal compared with the

Inside Access

267

alternatives. Misalignment by only one syllable leaves the bisyllabic foot straddling the MStem and MRoot domains. The only situation that could compel a foot to be realized fully to the left of the MRoot, as in losing candidate (39biii), would be a sequence of MStem prefixes the leftmost of which contained a full vowel. In such a case, Wgt-to-Str would overrule Al-L(Ft, MRt). However, the language conspires to avoid this scenario. The only source for a full vowel in an MStem prefix is reduplication; all realizations of a reduplicative full vowel appear immediately before the MRoot. The internal MRoot edge also plays a crucial role in stress foot alignment in forms where the MStem domain is polysyllabic but contains no full vowel. This situation is exemplified in the following data, where the Plural [- l-] @ [-l -] infix surfaces after the initial Progressive reduplicative prefix of an MRoot lacking a full vowel (Suttles 2004, 170): e

[prog- pl- MRoot [MRoot]] , ‘crack’ [s - l - [(s q) , ‘break’ [t’ - l - [(t’ q)

, [s l s ´q] , [t’ l t’ ´q]

‘several are cracking’ ‘several are breaking’

e ee e ee

e

e e e e

e

MStem , a. /sq/ , b. /t’q/

e

(40)

These forms have primary stress on the final odd-parity schwa-headed syllable, conspicuously inconsistent with the stress pattern previously encountered in trisyllabic words like those in (9), repeated in (41), that also have three schwa-headed syllables in a row: [ ´n x w y t] [´ ´q ll x w ] e e e e e

xw) y t w MRoot [(´ ´q l) l x MRoot [( ´n

e

e

e e e e

e

(41) a. b.

‘stop oneself ’ ‘know’

The stress di¤erences between (40) and (41) are predictable from their respective internal morphological structure, summarized below: .C . MRoot [(C ´C) MRoot [(C ´.C ) .C C

e e

MStem [

e e e

MStem [C

e

(42) a. Examples in (40): b. Examples in (41):

The apparent anomaly in (40) of having a degenerate (C ´C) foot at the right margin of a word thus follows straightforwardly from Al-L(Ft, MRt); what distinguishes (40) from (41) is that the final schwa syllable is initial in the MRoot. A tableau showing the optimal parse of data such as (40), following the established constraint ranking, is given in (43): , (43) s l s ´q ‘several are cracking’ (MRoot ¼ /sqª/) (¼ (40a)) , Wgt-to-Str Al-L(Ft, MRt) MStem [Prog-l- MRoot [sq , a. [(s `l ) [(s ´q) *!* , + b. [s l [(s ´q) e

e ee

e

ee

e

ee

268

Patricia A. Shaw

Finally, bringing together all of the claims of the proposed analysis, consider data where all three morphological domains have overt content. The Progressive reduplicative forms in (44a,b) show that when there is a full vowel in the MStem domain, it will attract stress through the ranking of Wgt-to-Str  Al-L(Ft, MRt). The form in (44b)28 a‰rms that stress footing is iterative, with Rightmost assigning relatively greater prominence to a foot at the right edge of the word. In the form in (44c), stress does not fall on the Progressive reduplicant prefix, since it has no full vowel to bring Wgt-to-Str into play. The apparent failure in (44c) to parse the initial sequence of two schwa-headed syllables under Parse-s-2 (15) falls out from the hypothesis that MStem prefixes, like MWord prefixes (see (28)–(30)), are external to the domain of stress footing. Parsing weightless prefix syllables into feet would incur unnecessary violations of Al-L(Ft, MRt).

e

e

e

e e

e

eee

e

e

e

e

e

e e e

e

e

12.5

[t wı´t t] ‘like it’s sleeping’ (WS, 255) , , , , [y pe`p ¨’xe´n m] ‘feeling along with feet’ [w sw wı´st x w ] ‘it was in plain view’ (WS, 542) [st l t tı´w n] ‘a¤ectionate (dim.) term for siblings or living cousin’s children’ (WS, 479) e

d.

e

c.

MRoot [t t) sleep , , , [p ¨’) (¼xe´n- m) feel¼foot-intr [(wı´-st- x w ) reveal-caus-3agr [(tı´w n) sibling e

b.

MStem [(ı´prog, [(pe`prog[w prog[t - l - t Pl- Pl- dim-

e

a.

MWord [t wlike[y along[w -sest-s[snom-

e

(44)

Conclusions

Several types of argumentation have been presented in support of the general claim ,, , , that in h nq min m, word-internal morphological structure plays an independent and critical role in stress assignment. The focus of the present analysis has been the argument that prosodic constraints on the parsing and alignment of stress feet must make explicit reference to the left edge of morphological domains, specifically the MRoot domain, as summarized in (45): e

e

e

(45) [MWord Nonredup prefixes [MStem Redup prefixes [MRoot Lexical root . . . j

Footing domain: Al-L(Ft, MRt) Parse-s-2 The prosodic e¤ects of morphological su‰xes or clitics at the right edge remain unaddressed here, and constitute an important area for further research.29

Inside Access

269

The principal theoretical issue addressed in this chapter is to what degree reference to internal morphological structure such as the MRoot can be constrained. The Generalized Alignment schema by definition (McCarthy and Prince 1993a) allows reference to the domain edge of any PCat or MCat in constraint evaluation, without restriction on access to internally embedded hierarchical domains. In comparison with earlier and much more restrictive theoretical hypotheses like Bracket Erasure and Strict Cyclicity, the relative freedom of access at the Output level to internal constituency permitted by Generalized Alignment might seem unduly powerful. In ,, , , this case, the complex morphoprosodic interactions of h nq min m have provided compelling empirical support for the necessity of recognizing the internal left-edge boundary of the internal major lexical category MRoot. Significantly, Kiparsky’s (2000c) model of the phonology-morphology interface in OT requires that the internal morphological and prosodic constituency of Stem and Word both be accessible to phonology, thus allowing cyclic e¤ects to be captured through IO correspondence. The argumentation for including the MRoot as a fundamental morphological category that may be recognized at the Output level is patently plausible, given that the MRoot constitutes the fundamental lexical unit within any theory of morphology. In fact, if constraints on accessibility are to be constructed with reference to the basic constituent hierarchies of morphology and prosody, it would seem highly implausible for MRoot to have to be excluded. The inclusion of the MRoot in the list of morphological constituents to which phonology may refer also connects to crosslinguistic findings regarding cognitive processing and learnability. It is well known that prefixes frequently do not belong to the principal phonological constitutent of a morphological word (CzaykowskaHiggins 1998, 180; see also Cohn 1989 on Indonesian and Szpyra 1989 on Polish, ,, , , among others). H nq min m prefixes pattern similarly: with the systematic exception ,, of the minimal alignment violations documented in section 12.4.2, prefixes in h nq , , min m are not parsed into stress feet and hence are not part of the Prosodic Word. ,, , , Stress is demarcative in such systems; in h nq min m, a polysynthetic language with complex prefixal concatenation, stress footing functions to identify the major internal morphological category, the MRoot. Any theoretical model must provide clear and constrained mechanisms to capture this kind of ‘‘inside access’’ to morphological constituency. e

e

e

e

-e

e

e

e

e

e

e

e

Notes I am particularly indebted to Darin Howe, Jill Campbell, and Larry Grant for their invaluable contributions to the Musqueam research program. I also thank Luigi Burzio, Laura Downing, Sharon Hargus, Sharon Inkelas, Rene´ Kager, and an anonymous reviewer for helpful discussion, as well as Dale Kinkade, Wayne Suttles, Larry Thompson, and Ewa CzaykowskaHiggins for sharing their thoughts about Salish on many occasions.

270

Patricia A. Shaw

1. For issues related to the endangered status of this and other First Nations languages in British Columbia, see Shaw 2001a and Shaw 2004. ,, , , 2. These data represent the ‘‘downriver’’ dialect of the Hul’q’umin’um’-h nq min mHalkomelem continuum, as traditionally spoken in Musqueam and other villages of the Fraser River delta in British Columbia. This research has been conducted under the auspices of the Musqueam-UBC First Nations Languages program. I am profoundly grateful to the late Adeline Point, as well as to the late Dominic Point, the late Edna Grant, and the late Arnold Guerin Sr. for their insightful teachings. I thank my Musqueam colleagues for their collaboration and deep commitment to language revitalization. Additional data are cited from Suttles 2004 and Guerin n.d. (abbreviated WS and AG respectively). The language content belongs to the Musqueam people and the scholarship belongs to the author; the views presented here are those of the author and not necessarily those of the Musqueam Indian Band. e

e

e

3. Sounds in parentheses are marginal within the native lexicon; [sˇ] occurs only as a rounded backed allophone of /s/. Following NAPA (North American Phonetic Alphabet: see American Anthropologist 36(4), 1934) orthographic conventions, the symbol ‘‘c’’ represents an alveolar a¤ricate [t s ], ‘‘cˇ’’ represents an alveopalatal a¤ricate [tS], and ‘‘sˇ’’ an alveopalatal fricative [S]. Glottalization is represented by an apostrophe directly over the symbol, and labialization by a raised w immediately after. 4. Since Proto-Salish *u > a in the dialect groups to which Musqueam belongs (Boas and Haeberlin 1927, 131–132; Shaw 2001b), the short [u]s that are left in the synchronic lexicon derive primarily from loanwords. 5. Phonemic distinctions in vowel length carry a very low functional load; most, if not all, long vowels are analyzable as derived. 6. Due to space limitations, the present analysis does not extend to right-edge domains. Note that van Eijk (1998, 472) presents a model of morphological constituency for Lillooet that, on the basis of quite independent criteria, is directly comparable in its left-edge demarcations to those posited here. Czaykowska-Higgins (1998, 154, 172) posits a more substantive mismatch in prosodic versus morphological domains in Moses-Columbian Nxaamxcin Salish, particularly within the MStem/PStem domains. However, her conclusion that ‘‘stress is never assigned further left than the root morpheme’’ correlates importantly with the present analysis, requiring for Nxaamxcı´n simply a higher (inviolable) ranking of the same Align-L constraint pro,, , , posed here for h nq min m (see section 12.4). Black 1998 and Bar-el and Watt 2001 explore domains in other Salish languages. e

e

e

7. A characterization of the limited set of contexts where schwa is seemingly lexical is beyond the scope of the present chapter. For discussion of schwa in other Salish languages, see among many others Blake 1992, 1999, 2000; Kinkade 1998; Roberts and Shaw 1994; Urbanczyk 1996; Czaykowska-Higgins and Willett 1997; Rowicka 2000; and references therein. For the diachronic complexities, see especially Thompson 1979. 8. Whereas syllable, nucleus, and mora are treated here as theoretical primatives (Shaw 1992, 1993), the terms onset and coda hold no formal theoretical status, and are simply used mnemonically to refer to nonnuclear segments at the left and right edges of the syllable. 9. Abbreviations: agr ¼ agreement; dim ¼ diminutive; intr ¼ intransitive; ls ¼ lexical su‰x; pl.name ¼ place name; pro ¼ pronoun; prog ¼ progressive; nom ¼ nominalizer; n.tr ¼ ntransitivizer (limited control); obj ¼ object; pl ¼ plural; red ¼ reduplicant; refl ¼ reflexive;

Inside Access

271

s ¼ singular; su ¼ subject; t.tr ¼ t-transitivizer (control); ind ¼ independent; 1s ¼ firstperson singular; 3 ¼ third person; past ¼ past tense; instr ¼ instrumental; 2p ¼ second person plural. 10. Certain higher-ranked constraints will allow a sequence of full vowels to surface—for example, the constraint * ]s against a tautosyllabic schwa-glottal sequence blocks -reduction in words like /s-k w aya/ [sk w a´ya] ‘squirrel’. The basic left-aligned trochaic system prevails; see Shaw et al. 1999. , , 11. Suttles (WS, 140, 177) records this as [xa´-x w k w -]; our consultant gives [xa´-x k w -]. e

e

e

e

12. Contra various claims (e.g., Hayes 1995) that the canonical disyllabic trochee is evenly ,, , , balanced—that is, LL ¼ (m:m)—the h nq min m system actively optimizes uneven trochees (m.Ø). Note, however, that under the hypothesis that schwa is nonmoraic, reduction of a moraic vowel to schwa creates an optimal Strong-Weak rhythmic contour within the foot (cf. ,, , , Kager 1999, 173¤. and earlier work). The present analysis of h nq min m therefore predicts that uneven trochees may in fact be the least marked foot canon in other trochaic systems sensitive to a moraic full vowel–versus–nonmoraic schwa distinction as well. e

e

e

e

e

e

13. As motivated in section 12.4, this holds within the domain initiated by the MRoot. 14. See (26) ¤. for an analysis of systematic exceptions to this generalization. 15. See Shaw et al. 1999 for arguments against a foot-reversal analysis. 16. Unfooted syllables violate Parse-s-to-Foot (not included in the tableaux here), ranked below Foot-Bin-s. 17. There are various formalizations of this generalization, the basic intent being that one syllable may remain unparsed, but a sequence of two cannot. See especially Selkirk 1984 and Elenbaas and Kager 1999, among others. 18. The crucial ranking of Al-L above Ft-Bin-s is established in (34a). 19. See Suttles (2004) for a characterization of aspectual, modal, and derivational properties of these prefixes. Questions of apparent homonymy and cliticization remain important areas for further research. 20. See note 10 for explanation of the final unstressed [e] here. , , 21. The heteromorphemic l- sequence merges to [l] in fast speech: [t l ´n c ]. The posttonic schwa is variably realized or deleted in fast speech. e e ee

22. Although Suttles 2004 and Guerin n.d. both regularly mark primary stress, neither documents secondary stress consistently. The stress patterns in (26) and elsewhere were attested in my consultants’ speech and/or in Guerin’s (n.d.) documentation. 23. See Shaw 2002 for an analysis of word-initial clusters and epenthesis. 24. Although the collocation of more than one reduplicative prefix ,is, highly constrained, , , examples of Diminutive with Progressive—for example, [til m] ‘sing’, [t tit l m] ‘a little one (child) is singing’—are recorded by Suttles (72¤.), as well as Plural plus Diminutive forms, such as (38aiii,bii), from my own database. Such forms obey MaxNucIR in the Progressive and follow the stress generalizations described here. ee

e

e

25. Only Progressives formed by reduplication are considered here. See Hukari 1978 and Suttles 2004 for classification of other Progressives. ,, , , 26. IR faithfulness is not a property of all reduplicative prefixes in h nq min m; (32a) applies to the Progressive as well as to the Plural, in (35). Discussion of the implications of these facts e

e

e

272

Patricia A. Shaw

for recent claims about reduplicative faithfulness by Struike (2000) and Inkelas and Zoll (2000) is unfortunately beyond the scope of the present chapter. 27. See (40). A full analysis of the constraints governing the locus of the Plural infix is beyond the scope of this chapter (see Shaw 2003). The basic generalization is that the Plural infix is aligned as close as possible to the beginning of the MStem domain, or to the beginning of the MRoot if there are no MStem prefixes, subject to higher ranking constraints on foot alignment and reduplicant anchoring. The epenthesis of an adjacent schwa follows from independent constraints on syllabification. 28. The full gloss from Arnold Guerin Sr.’s lexical files is ‘to be feeling along with the feet (in shallow water looking for cockles)’. 29. See Shaw et al. 1999 for preliminary hypotheses regarding prosodic roles of lexical su‰xes.

13

Morphosyntactic Correspondence in Bantu Reduplication

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

13.1

Introduction

Since the pioneering work of Marantz (1982), Kiparsky (1986b), McCarthy and Prince (1986), and others, the primary goal in the study of partial reduplication has been to construct a theory that insightfully captures the full range of considerations that speakers may invoke in determining how a reduplicant will relate to its base. Given that both phonology and morphology are potentially involved, this has meant two things. First, there has been an attempt to characterize the reduplicant in prosodic terms: the shape of a reduplicant is frequently defined by reference to foot, syllable, and/or moraic structure. Second, the literature has shown an increasing awareness of the role of morphological structure in determining the base-reduplicant relationship. Researchers such as Downing (1997a, etc.), Urbanczyk (1996), and McCarthy and Prince (1993b, 1995) have shown that, in addition to prosodic constraints, the realization of a reduplicant may be influenced by purely morphological conditions. In Bantu verb-stem reduplication, for example, simplex stems may reduplicate di¤erently from polymorphemic ones, which may show further di¤erentiations in turn, depending on whether the a‰xes are derivational versus inflectional in nature. Some of these morphological conditions can be quite subtle, and yet, as we will show, provide crucial evidence for our very conception of how and where (partial) reduplication takes place within a grammar. In short, reduplication provides an ideal testing ground for theories of morphology, phonology, and their interface. In this chapter we provide a detailed description of verb-stem reduplication in Ndebele, a Southern Bantu language of the Nguni group, which also includes Zulu, Xhosa, and Swati. We show that the reduplicant in Ndebele is not only conditioned by phonological and morphological factors, as in other Bantu languages, but that these factors are ‘‘abstract’’ in nature: despite surface appearances to the contrary, the reduplicant of an Ndebele verb stem must be analyzed as a verb stem itself (cf. Downing 1997a, etc.). Its surface form is obtained not by surface correspondence to the base output, but rather by direct spell-out of its own (identical) morphosyntactic

274

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

structure, which, in turn, is a direct copy from the base. As schematized in the (typically) prefixal reduplication in (1), (1) Morphosyntactic representation (base)

the input is a left-branching morphosyntactic structure, where the deepest embedded morpheme is the root—for example, a verb root in our study. The surface outputs to the right of the arrow are obtained in the following way. First, we copy this morphosyntatic representation as a reduplicant to the left of the input base. We then spell out both structures independently. What we have identified as the surface base is derived by the normal rules of word formation, constrained by (sometimes conflicting) considerations such as compositionality, morphotactic restrictions, and so on. The surface reduplicant is also subject to these general considerations, as well as those imposed specifically on the reduplicant. The reduplicant may thus be subject to prosodic as well as morphological constraints. In Ndebele, for instance, the reduplicant is limited to two syllables and inflectional su‰xes may not appear within it. Our conception of verb stem reduplication in Ndebele in (1) di¤ers from most other conceptions in two ways. First, we make explicit that reduplication is a morphological process. While no one would contest this conclusion, research on partial reduplication has mostly been conducted by phonologists who emphasize surface base/reduplicant correspondence, and hence view morphology in terms of surface morphs rather than morphosyntactic structure. With Inkelas and Zoll (2005) we do not assume a direct phonological correspondence between the reduplicant and the base. Second, we take the position that partial reduplication is derived from total reduplication. The conception in (1), however, goes beyond phonological ‘‘full-copy’’ theories such as Steriade 1988, in explicitly treating reduplication as the total copy of the abstract morphosyntactic structure of the base. If there are no special phonological or morphological conditions on the reduplicant beyond those characterizing the base, we in fact derive total reduplication on the surface—an apparent compounding of a base with itself. If there are special conditions, we obtain partial reduplication. We thus agree, in part, with Eulenberg (1971, 73), who states that ‘‘cases of so-called partial reduplication are simply phonological reductions, sometimes drastic, from cases of full reduplications.’’ As will be seen below, we propose to revise Eulenberg’s statement to read ‘‘phonological or morphological reductions.’’ In many, if not most cases of partial reduplication, there will be no di¤erence between our morphosyntactic approach versus the ‘‘morph’’ approaches that have

Morphosyntactic Correspondence in Bantu Reduplication

275

characterized the prosodic analysis of reduplication since Marantz 1982 and McCarthy and Prince 1986. This is because in most situations the properties (e.g., linear order) of surface morphs (specifically, a‰xes) generally mirror the underlying morphosyntactic representation. In Bantu, however, there are widespread instances where this is not the case. As we will see in section 13.5, when there is a mismatch between the surface order of su‰xes versus the underlying morphosyntactic representation, it is the latter that determines what can appear in the reduplicant. The chapter is organized as follows. In section 13.2 we provide a basic overview of verb-stem reduplication in Ndebele. In subsequent sections we treat complications arising in the reduplication of stems containing subminimal or ‘‘consonantal’’ verb roots (section 13.3), fusion or ‘‘imbrication’’ of perfective -ile (section 13.4), and the passive su‰x -w- (section 13.5). We then conclude by considering synchronic and diachronic implications of our findings. 13.2

Basic Overview

We begin by considering the basic properties of verb-stem reduplication in Ndebele. In (1) we first consider verb stems that consist of a ‘‘long’’ root (bCVC) and the default inflectional final vowel su‰x -a. The meaning of such reduplications generally is to do the action in little bits, here and there, perhaps not very well. The forms are given in their minimal citation form—that is, minus inflectional prefixes (and with underlying High tone marked on the first vowel of verb roots). With one major exception that we will examine in section 13.3, prefixes are irrelevant to stem reduplication. We will adhere to the notational convention of separating reduplicant and base with the ‘‘þ’’ symbol, reserving ‘‘’’ for internal morpheme breaks:1 (2)

Plain verb stem a. lim-a thum-a (H) b. nambith-a (H) thembuz-a (H/L)

Reduplicated verb stem lim-aþlim-a thum-aþthum-a nambiþnambith-a thembuþthembuz-a

‘cultivate’ ‘send’ ‘taste’ ‘go from wife to wife’

The pattern in (2a), in which the verb root ¼ CVC, shows a total reduplication of the verb stem. In (2b), however, where the verb root > CVC, we see that the reduplicant is limited to two syllables—that is, to a bisyllabic foot. In both sets of examples the reduplicant is identical to the first two syllables of the base verb stem. However, when we turn to cases of productively su‰xed roots in (3), we find that something more subtle is going on. In these examples, where -el- is the applicative su‰x and -is- is the causative su‰x, we see that there are two possible shapes of the reduplicant: it can either be identical to the first two syllables of the base, hence lim-e and lim-i, respectively, or it can end in [a], thus lim-a.

276

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

(3) a. lim-el-a b. lim-is-a

lim-eþlim-el-a lim-aþlim-el-a lim-iþlim-is-a lim-aþlim-is-a

‘cultivate for/at’

(applicative -el-)

‘make cultivate’

(causative -is-)

To account for this variation, consider the traditional view of the internal structure of the Bantu verb stem in (4). (4)

cf. Meeussen 1967

As seen, a verb stem consists of a base and an obligatory final vowel (FV) morpheme, which is [a] in most verb forms. Within what Bantuists refer to as the ‘‘base,’’ a root may be ‘‘extended’’ by derivational su‰xes (or extensions). Among the Ndebele extensions that we consider in this study are applicative -el-, causative -is-, and passive -w-. The second variants in (3a) and (3b) show that the extension vowel may optionally not appear in the reduplicant, in which case the reduplicant is pronounced with a final [a]. As seen in (5a), this second variant ending in [a] is not available if the verb root is polysyllabic: (5) a. nambith-a (H) *namb-aþnambith-a thembuz-a (H/L) *themb-aþthembuz-a b. casuk-a casuþcasuk-a *cas-aþcasuk-a casul-a casuþcasul-a *cas-aþcasul-a

‘taste’ ‘go from wife to wife’ ‘become nauseated’

/cas-uk-a/

‘nauseate’ (transitive)

/cas-ul-a/

A final [a] is also not possible in (5b), where the input verbs carry the nonproductive reversive su‰xes -uk- and -ul-. Finally, to complete this introduction to the basics, note in (6) that the reduplicant cannot include inflectional material occurring in the so-called final vowel slot: (6) a. lim-e

!

b. lim-i

!

c. lim-ile

!

lim-aþlim-e *lim-eþlim-e lim-aþlim-i *lim-iþlim-i lim-aþlim-ile *lim-iþlim-ile

(subjunctive -e) (negative -i) (perfective -ile)2

Morphosyntactic Correspondence in Bantu Reduplication

277

These examples are significant for two reasons. First, they show that inflectional suffixes are outside the scope of reduplication. Second, they show that the source of reduplicant-final [a] is not necessarily the base, as one might have presumed from [a]-final reduplicants in (3). The data in (6) make it clear that there is an independent source for reduplicant-final [a]. In all of the above respects, Ndebele verb-stem reduplication is equivalent to that in Swati, for which Downing (1997a, etc.) proposes the Bantu verb-stem structure in (7). (7) Bantu verb stem according to Downing; cf. Mayers 1987 a. I-stem ¼ Inflectional stem b. D-stem ¼ Derivational stem c. Root ¼ minimal D-stem d. Xþa ¼ ‘‘canonical stem’’ (CS)

As seen, the full verb stem is referred to as an inflected stem (or I-stem), which in turn has two parts: (i) a D-stem, which may be potentially extended by derivational suffixes, and (ii) an inflectional final su‰x. The root is identified as the minimal D-stem. In (7) this structure is exemplified with the root lim- ‘cultivate’, the applicative and causative extensions -el- and -is-, and the inflectional final vowels -e, -i, -ile, and -a. Downing refers to any verb stem that ends in the final vowel -a as a ‘‘canonical stem.’’ To summarize the facts we have seen thus far, and as indicated in the last line of (7), the root -lim- must be copied into the reduplicant, the vowels [e] and [i] of applicative -el- and causative may optionally copy, and the [e] and [i] of the inflectional su‰xes -e, -i, and -ile may not copy. If the structure of the input ‘‘base’’ stem is as in (7), the analytical challenges before us are the following. First, by what means do we ensure that only derivational su‰xes, not inflectional ones, are available to reduplicate?

278

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

Second, by what means do we ensure that the CVC-a reduplication pattern is permitted with CVC-VC-V stems but not with CVCVC-V stems? That is, why are forms such as lim-aþlim-el-a and lim-aþlim-is-a acceptable, while *namb-aþnambith-a and *themb-aþthembuz-a are not? Downing’s (1997a,c) position is that in the case of CVC-a reduplicants, the CVC- string must correspond to an independently existing minimal D-stem, which lim- clearly is. CVCa RED must be a corresponding word (cf. imperative lima!). On this account, the reduplicants *namb-a and *themb-a are disallowed because namb- and themb- do not exist as roots, and therefore the requisite minimal D-stems, *namb-a and *themb-a, do not independently exist. Steriade (1997) slightly amends the story to refer to existing versus nonexisting words, noting that lim-a exists as a corresponding imperative verb, while *namb-a and *themb-a do not. In this chapter we do not speak of a reduplicant as corresponding to a stem. Rather, it is a stem, which is in morphosyntactic featural agreement with the following, ‘‘normal’’ stem that it appears to reduplicate. As seen in (8), (8)

reduplication is stem juxtaposition, where Stem1 , the reduplicant, is subject to a bisyllabic size constraint, and, as indicated by the subscript on its morphosyntactic structure, is in perfect featural agreement with Stem2 , the base. At this point we note that there are two inviolable morphological properties of base verb stems: First, verb stems must contain a verb root. Second, verb stems must be morphologically complex. The first property is self-evident: one cannot have a stem of any sort that does not in turn consist of a root. The second property is what interests us here: Verb stems in Ndebele (and in most Bantu languages) must be bipartite (i.e., su‰xed). That is, there must not be a right alignment of the verb root with the verb stem: *[[verb]root ]stem . While it has been hypothesized that preProto-Bantu may not have required a su‰xal vowel on all verb roots (Gre´gorie 1979), verb stems in Ndebele must end in one of the inflectional endings—the most generally distributed one being -a. In this study we follow the tradition of identifying -e, -i, and -ile as inflectional endings—that is, as Downing’s IFS. We will depart from previous scholarship, however, in treating -a not as an inflectional ending, but rather as the default stem su‰x in Bantu in general:

Morphosyntactic Correspondence in Bantu Reduplication

279

(9)

As indicated, -a may be invoked, typically in final position, to ‘‘fill out’’ any kind of stem: D-stem, I-stem, R-stem. The su‰x -a di¤ers from other verb and noun endings in having no corresponding morphosyntactic feature(s); unlike -e, -i, and -ile, it does not directly realize tense/aspect, mood, or polarity distinctions on verbs. While these morphs spell out feature complexes that include [þsubjunctive], [þnegative], or [þperfective], -a appears only in their absence—that is, as a default. We will henceforth refer to -a as the FV, reserving IFS for the other final su‰xes. With this interpretation of -a established, we can now explain why -a can appear in the reduplicant, while IFSs may not: the latter may occur only in an I-stem, while the reduplicant is a D-stem in Ndebele. Since -a is devoid of inflectional features, it may appear as a default ending on any kind of stem—for example, the reduplicant D-stem, which, by definition, will not carry an IFS. Our proposal in (10) is that the R-stem is a daughter of Downing’s I-stem and sister to her IFS: (10)

Each of the stems under the R-stem in (10) is a D-stem, although the reduplicant is limited to two syllables. We assume, as is the case in D-stems, that -a is always available to reduplicants as well. As seen in (11), this gives rise to the variation seen in reduplication of D-stems that have extension su‰xes. We assume, for the sake of argument, that the FV is present even in (11a), where (due to disyllabic size constraints) it is not parsed: (11) Reduplicant D-stem of base -lim-is-a ‘cause to cultivate for’ a. b.

In (11a), the [i] of -lim-i comes from the causative su‰x -is-, while in (11b) it comes from the FV -a.

280

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

Now note with respect to the articulated structure of the reduplicated verb stem in (10) above that both D-stem2 and the I-stem should have to branch. In case D-stem2 is simplex (e.g., -lim-), it would appear to need the FV -a. However, this morpheme vies for the same ‘‘slot’’ as the IFSs -e, -i, and -ile. Therefore, since -a is featureless, the higher-ranked requirement that the IFS be spelled out takes precedence and, in the subjunctive, we get [[lim-] -e], rather than [[lim-a]]. The fully articulated structure of reduplicated verb stems is seen in (12). (12)

The crucial point in (12) is the disassociation of the FV morpheme -a from the IFS category to which all previous studies have assigned it. As seen, the FV -a is considered to belong to the D-stem, while the inflectional endings -e, -i, and -ile, which mark certain subjunctives, negatives, and perfectives, are a property of the I-stem. Since it is a featurally empty morph, the FV -a is able to occur in D-stemRED without disrupting the required morphosyntactic featural agreement between DstemRED and D-stemBASE. Assuming that the morphological makeup of the reduplicant is as given above, we now turn to its phonological requirements, also indicated in (12). We assume a set of statements roughly like the following, which are phrased for convenience in the terminology of Optimality Theory (Prince and Smolensky 1993). They are part of the cophonology (phonological ‘‘level’’; see, e.g., Inkelas, Orgun, and Zoll 1997) of Stem1 in the reduplication construction. They ensure that the reduplicant is of the proper size and determine which elements of the morphological material under the reduplicative D-stem will actually instantiate the bisyllabic reduplicant. (13) D-stemRED cophonology a. [ss] The reduplicant is bisyllabic (¼ a minimal prosodic word) b. MAX(Root) The reduplicant should parse the root (cf., e.g., McCarthy and Prince 1993; Urbanczyk 1995; Futagi 1997)

Morphosyntactic Correspondence in Bantu Reduplication

281

c. MAX(Ext) The reduplicant should parse extension su‰xes (if any) d. MAX(FV) The reduplicant should be faithful to FV -a (if any) The first constraint requires the reduplicant to assume the exact shape of the minimal prosodic word in Ndebele: one foot consisting of two syllables. This is imperative, and causes the truncation we have observed above. (Later we will see cases where this same constraint causes augmentation.) The second through the fourth constraints are violable, with Max(Root) ranked above the other two. The last two statements in (13) thus come into play when the root does not exhaust the bisyllabic reduplicant (i.e., when the root is smaller than CVCV, e.g., the CVC root lim-). As we have seen, there are two options for realizing the bisyllabic reduplicant in such cases: use material from extension su‰xes, if any (as is done in, e.g., lim-eþlim-ela), or use the FV morph, -a (e.g., lim-aþlim-el-a). The equal viability of these two options can be expressed in OT by freely ranking the corresponding constraints (as proposed by Anttila 1997 and Ito and Mester 1997, among others). The rankings of the constraints in (13) are given in (14). (14) [ss]  Max(Root)  Max(Ext), Max(FV) The tableaux in (15) and (16) show how these constraints correctly predict the CVCV- and CVCa- reduplicants of extended CVC roots. (Note our assumption that the filler morph [a] is present in the input.3) (15) Derivation of reduplicant in lim-eþlim-el-a lim-el-a

[ss]

Max(Root)

Max(Ext)

Max(FV)

*(l)

*

+ a. lim-e b. lim-a

**!(el)

(16) Derivation of reduplicant in lim-aþlim-el-a lim-el-a a. lim-e + b. lim-a

[ss]

Max(Root)

Max(FV)

Max(Ext)

*!

*(l) **(el)

In neither tableau is there a violation of [ss] or Max(Root), since the reduplicants are both bisyllabic and incorporate the entire verb root lim-. In (15), Max(Ext) is ranked higher than Max(FV). The winning candidate is thus (15a), lim-eþlimela, since the reduplicant in (15b), lim-a, could have parsed the [e] of the stem lim-el-a, but did not. In (16), the constraint Max(FV) is ranked higher than Max(Ext). In

282

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

this case the winning candidate is (16b), lim-aþlim-el-a, since (16a), lim-eþlim-el-a, violates the now higher-ranked constraint Max(FV). We thus see how the free ranking of the two lowest constraints produces the two reduplicant possibilities. Now compare these results with the corresponding tableaux in (17) and (18), where the input nambith-a has a bisyllabic root: (17) Derivation of reduplicant in nambiþnambitha nambith-a

[ss]

+ a. nambi

Max(Root)

Max(Ext)

*(th)

b. namb-a

Max(FV) *

**!(ith)

(18) Derivation of reduplicant in nambiþnambitha nambith-a

[ss]

Max(Root)

+ a. nambi b. namb-a

Max(FV)

Max(Ext)

* *!*(ith)

As seen, *namb-aþnambith-a is ruled out in both (17b) and (18b), because it violates the relatively high ranked Max(Root) constraint—that is, the [i] of the root nambithcould have been parsed but was not. To summarize, we analyze the reduplicant as a D-stem that is constrained in two ways: (i) it is bisyllabic, and (ii) it must match the base D-stem morphosyntactically. Our analysis resembles those of Downing (1997a, etc.) and Steriade (1997) in relating the reduplicant to an existing morphological constituent: like Downing, we take the D-stem to be the relevant level. These studies di¤er from ours, however, in relying on output-output correspondence to relate the reduplicant and base. Thus, in order to get the FV -a in the reduplicant, it is necessary that the preceding -CVC- be an existing verb root in the language. In the following sections we will see that this condition is neither necessary nor su‰cient to predict the full range of facts in Ndebele reduplication. We thus consider, in turn, subminimal (section 13.3), imbricated (section 13.4), and passivized (section 13.5) verb forms. In each case we will observe the overriding importance of reference to roots versus reference to other elements. The unmistakable generalization will be that as long as the root is fully parsed, one has great freedom in how the two syllables of the reduplicant are filled out in Ndebele. 13.3

Subminimal Roots

Up until now we have been considering cases in which the reduplicant must be phonologically truncated in order to achieve the bisyllabic size condition. In this section

Morphosyntactic Correspondence in Bantu Reduplication

283

we consider the reduplication of verb stems that contain phonologically subminimal verb roots such as those in (19a), in which entirely di¤erent behavior is observed. (19) a. ‘‘Consonantal roots’’ -dl- (H) ‘eat’ -m- (H) ‘stand’ b. Monosyllabic stems -dl-a (H) ‘eat’ -m-a (H) ‘stand’

-lw- (H/L) -z- (H/L)

‘fight’ ‘come’

-lw-a (H/L) -z-a (H/L)

‘fight’ ‘come’

As seen in (19b), when these so-called consonantal roots are followed by the FV -a, the resulting stems are monosyllabic. Normally these stems would occur with prefixes. Compare, however, the imperative forms in (20). The singular a‰rmative imperative is the only context in which a stem can occur without prefixes: (20) a. lim-a thum-a (H) b. nambith-a (H) thembuz-a (H/L) c. yi-dl-a (H) yi-m-a (H)

‘cultivate!’ ‘send!’ ‘taste!’ ‘go from wife to wife!’ ‘eat!’ (*dl-a) ‘stand!’ (*m-a)

bamb-a hlamb-a bhavum-a thembis-a (H) yi-lw-a (H/L) yi-z-a (H/L)

‘catch!’ ‘swim!’ ‘growl!’ ‘promise!’ ‘fight!’ (*lw-a) ‘come!’ (*z-a)

The examples in (20a,b) show that the unprefixed verb stem is used in the imperative (singular, a‰rmative), if the verb stem is at least two syllables long. In (24c), however, we see that a monosyllabic stem cannot occur in its bare form in the imperative, but rather acquires an augmentative syllable yi-, known as a ‘‘stabilizer’’ in the Bantu literature, which we analyze as a semantically empty ‘‘filler’’ morph, much like the FV -a. The motivation for augmentation with yi- in (24c) is clear: as in many other Bantu languages (Myers 1987a; Kanerva 1989; Mutaka and Hyman 1990; Downing 1998), Ndebele words are subject to the bisyllabic minimality condition in (21). (21) Minimal prosodic word (o) ¼ [s s]foot Verbs are required to meet the minimal prosodic word condition in (21) and must thus be augmented by the empty morph yi- where subminimality would otherwise result. This situation arises only with monosyllabic stems in the imperative, where there is no prefix to supply the required second syllable. This minimality condition also helps us make sense of the similar size condition on reduplicants: reduplicants must be minimal prosodic words as well. Reduplicated verbs thus display the nested prosodic word structure depicted in (22).

284

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

(22)

That is, the prosodic structure of a reduplicated verb is: [Prefixes [RED]o BASEþIFS]o . 13.3.1

Subminimality and Augmentative -yi

With this established, we now turn to the question of how monosyllabic verb stems are reduplicated. (23) shows that the augmentative [yi] makes its appearance here as well, this time as a su‰x. As shown in square brackets in (23), -yi augments the otherwise subminimal reduplicants of such stems: (23) [dl-a-yi]þdl-a (H) [m-a-yi]þm-a (L) [lw-a-yi]þlw-a (H/L) [z-a-yi]þz-a (H/L)

‘eat!’ ‘stand!’ ‘fight!’ ‘come!’

As seen, -yi appears between the two occurrences of the monosyllabic stem. The (bracketed) reduplicant is subject to the two-syllable requirement, to whose satisfaction the su‰xal -yi is a crucial contributor. The absence of a corresponding -yi in D-stemBASE is no mystery on this account, because D-stemBASE constituents are never subject to the minimality condition that motivates the presence of [yi] in the first place. We know that D-stemRED is required to be exactly bisyllabic. As was seen in (26), the verb as a whole is also required to be minimally bisyllabic, but in the above examples, the whole verb would be bisyllabic anyway even without [yi]. It is thus only the reduplicant that would otherwise be subminimal. The data in (24), which contain the infinitive prefix uku-, confirm that the minimality condition is imposed on the reduplicant and on the whole verb but not on ordinary verb stems. Here, the existence of a prefix blocks [yi] augmentation in the unreduplicated forms, because the verb as a whole is at least bisyllabic. The fact that the verb stem is monosyllabic is of no relevance. When reduplication is present, however, [yi] augmentation does occur. It is required to supplement the otherwise

Morphosyntactic Correspondence in Bantu Reduplication

285

subminimal reduplicant. In (28), constituents subject to prosodic minimality are show in brackets: (24) Verb with unreduplicated stem [uku-dl-a] [uku-m-a] [uku-lw-a] [uku-z-a]

Verb with reduplicated stem [uku-[dl-a-yi]þdl-a] [uku-[m-a-yi]þm-a] [uku-[lw-a-yi]þlw-a] [uku-[z-a-yi]þz-a]

‘to ‘to ‘to ‘to

eat’ stand’ fight’ come’

(*uku-yi-dla) (*uku-yi-ma) (*uku-yi-lwa) (*uku-yi-za)

The greatest relevance of consonantal roots for our analysis comes from their behavior under su‰xation. D-stems consisting of consonantal roots with extension suffixes show a dazzling variety of reduplication possibilities, as illustrated below: (25) a. uku-dl-a (H) b. uku-dl-el-a

c. uku-dl-is-a

‘to eat’ ‘to eat for/at’

‘to feed’

uku-[dl-a-yi]þdl-a uku-[dl-el-a]þdl-el-a uku-[dl-a-yi]þdl-el-a uku-[dl-e-yi]þdl-el-a uku-[dl-is-a]þdl-is-a uku-[dl-a-yi]þdl-is-a uku-[dl-i-yi]þdl-is-a

(cf. [lim-e]þlim-el-a) (cf. [lim-a]þlim-el-a) (cf. [lim-i]þlim-is-a) (cf. [lim-a]þlim-is-a)

Again, reduplicants are bracketed. In (25a), a form with no extension su‰xes, the input to the reduplicant cophonology consists of the consonantal verb root dl-, plus freely available empty morphs -yi and FV -a. As seen, there is only one possible outcome, which is to supplement the consonantal root with both empty morphs, yielding a reduplicant of the shape [dl-a-yi]. In (25b,c), in which the base contains a -VC- extension su‰x, the same reduplication pattern seen in (25a) is still possible: in the reduplicant, -yi supplements the root and final vowel, yielding in both cases [dl-a-yi]. But in addition, two further reduplication possibilities emerge. One possibility is to exhaustively parse the extension suffix, yielding, in (25b), [dl-el-a], and in (25c), [dl-is-a]. The other possibility is to take the vowel from the extension su‰x and use -yi to supply the remaining syllable, thus [dl-e-yi] and [dl-i-yi]. In summary, the use of -yi is on a par with the use of the FV -a and material from extension su‰xes. Any of these options (singly, or in combination) may be used to supplement a consonantal root and bring a reduplicant to the bisyllabic goal: -yi, -a, material from extension su‰xes. Our existing analysis needs to be supplemented with a constraint that makes reference to -yi. Our intuition is that -yi is a morphosyntactically empty morph that can fill an optional position in D-stemRED just like the FV morph -a.4 In parallel with our treatment of -a, for which we proposed the constraint Max[FV] in (13), we invoke the constraint in (26) which mandates the surfacing of the filler su‰x -yi.

286

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

(26) MAX(-yi) Parse the -yi su‰x. Ranked freely with Max(FV) and Max(Ext), this constraint accounts for the new reduplication patterns we have just seen. In the input in (27), both -a and -yi are assumed to be present in input, just to illustrate how the constraints pick and choose among daughters of D-stemRED: (27) Derivation of reduplicants in dlela-dlela, dleyi-dlela, and dlayi-dlela dl-el-a-yi

RED¼ [ss]

a. dl-el-a-yi

Max(Root)

Max(Ext)

Max(FV) Max(-yi)

*!

+ b. dl-el-a

*

+ c. dl-e-yi

*(l)

+ d. dl-a-yi

**(el)

*

The candidate in (27a) loses because it violates the highest-ranked constraint, RED¼[ss]. For each of the candidates in (37b–d), there is some ranking of the bottom three constraints such that that candidate will emerge as the winner: candidate (b) wins when Max(Ext), Max(FV)  Max(yi); candidate (c) wins when Max(yi)  Max(Ext)  Max(FV); candidate (d) wins when Max(FV), Max(yi)  Max(Ext). The analysis correctly predicts that -yi will not surface when the root is su‰ciently large. Even if -yi is present in the input, as indicated in (28), Max(Root) forbids it to supplant root material. (28)

RED¼ [ss]

Max (Ext)

Max (FV)

Max (-yi)

+ a. lim-eþlim-el-a

*(l)

*

*

+ b. lim-aþlim-el-a

**(el)

lim-el-a-yi

Max (Root)

c. li-yiþlim-el-a

*!(m)

***(mel)

d. l-a-yiþlim-el-a

*!**(lim)

****(lime)

* *

The candidates augmented with -yi in (28c,d) are ruled out immediately by Max(Root), leaving the freely ranked Max(Ext) and Max(FV) to decide between the two attested candidates, (28a,b). While yielding easily to our analysis, [yi]-augmentation poses problems for the canonical stem and word-based accounts of Downing and Steriade. First the redupli-

Morphosyntactic Correspondence in Bantu Reduplication

287

cant [dla-yi] in (27d) is clearly not an independent word, and is thus problematic for Steriade. [dla-yi] could be considered a canonical stem by Downing, if -yi is ignored. However, in (27c) there is no definition of either the canonical stem or the independent word that the truncating reduplicant [dle-yi] would meet. Even if -yi is stripped o¤, what remains behind, namely [dl-e], is not a possible corresponding stem or word, because it contains some but not all of the applicative su‰x, and no FV.5 It is because our analysis is sensitive to the internal structure of the D-stem, distinguishing roots, extension su‰xes, and filler su‰xes, that we are able to account for reduplication patterns in which the reduplicant contains material other than what is in the ‘‘base,’’ including cases where it does not correspond to a possible or existing stem or word in the language. 13.3.2

Subminimality and the Macrostem

Before leaving the topic of subminimal roots, we turn to two more sets of data that provide further evidence for the view of reduplication developed in this chapter. The first concerns an alternative realization of reduplicated imperatives of subminimal verbs. Consider the forms in (29). (29) Bare stem -dl-a -m-a -lw-a -z-a

Imperative yi-dl-a yi-m-a yi-lw-a yi-z-a

Reduplicated imperative yi-dl-aþyi-dl-a yi-m-aþyi-m-a yi-lw-aþyi-lw-a yi-z-aþyi-z-a

‘eat!’ ‘stand!’ ‘fight!’ ‘come!’

Cf. from (23) dl-a-yiþdl-a m-a-yiþm-a lw-a-yiþlw-a z-a-yiþz-a

This method of reduplicating imperatives di¤ers in two ways from that seen previously. First, augmentative yi- is initial, rather than final, in the reduplicant. Second, it also appears in the base. Similar facts occur in the reduplication of bases with consonantal roots that are su‰xed with the perfective -ile, which occupies the IFS position, outside the domain of reduplication. As (30a) shows, a prefixed, unreduplicated base of this sort—here, m-ile—is not augmentable with [yi]. Yet under reduplication, we find the same two alternatives as in the imperative: either the reduplicant is [yi]-final, and the base is not augmented, as in (30b), or the reduplicant and base are both [yi]-initial (30c). (30) a. (ba-) m-ile (H) b. (ba-) m-a-yiþm-ile c. (ba-) yi-m-aþyi-m-ile

‘(they) stood’

*(ba)-yi-m-ile (single, su‰xed -yi in reduplicant) (double, prefixed yi- in reduplicant and base)

There are two ways of thinking about the pattern in which both base and reduplicant show initial yi- of the type seen in (29) and (30c). One is that the presence of yi- is driven by a prosodic requirement on the base, being passively reflected in the

288

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

reduplicant by some kind of correspondence requirement. The other is that yi- is required by the reduplicant itself, and must then by correspondence appear in the base as well. The latter scenario can be termed morphological ‘‘backcopying,’’ on analogy to the transmission of phonological alternations from reduplicant to base, documented in a small number of reduplication examples by McCarthy and Prince (1995) (but see Kiparsky 1997d and Inkelas and Zoll 2005 for reanalyses). The evidence would appear to favor backcopying in this instance. First, it would be di‰cult to maintain the base-augmentation analysis, given our earlier finding that prosodic minimality is enforced on reduplicants and on whole words but not otherwise on verb stems, as was schematized in (22). Recall that the evidence for this was the behavior of unreduplicated monosyllabic verb stems, as in (24): these are not supplemented by [yi]. If the stem formed a domain for prosodic minimality, an assumption necessary to maintain the reduplication analysis, then the facts in (29) and (30) would be unexplained. The second piece of evidence for backcopying comes from the behavior of bases preceded by a prefixal object marker (OM), such as noun class 10 zi-, the one used in all examples herein. If, and only if, the root of a reduplicated stem is consonantal, then zi- (and other OMs) can be reduplicated, appearing in both reduplicant and base. In (31) we present a case of OM doubling where the consonantal root is not followed by extension su‰xes. The reduplicant is enclosed in brackets: (31) uku-zi-dl-a (H) ‘to eat them’ uku-zi-[dl-a-yi]þdl-a (OM is not reduplicated) uku-[zi-dl-a]þzi-dl-a (OM is reduplicated) These data are quite parallel to the yi-prefixation data seen earlier: exactly when there is room in the reduplicant for the OM, the OM may appear doubled in the base. Its appearance in the reduplicant is being driven by the bisyllabic prosodic size requirement on the reduplicant; the base (here, dl-a) would be well formed without the OM, but the reduplicant would not. Unlike prefixal yi-, OMs can be incorporated into the bisyllabic reduplicant, and ‘‘backcopied,’’ even when the consonantal root is followed by extension su‰xes. As (32) shows, in such cases, the reduplicant can be brought up to two syllables by incorporating either the OM, the extension su‰x, the FV -a (or both), or su‰xal -yi can be used to bring the reduplicant up to two syllables: (32) a. uku-zi-dl-el-a (H)

b. uku-zi-dl-is-a (H)

‘to eat them for/at’

‘to feed them’

uku-zi-[dl-el-a]þdl-el-a uku-zi-[dl-a-yi]þdl-el-a uku-zi-[dl-e-yi]þdl-el-a uku-[zi-dl-a]þzi-dl-el-a uku-[zi-dl-e]þzi-dl-el-a uku-zi-[dl-is-a]þdl-is-a uku-zi-[dl-a-yi]þdl-is-a uku-zi-[dl-i-yi]þdl-is-a

(OM not reduplicated)

(OM is reduplicated) (OM not reduplicated)

Morphosyntactic Correspondence in Bantu Reduplication

c. uku-zi-thum-a (H) ‘to send them’

uku-[zi-dl-a]þzi-dl-is-a uku-[zi-dl-i]þzi-dl-is-a uku-zi-[thum-a]þthum-a *uku-[zi-thu]þzi-thum-a

289

(OM is reduplicated) (OM not reduplicated)

As seen in the unacceptable form *uku-zi-[yi-dl-a]þyi-dl-a (intended: ‘to eat them’), the only option for fleshing out the reduplicant that is not available in these verbs with OMs is prefixal yi-. Prefixal yi- is in complementary distribution with OMs. We thus add the OM, if present in the verb, to the list of options for fleshing out the bisyllabic D-stemRED constituent: (33) Potential daughters of D-stemRED a. Root b. Extension su‰xes c. FV -a d. Su‰xal -yi e. Prefixal yiEntails backcopying; in complementary distribution with OM f. OMs Entails backcopying; in complementary distribution with prefixal yiNote in (32c) that the OM cannot appear in reduplicants whose root is CVC or longer (cf. the correct output, uku-zi-thuma-thuma, where the OM zi- is not included in the bisyllabic reduplicant, thum-a). This follows from our existing analysis. Since reduplicants are required by the high-ranking Max(Root) to exhaust the first CVC portion of the root, in such a case there is no room for any a‰xes in the reduplicant: (34)

zi-thum-a

[ss]

Max(Root)

+ a. zi-thum-aþthum-a b. zi-thuþzi-thum-a

*!(m)

At this point in the analysis, three questions arise: Why are OMs in complementary distribution with prefixal yi-? Why are only OMs and prefixal yi- ‘‘backcopied’’ to the base? What is the morphological analysis of backcopying? Our analysis is the following. First, we propose that OMs and prefixal yi- occupy the same morphological ‘‘slot.’’ They are prefixes that attach to D-stems to create a constituent that Bantuists refer to as the Macrostem (M-stem): (35)

M-stem Prefix

D-stem

where Prefix ¼ OM or yi-

290

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

Prefixal yi- may be semantically empty, but has a morphological e¤ect nonetheless: constituents containing it must be of the level M-stem.6 Given the structure in (35), our next step is to generalize the reduplication construction. In addition to consisting of juxtaposed D-stems, we now allow the R-stem to consist of juxtaposed M-stems. The juxtaposed stems in the reduplication construction must still agree in every feature—including morphological category (Dstem versus M-stem). (36)

R-stem Stem1

Stem2

where Stem1 and Stem2 can be either D-stems or M-stems, M-stems so long as they agree in every morphological feature (including category)

We can now understand why it is the prefixal supplements to the reduplicant that induce backcopying: they induce it by requiring the reduplicant to be an M-stem. Due to the highly agreeing nature of the reduplication construction, if the reduplicant is an M-stem, the base must be as well—and when the base is an M-stem, it of course must have a prefix. Once it is established that both reduplicant and base are M-stems, it is a simple matter of morphosyntactic feature checking to ensure that they each have the same prefix. Thus, OM doubling is not in fact backcopying, but rather simple reduplication. The components involved in the apparent backcopying of an OM are recapitulated as follows. First, the morphology allows a free choice of M-stem or D-stem reduplication constructions. Second, if the M-stem construction is chosen, the obligatory prefix (an OM or yi-) is parsed into the reduplicant. And finally, morphosyntactic feature agreement between reduplicant and base ensures that reduplicant and base end up with the same prefix. The M-stemRED cophonology is the same as the DstemRED cophonology, with one exception: with the M-stemRED cophonology Max(Pfx) and Max(Root) are inviolable. The inviolable nature of both constraints means that if the root is longer than C, the M-stem reduplication construction cannot be used. This is a correct prediction.7 In summary, we have examined further data that confirm our claim that the reduplicant is morphologically complex. We have shown that the truncation-agreement analysis of reduplication can extend nicely to these new data, including to the apparent ‘‘backcopying’’ phenomena that results when the reduplicant is prefixed. By contrast, the canonical stem and corresponding word analyses of Downing and Steriade have little to say about these data. Certainly reduplicants with initial yi(e.g., yi-dl-a) are not ‘‘canonical stems,’’ nor are the truncated reduplicants with initial zi- (e.g., zi-dl-i, zi-dl-a) possible words. We next consider imbricated verb stems, which provide further support for our analysis.

Morphosyntactic Correspondence in Bantu Reduplication

13.4

291

Imbrication

In this section we consider the reduplication of verb stems that have undergone imbrication, a special fusion process triggered by the perfective su‰x -ile. Whereas most CVC roots add -ile in the perfective, as seen to the right in (37a), (37) a. lim-a thum-a (H) b. bonakal-a (H/L) dumal-a (H/L) thath-a (H)

‘cultivate’ ‘send’ ‘appear, be visible’ ‘become depressed’ ‘take’

ba-lim-ile ba-thum-ile ba-bonakel-e ba-dumel-e ba-theth-e

‘they cultivated’ ‘they sent’ ‘they appeared, were visible’ ‘they became depressed’ ‘they took’

others such as those in (37b) instead fuse or ‘‘imbricate’’ the -il- part of the perfective ending -ile. In these examples, the penultimate root vowel [a] is realized as [e], and the stem also ends in the final vowel -e.8 The question is how the ‘‘umlaut’’ in (37b) a¤ects reduplication. As seen in (38a), (38) a. ba-bonakel-e (H/L) b. ba-dumel-e (H/L) c. ba-theth-e (H)

‘they appeared’ ‘they became depressed’ ‘they took’

ba-bonaþbonakel-e ba-duma-dumele ba-dume-dumele ba-thath-aþtheth-e ba-theth-aþtheth-e *ba-theth-eþtheth-e

since the reduplicant is bisyllabic, imbrication has no e¤ect on the reduplicant when the umlauted penultimate vowel does not occur in one of the first two syllables of the stem. In (38b), however, where imbrication a¤ects the second vowel of the root dumal-, converting it to dumel-, we see that the reduplicant can be either duma- or dume-. Similarly, in (38c), where imbrication converts the root thath- to theth-, there are again two possible reduplicants: thatha- and thetha-. The reduplicant thethe-, on the other hand, is not grammatical. The correct observation is that the ‘‘umlaut’’ triggered by the perfective can be reduplicated, while the final -e of the perfective cannot be. Note in the case of ba-dumaþdumel-e in (38b), that the reduplicant duma- is not an existing minimal canonical stem. So it cannot be the case that the sequence -ele of -dumele is simply being truncated. Instead, to account for the perfective, we recognize two ‘‘alloconstructions.’’ The first, schematized in (39), shows the nonimbricating alternate:

292

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

(39)

As shown, the nonimbricating -ile su‰x is unambiguously in the IFS position, hence unavailable for reduplication. Now consider the analysis of perfectives marked by -e with umlaut.9 The question is where the umlauting feature (. . .) should be within the morphosyntactic structure, since it optionally appears in the reduplicants in (38b,c). Our view is that Ndebele speakers themselves are not sure where to assign the umlaut, and hence have the two competing analyses, schematized in (40). (40) a. Imbricating perfective: Umlaut analyzed to be part of IFS

b. Imbricating perfective: Umlaut analyzed to be part of D-stem (e.g., root)

In (40a), the umlaut is assigned to the IFS, identically to -ile in (39), from which it clearly derives historically. In (40b), however, the umlaut is seen as an internal mod-

Morphosyntactic Correspondence in Bantu Reduplication

293

ification on the morpheme to which it attaches—here, the root.10 If assigned the structure in (40a), the umlaut will not be realized in the reduplicant, which bars the realization of inflectional features. If assigned the structure in (40b), the umlaut will be obligatorily realized because of the high-ranking constraint Max(Root). In neither case will the final su‰x -e make it into the reduplicant, since it is unambiguously an IFS. Finally, for the sake of completeness, note in (41) that reduplicated verbs optionally are exempt from imbrication even when their nonreduplicated forms must imbricate: (41) a. ba-bonaþbonakal-ile (H/L) ‘they appeared’ b. ba-dumaþdumal-ile (H/L) c. ba-thath-aþthath-ile (H)

‘they became depressed’ ‘they took’

*ba-bonakal-ile *ba-dumal-ile *ba-thath-ile

(cf. (38a)) (cf. (38b)) (cf. (38c))

We hypothesize that these forms arise in the following way. First, we note that imbrication is not automatic in Ndebele. For example, of the 148 CaCaC- verb roots in the Comparative Bantu On-Line Dictionary (CBOLD) version of Pelling (1971), 39 obligatorily undergo imbrication and 24 do so optionally, while 85 may not imbricate. Roots (and extension morphemes such as the reciprocal -an-) will therefore have to carry lexical marking for imbrication. This lexical marking only optionally percolates up to the R-stem level. When it is present, the appropriate match with the IFS -e will be made. When it is absent, -ile must be used. To summarize, with this last assumption, as well as the two ‘‘alloconstructions’’ in (38), imbrication supports the analysis we have developed thus far. Reduplication is juxtaposition of morphologically identical stems (either M-stems or D-stems), with the first subject to a bisyllabic size condition. 13.5

Palatalization

In the last two sections we have established that a‰xal material in the D-stem may surface in the reduplicant only when the verb root has been exhaustively parsed. We now consider passivized forms, which further support this conclusion and make more precise our proposals about the internal structure of the reduplicant. In (42a) we see that the passive is derived by means of the derivational su‰x -w-, here occurring directly after the root: (42) a. bal-a bik-a (H) phek-a b. boph-a (H) bumb-a (H) thum-a (H)

‘read’ ‘announce’ ‘cook’ ‘tie’ ‘mold’ ‘send’

bal-w-a bik-w-a phek-w-a botsh-w-a bunj-w-a thuny-w-a

‘be ‘be ‘be ‘be ‘be ‘be

read’ announced’ cooked’ tied’ molded’ sent’

294

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

The examples in (42b) show that this -w- palatalizes an immediately preceding labial consonant. However, as the examples in (43) show, palatalization is not restricted to the immediately preceding consonant, but can target any preceding labial, as long as it is not the first consonant in the constituent:11 (43) gombolozel-a vumbulul-a (H) fumbath-a (H)

‘encircle, surround’ ‘uncover, unearth’ ‘clench (hand)’

gonjolozel-w-a vunjulul-w-a funjath-w-a

‘be encircled’ ‘be uncovered’ ‘be clenched’

Like the imbricating alternant of the perfective, we follow Sibanda 2004 in analyzing the passive as having two phonological components. Basing himself on Zoll 1996, Sibanda characterizes these as subsegments, one palatalizing (call it p) and the other labializing (call it o). The palatalizing component attaches, potentially over some distance, to a preceding labial consonant and changes its place of articulation to palatal. The labializing component attaches to the immediately preceding consonant and gives it a labial o¤glide. Now consider the possible reduplicated forms of the verbs in (44). (44) a. bal-a

‘read’

bal-w-a

‘be read’

bik-a (H)

‘announce’

bik-w-a

‘be announced’

phek-a

‘cook’

phek-w-a

‘be cooked’

‘tie’

botsh-w-a

‘be tied’

bumb-a (H) ‘mold’

bunj-w-a

‘be molded’

thum-a (H)

thuny-w-a

‘be sent’

b. boph-a (H)

‘send’

bal-w-aþbal-w-a bal-aþbal-w-a bik-w-aþbik-w-a bik-aþbik-w-a phek-w-aþphek-w-a pek-aþphek-w-a botsh-w-aþbotsh-w-a botsh-aþbotsh-w-a *boph-a-botsh-w-a bunj-w-aþbunj-w-a bunj-aþbunj-w-a *bumb-aþbunj-w-a thuny-w-aþthuny-w-a thuny-aþthuny-w-a *thum-aþthuny-w-a

In each case the -w- may or may not appear in the reduplicant, exactly as we expect of a derivational su‰x. However, any labial palatalization induced by the passive obligatorily surfaces in the reduplicant (cf. the ungrammaticality of *thum-aþthunyw-a, etc.). The same facts are obtained when the final vowel is other than -a—for example, subjunctive -e in (45). (45) a. bal-w-e bik-w-e

‘be read’ (subjunctive) ‘be announced’

bal-w-aþbal-w-e bal-aþbal-w-e bik-w-aþbik-w-e bik-aþbik-w-e

Morphosyntactic Correspondence in Bantu Reduplication

phek-w-e b. botsh-w-e

‘be cooked’

phek-w-aþphek-w-e pek-aþphek-w-e botsh-w-aþbotsh-w-e botsh-aþbotsh-w-e *boph-a-botsh-w-a bunj-w-aþbunj-w-e bunj-aþbunj-w-e *bumb-aþbunj-w-e thuny-w-aþthuny-w-e thuny-aþthuny-w-e *thum-aþthuny-w-e

‘be tied’

bunj-w-e

‘be molded’

thuny-w-e

‘be sent’

295

Treating the passivizing p and o as subsegments under the D-stem parallels our analysis of the umlauting subsegment . . . that results in imbrication, as discussed earlier. Two important di¤erences subdivide this set of subsegments, however. First, while o and . . . are constrained to appear on the immediately preceding segment (the preceding consonant and /a/, respectively), p is permitted to find suitable hosts anywhere within the D-stem. That is, o and . . ., but not p, are subject to morphological adjacency (alignment), whereby the output segment on which it surfaces must correspond to the rightmost C or V of the immediately preceding morpheme in the input. This condition draws support from the data in (46), in which the root is longer than CVCV and must therefore be truncated in the reduplicant. (46) a. gombolozel-a vumbulul-a fumbath-a b. gonjolozel-w-a vunjulul-w-a funjath-w-a

‘encircle, surround’ ‘uncover, unearth’ ‘clench (hand)’ ‘be encircled’ ‘be uncovered’ ‘be clenched’

gomboþgombolozel-a vumbuþvumbulul-a fumbaþfumbath-a gonjoþgonjolozel-w-a vunjuþvunjulul-w-a funjaþfunjath-w-a *funj-w-aþfunjath-w-a

Although these are passive stems, passive o is not able to surface in the reduplicant. This follows straightforwardly from adjacency: the final consonant of the root does not surface in the reduplicant, and therefore, by the terms of adjacency, the passive -w- cannot surface either. As a control, the data in (47) show that labialization that is not from the passive morpheme must be preserved under reduplication: (47) a. yejwayel-a yejwayelek-a yejwayez-a yenway-a

‘get accustomed to’ ‘become customary’ ‘accustom’ ‘scratch an irritation’

yejwaþyejwayel-a yejwaþyejwayelek-a yejwaþyejwayez-a yenwaþenway-a

*yejaþyejwayel-a *yejaþyejwayelek-a *yejaþyejwayez-a *yenaþyenway-a

296

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

b.

yenwab-a @wonwab-a c. qwaqwaz-a tshwatshwaz-a

‘be happy’ ‘be happy’ ‘click (stick on stick)’ ‘make hissing noise’

yenwaþyenwab-a wonwaþwonwab-a qwaqwaþqwaqwaz-a tshwatshwaþ tshwatshwaz-a

*yenaþyenwab-a *wonaþwonwab-a *qwaqaþqwaqwazw-a *tshwatshaþ tshwatshwaz-a

It is not optional to copy only the C portion of a tautomorphemic Cw structure.12 The second dimension of di¤erence among the three subsegments is why imbricating . . . and passivizing o are only optionally parsed in the reduplicant, while passivizing p must surface (assuming, of course, that a suitable host exists in the output). The approach taken to imbrication in section 13.4 readily provides an explanation. Optionality in the case of . . . was said to result from the availability of the two alloconstructions in (40): when . . . is taken to be part of the IFS, it cannot appear in the reduplicant. When it is taken, instead, to belong to the same constituent as the segment on which it is realized (i.e., the root), it must be parsed into the reduplicant. In the passive case, there is only one construction: when a root labial is palatalized, p belongs to the root and hence must appear in the reduplicant because of high-ranked Max(Root). To recapitulate what we have said about the passive thus far, p obligatorily appears in the reduplicant because it is parsed with the root on which its palatalizing e¤ect is realized. On the other hand, o was said to optionally occur in the reduplicants in (44) and (45) because it is a derivational a‰x that can be truncated as in the case of applicative -el- and causative -is-. However, close inspection of more morphologically complex forms shows that additional, and more interesting, conditions hold on the surfacing of the passive o subsegment. Consider in particular the passivized applicative stems in (48). The applicative extension su‰x -el- occurs between the root and the passive su‰x. Note the expected long-distance action of the palatalizing component p of the passive in the nonreduplicated stems in (48b): (48) a. bal-el-a bik-el-a (H) phek-el-a b. boph-el-a (H) bumb-el-a (H) thum-el-a (H)

‘read for/at’ ‘announce for/at’ ‘cook for/at’ ‘tie for/at’ ‘mold for/at’ ‘send for/at’

bal-el-w-a bik-el-w-a phek-el-w-a botsh-el-w-a bunj-el-w-a thuny-el-w-a

‘be read for/at’ ‘be announced for/at’ ‘be cooked for/at’ ‘be tied for/at’ ‘mold for/at’ ‘be sent for/at’

Up to now we have not seen a case in which two semantically contentful su‰xes jockey for a position in the reduplicant. Which will take precedence? Will there be free variation, as we have seen between extension su‰xes and the ‘‘filler’’ morphs [yi] and [a]? The forms in (49) show the reduplicated versions of the passivized applicatives in (48). As seen earlier in the unextended stems, the palatization component p of the passive surfaces in the reduplicant, while the labial component o is optional:

Morphosyntactic Correspondence in Bantu Reduplication

(49) a. bal-el-w-a

‘be read for/at’

bik-el-w-a

‘be announced for/at’

phek-el-w-a

‘be cooked for/at’

b. botsh-el-w-a

‘be tied for/at’

bunj-el-w-a

‘be molded for/at’

thuny-el-w-a

‘be sent for/at’

297

bal-eþbal-el-w-a bal-aþbal-el-w-a bal-w-aþbal-el-w-a bik-eþbik-el-w-a bik-aþbik-el-w-a bik-w-aþbik-el-w-a phek-eþphek-el-w-a phek-aþphek-el-w-a phek-w-aþphek-el-w-a botsh-eþbotsh-el-w-a botsh-aþbotsh-el-w-a botsh-w-aþbotsh-el-w-a bunj-eþbunj-el-w-a bunj-aþbunj-el-w-a bunj-w-aþbunj-el-w-a thuny-eþthuny-el-w-a thuny-aþthuny-el-w-a thuny-w-aþthuny-el-w-a

Note that when the passive o does surface in these reduplicants, it surfaces on a segment to which it is not adjacent in the input, in apparent violation of the Adjacency constraint proposed earlier. That is, it appears that a mapping such as that in (50) is occurring: the passive is surfacing not on the final segment of the immediately preceding morpheme, but on the final segment of the morpheme before that. The applicative su‰x is being skipped over, even though it is closer to the beginning of the stem and (under standard assumptions about multiply extended D-stems) forms a closer constituent with the root than the passive does: (50)

Is Adjacency violable in Ndebele? We suggest that it is not. Rather, we contend that the appearance of ‘‘at-a-distance’’ linking of passive o in (50) is illusory, and that the applicative is not in fact being skipped over in these cases.

298

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

To appreciate this contention, consider the data in (51), which serve as background to a previously undetected morphological condition on the surfacing of passive o in reduplicants. The verbs in both sentences in (51) contain the sequence -el-w-, composed of applicative and passive extension su‰xes: (51) a. abantwana b-a-phek-el-w-a ukudla children they-past-cook-appl-pass food ‘The children were cooked food’ b. ukudla kw-a-phek-el-w-a abantwana food it-past-cook-appl-pass children ‘The food was cooked (for) the children’ Although possessing the same surface morphs in the same linear order, these verb stems realize two di¤erent argument structures. The sentence in (51a) is a passivized applicative, and its verb has the argument structure in (52a). (52) a.

b.

c.

! -el-wThis morphological structure and this order of the su‰xes is what is expected from the scope relations or ‘‘mirror principle’’ (Baker 1985; Alsina 1999; Hyman and Mchombo 1992; Hyman 1994a). The sentence in (51b), on the other hand, is an applicativized passive, and its verb has the argument structure in (52b). In this case, by the mirror principle, the order of the morphs is expected to be passive -w- followed by applicative -el-. However, in Ndebele, as in most Bantu languages, the applicative cannot follow the passive.13 Hence, a number of linguists, particularly of the Tervuren school inspired by Meeussen (1967), have proposed rules that metathesize morphs, as when (52b) is transformed into (52c). It is thus not the surface order of the morphs that interests us, but rather the underlying morphosyntactic structure. Given this, now consider how the verbs in (51) reduplicate: (53) a. abantwana

b. ukudla

b-a-phek-eþphek-el-w-a b-a-phek-aþphek-el-w-a *b-a-phek-w-aþphek-el-w-a kw-a-phek-eþphek-el-w-a kw-a-phek-aþphek-el-w-a kw-a-phek-w-aþphek-el-w-a

ukudla

abantwana

‘The children were cooked food’ ‘Food was cooked for the children’

Morphosyntactic Correspondence in Bantu Reduplication

299

In (53a), the passivized applicative, we see that the reduplicant can be pheke- or pheka-, but not *phekwa. Passive o can not be parsed in the reduplicant. In (53b), by contrast, the reduplicant can be phek-eþ, phek-aþ, or phek-w-aþ. The reduplicant is free to parse o or not, as we are used to seeing. The reason for this surprising discrepancy jumps out once one compares the argument structures of the verb stems in (52): the passive surfaces under reduplication only if it is sister to the root. Thus, phek-w-aþ is well-formed as the reduplicant of -phek-el-w-a only if the subtree rootþpassive occurs in the base to which it corresponds, as in (52b). We conclude that reference to internal morphosyntactic structure is critical to determining the well-formedness of reduplicants. How exactly is this reference to be made? We posit that it follows as a direct consequence from the internal structure of the Ndebele verb stem. Making the standard assumption that D-stems are binary branching, (54) gives the structure of the reduplicant of an extended verb stem of the kind seen above: (54)

Whether it is D-stemREDj or D-stemREDi that is spelled out, only two reduplicants are possible: phek-e and phek-a. The candidate phek-w-a is ruled out because of the adjacency violation that results from skipping over the applicative -el-. On the other hand, in the case of applicativized passives, we would have a structure like the one in (55), in which the lowest D-stem node has the passive su‰x adjacent to the root: (55)

As seen, both phek-w-a and phek-a can be spelled out as the reduplicant at both the D-stemREDj and D-stemREDi levels, while phek-e can result only as the realization of D-stemREDj . Note that phek-w-e is not a possible reduplicant, since the morphotactics prohibit the [e] of applicative -el- from following passive -w- on the surface.

300

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

As seen, in the D-stemRED cophonology in both (54) and (55), copying of labialization is optional, but adjacency is always respected. Thus the passive -w- can, but need not, be parsed at the lowest D-stem RED node, where the segment to which it would be adjacent in output ends the immediately preceding morpheme in the input. Observe that the branching structure we hypothesize for the Bantu verb stem, both ‘‘base’’ and ‘‘reduplicant,’’ is tantamount to a cyclic analysis. Constraints are enforced on every incarnation of a D-stem (or M-stem), regardless of whether it is a daughter or a mother in the verb structure. This e¤ect is predicted by the theory of Sign-Based Morphology (Orgun 1996) and is particularly evident in cases where one valency demands one versus another order of su‰xation. In the example in (56a), for instance, the verb fik-a ‘come’ is intransitive: (56) a. fik-a b. fik-is-a c. fik-is-w-a

‘come’ ‘make come’ ‘be made to come’

In (57b), fik-is-a ‘make come’ is made transitive by the causative su‰x -is-, and, as a result, can be passivized as in (57c). Now note the possible reduplications of (56c) in (57a). (57) a.

fik-iþfik-is-w-a fik-aþfik-is-w-a b. *fik-w-aþfik-is-w-a

As seen, the reduplicant consists of the root fik- plus either the [i] of the causative su‰x -is- or the default FV -a. In (57b) we see that it is not possible to parse the passive -w- in the reduplicant because this would imply that fik- ‘come’ had become passivized in the first cycle—which is, of course, not the case, since the verb is intransitive. This e¤ect is automatically obtained from the approach taken here: rather than predicting the properties of the reduplicant by reference to the surface output of the corresponding base, the full range of facts is accounted for by attributing to the D-stemRED the full morphosyntactic structure of the base. This morphosyntactic structure is in turn spelled out in the normal way, subject to the cophonology of the reduplicant—for example, the two-syllable constraint. Most interesting is the role played by this constraint in producing reduplications such as phek-w-aþphek-el-w-a in (55). While a superficial comparison of base and reduplicant suggests a violation of Adjacency, it is the underlying structure in (55) combined with [s s] that make the reduplicant phek-w-aþ possible. Because of the prosodic constraint, both -wand -el- cannot be parsed. In addition, *phek-w-eþ is not a possible reduplicant because the passive morph -w- is not allowed to precede the applicative morph -el- (or its subpart, as -e would be in this case). The reduplicant phek-w-aþ is possible because it respects the hierarchical morphosyntactic structure in (55), and, crucially,

Morphosyntactic Correspondence in Bantu Reduplication

301

because there is no requirement that the feature applicative be spelled out in the reduplicant. If that option is taken, however, then the only possible grammatical output is phek-eþphek-el-w-a, which more closely resembles the surface order of the -el-w- sequence. Examples of this sort can be produced with respect to other su‰xes and other morphosyntactic situations. Ndebele thus provides strong evidence that the potential spell-outs of a reduplicant are governed by more than superficial resemblances with its base—and, once again, that as concerns the Bantu verb stem, morphology must proceed in a cyclic fashion (Hyman 1994b). 13.6

Conclusion

At the outset of this study, we characterized the study of partial reduplication as having as its goal to construct a theory that insightfully captures the full range of considerations that speakers ‘‘care about’’ in determining how a reduplicant will relate to its base. What has the study of Bantu reduplication contributed toward this goal? Certainly one very important contribution has been made in the work of Downing (1997a, etc.), in recognizing the important role that morphological structure plays in Bantu reduplication, which, therefore, is not solely a prosodic phenomenon. Mutaka and Hyman (1990) showed for Kinande the role that morphological considerations can play, because morpheme integrity can inhibit prosodic templatic e¤ects in certain languages. Downing went further and showed the role that morphological constituent structure plays, noting e¤ects of the derivational stem constituent on Bantu reduplication. In this chapter we have taken this important insight a step further, finding that not only the D-stem as such but also its internal constituents—root, extension su‰xes, and filler su‰xes—play a role in defining the variety of shapes of the Ndebele reduplicant. In making reference to D-stems and their constituents, the ‘‘canonical stem’’ shape to which Downing observed most reduplicants to adhere can actually be derived, rather than stipulated. This is an especially desirable outcome in Ndebele, where not all reduplicants are canonical stems in any case. In fact, the comparative study of verb reduplication in Bantu shows that the phonological and morphological constraints on the reduplicant can be independently controlled to produce most of the possible combinations. In our introduction we cited Eulenberg 1971 as expressing the view that partial reduplication derives from total reduplication—that is, by phonological and morphological paring down of the full base. Diachronic evidence for this position is quite clear as concerns the Bantu verb stem. Numerous Bantu languages require full verb-stem reduplication including all extensions and the IFS—for example, Luganda lim-il-agan-aþlim-il-agan-a ‘cultivate for each other here and there’, and its subjunctive, lim-il-agan-eþlim-il-agan-e.

302

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

It is not di‰cult to explain why this original state of total reduplication should become modified over time: it is hardly necessary to repeat the whole verbal I-stem in order to convey the aspectual idea that the action is done a little here and there. It is therefore possible to economize and limit the reduplicant to a subpart of the base. This can be done either phonologically and/or morphologically, as we have seen. Historical developments can follow one or both of the following scales: (58) a. Phonological scale full > foot > syllable (> mora > gemination) b. Morphological scale full > derivational stem > root The phonological scale consists of a gradual narrowing down of the reduplicant to fit a prosodic template, frequently a bisyllabic foot. The morphological scale consists of first restricting the reduplicant to derivational material, and ultimately to root material. As is known from Kikerewe (Odden 1996; Downing 1997b), there can be ‘‘intermediate’’ stages. In this language, a verb that has two or more extensions can be reduplicated fully, or can copy zero, one or more of these extensions—for instance, lim-il-anþa ‘cultivate for each other’ can reduplicate as lim-il-an-aþlim-il-an-a, lim-ilaþlim-il-an-a, or lim-aþlim-il-an-a. Restricting ourselves to verb-stem reduplication in Bantu, the specific choices speakers make can be listed as parameter settings, as in (59). (59) a. Phonology i. Size constraint: yes/no (1, 2 s’s) ii. Tone transfer: yes/no14 iii. Length transfer: yes/no b. Morphology i. IFS: yes/no ii. OM: yes/no iii. FV ¼ -a: yes/no iv. Max(Root): yes/no Concerning the morphological parameters, we hypothesize that copying the OM is not original. Rather, verb reduplication was originally limited to the I-stem. Reduplication of the OM is thus a subsequent development motivated by two separate phonological conditions. The first, seen in Ndebele, occurs when the verb stem is monosyllabic and the OM is copied in order to fill out the bisyllabic foot condition on the reduplicant.15 This is one of three strategies that have been noted to fill out a [s s] condition on the reduplicant: (i) ‘‘move up’’ to M-stem and copy the OM, if there is one (e.g., Ndebele class 10 zi-); (ii) fill with a ‘‘dummy’’ syllable (e.g., Ndebele yi-); (iii) double reduplicate the monosyllabic stem (e.g., sw-a-sw-aþsw-a ‘grind here and there’ in Kinande) (Mutaka and Hyman 1990).

Morphosyntactic Correspondence in Bantu Reduplication

303

Turning now to the tendency for the reduplicant to end in -a, we note, as does Downing (1997c, etc.), that this is the most frequent verb ending in most Bantu languages. However, rather than seeing this as an indication of a special ‘‘canonical verb stem,’’ we believe it is necessary to have a Bantu-specific constraint FV[-a], whose e¤ects are, in fact, not restricted to verb reduplication—or even verbs—in Bantu.16 There is in fact considerable evidence that -a is the default stem-final vowel, whether its function is inflectional or derivational, and whether applying to nouns or verbs. First, it is true, as Downing points out, that -a is used as an inflectional default FV in many Bantu languages. In Giphende, for example, the recent past tense is marked by the IFS -ı´ if immediately preceded by a CV-, CVC- or CGVC- verb root, as in (60a,b,c) respectively:17 (60) a. tw-a-mb-ı´ tw-a-y-ı´ b. tw-a-som-ı´ tw-a-meng-ı´ c. tw-a-kwec-ı´ tw-a-mwang-ı´ d. tw-a-vumbı´g-a´ tw-a-digı´m-a´ e. tw-a-som-e´l-a´ tw-a-som-e´s-a´ f. tw-a-mb-e´l-a´ tw-a-mb-e´s-a´

‘we ‘we ‘we ‘we ‘we ‘we ‘we ‘we ‘we ‘we ‘we ‘we

put’ went’ loaded (a gun)’ detested’ tied’ scattered’ buried’ were afraid’ loaded for/at’ made load’ put for/at’ made put’

In (60d) we see that when the stem consists of three full syllables, the IFS must be -a´. In these examples the root is morphologically unanalyzable. In (60e), where a -CVCroot is followed by a -VC- extension, the same -a´ is required. Finally, in (60f ), where the first stem syllable consists of a CV- root which has fused with the V of a -VCextension, we also must get the default FV -a´ rather than -ı´. The generalization is, therefore, that -ı´ can occur only within the second syllable of a verb stem whose base is morphologically simplex. Otherwise the recent past tense reverts to default -a´. It is clear from such evidence that the use of the FV -a is logically independent of reduplication. We also suggest that it is independent of the bisyllabic CVC-a shape that Downing has referred to as a the ‘‘minimal canonical stem.’’ To see this, consider the following forms from Lengola (Stappers 1971, 268), where verb-stem reduplication marks the habitual: (61) a. i-kul-a i-£on-a i-tu´m-a i-lı´mb-a

‘acheter’ ‘regarder’ ‘envoyer’ ‘chanter’

i-kul-aþkul-a i-£on-aþ£on-a i-tu´m-aþtum-a i-lı´mb-aþlimb-a

304

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

b. i-kpet-a i-ki-a i-gbok-a c. i-£ı´-a i-Bıˆ-a i-„a´m-a

‘couper’ ‘faire’ ‘trouver’ ‘manger’ ‘parler’ ‘crier’

i-kp-aþkpet-a i-k-aþki-a i-gb-aþgbok-a i-£-aˆþ£i-a i-B-aˆþBi-a i-„-aˆþ„am-a

In (61a) we observe full rootþFV reduplication (preceded by the infinitive prefix i-). However, Stappers also indicates that verbs such as those in (61b,c) form their habitual by means of a [Ca] reduplicant. The most straightforward analysis is that in these forms D-stemRED is limited to one syllable (versus the more usual two) and must end in -a. A similar conclusion is drawn from gerundive reduplication in Lomongo (Hulstaert 1965):

‘lancer’ ‘duper’ ‘eˆtre vivant’ ‘couvrir’ ‘insulter’ ‘piler’

c

-lı´k-le´ng-kng-ku´k-to´l-t ´k-

Gerundive Bombwanja dialect Coquilhatville and Northern dialects n-dı´þlı´k-a n-d-a´þlı´k-a n-de´þle´ng-a n-d-a´þle´ng-a n-kþkng-a n-k-a´þkng-a n-ku´þku´k-a n-k-a´þku´k-a n-to´þto´l-a n-t-a´þto´l-a n-t ´þt ´k-a n-t-a´þt ´k-a c

Verb

c

(62)

c

In this case of verb reduplication, the reduplicant again consists of a single CV syllable. In Bombwanja dialect, the CV corresponds to the initial CV of the base. In Coquilhatville and Northern dialects of Lomongo, however, it has the shape [Ca]. As in Lengola, the fixed vocalism is presumably a reflection of the FV -a, hence the reduplicant can be analyzed as C-aþ. If we were to extend Downing’s analysis of the Kinande ‘‘minimal canonical D-stem’’ CVC-a, we would have to say that in Lengola and Lomongo the minimal canonical D-stem is C-aþ. However, we know of no evidence for this. Rather, in this construction we would simply say that RED ¼ s and the Bantu-specific constraint FV[-a] is ranked high. Finally, note that even in languages where there is a bisyllabic minimum in e¤ect, CVC-a can be a possible reduplicant without there having to be a corresponding CVC- root. The example comes from Kikuyu (Peng 1991; Downing 1999): (63) a. kor-a cin-a b. koor-a Buut-a

‘grow’ ‘burn’ ‘pull out’ ‘depose’

! ! ! !

kor-aþkor-a cin-aþcin-a koor-aþkoor-a Buut-aþBuut-a

Morphosyntactic Correspondence in Bantu Reduplication

! ! ! !

Boc-aþBocor-a h r-aþh rr-a ci'-aþci'erer-a hwr-aþhwrrk-a cc

‘be indented’ ‘be quiet’ ‘encircle’ ‘tilt’

cc

c. Bocor-a h rr-a d. ci'erer-a hwrrk-a

305

cc

When the base is CVC-a or CVVC-a, as in (63a,b), reduplication appears to be total.18 The longer stems in (63c,d) indicate, however, that only the first CV(V)Cmay appear in the reduplicant, and that the FV -a is required. To account for this, Peng refers to the initial CV(V)C- of verbs as a prosodic minimal root (Rmin ) that has the shape [s.C]. In our framework, the FV[-a] simply outranks Max(Root). What these data from Lengola, Lomongo, and Kikuyu clearly show is that each of the properties of reduplicants can be independently manipulated and reranked in terms of the appropriate constraints. Thus, although Max(Root) is very important crosslinguistically, it can be outranked by RED[ss] in Ndebele, by RED[s] in Lengola and Lomongo, or by FV[-a] in Kikuyu. In fact, Max(Root) appears to be even more seriously violated in at least some realizations in Bukusu in (66) below. To appreciate this, however, we need to seek explanations for two crucial—and, as we will suggest—related questions: Why should there be a tendency to realize as much of the root as possible in Bantu verb-stem reduplication? Why is Bantu verb-stem reduplication prefixal? The most immediate answer that linguists will probably advance to the first question is that roots—an open morphological class—are universally more contrastive and salient than grammatical a‰xes. However, a moment’s reflection will reveal that this answer is insu‰cient. We suggest a second contributor to relative high ranking of Max(Root), namely that a Bantu verb-stem reduplication is prefixal. As originally observed by Marantz (1982), prefixal reduplication tends to be left-anchored, while su‰xal reduplication tends to be right-anchored. Schematically, the unmarked mappings (in Marantz’s framework) or Base-Reduplicant correspondences (McCarthy and Prince 1995) are as in (64a), their marked counterparts as in (64b). (64) a. Unmarked Prefixal: Su‰xal: b. Marked Prefixal: Su‰xal:

tuki tuki

! !

tuþtuki tukiþki

tuki tuki

! !

kiþtuki tukiþtu

Niepokuj (1991) speculates that the more widespread unmarked situation in (64a) has a functional explanation: in these outputs the copied sequence in the reduplicant appears adjacent to the same sequence in the base. As a result, the reduplicative process is more transparent—that is, not requiring a long-distance identification as in

306

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

(64b). It is therefore to be expected that there will be greater faithfulness to the root in prefixal reduplication than there will be to su‰xes. To show that this is the correct interpretation, consider what Bantu verb-stem reduplication would look like if it were su‰xal. Let us take the same forms we considered from Ndebele in (3) and assume that the su‰xed reduplicant is subject to Red(ss). Possible outputs might be those indicated in (65). (65) a. lim-el-a lim-is-a b. lim-el-a lim-is-a

lim-el-aþlim-e lim-el-aþlim-a lim-is-aþlim-i lim-is-aþlim-a lim-el-aþm-el-a lim-is-aþm-is-a

‘cultivate for/at’

(applicative -el-)

‘make cultivate’

(causative -is-)

‘cultivate for/at’ ‘make cultivate’

In the mirror-image realizations of Ndebele in (65a), we provide two variants for each reduplication: the first copies the first two syllables of the base, while the second uses the FV -a. Such realizations, which represent the marked base-reduplicant correspondence in (64b), are unattested in Bantu. The unmarked BR correspondence with right-anchoring is given in (65b), where su‰xes are preserved at the expense of (part of ) the root -lim-. The realizations in (65b) violate Max(Root) in a way that (65a) does not, but can we say that they are any less ‘‘natural’’ or expected? While we have not (yet?) found exact outputs as in (65b) in any Bantu language, we have noted some variants in Bukusu that are highly suggestive.19 Thus, consider the reduplications in (66). (66) a. lim-a re´m-a b. lim-il-a re´m-er-a c. kacul-a mulix-a

‘cultivate’ ‘cut’ ‘cultivate for/at’ ‘cut for/at’ ‘chat, talk’ ‘flash’

! ! ! ! ! !

lim-aþlim-a re´m-aþrem-a lim-aþlim-il-a re´m-aþrem-er-a kacul-aþcul-a mulix-aþlix-a

In (66b), where the applicative su‰x has been added, reduplication is prefixal. Consider, however, the forms in (66c), where the stem contains an unanalyzable CVCVC- verb base. While some such verbs reduplicate fully, these reduplicate by means of a truncated ‘‘second part’’—that is, by apparent su‰xation.20 The generalization is that the speaker chooses to isolate the root at the left and the su‰xes at the right of the total form. Why should this be? We suggest that the answer is the same as to our second question: Why should Bantu verb-stem reduplication be prefixal? The Bantu verb stem is su‰xal, as we have seen. Speakers are therefore accustomed to both derivational and inflectional grammatical marking at its right edge. If reduplication were to proceed as in (65a), this general organizing principle would be violated. The resolution is thus as in (66b):

Morphosyntactic Correspondence in Bantu Reduplication

307

the root appears at the left of the structure, where it is expected, and the su‰xes appear at the right, where they too are expected. The resolution in (66c) is thus ingenious in that the second part of the R-stem includes all of the su‰xal material—and only a minimum of root material. The fact that any of the root occurs in the second part at all is presumably due to the otherwise exceptionless property that a‰xes cannot occur without a root. Since this is but one of the reduplicative strategies employed by our consultant, it would be much more interesting if some Bantu language had only this pattern. To summarize this last result, we suggest that Max(Root) is driven by the su‰xal nature of the Bantu verb stem. We therefore make the following general markedness claims concerning the a‰x orientation of the reduplicant crosslinguistically: (i) the reduplicant will tend to be prefixed when the base has a su‰xing structure, and (ii) the reduplicant will tend to be su‰xed when the base has a prefixing structure. We have already seen the evidence for prefixal reduplication in Bantu verb-stem reduplication. On the other hand, Mutaka and Hyman (1990) demonstrate that noun reduplication is a word-level su‰xal process in Kinande: (67) a. ku-gulu mu-go´ngo` ku´-boko b. o.mu´-twe bi-la ka´-tı`

‘leg’ ‘back’ ‘arm’ ‘head’ ‘intestines’ ‘stick’

ku-gulu.gulu mu-go´ngo.go´ngo` ku´-boko´.boko mu´-twe´.mu´-twe bi-la.bi-la ka´-tı´.ka´-tı`

‘a real leg’ (etc.)

The direction of a‰xation is not evident in (67a), where we might think that it is the stem that has been totally reduplicated. However, the nouns in (67b) provide the answer. Here, the noun class prefix is also reduplicated to fill out the bisyllabic prefixal template. Mutaka and Hyman’s analysis is that the prefixþstem is reduplicated as a su‰x, and the noun prefix is copied in (67b) because the stem is monosyllabic. In our view, su‰xal reduplication is possible in noun reduplication because nouns are not characterized by extensive su‰xation. Instead, they are most notably marked by noun class prefixes. As such, they are good candidates for su‰xal reduplication, as per (67b). In this study we have had several goals. A first goal has been to document in some detail the nature of verb-stem reduplication in Ndebele. A second goal has been to contribute to discovering the full class of reduplicative parameters that can be set by means of di¤erent constraint rankings. In so doing, a third goal has thus been to contribute to Optimality Theory in providing yet another example of language-internal variation that can be attributed to free constraint ranking. Finally, we have illustrated the need for enforcement of constraints on subtrees (cyclicity) and have provided a morphological reanalysis of apparent backcopying in reduplication.

308

Larry M. Hyman, Sharon Inkelas, and Galen Sibanda

Notes This is a shortened (and only slightly updated) version of a paper prepared on April 25, 1998, for the Trilateral Phonology Weekend (TREND) meeting in Berkeley. We are grateful for several helpful discussions, especially with Laura Downing. Research on the Ndebele lexicon, supported by the Comparative Bantu On-Line Dictonary (CBOLD) project, was funded in part by National Science Foundation grant SBR96-16330. 1. Verb tone is indicated throughout by (H) following verb stems with High-tone roots and (H/L) following verbs in which the root can take both High and Low tone. Unmarked forms have Low-tone roots. High tone on consonantal roots is realized only in the presence of extension su‰xes. 2. Technically, -ile can be demonstrated to consist of two su‰xes, -il-e, since the passive -wappears between them—for example, si-thu´ny-i-w-e ‘it was sent’ (cf. corresponding active stem: /thum-il-e/). 3. Completeness would require us to consider, in addition, reduplicants lacking the FV -a. This would force us to use the additional notational complication of tableaux des tableaux (Itoˆ, Mester, and Padgett 1995) for dealing with allomorphy. We leave this to the reader. 4. We assume that -yi is disallowed in regular D-stems because it serves no function and thus gratuitously violates *Struc (the general ban on structure of all kinds). Below we relate the presence of prefixal yi- in imperatives such as yi-dla ‘eat!’ to the macrostem level. 5. We are not considering the possibility that the [e] of [dl-e] is interpreted as the subjunctive final vowel -e, since the base in this case is not subjunctive, and since the IFS -e never reduplicates under any condition. 6. In proposing that OMs belong to a Macrostem constituent that excludes the IFS, we depart from the tradition of analyzing them as sisters to the I-stem. In a number of Bantu languages that otherwise restrict the prefix slot to one OM, the first-person singular prefix n- can still cooccur with a second OM. A reasonable conclusion to draw is that it joins the stem, thereby freeing up the one prefix slot for a second OM (Schlindwein 1986). 7. Another option for generating backcopying would be to assume a phonological correspondence between reduplicant and base, following McCarthy and Prince 1993b, 1995, and use an anchoring constraint. This would capture the generalization that the a‰xes that force backcopying are both prefixes: Anchor-L: The initial elements of RED and Base must correspond. 8. The imbricated forms derive historically from *bonakail-e, *dumail-e, and *thaith-e. See Bastin 1983 and Hyman 1995 for further evidence and discussion. 9. It should be pointed out that some verbs do not accept the imbricated form, while still others show variability. In addition, verbs whose last vowel is other than /a/ may also take the ‘‘short’’ -e form in the perfective, but in these cases there is no umlaut, e.g. a-dabul-ile ¼ a-dabul-e ‘he tore’. 10. An alternative would be to analyze the umlaut as an ‘‘extension,’’ where it would be optionally parsed under reduplication, exactly like applicative -el- and causative -is-. While this works for the perfective, it does not generalize to the passive construction analyzed in section 13.5. We therefore prefer to say that speakers are inconsistent in assigning the umlauting feature to either the IFS or to the D-stem, as in (40a,b). The one exception is where the umlaut is assigned to an extension with /a/ (e.g., thum-an-a ‘send each other’ ! ba-thum-en-e ‘they sent

Morphosyntactic Correspondence in Bantu Reduplication

309

each other’). In this case the umlaut is assigned to -an-—that is, under the extension node in (40b). 11. The imperviousness of morpheme-initial labials to palatalization has been addressed by Chen and Malambe (1998), Downing (1998), and Sibanda (1999a); we will pay it no further attention here. 12. The forms in (47) are the only examples we know where tautomorphemic Cw is found in post-stem-initial position. Since the glides of initial [ye] and [wo] in (47a,b) can be treated as epenthetic, the correct generalization may be that tautomorphemic /w/ appears only after the first consonant of a stem. See Sibanda 2004. 13. For a comprehensive study of verb extension combinations in Ndebele, see Sibanda 2004, where it is shown that the ‘‘morphotactically unmarked’’ linear order is -is-el-an-w- (causativeapplicative-reciprocal-passive); also Hyman 2003 for an overview of templatic su‰xing ordering in Bantu. 14. The only Central Bantu language we know with tone transfer in verb-stem reduplication is Chichewa, where each stem is a phonological word (Mtenje 1988; Kanerva 1989; Hyman and Mtenje 1999; Myers and Carleton 1996). 15. A second situation in which the OM sometimes becomes reduplicated is when it fuses with the root—that is, when the OM is either a homorganic nasal, or a CV- prefix followed by Vinitial root, as in Kihehe (Odden and Odden 1985). 16. In Basaa, for instance, nouns are diminutivized by CV prefixal reduplication and a shift to class 19/13 (prefixes hi-/di-). As seen in the following examples provided by Deborah Schmidt, the full form typically ends in -a´: mim ‘body’ ! hi-miþmı´m-a´, £l ‘thigh’ ! hi-£þ£l-a´, l Ð ‘country’ ! hi-l þl ´Ð-a´. c

c

c

17. Study of Giphende was conducted with Mwatha Ngalasso. The final -i of the recent past optionally assimilates to the preceding vowel. Thus, the two forms in (60b) may also be realized tw-a-som-o´ and tw-a-meng-e´. 18. Peng cites forms without tones and without FV. We have added -a to all verb bases to make them more comparable to those seen in Ndebele and elsewhere in this study. 19. These forms were collected by Julia Hill in an undergraduate field methods course at UC Berkeley in fall 1997. The consultant was Wanjala Khisa, who showed a great deal of variation (and with whom Downing 2004, pursued a more detailed study). 20. We use the designations ‘‘first part’’ and ‘‘second part’’ instead of reduplicant and base, since it is not evident which is which in these forms. This problem, anticipated in Niepokuj 1991, is further explored in Inkelas and Zoll 2005.

14

Patterns of Reduplication in Yoruba

Douglas Pulleyblank

The task is thus to harmonize the theory of reduplication with the rest of non-linear morphology as far as possible, in a way which also solves the descriptive problems, while maintaining, and if possible increasing, its desirable restrictiveness. —Kiparsky 1986b, 8 14.1

Introduction

Cases of reduplication can be located on a scale where one logical extreme is to ‘‘compound’’ two completely accurate copies of a base. Moving down the scale, one observes cases involving deviations from such accuracy—cases where only a subset of the base is copied, where the copy is not fully accurate, or where reduplication mixes copied and specified material. The logical extreme as one moves down this scale is an ‘‘ordinary’’ a‰x whose segmental and prosodic makeup is fully independent of the base to which it attaches. An adequate theory of morphology and phonology must be able to account for the entire range of such cases. This chapter is a case study of four types of reduplication in Yoruba, a NigerCongo language of Nigeria, in which reduplication is diverse and productive (see also Abı´o´du´n 1997; Adewole 1997).1 Each of the four reduplication types exhibits ˙ rather di¤erent properties. At one end of the scale, agentive reduplication involves two identical copies of a base, with both segmental and tonal material copied. Partway down the scale, infixing reduplication involves a complete copy of the base supplemented by a segmentally and tonally fixed infix. Further down the scale, distributive reduplication copies part of a base, restricting its copying to segmental material. At the end of the scale, gerundive reduplication copies only part of a base, mixing the copied material with lexically specified material. Much recent work has stipulated that base-reduplicant identity results from special constraints on correspondence relations for reduplication (e.g., McCarthy and Prince 1995). This chapter argues that, at least in Yoruba, such special constraints are unnecessary and redundant. Complete or partial identity between a base and a reduplicant results from either (i) morphological copying or (ii) the satisfaction of

312

Douglas Pulleyblank

prosodic requirements. While special correspondence relations do occur in reduplicative constructions, these relations are shown to be by-products of the phonology seeking to minimize violations of input-output faithfulness, not the result of constraints specifically tailored for reduplication. 14.2

Complete Reduplication: The Agentive

At the ‘‘complete accurate copy’’ end of the reduplicative scale, agentives and instrumentals are productively derived by faithfully reduplicating a verb þ noun combination as in (1) (Folarin 1987; Ola 1995).2 Segmentally, both consonants and vowels ˙ ˙ are copied intact, including vowel height, nasality, and tone. (1) Agentive reduplication a. wole´wole´ ‘sanitary inspector’ b. pejapeja ‘fisherman’ ˙ ˙ c. yoyı´nyoyı´n ‘dentist’ ˙ ˙ d. jedı´jedı´ ‘piles’ ˙ ˙ e. yı´nru`nyı´nru`n ‘meningitis’ ´ ´ ´ ´ f. nawonawo ‘extravagant person’ g. jaye´jaye´ ‘lover of pleasure’

wole´ peja ˙ yoyı´n ˙ jedı´ ˙ yı´nru`n na´wo´ jaye´

‘look at the house’ ‘kill fish’ ‘extract tooth’ ‘eat anus’ ‘twist neck’ ‘spend money’ ‘enjoy life’

In analyzing total reduplication, there are two basic approaches. One possibility is to analyze one half of the reduplicated form as the ‘‘base,’’ the other half as the ‘‘reduplicant,’’ and to require correspondence between the two (McCarthy and Prince 1995). If no constraint interferes with perfect correspondence, then total reduplication results. A second possibility is to attribute reduplication more directly to the morphology. Specifically, reduplication can be analyzed as a form of compounding with the morphological requirement that the two halves of the compound be identical (Yip 1998; Hyman et al., this volume; Inkelas and Zoll 2000). Although both approaches can derive the Yoruba agentives in (1), there are significant di¤erences between them. 14.2.1

Morphological Identity

Under the morphological identity approach, the morphological structure of the reduplicated form is one in which the two halves are identical: (2) Agentive reduplication [ Agentive [VPi ][VPi ]] This approach bears a clear relation to the proposal of Steriade (1988) whereby reduplication results from a complete copy that can be pared down by the application of phonological processes. The crucial phonological constraint under the morphological identity approach is input-output correspondence (McCarthy and Prince 1995):

Patterns of Reduplication in Yoruba

313

(3) Input-output faithfulness a. MAXIO Every segment of the input has a correspondent in the output. b. DEPIO Every segment of the output has a correspondent in the input. Since the postulated morphological structure involves two identical compounded bases, input-output faithfulness will ensure that two phonologically identical strings surface. The base forms in (1) are a simplification of the actual, morphologically complex input forms: wole´ ‘look at the house’ derives from wo ‘look at’ þ ile´ ‘house’, peja ‘kill ˙ fish’ derives from pa ‘kill’ þ eja ‘fish’, and so on. As shown in Pulleyblank and Akin˙ labi 1988, the morphological base for this type of reduplication is a verb phrase. To obtain the correct surface forms in such cases, constraints independent of reduplication are required to ensure vowel deletion, retention of H tones, and so forth (Rosenthall 1997; Pulleyblank in press; Orie and Pulleyblank 2002—see also below). Each half of the reduplicated form will be subject to the same constraints. To see this, consider the form pejapeja ‘fisherman’, derived from the morphological form ˙ ˙ [[paþeja][paþeja]]. A constraint prohibiting vowel hiatus rules out sequences of vow˙ ˙ els (Pulleyblank in press; Orie and Pulleyblank 2002) and a left-edge anchoring constraint favors retention of the second vowel in a VV sequence (Pulleyblank in press).3 (4) a. NOHIATUS It is ill-formed to have two adjacent vowels linked to di¤erent sets of features.4 b. ANCHORL The segment at the left edge of a morpheme in the input has a correspondent at the left edge of the morpheme in the output.5 Together these constraints eliminate candidates (5a–c). (5) Vowel sequences in the input

a. b. c. + d. e. f.

/pa-ejaþpa-eja/ ˙ ˙ [paejaþpaeja] ˙ ˙ [paejaþpeja] ˙ ˙ [pejaþpaeja] ˙ ˙ [pejaþpeja] ˙ ˙ [pahejaþpaheja] ˙ ˙ [pajaþpaja]

NoHiatus

DepIO

MaxIO

AnchorL

*!* *!

*

*!

* ** *!* **

*!*

314

Douglas Pulleyblank

In (5d), the deletion of a vowel to satisfy NoHiatus occurs in both halves of the reduplicated form, while in (5e), a similarly motivated consonant epenthesis also occurs in both halves. Such apparent mimicking of one half by the other follows from the requirement that the morphologically defined input of the reduplicated form contain two identical copies of the same morphosyntactic input, each subject to the same markedness constraints (NoHiatus in this case) and to the same inputoutput faithfulness. Note that this type of example establishes that DepIO is ranked above MaxIO in Yoruba: consonant epenthesis (5e) is a more costly faithfulness violation than vowel deletion (5d). Orthogonal to the issue of reduplication, AnchorL selects the candidate where V2 is retained in a heteromorphemic V1 V2 sequence (candidate (5d) over (5f )).6 14.2.2

Base-Reduplicant Correspondence

Consider now the approach to such reduplication based on base-reduplicant correspondence. Under this analysis, the morphology of the Yoruba agentive would be assumed to have a base to which a reduplicative a‰x is attached.7 (6) Agentive reduplication [RedAgentive [VP]] Identity between the reduplicant and the base would be ensured by a set of basereduplicant faithfulness constraints that parallel the input-output constraints of (3). (7) Base-reduplicant faithfulness a. MAXBR Every segment of the base has a correspondent in the reduplicant. b. DEPBR Every segment of the reduplicant has a correspondent in the base. As with the purely morphological analysis sketched above, NoHiatus, DepIO, and MaxIO would ensure that VV sequences in the base are resolved in favor of vowel deletion. This e¤ect would then be transferred to the reduplicant via the basereduplicant faithfulness conditions. This is illustrated by the same case seen above; the reduplicant is underlined for identification:

Patterns of Reduplication in Yoruba

315

(8) Vowel sequence resolution and base-reduplicant correspondence AnchorL

MaxBR

DepBR

MaxIO

DepIO

a. [paejaþpaeja] ˙ ˙ + b. [pejaþpeja] ˙ ˙ c. [pahejaþpaheja] ˙ ˙ d. [paejaþpeja] ˙ ˙ e. [pahejaþpeja] ˙ ˙ f. [peþpeja] ˙ ˙

NoHiatus

/Red þ pa-e: ja/

*!* * *! *!

*

*

*

*!*

*

*!*

The first three candidates (8a–c) show perfect reduplicative copying: of the three, (8b) has the optimal base since the inevitable violation of faithfulness needed to satisfy NoHiatus (cf. (8a)) is achieved through violation of the more lowly ranked MaxIO rather than the higher ranked DepIO (cf. (8c)). The last three candidates (8d–f ) show imperfect copying (violating either DepBR or MaxBR) in spite of an optimal base; as such all three are nonoptimal. Both the morphological identity approach and the base-reduplicant correspondence approach thus successfully derive simple patterns of total reduplication such as the Yoruba agentive. 14.2.3

Comparison: Conceptual Considerations

Before considering empirical reasons for dispreferring the base-reduplicant faithfulness approach, note a curious conceptual redundancy inherent in that theory. As expressed by McCarthy and Prince (1995, 249), ‘‘Constraints of faithfulness demand that the output be as close as possible to the input, along all the dimensions upon which structures may vary.’’ They go on to argue that ‘‘input-output faithfulness and base-reduplicant identity . . . are controlled by exactly the same set of formal considerations, played out over di¤erent pairs of compared structures’’ (p. 250; italics in original). The leading idea of this approach is thus to account for patterns of reduplication through recourse to an independently motivated constraint structure. If faithfulness constraints such as Max, Dep, Contiguity, Linearity, Uniformity, Integrity (McCarthy and Prince 1995) are required to govern input-output relations, then the same constraints should be applicable to the faithfulness relations holding in reduplication. Phrased di¤erently, the goal is to account for reduplication without postulating special reduplication-specific processes or constraints.

316

Douglas Pulleyblank

One could imagine, therefore, that a grammar in which the constraints governing reduplication would literally be the same as those governing nonreduplicative inputoutput configurations would be the best possible realization of this goal. In the proposal that McCarthy and Prince (1995) present, however, this is not the case. Rather, a wholesale reduplication of faithfulness constraints is proposed, where there is a set of faithfulness constraints to govern input-output relations {MaxIO, DepIO, ContiguityIO, . . .}, a complete copy of this set to govern base-reduplicant relations {MaxBR, DepBR, ContiguityBR, . . .}, and a second copy of the set to govern input-reduplicant relations {MaxIR, DepIR, ContiguityIR, . . .}. The basereduplicant set is referred to as the ‘‘basic model,’’ while the inclusion of the inputreduplicant set gives the ‘‘full model.’’ The irony is that while attempting to account for reduplication via independently motivated constraint types, the basic model actually ends up with twice as many faithfulness constraints as are motivated independently of reduplication, and the full model ends up with three times as many constraints. While the correspondence relations postulated fall out from the free assignment of correspondence between any specified phonological elements, the actual number of correspondence constraints needed to deal with the postulated relations triples—with fully two thirds of the full set of correspondence constraints referring specifically to reduplication. Moreover, these reduplication-specific constraints depend on the establishment of an abstract morphological category ‘‘Reduplicant’’ (¼ ‘‘Red’’) to which the reduplication-specific faithfulness constraints refer. As will be seen shortly, this multiplicity of constraints is unnecessary. It is important to distinguish between the structural phonological relations posited by the theory and the constraint set that is proposed to govern those relations. There is no question that two corresponding representations can di¤er in terms of the particular elements contained, the order of such elements, and so on. These relations can then be referenced by constraints such as Max, Dep, Linearity, and so forth. Given multiple representations, these relations can be examined between some input and some corresponding output, between some output and some other corresponding output, between some a‰x and some corresponding base, and so on. At issue for reduplication is whether the substantive set of faithfulness constraints that govern such relations should include constraints specifically referring to reduplicative structures. Allowing construction-specific constraints referring to reduplication not only results in the proliferation of faithfulness constraints just noted, but also raises the possibility of other types of construction-specific constraints. I argue below that it is therefore conceptually desirable to eliminate the class of reduplication-specific faithfulness constraints. For the agentive in Yoruba, this means adopting the morphological identity analysis sketched in section 14.2.2. I therefore also show below how such an approach to reduplication can be applied to cases of partial reduplication.

Patterns of Reduplication in Yoruba

14.3

317

Reduplication with Morphologically Fixed Segmentism: kı´ Infixation

A fundamental property of the proposed analysis is that a single constraint ranking is posited for the full range of reduplicative constructions. While particular reduplicative patterns exhibit di¤erent surface properties (copying of tone, hiatus resolution, and so on), it will be shown that these di¤erences do not require the postulation of di¤erent constraint rankings for di¤erent reduplicative types. In this section, a reduplicative pattern is considered that involves a concurrent ‘‘infix.’’ Exactly the same constraints seen in the previous section account for these ‘‘infixing’’ patterns. The observed patterns di¤er somewhat, however, due to phonological properties of the input forms. 14.3.1

Data

Awoyale (1974) describes a class of reduplicative cases as involving word linkers between two identical noun components. The most widely discussed case involves the morpheme /kı´/. The derived noun that results has the meaning ‘any X’ or pejoratively ‘any old/bad X’. In addition to Awoyale’s treatment, the pattern has been discussed in various places including Ogunbowale 1970, Awobuluyi 1985, Bamgbose ˙ ˙ 1986, and Ola 1995. The following data are taken from Ola 1995. ˙ ˙ (9) /kı´/ infixation Noun Gloss /kı´/ form Gloss a. fı`la` ‘cap’ fı`la`kı´fı`la` ‘any type of cap’ pa´ta´ko´ ‘wood’ pa´ta´ko´kı´pa´ta´ko´ ‘any type of wood’ ja`gu`da ‘thief ’ ja`gu`dakı´ja`gu`da ‘any thief ’ ja`n`du`ku´ ‘dubious person’ ja`n`du`ku´kı´ja`n`du`ku´ ‘any dubious person’ so`we´dowo´ ‘check’ so`we´dowo´kı´so`we´dowo´ ‘any check’ ˙ ˙ ˙ b. omo ‘child’ omoko´mo ‘any child’ ˙ ˙ ˙ ˙ ˙ ˙ ere´ ‘play’ ere´ke´re´ ‘any play’ olorı` ‘queen’ olorı`ko´lorı` ‘any queen’ ake´koˇ ‘student’ ake´koˇka´ke´koˇ ‘any student’ ˙ ˙ ˙ ˙ ˙ ˙ ´ ` ´ ´ko ` ´ ´la ´ ` ´ alakowe ‘educated person’ ala weka kowe ‘any educated person’ ˙ ˙ ˙ It is clear that the presence of kı´ is due to the morphology, not to the phonology. Phonologically, there is no reason a reduplicant cannot occur immediately adjacent to a base, as seen with the agentive forms of section 14.2. If the morphology were to create a sequence such as /pa´ta´ko´þpa´ta´ko´/, there would be no reason to expect it to undergo any phonological modification (cf. ko´le´ko´le´ ‘thief ’). Morphologically, kı´ (plus reduplication) indicates the meaning of ‘any’ or ‘whichever’. Other a‰xes occur in this type of construction with di¤erent meanings. Ogunbowale (1970) lists a vari˙ ety of examples including the following:

318

Douglas Pulleyblank

(10) Various reduplication-triggering morphemes Noun Gloss Reduplicated form a. de´ omo ‘child’ omodo´mo ˙ ˙ ˙ ˙ ˙ ˙ owo´ ‘hand’ owo´do´wo´ ˙ ˙ ˙ ˙ ˙ ˙ b. ko` e`gbe´ ‘side’ e`gbe´ke`gbe´ ˙ ˙ ˙ ˙ ˙ ˙ oju´ ‘face’ oju´koju´ c. nı´ eran ‘animal’ eranle´ran ˙ ˙ ˙ omo ‘child’ omolo´mo ˙ ˙ ˙ ˙ ˙ ˙ d. sı´ ile´ ‘house’ ile´sı´le´ abu´le´ ‘village’ abu´le´sa´bu´le´

Gloss ‘from child to child’ ‘from hand to hand’ ‘side by side’ ‘face to face’ ‘another person’s animal’ ‘another man’s child’ ‘house to house’ ‘from village to village’

Each of these a‰xes can be related semantically, at least diachronically, to a full verb. Ogunbowale (1970) notes the relation of the de´ forms to de´ ‘arrive/reach/extend ˙ to’, the ko` forms to ko` ‘meet’, the nı´ forms to nı´ ‘have/possess’, and the sı´ forms to sı´ ‘to’ (see also Awoyale 1974). Each morpheme therefore has a particular meaning that it contributes to the reduplicative construction. There is an important phonological point to note about these forms, namely that they involve exact copying of the base in conjunction with a lexically idiosyncratic linker. There is no necessary or systematically recurring property in the phonological shape of these a‰xes. Neither their segmental makeup nor their tonal makeup is predictable: segmentally, coronals and noncoronals are observed, as are voiceless and voiced segments, stops and fricatives, oral and nasal segments; tonally, both H and L tones are attested. This variety in specified phonological properties is consistent with the analysis of these segments as morphologically determined, not as representing some sort of emergent phonological property—a point that will be important for cases of reduplication discussed below. 14.3.2

Analysis

The proposed analysis of kı´-reduplication is analogous to that already presented for agentive reduplication where the morphology specifies a compounding structure requiring that both conjuncts be identical. The important di¤erence is that kı´reduplication also stipulates a morphological linker: (11) kı´-reduplication, ‘any old X’ [[Ni ] þ kı´ þ [Ni ]] In its essence, this analysis is one where there is simultaneous morphological specification of reduplication and of a specified a‰x—an overwriting case of the ‘‘table-schmable’’ type (McCarthy and Prince 1995; Alderete et al. 1999, etc.). The di¤erence is that the entire base is copied in addition to the prespecified material. That is, the prefixal kı´ does not get substituted for base material to produce a form like *ja`gu`kı´ja`gu`da or *kı´gu`daja`gu`da. This follows directly from the proposed mor-

Patterns of Reduplication in Yoruba

319

phological structure and an absence of any prosodic shape constraints governing kı´reduplication.8 Since there is no pressure to have the output form conform to a particular prosodic shape, the constraints combine to retain the morphologically specified kı´, except where it is a¤ected by hiatus avoidance, implemented according to the ranking already established in (5). Cases involving a C-initial noun are phonologically transparent: an underlying structure [[ ja`gu`da] þ kı´ þ [ ja`gu`da]] results in the surface form [ ja`gu`dakı´ja`gu`da]. Of more phonological interest are those cases where the base is vowel-initial (9b). In such cases, retention of the full prefix in conjunction with maximal base copying would result in forms like *omokı´omo ‘any child’, *omode´omo ‘from child to child’, ˙ ˙ ˙ ˙ ˙ ˙ ˙ ˙ and so forth instead of the forms in (9) and (10). Outputs with VV sequences are illformed because they violate the prohibition on hiatus (12b); an optimal form like omoko´mo ‘any child’ exhibits vowel deletion as in (12a), where by deleting the first ˙ ˙ ˙ ˙ vowel in the sequence a violation of AnchorL (12c) is avoided. (12) Vowel deletion in reduplication /omo-kı´-omo/ ˙ ˙ ˙ ˙ + a. omoko´mo ˙ ˙ ˙ ˙ b. omokı´omo ˙ ˙ ˙ ˙ c. omokı´mo ˙ ˙ ˙

NoHiatus

DepIO

MaxIO

AnchorL

* *! *

*!

One issue needs to be addressed with regard to the tonal properties of kı´reduplication. As seen in (12), the H tone of the reduplicative prefix is retained, even though it surfaces on the first base vowel rather than the prefix itself. This point is of some importance in the general context of fixed segmentism, particularly relevant for reduplicative patterns such as the gerundive discussed below. The issue is whether the H tone is a lexically specified tonal property or a redundant unmarked tonal category. Independent evidence squarely points to an analysis of the H tone as a lexically specified property (Akinlabi 1984; Pulleyblank 1986, 2004). Evidence from a variety of phenomena—including tonal stability, tones under epenthesis, OCP e¤ects— argues for an analysis where the retention of H and L tones dominates the retention of an M tone, and where an M is the preferred default. For kı´-reduplication, this means that the fixed H must be lexically specified, a pattern that we will see below is paralleled tonally in gerundive reduplication. 14.3.3

Alternations between [i] and [u]

One final point is relevant for this particular type of reduplication. There is a class of cases involving base-initial i@u alternations. These cases illustrate an important

320

Douglas Pulleyblank

diagnostic for distinguishing between reduplication induced by morphological copying and reduplication induced by the satisfaction of prosody. As seen in the examples of (13), the initial [i] of the unreduplicated base sometimes appears as [u] after the kı´ prefix (Bamgbose 1986). Some reduplicated forms are typically produced with the [u] ˙ (13a), while others exhibit variants with [i] and with [u] (13b).9 (13) [i]@[u] alternations Noun Gloss a. `ıse ‘doing’ (ı`-se) ˙ ˙ `ıje ‘eating’ (ı`-je) ˙ ˙ `ıta` ‘selling’ (ı`-ta`) b. `ılo` ‘use’ (ı`-lo`)

/kı´/ form Gloss `ıseku´se ‘bad doing’ ˙ ˙ `ıjeku´je ‘bad eating’ ˙ ˙ `ıta`kuta` ‘bad selling’ `ılo`kulo` ‘bad use’ `ılo`kilo` ‘any use, bad use’ `ırı`n `ırı`nkurı`n ‘walking’ (ı`-rı`n) ‘bad walking’ `ırı`nkirı`n ‘any walking, bad walking’ `ıfe´ `ıfe´ku´feˇ ‘loving’ (ı`-fe´) ‘bad loving’ ˙ ˙ ˙ ˙ `ıfe´kı´feˇ ‘any loving, bad loving’ ˙ ˙ Bamgbose (1986) observes that a pejorative reading is particularly observed with the ˙ forms exhibiting [u]. In addition, he observes that the appearance of [u] in these reduplicative forms is dependent on the form being derived by the prefixation of the abstract nominal prefix `ı-. In underived nouns with an initial [i], only a reduplicative form with [i] is possible (14a); he notes a single exception to this generalization, namely the form for ‘‘anytime’’ in (14b). (14) Underived forms with initial [i] Noun Gloss /kı´/ form a. isu ‘yam’ isukı´su ˙ ˙ ˙ ile´ ‘house’ ile´kı´le´ `ılu´ `ılu´kı´luˇ ‘town’ `ıgba`kugba` b. `ıgba` ‘time’ `ıgba`kigba`

Gloss ‘any yam, bad yam’ ‘any house, bad house’ ‘any town, bad town’ ‘anytime’ ‘anytime’

Two types of analysis have been proposed for these data. One possibility is to attribute the change of [i] to [u] to a process of coalescence (Awobuluyi 1985) or consonant-induced assimilation (Pulleyblank 1988; Ola 1995). This type of solution ˙ is essentially phonological in nature. A purely phonological solution raises a problem, however, in that the alternation in question takes place robustly only in a single environment, an environment that is defined morphologically, not phonologically. Specifically, as already seen, the only robust environment for the i@u alternation is when the second noun is a deverbal nominative, formed by prefixation of `ı-. If

Patterns of Reduplication in Yoruba

321

framed in phonological terms, there is no explanation for why [i] does not change to [u] in the large number of i-initial nouns that do not involve the abstract nominal prefix as in (14a). An alternative is Bamgbose’s (1986) proposal that the u-initial forms involve the ˙ retention of an u`- prefix found in various Southeastern dialects of Yoruba. He notes forms such as the following in the Ijebu dialect: (15) Ijebu dialect: u-initial Noun Gloss u`-lo` ‘using’ u`-se ‘doing’ ˙ u`-je ‘eating’ ˙ u`-ta` ‘selling’

forms permitted /kı´/ form Gloss u`lo`ku´lo` ‘bad use’ u`seku´se ‘bad doing’ ˙ ˙ u`jeku´je ‘bad eating’ ˙ ˙ u`ta`ku´ta` ‘bad selling’

According to this analysis, Standard Yoruba forms like `ılo`kulo` ‘bad use’ derive from /ı`lo`þkı´þu`lo`/. That is, the presence of [u] in the reduplicated forms is attributed to the preservation of the u-prefix on the second noun of the reduplicant, as observed in dialects like Ijebu. A variant of this proposal is to analyze the Standard Yoruba forms as derived directly from an input analogous to the dialectal form: /u`lo`þkı´þu`lo`/. Given the independent prohibition on u-initial forms in Standard Yoruba discussed by Bamgbose, such inputs would be minimally changed into the ˙ actually attested output `ılo`kulo`. Such an approach is not adopted here since it too encounters serious di‰culties. Like the assimilation and coalescence approaches, this analysis fails to account for the limited distribution of the i@u alternation. An initial [u] is common in Southeastern dialects, by no means restricted to nouns involving the abstract deverbal prefix. If u-initial nouns are carried over from such dialects, then why do more cases of the ‘‘u’’ type not occur? Why are forms such as those in (16a) systematically bad if the attested [i] is replaced by [u]? In addition, Olanike Ola Orie (personal communi˙ ˙ ˙ cation) has noted that Standard Yoruba forms of the `ılo`kulo` type may never have polysyllabic bases. Polysyllabic forms are only possible with kı´, not ku´: (16) Polysyllabic Noun a. `ımo`ra`n ˙ b. `ıgba`gbo´ ˙ c. `ıle´rı´

bases: Only Gloss ‘advice’ ‘belief ’ ‘promise’

kı´ possible /kı´/ form `ımo`ra`nkı´mo`ra`n ˙ ˙ `ıgba`gbo´kı´gba`gbo´ ˙ ˙ `ıle´rı´kı´leˇrı´

*/ku´/ form *ı`mo`ra`nku´mo`ra`n ˙ ˙ *ı`gba`gbo´ku´gba`gbo´ ˙ ˙ *ı`le´rı´ku´leˇrı´

Gloss ‘any advice’ ‘any belief ’ ‘any promise’

The polysyllabic restriction is not found in the Southeastern dialects that constitute the putative source for the ku´ forms. An additional problem with the appeal to u-initial nouns is that such forms do not appear to be possible with any other type of reduplication. Recall from (10) that kı´

322

Douglas Pulleyblank

is only one of the possible word linkers in the NþN reduplication type. With other formatives, both nouns must be i-initial; the second noun cannot be u-initial. This is illustrated with forms involving sı´ (cf. 10d). (17) Impossibility of u-forms with sı´ Noun a. igbo´ `ılu´ ipo` `ıjo ˙ b. `ıpa`de´ `ıgba`gbo´ ˙ `ıgbı`mo` ˙ `ıjo`ye`

Gloss ‘forest’ ‘town’ ‘position’ ‘church’ ‘meeting’ (cf. pa`de´ ‘meet’) ‘belief ’ (cf. gba`gbo´ ‘believe’) ˙ ‘committee’ (cf. gbı`mo` ‘confer’) ˙ ‘chief ’ (cf. joye` ‘appoint to o‰cial position’)

/sı´/ form: ‘from X to X’ igbo´sı´gbo´ `ılu´sı´luˇ ipo`sı´po` `ıjosı´jo ˙ ˙ `ıpa`de´sı´pa`de´ `ıgba`gbo´sı´gba`gbo´ ˙ ˙ `ıgba`gbo´sı´gba`gbo´ ˙ ˙ `ıjo`ye`sı´jo`ye`

*/su´/ form *igbo´su´gbo´ *ı`lu´su´luˇ *ipo`su´po` *ı`josu´jo ˙ ˙ *ı`pa`de´su´pa`de´ *ı`gba`gbo´su´gba`gbo´ ˙ ˙ *ı`gba`gbo´su´gba`gbo´ ˙ ˙ *ı`jo`ye`su´jo`ye`

Whether the base of reduplication is an underived noun (17a) or a noun derived by a‰xation of the abstract nominal prefix `ı- (17b), reduplication with sı´ never involves the replacement of [i] by [u]. I propose therefore that the special u-forms observed in kı´-reduplication are a property of the linking prefix itself. That is, the ‘any (old) X’ morpheme has two allomorphs: kı´- and ku´-. The kı´- allomorph has both a neutral ‘any X’ reading and a pejorative ‘any old X’ reading, while the allomorph ku´- is restricted to the pejorative reading. In addition, the kı´- allomorph subcategorizes for the category ‘‘noun’’ while the ku´- allomorph has two more specific restrictions: (i) the noun that ku´- attaches to must be exactly one minimal word, and (ii) ku´- subcategorizes for the abstract nominal prefix ´ı-. The first restriction rules out reduplicative forms with [u] that are polysyllabic (16). The second restriction rules out reduplicative forms with [u] that involve underived nouns (14a) and would also rule out ku´-forms involving consonant-initial bases. The impossibility of u-forms with other linking morphemes such as sı´ is guaranteed by the absence of allomorphy for such morphemes. This analysis interacts with hiatus avoidance. Awobuluyi (1967, 1978) and Pulleyblank (1988, in press) show that the vowels of Yoruba behave asymmetrically with respect to processes of assimilation, deletion, epenthesis, and so on. In the type of contact situation illustrated in (19), nonhigh vowels and back vowels win out over the high front vowel [i] even at the expense of anchoring violations. Following Pulleyblank in press, this asymmetry is attributed to the encoding of sonoritybased distinctions directly into the set of faithfulness constraints. I assume that along the high-nonhigh dimension, nonhigh vowels are more sonorous than high vowels; along the front-back dimension, back vowels are more sonorous than front vowels:10

Patterns of Reduplication in Yoruba

323

(18) Sonority through faithfulness a. Constraint set i. MAXNONHIBK Every NonHi Back element of the input has a correspondent in the output ii. MAXNONHIFR Every NonHi Front element of the input has a correspondent in the output. iii. MAXHIBK Every Hi Bk element of the input has a correspondent in the output. iv. MAXHIFR Every Hi Fr element of the input has a correspondent in the output. b. Harmonic ranking MaxNonHiBk, MaxNonHiFr  MaxHiBk  MaxHiFr The e¤ect of these constraints is illustrated in (19), setting aside tones that are the focus of the next section. (19) Vowel deletion in reduplicated forms with ku´/ı`lo`-ku´-ı`lo`/ a. `ılo`ku´`ılo`

No Hiatus

DepIO

MaxIO HiBk

HiFr

AnchorL

*

*

*!*

**

**

**

*!

b. `ılo`kilo`

*!

+ c. `ılo`kulo` d. lo`kulo` e. ku´lo`kulo`

NonHi

*!*

The optimal candidate must respect NoHiatus. Deletion of the initial high front vowel of `ılo` (19c) is preferable to deletion of the high back vowel of ku´- (19b) because of the articulated MaxIO constraints. As candidates (19d,e) show, the ku´-forms also have interesting implications for the treatment of reduplication itself. In (19), the surface forms are straightforwardly derived under the assumption that reduplication is the result of morphologically imposed identity. The alternative, base-reduplicant correspondence, is less straightforward. Base-reduplicant correspondence per se is inadequate to account for forms like `ılo`kulo`. If ku´ were analyzed as part of the base then it should be copied: this would incorrectly derive a form like *ku´lo`kulo` (19e). If ku´ were not analyzed as part of the base then it should not be copied: this would incorrectly derive a form

324

Douglas Pulleyblank

like *lo`kulo` (19d). Since part of the input is copied but does not survive on the surface, it would be crucial to invoke input-reduplicant correspondence in addition to the more general base-reduplicant correspondence. Hence this pattern of reduplication would require the ‘‘full model’’ of reduplication, with its proliferation of reduplicant-specific constraints. 14.4

Prosodically Limited Reduplication of Segments Alone: The Distributive

In the two types of reduplication discussed so far, reduplication is total. The next two types to be discussed are partial reduplication, and raise the interesting question of whether reduplication of such a partial nature is consistent with the requirements of morphological identity. Hyman et al. (this volume) and Inkelas and Zoll (2000) argue that it is; it will be argued here, however, that the cases of partial reduplication in Yoruba do not involve morphological identity. But it will also be argued that base-reduplicant and input-reduplicant correspondence is unnecessary. I begin by outlining the facts of distributive reduplication, and then develop the theoretical approach as the analysis of this case is presented. The proposal builds on Kiparsky’s (1986b) proposal that total reduplication corresponds to compounding (the reduplicative types seen in sections 14.2 and 14.3), while partial reduplication corresponds to a‰xation (the reduplicative types about to be examined).11 14.4.1

Distributive Reduplication

Distributive reduplication is a foot-based pattern, involving the copy of an initial VCV sequence (Folarin 1987; Pulleyblank 1988; Ola 1995; among others).12 ˙ ˙ (20) Distributive reduplication a. eweewe´ ‘every leaf ’ ewe´ ‘leaf ’ odoodu´n ‘every year’ odu´n ‘year’ ˙ ˙˙ ˙ osoosu` ‘every month’ osu` ‘month’ ˙ ˙ ˙ ojooju´mo´ ‘every day’ oju´mo´ ‘day’ ˙ ˙ `ıla` b. `ılı``ıla` ‘every line’ ‘line’ o`ro`o`ru ‘every midnight’ o`ru ‘midnight’ `ırı``ıro`le´ `ıro`le´ ‘every evening’ ‘evening’ ˙ ˙ ˙ ˙ o`wu´ro` ‘morning’ o`wo`o`wu´ro` ‘every morning’ ˙ ˙ The distributive pattern exhibits interesting properties. First, it rigidly involves the copy of a single foot.13 If the base constitutes a single binary foot, then the entire foot is copied—for example, eweewe´ ‘every leaf ’. In contrast, if the base is longer than a binary foot, then only the initial VCV sequence is copied, as in ojooju´mo´ ˙ ‘every day’. Second, the distributive exhibits tonal neutralization. If the first mora of the base is M, then both vowels of the distributive are M; if the first mora of the base

Patterns of Reduplication in Yoruba

325

is L, then both vowels of the distributive are L.14 These two core properties of the distributive will be considered in turn, showing also that the vowel assimilation these forms exhibit follows from the analysis proposed. 14.4.2

Integrity Violations

The distributive manifests itself as a single foot, the Yoruba minimal word. Rather than have this constitute a formal condition on some category ‘‘reduplicant,’’ I suggest here that this characterization is by itself su‰cient to derive reduplication without any formal constraints actually referring to reduplication—a return to the type of prosodic characterization of reduplicants found in Marantz 1982, McCarthy and Prince 1986, and so on. Specifically, I propose that the phonological representation of the distributive morpheme is underlyingly /F/ (‘foot’) and that /F/ must be featurally specified on the surface in order to satisfy a family of constraints requiring that segments be featurally specified (Itoˆ and Mester 1993; Padgett 1995; Pulleyblank 1997; and so forth). (21) HAVESPEC A prosodic category must dominate featural specifications. Itoˆ and Mester (1993) explore the possibility that this type of constraint is part of the family of properheadedness constraints. Just as a foot must be headed by a syllable and a syllable headed by a mora, so must a mora be headed by a root node, a root node headed by a place specification, and so on. The precise formulation of this constraint type is not crucial here; what matters is only that they supply pressure to assign features to prosodic categories such as foot and syllable. To see how the analysis works, consider the derivation of a form like ojooju´mo´ ˙ from the input /F-oju´mo´/. HaveSpec rules out the candidate output in which the ˙ foot surfaces without any phonological content at all (22a). MaxIO bans deletion of the a‰xal foot (22b). A third possibility would be to insert default phonological content, a violation of DepIO (22c): (22) Basic options for realization /F-oju´mo´/ ˙ a. F [ ]-oju´mo´ ˙ b. oju´mo´ ˙ c. F [ihi]-oju´mo´ ˙

HaveSpec

DepIO

MaxIO

* * ***

Clearly, none of the outcomes entertained in (22) is reduplicative. A third class of options exists, however. Consider candidates that satisfy the requirement of being a

326

Douglas Pulleyblank

foot, but where each output segment corresponds directly (23a) or indirectly (23b) to some input segment: (23) a.

b.

Such candidates would satisfy the lexical specification that the distributive is a foot, and would do so without incurring any violations of DepIO. DepIO is satisfied in (23a) because every segment in the output directly corresponds to a segment in the input. Struijke (1998) refers to this type of faithfulness as ‘‘broad’’ input-output faithfulness.15 In (23b), DepIO is indirectly satisfied: the reduplicated form corresponds to part of the output base form, which in turn corresponds to part of the input. If the correspondence relations of (23b) are interpreted transitively, then all surface segments correspond to some input segment, hence DepIO is satisfied (see Spaelti 1997). E¤ectively, therefore, there is no need for a special set of constraints to ensure correspondence between the ‘‘Base’’ and the ‘‘Reduplicant’’—such correspondence is simply one way of satisfying the prosodic condition on the distributive without incurring violations of DepIO. An output form like (23a) involves correspondence between the input oju´mo´ and ˙ the stem output oju´mo´, as well as between the input oju´mo´ and the output redupli˙ ˙ cant oju. While this representation is essentially the same as the one that would result from invoking input-output and input-reduplicant faithfulness (McCarthy and Prince 1995, 1999), the nature of the form is actually quite di¤erent. To achieve the result in (23a), input-reduplicant faithfulness would have to (i) posit a category of morpheme ‘‘reduplicant’’ (Red) that is subject to a special class of faithfulness constraints, and (ii) posit a special class of faithfulness constraints relating inputs to the ‘‘reduplicant’’ category so established. The input-reduplicant correspondence relation is directly mediated by the special constraints. In contrast, the analysis proposed here accomplishes a similar relation indirectly. The correspondence relation between the input oju´mo´ and the prefix oju exists not to satisfy some special class of input-reduplicant ˙ faithfulness, but to satisfy the requirement that the prefix have featural content while avoiding violations of the input-output faithfulness requirement Dep. What in general then would prevent such candidates from occurring, unmotivated by any ‘‘reduplicative’’ morpheme? Why do we not observe unprovoked duplication of base material willy-nilly? An answer to this question noted by Spaelti (1997) is to

Patterns of Reduplication in Yoruba

327

be found in the constraint Integrity, proposed and defined by McCarthy and Prince (1995):16 (24) INTEGRITY No element of the input has multiple correspondents in the output. Representations such as (23) violate Integrity since the three input segments /oju/ each have two surface correspondents. The resulting Integrity violations will be tolerated only if forced by some higher constraint—the standard OT scenario. In the case of the distributive morpheme, Integrity violations are forced by the need to satisfy the constraint requiring that the lexically specified foot of the distributive dominate featural content. Consider three possible candidates for the distributive form ojooju´mo´ (base oju´mo´ ), ˙ ˙ first under the assumption of indirect correspondence (comparable to (23b)): (25) Basic options for realization (continued) /F-oju´mo´/ ˙ a.

HaveSpec

DepIO

FootBin

*!**

+ b.

c.

Integ

***

*!

*

328

Douglas Pulleyblank

/F-oju´mo´/ ˙

HaveSpec

DepIO

d.

FootBin

Integ

*!

*****

In (25a), the first three segments of the distributive are accidentally homophonous with segments of the base. Accidental homophony, however, does not constitute formal correspondence, hence each such segment constitutes an unforced violation of DepIO (fatal violations given the ranking). A candidate such as *ihioju´mo´ would be ˙ similarly excluded by DepIO. Comparing candidates with partial copying (25b,c) versus complete copying (25d), we see that copying will take place to the point where prosodic constraints are satisfied—in this case, the requirement that a foot be binary (FootBin—see Ola 1995; Orie and Pulleyblank 2002). ˙ The tableau in (25) indicates only the ‘‘reduplicative’’ core. All four candidates shown violate NoHiatus. These vowel contact violations can be avoided by epenthesizing a consonant (26a), by copying an additional segment (26b), or by copying less and postulating a multiply linked segment (26c). Note that a ‘‘V-deletion’’ option (*F [oj]-oju´mo´)/*F [ojo]-ju´mo´) would not be viable since it would either violate left˙ ˙ edge anchoring or prosodic conditions on feet. (26) Resolving the hiatus problem /F-oju´mo´/ ˙ a. F [oju]h-oju´mo´ ˙ b. F [oju]m-oju´mo´ ˙ + c. F [ojo]-oju´mo´ ˙

NoHiatus

DepIO *!

MaxIO

AnchorL

Integ *** ***!* **

MaxIO plays no role in distinguishing (26b) and (26c) since all input segments have output correspondents in both. The basic conclusion is that constraints with specific reference to a category type ‘‘reduplicant’’ are redundant. Given prosodically specified morphemes with incomplete specifications (Marantz 1982), appropriately ranked constraints on inputoutput correspondence will evaluate reduplicative forms as optimal. Under this view,

Patterns of Reduplication in Yoruba

329

reduplication is an accidental by-product of underspecified templates with independently motivated constraints. A formal issue arises under this view as to whether reduplication should be treated as involving direct (23a) or indirect (23b) correspondence to the input form. In (25), indirect correspondence was assumed, but the alternative of direct correspondence was not considered. Abstracting away from all aspects of (26c) other than the two segments that are in correspondence, consider how the constraint set would evaluate the two forms in (23). (27) Direct versus indirect correspondence in reduplication /F-oju´mo´/ ˙

DepIO

MaxIO

AnchorL

Integ

a.

**

b.

**

As can be seen, the constraint set does not distinguish between the two patterns of correspondence.17 This changes, however, if Linearity is brought into play (McCarthy and Prince 1995). (28) LINEARITY The linear order of elements in S1 is identical to the linear order of their corresponding elements in S2 . Because of the transitive correspondence relation between the elements in correspondence in (29b), Linearity is fully respected. In contrast, the direct correspondence relation observed in (29a) brings about Linearity violations. For example, in the input oju´mo´, o precedes j but in the output, a direct correspondent of the input o ˙ follows j (ojuoju´mo´). Such contradictory precedence relations are an indication that ˙ Linearity has been violated.18

330

Douglas Pulleyblank

(29) The relevance of linearity /F-oju´mo´/ ˙

DepIO

MaxIO

AnchorL

Integ

Lin

a.

**

*!

+ b.

**

Note that the ranking of Linearity with respect to the other constraints is not crucial here because the two candidates di¤er only in their satisfaction/violation of this one constraint. While the correspondence relations of (29a) are comparable to those of inputreduplicant correspondence, and those of (29b) are comparable to those of basereduplicant correspondence, it is important to keep in mind that the status of such relations is fundamentally di¤erent in this approach to that proposed in McCarthy and Prince 1995. Unlike the McCarthy and Prince proposal, which posits faithfulness constraints that explicitly govern these relations, in the approach taken here the correspondence relations are indirectly governed through reference to nonreduplicative constraints such as Integrity and Linearity. To summarize, the proposal made here is that phonological representations do not need to postulate an ad hoc category ‘‘Reduplicant’’ in order to derive reduplicated forms. There is no need for special reduplication-specific classes of correspondence constraints of the Base-Reduplicant and Input-Reduplicant types as proposed in McCarthy and Prince 1995. Instead, forms with correspondence relations between one morpheme and another are possible candidates produced by Gen, and they will be analyzed as optimal in situations where a morpheme is prosodically characterized, incompletely specified, and where Dep and Integrity are appropriately ranked. This proposed elimination of the class of reduplication-specific faithfulness constraints requires, however, consideration of several further issues, to which I now turn: (i) why prosody-induced reduplication cannot be reduced to reduplication by morphological identity, (ii) why a return to a prosodic characterization of reduplication does not induce base erosion, and (iii) how one accounts for e¤ects of the emergence of the unmarked.19

Patterns of Reduplication in Yoruba

14.4.3

331

Reduplication Not Morphologically Induced

Vowel contact arising from distributive reduplication is unlike the general case where two vowels abut, and also unlike cases of abutting vowels in kı´-reduplication. As shown in this section, the observed behavior is, however, precisely the expected behavior where reduplication results from the satisfaction of prosodic requirements, not from morphological identity. The crucial point for hiatus resolution is that the reduplicant is defined by a particular prosodic shape, as argued for the distributive, and not by morphological identity in conjunction with a prosodic delimiter.20 Consider distributive reduplication cases where the initial vowel of the base is [i] (reproduced from (20)). (30) Distributive reduplication with i-initial bases `ılı``ıla` `ıla` ‘every line’ ‘line’ `ırı``ıro`le´ `ıro`le´ ‘every evening’ ‘evening’ ˙ ˙ ˙ ˙ If reduplication were morphologically induced, we would have inputs such as /ı`la`-ı`la`/. Given the phonological grammar of Yoruba developed in sections 14.2 and 14.3,21 this would lead to the incorrect prediction that the surface form should be *ı`la`a`la`. The tableau in (31) illustrates this and should be compared with the tableau in (19); here and elsewhere, a bomb (M) in a tableau designates a candidate that is evaluated as optimal by the analysis but is in fact ungrammatical. (31) Retention of base vowel in distributive reduplication: Morphological identity /ı`la`-ı`la`/ a. `ıla``ıla`

NoHiatus

DepIO

NonHi

HiFr

AnchorL

*

*

*!

M b. `ıla`a`la` c. `ılı``ıla`

MaxIO HiBk

*!

For additional discussion of hiatus resolution in sequences of underlyingly specified vowels, see Pulleyblank 1988, in press; Rosenthall 1997; Orie and Pulleyblank 2002. If reduplication is the result of prosodic specification of a foot, then the correct result is predicted:

332

Douglas Pulleyblank

(32) Retention of base vowel in distributive reduplication: Prosodically specified a‰x /F-ı`la`/ a. `ıla`ı`la`

NoHiatus

DepIO

NonHi

MaxIO HiBk

HiFr

AnchorL

*!

b. `ıla`a`la`

*!

+ c. `ılı``ıla` Because the distributive has no segmental specifications in the input, MaxIO is irrelevant for the prefixal foot;22 because each of the segments of the input `ıla` has one or more surface correspondents in the three output candidates of (32), MaxIO is fully satisfied by all three candidates. As a result, MaxIO is irrelevant and left-edge anchoring of the morpheme `ıla` ensures selection of `ılı``ıla` as optimal rather than *ı`la`a`la`. Morphologically induced reduplication thus must be distinguished from reduplication that is prosodically induced. While this is a return in spirit to the proposal of Marantz 1982, it is a return mediated by the theory of correspondence proposed by McCarthy and Prince 1995, yet not requiring the reduplication-specific constraints proposed there. 14.4.4

Why the Base Does Not Erode

It is argued in Spaelti 1997 and McCarthy and Prince 1999 that it is undesirable to characterize reduplicative morphemes prosodically due to the potential for base erosion—a problem referred to by McCarthy and Prince as the ‘‘Kager-Hamilton problem.’’ If constraints of base-reduplicant correspondence are posited, then it should be possible for these to be in a language in which MaxBR outranks MaxIO. Given such a ranking, if reduplicants are prosodically characterized, then it should be possible for satisfaction of a prosodic Red constraint (e.g., Red ¼ Foot) and MaxBR to force erosion of the base. For example, if this ranking were to apply to an input like RED-oju´mo´, then the optimal output (ignoring constraints such as ˙ NoHiatus) would be ojuoju´ (the reduplicant is underlined): (33) Base-reduplicant correspondence resulting in base erosion /Red-oju´mo´/ ˙ a. [ojumo-oju´mo´] ˙ ˙ b. [oju-oju´mo´] ˙ + c. [oju-oju´]

MaxBR

Red ¼ Foot

MaxIO

*! *!* **

Patterns of Reduplication in Yoruba

333

Spaelti (1997) and McCarthy and Prince (1999) assume that base erosion is nonexistent and seek to eliminate it as a possibility by prohibiting constraints like Red¼ Foot from referring to prosodic constituents. ‘‘Generalized template’’ theory (see McCarthy and Prince 1999 and references therein) is one approach to this type of restriction. Two points are important in addressing this issue. First, it must be established whether there are valid cases of base erosion. If, for example, Hausa is correctly considered to exhibit base erosion (Downing 2000), then there is a need for an analysis of base erosion. Second, whatever the answer to the question of whether base erosion exists, it would appear that such erosion should not be the result of phonological constraints. That is, if base erosion exists, then it must be of a type that is morphologically rare. The proposal elaborated here is that base erosion should be accounted for by the sorts of morphological mechanisms responsible for subtractive morphology. Just as subtractive morphology in general is rare, so then should base erosion be rare. The crucial aspect of the current phonological proposal is that base erosion cannot result from the properties of the purely phonological constraints. By eliminating from the theory all faithfulness constraints referring explicitly to the category of ‘‘reduplicant’’—that is, by eliminating all constraints of the BR and IR classes— base erosion is completely excluded as a phonologically induced phenomenon. Consider candidates such as oju-oju´mo´ and oju-oju´ from the perspective of the ˙ analysis presented here, abstracting away from issues such as hiatus avoidance: (34) Absence of erosion in a theory without reduplicant-specific constraints /F-oju´mo´/ ˙ a. [ojumo-oju´mo´] ˙ ˙ + b. [oju-oju´mo´] ˙ c. [oju-oju´]

DepIO

FootBin

MaxIO

*!

Integ ***** ***

*!*

***

MaxIO is violated by deletion of base segments but is una¤ected by the number of corresponding segments found in the reduplicant (each of which violates Integrity). The relative ranking of MaxIO and Integrity is not crucial since a candidate with an eroded base (34c) incurs as many violations of Integrity as a form that is faithful to the base (34b). As a simple consequence of not having a category ‘‘reduplicant,’’ base-reduplicant correspondence cannot be ranked above Input-Output correspondence since there is no base-reduplicant correspondence at all. Phonologically induced erosion is eliminated.

334

14.4.5

Douglas Pulleyblank

Tonal Properties: Emergence of the Unmarked

The final point to be considered with regard to distributive reduplication is the construction’s tonal properties. As pointed out with reference to (20), the tone of the distributive reduplicant is completely predictable: both moras of the reduplicant are M if the first mora of the base is M (20a) and L if the first mora of the base is L (20b); recall from section 14.3.4 that H tones may not occur on a word-initial vowel, hence will not occur on a relevant reduplicative base. In earlier accounts of the distributive within a theory of tonal underspecification (Akinlabi 1984; Pulleyblank 1986), this pattern has been accounted for by (i) not copying tone during the reduplicative process, (ii) postulating a rule of leftward L tone spreading, and (iii) assigning midtones by default. The crucial aspect of these analyses is the rule of leftward spreading. Whether tone is copied or not during reduplication would determine whether the rule must be feature-changing or featurefilling, but would not a¤ect the necessity of the rule itself. Similarly, if M is a default value then the rule could spread L tone only, whereas if M was a specified value then the rule could be formulated so as to spread all tones. While the exact treatment of tone spreading may not be crucial, let us assume for concreteness that the grammar of Yoruba has a constraint-based equivalent of tone spreading as in (35). (35) ALIGNLEFT[TONE] AlignLeft(Tone; Word) The e¤ects of this constraint will be superseded in nonreduplicative forms by the exigencies of regular input-output faithfulness. Following work such as Lombardi 1995, 1998, Pulleyblank 1996, and Myers 1997, I assume here feature-based faithfulness and reference in faithfulness constraints to feature paths (Archangeli and Pulleyblank 1994). With regard to tone, therefore, the following constraints are relevant for an evaluation of faithfulness: (36) a. MAXTONE Every tone of the input has a correspondent in the output. b. DEPTONE Every tone of the output has a correspondent in the input. c. MAXPATHTONE Every path between a tone and an anchor in the input has a correspondent in the output. d. DEPPATHTONE Every path between a tone and an anchor in the output has a correspondent in the input. MaxTone will prevent the elimination of tones and MaxPathTone will prevent the redistribution of tones. DepTone will prevent the insertion of tones and DepPath-

Patterns of Reduplication in Yoruba

335

Tone will prevent the spreading of tones. If the tonal faithfulness constraints outrank AlignLeft[Tone], then there will be no leftward spreading. This can be seen by consideration of a word like `ıro`le´ ‘evening’ (from (20)). ˙ ˙ (37) Tonal faithfulness irole ˙ ˙ l h

MaxTone

DepTone

MaxPath Tone

+ a. irole ˙ ˙

AlignL Tone

DepPath Tone

**(H)

l h b. irole ˙ ˙

*!

*!

*

l *!

c. irole ˙ ˙

*(H)

*

l h d. irole ˙ ˙

*!

*!*

**

h Note in particular that the attested surface form `ıro`le´ violates AlignLeft twice. ˙ ˙ Since this form is nevertheless optimal, this means candidates respecting or minimally violating AlignLeft must be ruled out by more highly ranked constraints. Specifically, candidates (37b,d) must be ruled out by the higher ranking of at least one of MaxTone, MaxPathTone, and DepPathTone, while candidate (37c) must be ruled out by the higher ranking of either MaxPathTone or DepPathTone. The actual assumption made in (37) is that MaxTone and MaxPathTone outrank AlignLeft and that AlignLeft in turn outranks DepPathTone. While the low ranking of DepPathTone would have no e¤ect on simple forms, it would have a significant e¤ect on a reduplicated form. Consider the following tableau illustrating the reduplicated form `ırı``ıro`le´ ‘every evening’. (Consistent with the results of (37), only ˙ ˙ candidates that respect root tone assignments are illustrated.)

336

Douglas Pulleyblank

(38) Tonal alignment in reduplication Fþirole ˙ ˙ l h

MaxTone

DepTone

MaxPath Tone

AlignL Tone

DepPath Tone

*!*(L) ****(H)

a. iriirole ˙ ˙ l l h + b. iriirole ˙ ˙

****(H)

**

**(L) ****(H)

**

l h *!

c. iriirole ˙ ˙ m l h

The first candidate (38a) contains a tone in correspondence with the root; the second candidate (38b) spreads the tone of the root; the third candidate (38c) assigns a default tone to the reduplicant. Note that the associations of the initial L in (38a) do not constitute DepPathTone violations because both the L and the associations have correspondents in the input. Although the choice of candidate (38b) over (38a) does not result in a surface tonal di¤erence, the comparable choice in a form like ojooju´mo´ ‘every day’ is crucial: ˙ (39) Tonal alignment in reduplication (continued . . .) Fþojumo ˙ m h

MaxTone

DepTone

MaxPath Tone

AlignL Tone

DepPath Tone

*!(H) **(M) ***(H)

a. ojo ojumo ˙ m hm h + b. ojo ojumo ˙

***(H)

**

*(H)

***

m h c. ojo ojumo ˙ m

h

*!(M)

Patterns of Reduplication in Yoruba

337

By spreading the initial M of oju´mo´ (39b), AlignLeft is better satisfied than if tones ˙ are in correspondence (39a). Candidate (39c) shows that in a derived form just as in an underived form, spreading is constrained by faithfulness so as not to a¤ect root specifications. The failure to copy tone has been derived by ranking a constraint on tonal distribution (AlignLeft[Tone]) above a relevant faithfulness constraint. As such, this case exhibits a pattern of the ‘‘emergence of the unmarked’’ (McCarthy and Prince 1994, etc.). Note that many languages systematically require tonal specifications to be leftaligned—the left-edge asymmetry is built into early autosegmental treatments of tone such as Leben 1973 and Goldsmith 1976. While not generally true of Yoruba due to the high ranking of particular tonal faithfulness conditions, we see left-edge preference emerging under reduplication.23 The logic of the proposed analysis of tone in the distributive is that a set of faithfulness constraints (Faithi ) prevents some (markedness) constraint (C) from having any e¤ect in basic forms. (40) Faithi  C With regard to some more lowly ranked faithfulness constraint set (Faithj ), however, C is dominant. (41) C  Faithj Putting the two subrankings together, we get the following: (42) Faithi  C  Faithj This schema is precisely the ranking that derives emergence of the unmarked: the marked case is tolerated generally, but the unmarked case ‘‘emerges’’ in contexts governed only by Faithj . In the case of the Yoruba distributive, Faithi is the set {MaxTone, DepTone, MaxPathTone}, C is {AlignL[Tone]}, and Faithj is {DepPathTone}. The general point is that emergence of the unmarked e¤ects in reduplication does not depend on the postulation of a special set of base-reduplicant correspondence constraints. Even in a theory that does not include such construction-specific constraints, unmarked e¤ects may emerge from the way di¤erent input-output faithfulness constraints are ranked with respect to particular markedness constraints. Note also that whether tone is copied in a particular type of reduplication or not is straightforwardly derived from the constraint grammar. Contra Walsh-Dickey (1992, 1993), to distinguish between cases where tone is or is not copied, there is no need to assign tone to di¤erent structural tiers. Moreover, it is not the case that all reduplicative processes will behave the same way in the same language. We have seen that agentive reduplication and kı´-reduplication copy tone, for example, while distributive

338

Douglas Pulleyblank

reduplication does not. Such di¤erences in whether tone is copied or not are consistent with the constraint-based approach taken here but inconsistent with a theory where tone must be represented on di¤erent tiers to account for its variable behavior in reduplication. 14.5

Gerundive Reduplication

The last case of reduplication to be considered in this chapter is the gerundive. In addition to issues of descriptive interest that arise in connection with this pattern of reduplication, two issues of theoretical significance are addressed, namely fixed default segmentism and underapplication. These issues are of particular interest because the Yoruba gerundive has been taken as a paradigm case of base-reduplicant correspondence in Alderete et al. 1999. Gerundive reduplication is illustrated in (43):24 (43) Gerundive reduplication Verb Verbal noun/gerund a. rı´ [rı´] rı´rı´ [rı´rı´] ´ ´ ´ ´ b. gbe [gbe] gbıgbe [gbı´gbe´] c. je [J] jı´je [Jı´J] ˙ ˙ d. da´ra [da´ra] dı´da´ra [dı´da´ra] fı´fo´ [fı´f ´] e. fo´ [f ´] ˙ ˙ f. to´bi [to´bi] tı´to´bi [tı´to´bi] g. bu` [bu`] bı´bu` [bı´bu`] h. sı´n [sı˜´ ] sı´sı´n [sı´sı´˜ ] i. gba´n [gba˜´] gbı´gba´n [gbı´gba˜´] j. du`n [du`˜ ] dı´du`n [dı´du˜`]

Gloss ‘see’ ‘take’ ‘eat’ ‘be good’ ‘shatter’ ‘be big’ ‘dip out’ ‘sneeze’ ‘scoop out’ ‘be sweet’

c

c

This reduplicative pattern has been used to illustrate the analysis of reduplication with fixed segmentism in various prominent treatments of reduplication within prosodic models—for example, Marantz 1982, Kenstowicz 1994, and Alderete et al. 1999. One of the reasons for choosing the Yoruba pattern to illustrate fixed segmentism is its apparent regularity and simplicity: the gerundive forms are derived by a‰xing a [Cı´-] prefix, where the ‘‘C’’ is a copy of the first consonant of the verbal base. The fixed segmentism is the H-toned high front oral vowel [ı´]. Deeper investigation into the prosodic, tonal, and nasal properties of the gerundive prefix, however, reveals a much more complex situation, and a useful test of the approach to reduplication advocated in this chapter. This section is structured as follows. First, the prosodic properties of gerundive reduplication are examined. Second, the marked tonal features are incorporated into the analysis. Third, the vocalic properties, though unmarked, are argued to require

Patterns of Reduplication in Yoruba

339

prespecification. Finally, general properties of nasality in Yoruba are laid out, followed by a treatment of nasality in reduplication. 14.5.1

Prosodic Aspects of Gerundive Reduplication

Prosodically, the gerundive consists of a single syllable. There are three types of explanations for this shape: (i) it could be specified as part of the lexical entry of the gerundive that it be one syllable, (ii) there could be morphological reasons why the gerundive a‰x might be one syllable, and (iii) there could be phonological reasons why the morpheme might be one syllable. The possibility that the gerundive is a syllable due to the lexical specification ‘‘s’’ (Ola 1995) might appear to be the most straightforward approach. In fact, however, ˙ it seems redundant. In order for the template ‘‘s’’ to result in a CV surface form, it would be necessary for some constraint to force the presence of an onset consonant. The constraint ‘‘Onset,’’ requiring that all syllables have an onset, would be a plausible source of pressure. As argued by Orie and Pulleyblank (2002), however, general considerations of hiatus resolution in Yoruba argue that Onset is ranked below both Max and Dep—too low to force the presence of an onset in a morpheme specified only by a prosodic template. It might nevertheless be imagined that Onset could force the presence of an onset in the gerundive if we were to assume that it outranks Integrity. (44) Prosodic shape of the gerundive /s-to´bi/

HaveSpec

DepIO

a. to´bi b. s-to´bi c.

MaxIO

Onset

*! *!

*

´ ]bito´bi s [to

***!*

+ d. s [to´]to´bi e.

´ bi s [hı´]to

f.

´ ]to´bi s [o

Integ

** *!* *!

*

The optimal result would be a candidate filling the gerundive syllable with features in correspondence to the base (44c,d); since the specification of such features involves Integrity violations, however, there would be a preference for including the minimal number of specifications necessary to fill the lexically assigned syllable—hence (44d) would be preferred to (44c). (Note that the actual output, tı´to´bi, has fixed segmentism; this is addressed below.)

340

Douglas Pulleyblank

Attributing the impossibility of *o´to´bi to Onset, however, produces an immediate problem. If Onset is ranked above Integrity, then vowel-initial forms should be ruled out in general—a prediction roundly falsified in Yoruba (consider (9), (13), (20), and so on). To prevent such overproduction of reduplicative onsets, it is crucial that Integrity outrank Onset, and therefore that some other constraint rule out a form like *o´to´bi in (44f ).25 As will be seen below, the identification of the relevant constraint renders the lexical postulation of a syllable for the gerundive redundant. Turning to other possible explanations for the gerundive shape, one might imagine, following generalized template theory (Urbanczyk 1996; McCarthy and Prince 1999, etc.), that it falls out of general considerations of a‰x shape. Attempting to derive the CV form of the gerundive by stipulating that ‘‘Gerundive ¼ A‰x’’ turns out to be somewhat problematic, however. First, note that one of the strongest motivations behind generalized templates (namely how to prevent base erosion) has been removed by the absence of base-reduplicant correspondence constraints. As discussed in section 14.4.3, the problem of base erosion does not arise if there is no basereduplicant faithfulness. Without base erosion as a problem, a central reason for generalized templates disappears. In any event, a more direct problem arises when we actually consider the prosodic properties of Yoruba a‰xes. Consider the list in (45) (Bamgbose 1967; Rowlands ˙ 1969):26 (45) Yoruba prefixes a`nominalizer a`lo ‘going’ < lo ‘go’ ˙ ˙ aagentive akorin ‘singer’ < korin ‘sing’ ˙ ˙ `ıabstract nominal `ıda´wo´ ‘subscription’ < da´wo´ ‘subscribe’ a``ınegative a``ıda´ra ‘not being good’ < da´ra ‘be good’ a`tisise´ ‘to work’ a`ti- infinitival ‘work’ < sise´ ˙ ˙˙ ˙ ˙˙ onı´le´ ‘landlord’ < ile´ ‘house’ onı´- owner of X o`/o`- agentive o`sise´ ‘worker’ < sise´ ‘work’ ˙ ˙ ˙˙ ˙ ˙˙ olu`- agentive olu`ko´ ‘teacher’ < ko´ ‘teach’ ˙ ˙ On the basis of this list, the primary generalizations governing Yoruba a‰xes are as follows: (i) all a‰xes in nonreduplicative forms are vowel-initial, and (ii) the first (and sometimes only) mora of an a‰x is nonhigh (M or L). Phrased negatively, a‰xes are never consonant-initial and the initial mora of an a‰x is never H.27 These requirements mirror general conditions on the optimal shape of a noun, conditions that define that shape as [VCV . . . ] (Ola 1995). But if being vowel-initial and bearing ˙ a nonhigh tone are the optimal properties for an a‰x, then why does the gerundive reduplicant do exactly the opposite? Why does the gerundive begin with a consonant and bear an H tone? Moreover, there would be only weak language-internal grounds

Patterns of Reduplication in Yoruba

341

for analyzing a canonical a‰x in Yoruba as monomoraic when roughly 50 percent of the attested a‰xes are bimoraic. The basic point is that the prosodic properties of a‰xes in Yoruba, however ensured by constraints, cannot be responsible for deriving the gerundive shapes in (43) since the gerundive shape is not canonical. The gerundive shape could only arise in spite of general constraints on the prosodic realization of a‰xes. Hence generalized template theory must look to constraints other than those defining a‰x shape to account for the prosodic properties of the gerundive. We are therefore left with the possibility that the syllable shape of the gerundive is due to phonological properties interacting with its lexical specification. The crucial observation, made by Ola (1995), is that a word-initial bare V cannot carry an H ˙ tone. She suggests that this constraint accounts for the initial C of gerundive reduplication. By assuming (i) that the gerundive morpheme underlyingly contains an H tone, (ii) that the constraint demanding faithfulness to such an H (MaxH) is undominated, and (iii) that the constraint prohibiting an H-toned bare initial V is also un´ ’’), it follows that some modification dominated (a constraint that will be labeled ‘‘*[V of an H tone input is required. In exploring the implications of a lexical H tone specification, I assume an input for the gerundive that consists simply of an H tone mora; assuming the presence of a syllable as well would play no crucial role though it would be fully compatible with the surface pattern observed. (46) Underlying H tone initial vowel /mH-to´bi/

MaxH

a. [ı´-to´bi] b. [i-to´bi]

´ *[V *

*

Ola proposes that the necessary modification involves reduplication. By adding a ˙ reduplicative consonant to the H-toned prefix, both violations of (46) disappear: (47) Inclusion of reduplicative consonant /mH-to´bi/

MaxH

´ *[V

a. [tı´-to´bi] Ola suggests that this analysis fills out an apparent gap in the paradigm of tonal ˙ possibilities on a‰xes. Data such as in (45) show that of the three lexical tones of Yoruba, H, M, L, only two, M and L, appear on prefixes. Given the combinatorial possibilities that such tones would allow, we would expect H-toned prefixes to be possible—the gerundive reduplicative prefix constitutes exactly that case. Since an

342

Douglas Pulleyblank

H tone is marked (section 14.3.4), its presence in the gerundive must be lexically specified—an ‘‘overwriting string’’ in the terminology of Alderete et al. 1999. To summarize, it has been shown that neither the lexical specification of ‘‘s’’ nor the general properties of being an a‰x are su‰cient to account for the prosodic properties of the gerundive prefix. To derive the observed CV shape of the gerundive, it is proposed that the reduplicant bears an H tone underlyingly, a feature whose presence on a vowel forces the appearance of an onset due to the particular con´ . Of direct importance in this context is the conclusion reached in section straint *[V 14.3.2 concerning the markedness relation between H, M, and L tones. If, as concluded there, the H tone is the most marked tone (at least for Yoruba), then its presence in the gerundive reduplicant can only be due to the lexical presence of the H: it cannot be a default or unmarked realization if H is the most marked tone, and there is no general rule of H-insertion that could be considered to apply to the reduplicative prefix of the gerundive. The lexical representation of the gerundive must therefore include an H tone mora (mH). 14.5.2

The Segmental Nature of the Gerundive

As noted above, the gerundive exhibits the fixed vowel quality [i], a quality that Alderete et al. (1999) attribute to emergent unmarkedness in the reduplicant. Their proposal for the gerundive reflects the general schema of ‘‘emergence of the unmarked’’ e¤ects (McCarthy and Prince 1994, etc.), where a markedness constraint is ranked between input-output faithfulness and base-reduplicant faithfulness: FaithIO  Markedness  FaithBR. Specifically for the Yoruba case, constraints favoring [i] over other vowels are ranked below MaxIO but above MaxBR. Because they are ranked below MaxIO, they have no e¤ect on simple inputs; because they are ranked above MaxBR, they cause the vowel of the gerundive to be [i]. Just as they can within a theory assuming base-reduplicant faithfulness, patterns of ‘‘emergence of the unmarked’’ can be derived within the approach taken here. Nevertheless, when the large set of reduplicative patterns in Yoruba is analyzed as a whole, it is implausible that the [i] of the gerundive is emergent. To understand these points, consider the typology generated by DepIO above Integrity: in a grammar that ranks Integrity above DepIO, epenthesis is the optimal way to resolve the problem of an incompletely specified morpheme; in a grammar that ranks DepIO above Integrity, reduplication is the optimal strategy. Since Yoruba adopts the reduplicative strategy, it is necessary to rank DepIO above Integrity. Consider an example of the distributive, ojooju´mo´, along with an example of the gerundive, ˙ tı´to´bi.

Patterns of Reduplication in Yoruba

343

(48) Divergent reduplicative patterns /F-oju´mo´/ ˙ + a. ojo-oju´mo´ ˙ b. ijo-oju´mo´ ˙ c. iho-oju´mo´ ˙ /mH-to´bi/

DepIO

Integ **

*!

*

*!*

M d. to´-to´bi

**

e. tı´-to´bi

*!

f. hı´-to´bi

*!*

*

The problem with such pairs of forms is that the ranking of DepIO above Integrity ensures reduplication, and also results in the reduplication of both consonants and vowels. This is correct for the distributive, but incorrect for the gerundive. To resolve this problem, two approaches are possible. The first is to lexically specify the gerundive for the vowel [i], rather than just for a mora (Marantz 1982). This is the solution adopted here: (49) Divergent reduplicative patterns /ı´-to´bi/ a. to´-to´bi + b. tı´-to´bi c. hı´-to´bi

MaxIO

DepIO

*!

Integ **

*

*

*!*

This prespecification approach correctly derives the gerundive without jeopardizing the analysis already given of the distributive. Alderete et al. (1999, 358) mention this possibility but note that ‘‘a‰xation would fail to account for the correlations with independently motivated defaults.’’ Correlations, of course, may be accidental. Interestingly, Akinlabi (2000) argues for independent reasons that the [i] of the gerundive must be underlying, not default. He notes that assimilatory processes treat the gerundive vowel di¤erently than cases that are unambiguous instantiations of default [i]. Abı´o´du´n (1997) also suggests that the [i] of the gerundive is underlyingly specified. ˙ He observes that a pattern of ideophonic reduplication systematically specifies the vowel [a], noting therefore that the quality of the vowel in a context of fixed segmentism is not predictable.

344

Douglas Pulleyblank

(50) Fixed [a] in ideophonic reduplication a. fakafı`kı` ‘up and down movement’ b. wa`ra`we´re´ ‘very quickly’ c. ya`da`yo`do` ‘in a joyful mood’ d. gagagu´gu´ ‘plenty of load’ In spite of these arguments in favor of prespecification, it is worth considering an alternative whereby faithfulness is divided into faithfulness to vowel features and faithfulness to consonant features (Alderete et al. 1999). This alternative could derive the gerundive by ranking Integrity below input-output faithfulness to consonants (deriving the reduplication of consonants) and above input-output faithfulness to vowels (deriving default vowel ‘‘epenthesis’’). The split-faithfulness approach fails, however, when attempting to derive both the gerundive and the distributive since to succeed on the gerundive (copying consonants but not vowels) means to fail on the distributive (which copies both consonants and vowels). In spite of its overall failure, it is worth considering the split-faithfulness approach briefly since it illustrates one way of achieving emergence of the unmarked e¤ects even without base-reduplicant faithfulness constraints. Consider the tableau in (51): (51) Emergence of the unmarked /mH-to´bi/

DepConsonant InputOutput

a. to´-to´bi

Integ

DepVowel InputOutput

**!

+ b. tı´-to´bi

*

c. ho´-to´bi

*!

d. hı´-to´bi

*!

*

* *

The optimal candidate would reduplicate a consonant to avoid a violation of highranking input-output constraints on faithfulness to consonants, but would insert an epenthetic vowel to prevent gratuitous violations of Integrity. Note, however, that (51b) actually stands for a class of candidates where the vowel of the gerundive consists of [i] (tı´-to´bi), [e] (te´-to´bi), [a] (ta´-to´bi), and so on. However they are formulated, constraints on vowel markedness would intervene to select the candidate with the least marked vowel (tı´-to´bi) since all such constraints tie on all other relevant constraints. The nonviability of the emergence of the unmarked analysis of the gerundive lies not in some inability to produce such e¤ects, but in independent properties of reduplication in general in Yoruba and of the gerundive in particular.28 In conclusion, I analyze the fixed [i] of the gerundive as involving the prespecification of an a‰xal, H tone high front vowel, returning us to the traditional view that

Patterns of Reduplication in Yoruba

345

the gerundive involves a tonally and vocalically specified vowel along with a reduplicated consonant (Marantz 1982). This analysis is inconsistent with Alderete et al.’s (1999) specific claim that the Yoruba gerundive involves fixed default segmentism, although it is consistent with their overall proposal that fixed segmentism may result either from the emergence of the unmarked or from a‰xation. Note also that this analysis is incompatible with views of underspecification that would force the lexical absence of unmarked or default values (e.g., Kiparsky 1982d; Pulleyblank 1986), while it is consistent with a proposal such as combinatorial specification (Archangeli and Pulleyblank 1994). 14.5.3

Underapplication

The gerundive exhibits interesting and complex patterns with regard to nasalization. First, the vowel of the gerundive is oral whether the base vowel is oral or nasal. Second, the reduplicated consonant of the gerundive is oral if the base consonant is oral, nasal if the base consonant is a nasal stop, and oral if the base consonant is a nasalized continuant. The first two cases constitute patterns of accurate matching between the reduplicant and the base, while the third type is a nasal-oral mismatch. Third, the gerundive exhibits a pattern of underapplication with respect to a process of liquid nasalization: a rule that nasalizes a lateral generally before high front vowels is blocked from applying before the high front vowel of the gerundive (Clements and Sonaiya 1990). In this section, background regarding the analysis of nasality in ˙ Yoruba is presented, followed by a discussion of the reduplicative data. Background on Nasality Standard Yoruba exhibits seven oral vowels: {i, e, , a, , o, u}. Only three, the high and low vowels, exhibit contrastive nasality: {ı˜, a˜, u˜}.29 Both oral and nasal vowels are common, and minimal pairs are frequent.

14.5.3.1

c

(52) Contrastive nasality a. dı´n [dı˜´ ] ‘fry’ b. ku´n [ku˜´] ‘fill’ c. san [sa˜] ‘pay’

(cf. dı´ ‘close up holes (as of a net)’) (cf. ku´ ‘die’)

The prohibition on nasalized midvowels is widely attested in Kwa (Hyman 1972) and will be analyzed here through a markedness scale (Prince and Smolensky 1993) defining the relation of nasality to vowels of particular heights.30 In addition, we will see that it is important to further divide the constraint on high vowels, incorporating a front/back distinction. (53) Prohibition on nasalized vowels *Mid/Nas  *Lo/Nas  *HiBk/Nas  *HiFr/Nas Midvowels Low vowels High back vowels High front vowels cannot be nasal. cannot be nasal. cannot be nasal. cannot be nasal.

346

Douglas Pulleyblank

To derive the contrastive vowels of Yoruba, MaxNas must be ranked below *Mid/Nas and above the three other constraints:31 (54) Nasal faithfulness and ranking a. MAXNAS Every nasal specification in the input has a correspondent in the output. b. *Mid/Nas  MaxNas  *Lo/Nas  *HiBk/Nas  *HiFr/Nas Nasality is realized on all sonorants (vowels and consonants) in any syllable in which it is found. The data in (55a) show nasality a¤ecting a sonorant onset. (55b) shows both that obstruents are una¤ected by a following nasal vowel (already seen in (52)) and that a syllable preceding a nasal vowel is una¤ected. (55c–e) continue to make the point that nasal harmony is syllable-bounded; even in a string of sonorants, nasality does not a¤ect preceding or following syllables. (55) Syllable-bounded nasal harmony a. rı`n [r˜`˜ı ] ‘to walk’ ‘to dispense’ yu´n [y˜u˜´] ´ ´ b. orogun [orogu˜] *[or˜o˜gu˜] ‘cowife’ a`gu`ta`n [a`gu`ta`˜ ] *[agu˜ta˜] ‘sheep’ c. `ıbı´nu´ [ı`bı´nu´˜ ] *[ibı˜nu˜] ‘anger’ a`wo`ra´n [a`wo`r˜a˜´] *[aw ‘picture’ ˜ o˜r˜a˜] d. o`yı`nbo´ [o`y˜˜`ı bo´] *[oy˜ı˜bo˜] ‘European’ ´ ra´ntı´ [ra˜tı´] *[ra˜tı˜] ‘remember’ e. `ıda´nwo` [ı`da´˜ wo`] *[ida˜w ‘examination’ ˜ o˜] o`mo`we´ [ `m ˜` we´] *[ `m ˜` w ‘educated person’ ˜ e˜] ˙ ˙ This pattern of nasal harmony can be captured by the high ranking of constraints requiring the alignment of nasality with syllable edges: c c

c c

(56) Nasal harmony a. ALIGNRIGHT[NASAL] Align[Nasal, Right; s, Right] b. ALIGNLEFT[NASAL] Align[Nasal, Left; s, Left] These constraints interact crucially with MaxNasal. The ideal syllable in Yoruba must contain segments that agree in nasality, both segments oral or both nasal. Where an onset consonant is not nasalizable (e.g., ku˜´ ‘fill’ (53b)), nasality is retained on the vowel at the expense of an AlignLeft[Nasal] violation. That is, MaxNasal must outrank AlignLeft[Nasal]. In contrast, syllables of the type *ne, *no, and so forth are unattested: when the vowel is not nasalizable, the onset consonant cannot itself be nasal. This gap is accounted for by requiring that AlignRight[Nasal] outrank MaxNasal. Hence an input like /ne/ would be optimally realized as [le]

Patterns of Reduplication in Yoruba

347

(violating MaxNasal) rather than [ne] (violating AlignRight[Nasal]). Hence the following ranking is established: (57) Retention/loss of nasality in harmonically imperfect syllables AlignRight[Nasal]  MaxNas  AlignLeft[Nasal] Lateral Nasalization Given this preliminary background, we can now consider patterns directly involving lateral nasalization. As discussed in Clements and Sonaiya 1990 and elsewhere, the full range of vowels is attested with an obstruent ˙ onset (59a), the full range with the liquid /r/ (59b), but a restricted range with the liquid /l/ (59c). Specifically, there is a gap with /l/: the sequence li is not found (except in restricted contexts to be discussed below).

14.5.3.2

(58) Lateral nasalization a. kı´ ke´ ke´ ka˙` ko ko´˙ ku´ kı´n kan ku´n

‘greet’ ‘shout’ ‘pet’ ‘read’ ‘write’ ‘take’ ‘die’ [kı˜´ ] ‘scrub’ [ka˜] ‘be sour’ [ku˜´] ‘be full’

b. rı´ ‘see’ re´ ‘pare o¤ ’ re ‘soak’ ra˙` ‘buy’ ro` ‘be soft’ ro`˙ ‘think’ ru´ ‘sprout’ rı`n [r˜`˜ı ] ‘walk’ ra`n [r˜a˜`] ‘help’ run [r˜u˜] ‘perish’

c. *li le´ le la˙´ lo lo`˙ lu` lin [nı˜ ] la´n [na˜] lu`n [nu˜]

GAP ‘drive away’ ‘be lazy’ ‘lick’ ‘go’ ‘use’ ‘hit’ ‘worry someone’ ‘spend’ ‘become lost’

Adapting the analysis proposed by Clements and Sonaiya, this gap is accounted ˙ for by constraints whose combined e¤ect is to cause a lateral to be nasal when the syllable containing it can be harmonically nasal. That is, a lateral will nasalize when the following vowel can also nasalize. Crucial to the account is the phonologization of an observation made in Piggott (1992, 48): ‘‘A spontaneously voiced segment contains a nasal phase, if it is also characterized by complete oral occlusion.’’ Since a sonorant requires an open vocal tract, it follows that if there is a closure in the oral cavity ([continuant]) then the nasal cavity must be open.32 (59) NASALPHASE If [þsonorant, continuant] then [þnasal]. The relative ranking of NasalPhase and DepNasal (DepNasal: Every nasal specification in the output has a correspondent in the input) will determine whether a language tolerates oral sonorant stops. Unconstrained by anything else, where NasalPhase outranks DepNasal, [l] is impossible; where DepNasal outranks NasalPhase, [l] is tolerated. For Yoruba, the basic ranking is one where NasalPhase outranks DepNasal, but there are two additional conditions. First, the syllable containing the derived nasal must be nasalized throughout. That is, NasalPhase has an e¤ect only when it is also possible to satisfy AlignRight[Nasal]. Second, the

348

Douglas Pulleyblank

impact of NasalPhase is restricted by the markedness constraints on nasalized vowels, as follows. (60) *MidNas  *LoNas  *HiBkNas  NasalPhase  *HiFrNas Since all but *HiFrNas are ranked above NasalPhase, the introduction of nasality will only be possible when a lateral is followed by a high front vowel. For an input like /li/, the optimal output nasalizes the consonant and extends the nasality throughout the syllable. (61) Appearance of nasality where not in input /li/

AlignR [Nas]

Max Nas

Nasal Phase

a. [li]

*!

b. [lı˜ ]

*!

c. [ni]

*!

Dep Nas

*HiFr Nas

*

*

*

+ d. [nı˜ ]

*

*

AlignR[Nasal] is violated if the syllable contains a nasal-oral sequence (61c) and NasalPhase is violated if the syllable contains a lateral (61a,b). The only relevant markedness constraint on nasal vowels is *HiFr/Nas, which is ranked too low to have any e¤ect on the outcome. Note that NasalPhase will have no e¤ect on a syllable containing a consonant other than /l/ since only /l/ is both a sonorant and a stop. For example, NasalPhase does not exert pressure on syllables containing /k/ (58a) or /r/ to nasalize (58b). When /l/ precedes any other vowel (high back, mid, or low), the appropriate markedness constraint will rule out nasalization. To exemplify this, consider a case involving a midvowel. As discussed above, mid–nasalized vowels are impossible, prohibited by ranking *Mid/Nas above MaxNas (see (54)). (62) Maintenance of orality due to a midvowel /le/

*Mid Nas

AlignR [Nas]

+ a. [le] b. [le˜]

Nasal Phase

Dep Nas

* *!

c. [ne] d. [ne˜]

Max Nas

* *!

*!

* * *

Patterns of Reduplication in Yoruba

349

Nasalizing the midvowel is ungrammatical (62b,d); nasalizing the sonorant consonant without harmony is ungrammatical (62c). The optimal output is therefore fully oral: [le] (62a). Nasalized high back vowels and nasalized low vowels are possible in Yoruba, but they cannot be derived by the pressure of NasalPhase. MaxNas exerts no pressure to retain nasality when there is no lexically specified nasal value, and the constraints against nasality on a vowel (*HiBk/Nas & *Lo/Nas) prevent the introduction of nasality since they outrank NasalPhase. Note that [nu˜] and [na˜] are both fully grammatical surface forms, but only for inputs where nasality is specified—that is, /lu˜/ and /la˜/. 14.5.3.3 Orality of the Reduplicant Vowel As expected given the proposed analysis of the gerundive, the prefix vowel is oral, even if the base vowel is nasal.

(63) Orality of reduplicant a. sı´n [sı˜´ ] sı´sı´n b. gba´n [gba˜´] gbı´gba´n c. du`n [du`˜ ] dı´du`n

[sı´sı´˜ ] [gbı´gba˜´] [dı´du`˜ ]

‘sneeze’ ‘scoop out’ ‘be sweet’

This follows from the analysis developed here, in which the vowel of the gerundive prefix is underlying, rather than copied from the base. The optimal candidate simply maintains the gerundive vowel. As we will see in the next section, however, the situation with the consonant of the gerundive is more variable. Variable Nasality of the Reduplicant Consonant A stem-initial oral consonant reduplicates faithfully, but the situation is di¤erent for stem-initial nasals. Nasal stops reduplicate as nasal (64a), but, as noted in Clements and Sonaiya 1990, nasal ˙ continuants reduplicate as oral (64b).33

14.5.3.4

(64) Variable nasality of the consonant of the reduplicant a. na` [na˜`] nı´na` [nı˜´na˜`] ‘flog’ mı´mo` [mı˜´m ˜` ] ‘know, understand’ mo` [m ˜` ] ˙ ˙ ´ ´ b. wı´n [w wı´wı´n [wı´w ‘borrow’ ˜ ˜ı ] ˜ ˜ı ] yı´ya´n [yı´y˜ ´˜ ] ‘yawn’ ya´n [y˜ ´˜ ] ra´ntı´ [r˜a´˜ tı´] rı´ra´ntı´ [rı´r˜a˜´tı´] ‘remember’ c

c

c

c

Although both nasal stops and nasalized continuants are derived in Yoruba, the two types of segments behave in di¤erent ways. Clements and Sonaiya approach ˙ this problem by assigning the rule that derives [n, m] to a lexical stratum, while the rule that derives nasalized continuants is assigned to the postlexical stratum. We may treat the di¤erence between the two types of nasal consonants as emergence of the unmarked—assuming, quite plausibly, that nasal stops are less marked

350

Douglas Pulleyblank

than nasalized sonorant continuants. Within a theory postulating reduplicationspecific faithfulness constraints, the pattern can be derived by ranking a constraint prohibiting nasalized continuants below input-output faithfulness and above basereduplicant faithfulness. In the approach taken here, without reduplication-specific faithfulness, the same e¤ect is a by-product of the fact that NasalPhase constitutes pressure for sonorant stops to be nasal while it does not constitute pressure for a similar nasalization of continuants: (65) Nonretention of nasality on sonorant continuant consonant

*!

+ c. [yı´-y˜a˜´]

* *

d. [yı˜´-y˜a˜´]

f. [y˜˜´ı-y˜a´˜ ]

*[HiFr Nas

*

b. [yı´-ya´˜ ]

e. [y˜´ı-y˜a´˜ ]

Integ

*!

DepIO Seg, Nas

a. [ı´-ya´˜ ]

Nasal Phase

AlignL[Nas]

Max Nas

AlignR[Nas]

´ *[ V /ı´-ya´˜ /

*! *!

*

*

* *

*!

Candidates involving tautosyllabic nasal-oral or oral-nasal sequences are excluded by AlignR[Nasal] (65e) or AlignL[Nasal] (65b,d). The ultimate factor therefore is whether some constraint forces a violation of the constraint prohibiting nasalized high front vowels. With a sonorant continuant, neither NasalPhase nor any other constraint applies such pressure. The result is consequently that the *HiFr/Nas constraint rules out a candidate with the marked property of nasality in the reduplicant (65f ). The less marked candidate with an oral reduplicant wins (65c).34 In marked contrast with such cases, forms with a sonorant stop induce nasality in the reduplicant. Consider the case of nı´na` ‘flog’, where the relevant assessment will be identical whether the input verb is considered to contain /n/ (na˜`) or /l/ (la`˜ ). (For concreteness, the input with a lateral is assumed.)

Patterns of Reduplication in Yoruba

351

(66) Retention of nasal consonant

*

*

*

c. [lı´-la`]

*! *!

*

+ e. [nı˜´-na`˜ ] f. [lı˜´-na`˜ ] g. [lı´-na`˜ ]

*[HiFr Nas

*!

*!

b. [lı´-la˜`]

d. [nı´-na`˜ ]

Integ

Nasal Phase

a. [ı´-na˜`]

DepIO Seg, Nas

AlignL[Nas]

Max Nas

AlignR[Nas]

´ *[ V

*!

/ı´-la`˜ /

*!

*

*

*!

**

*

*!

*

Retention of nasality on the low vowel is ensured by the high ranking of MaxNas (66c); AlignL[Nasal] and NasalPhase ensure that the nasality so retained occurs on both the vowel and the consonant of the base (66b). That is, the base must be [na`˜ ], as it would be in isolation. This leaves four prefix possibilities: nasal-oral, nasal-nasal, oral-nasal, and oral-oral—the last four candidates of (66). Of the four, *[nı´-na`˜ ] (66d) is ruled out by the high-ranking constraint, AlignR[Nasal], which prohibits an oral-nasal sequence within a syllable. Both [lı´˜-na`˜ ] (66f ) and [lı´-na˜`] (66g) are nonoptimal because they violate NasalPhase; [lı˜´-na˜`] also violates AlignL[Nasal]. The optimal candidate is therefore the fully nasal nı˜´na˜` (66e). Underapplication of Lateral Nasalization Of particular interest are the gerundives of lateral-initial bases where the vowel is oral but not [i].

14.5.3.5

(67) Underapplication a. le´ lı´le´ b. le lı´le ˙ ˙ c. la´ lı´la´ d. lo lı´lo ˙ ˙ e. lo` lı´lo` f. lu` lı´lu`

of lateral nasalization ‘drive away’ ‘be lazy’ ‘lick’ ‘go’ ‘use’ ‘hit’

Since the base contains a vowel resistant to nasality, we correctly expect that such forms should be immune to NasalPhase. Perhaps surprisingly, however, we observe

352

Douglas Pulleyblank

that the gerundive forms are similarly immune, in spite of the fact that the reduplicant contains a sequence of a lateral followed by a high front vowel. Lateral assimilation underapplies in such cases. Within an approach positing reduplication-specific constraints, such forms could be derived by requiring that base-reduplicant correspondence outrank NasalPhase: MaxBR  NasalPhase. However, BR correspondence is not necessary to the analysis of such cases. Consider the analysis of such forms according to the proposals outlined here. A tableau for lı´lo` is used for illustration. (68) Underapplication of lateral nasalization

*!

*

+ b. [lı´-lo`] c. [lı´-no`]

* *!

d. [lı´-no˜`]

*!

e. [nı˜´-lo`] f. [nı˜´-no`] g. [nı˜´-no`˜ ]

Integ

a. [ı´-lo`]

DepIO Seg, Nas

Nasal Phase

AlignL[Nas]

Max Nas

*Mid Nas

AlignR[Nas]

´ *[ V /ı´-lo`/

*! *!

*

*

*

*

*

*

*

*

*!

*

*

*

*

*

*MidNas rules out forms involving a nasalized midvowel (68g), (68d). As always, a base is ruled out if it contains a nasal-oral sequence, eliminated by AlignR[Nasal] (68c,f ). This leaves two plausible candidates: [lı´-lo`] (68b) and *[nı˜´-lo`] (68e). Crucially, NasalPhase is violated in both and therefore serves no selective function. With the e¤ect of NasalPhase neutralized, the optimal form becomes the one best respecting faithfulness: lı´lo`, the form that maintains the orality of the input. One might ask, however, whether the optimal form lı´lo` should be assessed two violations of NasalPhase. The assumption made here is that feature specifications in correspondence with each other are assessed as a unit for the determination of constraint violations. For a correspondence set to satisfy a constraint, all members of the set must satisfy the constraint. The consonants in correspondence in [ni ´˜ı-ni o`] satisfy NasalPhase because both instances of ni satisfy NasalPhase; the correspondence

Patterns of Reduplication in Yoruba

353

pairs in candidates like [li ´ı-li o`], [li´ı-ni o`], and [ni ´˜ı-li o`] all violate NasalPhase because it is not the case that the two segments in correspondence both satisfy NasalPhase. In conclusion, the gerundive illustrates an important general point about the proposed treatment of reduplication. The analysis of nasality in reduplication follows straightforwardly from the treatment of nasality motivated generally. That is, the complexities of nasality under reduplication fall out from general properties of the grammar. In particular, the underapplication of nasalization in the gerundive provide no compelling evidence for reduplication-specific correspondence constraints. Input-output correspondence constraints are adequate for the evaluation of both nonreduplicative and reduplicative forms. 14.6

Conclusion

This chapter has examined a range of reduplication processes in Yoruba. As noted in the introduction, the four cases considered constitute only a subset of the large class of reduplicative patterns observed in Yoruba. Nevertheless, they serve to exemplify a range of behavior from total copy to fairly minimal copying in conjunction with lexical prespecification. Two points of particular importance emerge from the analysis of these forms. At a theoretical level, it has been demonstrated that complex reduplicative patterns taken to require ‘‘base-reduplicant’’ and ‘‘input-reduplicant’’ correspondence can be accounted for without the postulation of constraints that formally refer to the abstract category ‘‘reduplicant.’’ Returning to the sort of prosodic characterization of reduplicants proposed in Marantz 1982, it has been shown that partial reduplication can be accounted for without reduplication-specific constraints, but also without the negative theoretical consequences of base-reduplicant approaches to templatic reduplication. In particular, the problem of ‘‘base erosion’’ does not arise when there are no constraints forcing faithfulness of base and reduplicant. The proposed theory postulates the possibility of two fundamentally di¤erent types of reduplication, echoing an earlier observation by Kiparsky (1986b). ‘‘Compounding’’ reduplication involves the imposition of morphological identity (Hyman et al., this volume; Inkelas and Zoll 2000), while ‘‘a‰xing’’ reduplication involves minimal violations of Integrity. At an analytic level, the importance of considering reduplication within the broad context of a language’s phonology has been argued to be crucial for an adequate analysis. The Yoruba gerundive, for example, can be analyzed in the abstract as involving either a prespecified /i/ or as involving the emergence of an unmarked default [i]. Consideration of general properties of the phonology of Yoruba in conjunction with consideration of additional reduplicative types demonstrates the superiority of the former hypothesis over the latter.

354

Douglas Pulleyblank

Notes Thanks to Oladiipo Ajiboye, Akinbiyi Akinlabi, Rose-Marie De´chaine, Laura Downing, ˙ Masaru Kiyota, Uyi Stewart, and Suzanne Urbanczyk for discussion of variEun-Sook Kim, ous aspects of this chapter, as well as to an anonymous reviewer. Special thanks to Sharon Inkelas and Kristin Hanson for their extensive comments, both substantive and editorial, on an earlier version of the chapter. 1. Yoruba has numerous additional reduplicative constructions not discussed here—for example, adverbial reduplication (da´rada´ra ‘very well’, dı´e`dı´e` ‘slightly’, etc.); ‘‘all X’’ reduplication ˙ (me´je`e`jı` ‘both’, me´te`e`ta ‘all three’, etc.); ‘‘groups of˙ so-and-so many’’ reduplication (me´jı`me´jı` ˙ ˙ ˙ ‘two by two’, me´tame´ta ‘three by three’, etc.); inherent reduplication (du´du´ ‘black’, gbogbo ˙ ˙ form of banking’, agogo ‘bell’, etc.); ideophonic reduplication (mo`no`‘every’, e`su´su´ ‘traditional ˙` ‘of ˙ mo`no` ‘of repeated flashing’, ro`go`do`ro`go`do` ‘of several items big and round’, fe`re`ge`de`fe`re`ge`de ˙ ˙ ˙ ˙ ˙ ˙ ˙ ˙ ˙ ˙ several things being wide’, ke´e´ke`e`ke´e´ ‘in small bits’, ko´gbe´ko`gbe`ko´gbe´ ‘of clumsy, slow move˙ ˙ ˙ 1958; ˙ ˙ Bamgbos ˙ ment’, la´n´ko´nla`n`ko`nla´n´ko´n ‘of limping’, etc.). See Abraham e 1967; Ogunbo˙ ˙ ˙ ˙ ˙ wale 1970; Awoyale 1974; Awobuluyi 1978; Akinlabi 1984; and so on. 2. Transcriptions are in standard Yoruba orthography throughout, except where specially indicated. In Yoruba orthography e ¼ [], o ¼ [ ], Vn ¼ nasalized vowel, s ¼ [S], p ¼ [kp], ˙  ¼ H,  ¼ L,  or unmarked for tone˙ ¼ M, a˙ tone-marked nasal ¼ syllabic nasal. Certain entirely predictable tonal processes are not indicated or addressed. For example, a lexical H tone systematically appears as rising after an L tone, and an L tone appears as falling after an H tone. c

3. See Rosenthall 1997 and Orie and Pulleyblank 2002 for further discussion of vowel deletion, a detailed account of which is orthogonal to the concerns of this chapter. 4. There are foot-based restrictions on NoHiatus that are peripheral to the concerns of this chapter. See Pulleyblank in press for discussion. 5. Throughout this chapter, I assume unless otherwise indicated that when two segments are in correspondence they are identical. Patterns of coalescence and so on play no role in the cases considered. The assumed identity could be the result of undominated Ident or by featuresensitive Max and Dep (McCarthy and Prince 1995). 6. For at least some dialects/idiolects, there is evidence for a shape delimiter on agentive reduplication. Contrast the data presented in this section—all of which involve a disyllabic base— with longer examples (Pulleyblank and Akinlabi 1988): so`dodoso`dodo ‘truthful person’ ˙ evil’), ˙ ya´nila´soya´nila´so ‘cloth (< so`dodo ‘be truthful’), se`ba`je´se`ba`je´ ‘evil-doer’ (< se`ba`je´ ‘do ˙ ˙˙ ˙cloth’), fe´nilo´mo˙fe´nilo˙´ mo ‘someone who ˙ ˙ takes ˙ ˙people’s lender’ (< ya´nila´so ‘lend ˙ someone ˙ ˙ ˙ ˙ ˙ ˙ ˙ ˙ daughters and marries them’ (< fe´nilo´mo ‘marry someone’s child’). For some speakers, such ˙ reduplicated forms are acceptable;˙ for˙ others, they are not. Speakers who do not accept such long reduplicated forms derive the relevant agentives by simple prefixation: aso`dodo, ase`ba`je´, ˙ ˙ re˙ aya´nila´so, afe´nilo´mo. For such speakers, agentive reduplication exhibits a phonological ˙ ˙ ˙ ˙ ˙ quirement that each half of the reduplicated form must constitute a minimal word—that is, a disyllabic foot (Ola 1995). To block the reduplication of forms that are more than two syllables long, rather ˙than truncate them, the minimal word requirement for agentive reduplication and input-output faithfulness would have to outrank the requirement of morphological parsing (Prince and Smolensky 1993).

Patterns of Reduplication in Yoruba

355

7. For concreteness, I assume that Red is a prefix, although a su‰xal analysis would also be possible. Possible motivation for a prefixal analysis would be that Yoruba morphology is uniformly prefixing; see below. 8. While there appears to be variability as to whether a minimal word requirement is observed as a contingent condition for agentive reduplication, there seems to be no such variability with kı´-reduplication. 9. Principal tonal alternations are discussed in the next section. One alternation not discussed is the realization of an input HLL sequence as ML in forms like `ılo`kulo` ‘bad use’ ( cik h i‘to keep’ moti- > moci‘to be cruel’ tyoh- > cyoh-, coh‘to be good’ ty psi > cy psi, c psi ‘plate’ b. Across a morpheme boundary k t-hi > k c h i ‘like, as’ ‘friend-nom’ p t-i > p ci e

e

e v

e

e v

To model this sound changes formally, let us first characterize the grammar of Middle Korean of the fifteenth and sixteenth centuries, when Palatalization was dormant. The anterior value in the underlying representation of TI sequences surfaced intact in the output, as guaranteed by a ranking in which the Pal constraint ranks below Ident-[ant]:

472

Young-mee Yu Cho

(19) Stage 1: Fifteenth–sixteenth centuries (no Palatalization) /ti/

Ident-[ant]

+ a. ti b. ci

Pal *

*!

/tsi/ a. ci

*!

+ b. tsi

*

/tþi/ + a. ti b. ci

* *!

The second stage came about sometime in the seventeenth century, when Palatalization began to apply optionally. The strict ranking between Ident-[ant] and Pal did not hold any longer. With the undoing of this ranking, an input TI sequence could surface either as TI or as CI, regardless of whether the environment was morphologically derived. We may assume that by this time /ts/ had changed to /c/. (20) Stage 2: Seventeenth century (optional Palatalization) /ti/

Ident-[ant]

+ a. ti + b. ci

Pal *

*

/ci/ a. ti

*!

*

+ b. ci /tþi/ + a. ti + b. ci

* *

The literature notes that Palatalization was an optional rule for some time (K.-M. Lee 1980; Choi 1992), supporting an analysis of language change in which

A Historical Perspective on Nonderived Environment Blocking

473

ranked constraints become unranked (Anttila and Cho 1998). Importantly, both morpheme-internal and morpheme-final /t/’s were subject to optional Palatalization. The next stage is characterized by obligatory Palatalization. At this stage Pal is ranked above Faith (represented by Ident-[ant], as shown in (21)): (21) Stage 3: Early nineteenth century (obligatory Palatalization) /ti/ a. ti

Pal

Ident-[ant]

*!

+ b. ci

*

/ci/ a. ti

*!

*

+ b. ci /tþi/ a. ti

*!

+ b. ci

*

There were apparently some di¤erences in the way Palatalization targeted di¤erent strata of the lexicon. As illustrated in (22), obligatory Palatalization was established much faster in the Native Korean stratum than in the Sino-Korean stratum, where variation lasted for some time longer; this scenario supports Itoˆ and Mester’s (1995b, 1999) approach to lexical stratification solely in terms of constraint ranking. Palatalization ultimately spread from the core of the lexicon to all lexical strata (Cho 2001): (22) Palatalization in di¤erent strata (K.-M. Lee 1980; Choi 1992) (eighteenth century) a. Native Korean vocabulary ti-ta ! ci-ta ‘to lose’ ‘collapse’ munh ti-ta ! munh ci-ta tut-ti ! tut-ci ‘to hear’ b. Sino-Korean vocabulary tyosy n @ cos n ‘Chosun’ koktyo @ kokco ‘melody’ tyoty k @ toc k ‘thief ’ e

e

e

e

e

e

When Palatalization was completed, morpheme-final etymological T was realized either as T or C, depending on whether the following su‰x started with /i/.

474

Young-mee Yu Cho

Morpheme-internally, however, Palatalization had the e¤ect of absolute neutralization. Just as children learning Modern English have no reason to posit [x] in nightingale or a [kn] cluster in knife, post-Palatalization Korean learners had no reason to posit an underlying di¤erence between morpheme-internal TI and CI sequences, since both would always surface as CI. (23) represents Stage 4, at which inputs simplified due to the completion of the Palatalization sound change. (Note that certain OT assumptions about the Richness of the Base (Smolensky 1996) would render this stage vacuous.) (23) Stage 4: Early nineteenth century (reanalysis of /TI/ as /CI/) /ci/

Pal

Ident-[ant]

*!

*

a. ti + b. ci /tþi/ a. ti

*!

+ b. ci

*

With Palatalization firmly established by the end of the eighteenth century, the nineteenth century witnessed another sound change that interacted with Palatalization by providing it with a potential new target. In the nineteenth century the diphthong / ‘i/ began to lose its diphthongal quality and to surface as [i], as shown in (24). (24) Nineteenth century: Monophthongization of /i ‘i/ to [i] mun ‘i > muni ‘pattern’ pon ‘i > poni ‘inside skin of a nut’ > ky nti‘to endure’ ky nt ‘imut ‘i> muti‘to be dull’ t’ ‘i > t’i ‘belt’ t ‘i > ti ‘where’ n t h ‘i-namu > n t h i-namu ‘zelkova tree’ e

e

e

e

Stage 5 is modeled in (25) as a constraint reranking, assuming a constraint *Diph( ‘i) to ban the relevant diphthongal sequence. After some period of fluctuation (again obtained by the unranking of the two constraints), Stage 5 is captured as the ranking of *Diph( ‘i) above the faithfulness constraint preserving input / /. (25) Stage 5: Nineteenth century (monophthongization of /i ‘i/ to [i]) a. *DIPH(i ‘i) The tautosyllabic sequence / ‘i/ is prohibited.

A Historical Perspective on Nonderived Environment Blocking

475

b. MAX(i ‘) Do not delete the input / ‘/. c. Ranking *Diph( ‘i)  Max( ‘) The tableau in (26) shows how the new ranking simplifies input diphthongs: (26)

/p ‘i/ a. p ‘i

*Diph( ‘i)

Max( ‘)

*!

+ b. pi

*

Diphthongs preceded by a dental onset present a more challenging case. Given the ranking established for Palatalization in (23), the predicted interaction between Monophthongization and Palatalization is one of feeding: the surface [i]s created by Monophthongization should trigger palatalization of a preceding dental. This predicted interaction is illustrated in (27), where [ci] emerges as the winning output for input /ti/. Given that *Diph( ‘i) must outrank Max( ‘), and given that Pal must outrank Ident-[ant], there is no way for the candidate [ti] to win. (27) *Diph(i ‘i)  Max(i ‘), Pal  Ident-[ant] /t ‘i/

Max( ‘)

Pal

b. ti

*

*!

+ c. ci

*

a. t ‘i

*Diph( ‘i)

Ident-[ant]

*!

*

The prediction that [ci] is the correct outcome for input /t ‘i/ does not, however, correspond to what actually happened in Korean; as shown in (28), the result of Monophthongization following T was a surface TI sequence, opaque from the perspective of Palatalization. The interesting result is that Palatalization applies in morphologically derived environments but not in phonologically derived environments. In the derivational framework, this represents counterfeeding: the structural context of one rule (Palatalization) is potentially satisfied due to the application of a prior rule (Monophthongization), but cannot apply due to rule ordering. (28) Counterfeeding relationship between Monophthongization and Palatalization /t ‘i/ — Palatalization ti Monophthongization [ti] (opaque surface form)

476

Young-mee Yu Cho

Numerous approaches to opacity have been proposed; here we resort to an analysis utilizing sympathy (McCarthy 1999). In sympathy theory, a selector (‘‘¶’’) constraint identifies one ‘‘sympathy’’ candidate to which all others are required, by a sympathy constraint, to be faithful in some respect. The selector constraint in this case is ¶Max-V, which is satisfied by output candidates that preserve input diphthongs. In (29), the input is /t ‘i/. The sympathy (‘‘{’’) candidate is the most optimal (according to the rest of the constraint ranking) of those candidates preserving the input diphthong, namely [t ‘i] and [c ‘i]. Of these, [t ‘i] best satisfies faithfulness; they tie on Pal, since neither contains a Palatalization environment. Thus [t ‘i] is the sympathy candidate. The sympathy constraint, {Ident-([ant]), requires output candidates to be identical in consonant anteriority to the sympathy candidate ([t ‘i]). The candidates satisfying the sympathy constraint are [ti] and [t ‘i]; of the two, [ti] is optimal since it best satisfies *Diph( ‘i), and [ti] is therefore judged the winning candidate. Because Pal and Ident-[ant] are both ranked below the sympathy constraint, neither plays a prominent role in selecting the winner in this tableau. (29) Stage 5: Opaque interaction of Palatalization and Monophthongization /t ‘i/ { a. t ‘i

*Diph( ‘i)

¶Max( ‘)

{Ident-[ant]

Pal

Ident-[ant]

*!

+ b. ti

*

c. ci

*

* *!

*

In Contemporary Korean there is no longer any trace of / ‘i/ in most syllables originally containing it; Monophthongization has occurred in all environments except in word-initial onsetless syllables, as illustrated for Standard Korean in (30). (30) Tautosyllabic etymological /i ‘i/ across dialects Standard Southwestern Southeastern sa ‘i-sa isa hok ‘i-hok ihok ‘i-mi imi mi

‘doctor’ ‘doubt’ ‘meaning’

e e e

Some dialects, mainly the Southern ones, have simplified even these forms to yield monophthongal forms. Assuming that the first half of the [ ‘i] diphthong constitutes an onset to the second half, these dialectal di¤erences can be modeled by two di¤erent rankings: Onset  *Diph( ‘i) versus *Diph( ‘i)  Onset. When Monophthongization had run its course, there was no longer any reason to assume an underlying / ‘i/ diphthong anywhere except, of course, in initial onsetless syllables. The residue of historical Monophthongization is thus the source of one of the three types of synchronic exceptions to Palatalization. The first type, which we

A Historical Perspective on Nonderived Environment Blocking

477

have just seen, are the Native Korean words that etymologically contained the sequences /t ‘i/ or /t h ‘i/, as shown in (31). (31) Morpheme-initial /ti, t h i/ from ti < t ‘i titi-ta < t ‘it ‘i-ta < t h ‘i t hi h n t i-namu < n t h ‘i-namu mati < mat ‘i canti < cant ‘i

/ti ‘i, t h i ‘i/ ‘where’ ‘to tread’ ‘blemish’ ‘zelkova tree’ ‘joint’ ‘grass’

e

e

All the tautomorphemic nonpalatalizing TI sequences mentioned thus far in the chapter belong to this class of words. While for the forms in (31) there is no concrete synchronic evidence of an underlying diphthong, there is another set of data for which the sympathy analysis is still plausible synchronically. When the two vowels, / / and /i/, are concatenated across a morpheme boundary, it is possible to pronounce these as two syllables (i.i) or coalesce two vowels into one syllable (i); in this one context, there is still synchronic evidence for Monophthongization as an active process. Synchronic Monophthongization gives rise to the second type of counterexample to Palatalization. As shown in (32), a /t -i/ sequence that Monophthongizes to /t-i/ does not Palatalize: (32) Synchronic variation /t’ -i-ta/ t’ ita @ t’ita /t h -i-ta/ t h ita @ t h ita

‘to catch the eye’ ‘to be open’

The sympathy analysis of the diachronic counterfeeding relationship between Monophthongization and Palatalization in (29) (or its equivalent) is also required in a synchronic grammar of Korean to account for the failure of Palatalization in the monophthongal variants of the words in (32). The third set of exceptions to Palatalization consists of foreign loans introduced in the twentieth century; some examples are presented in (33). Anterior consonants in the source language (here, English) are systematically realized as dentals in Korean, regardless of context, resulting in unpalatalized sequences such as [ti, t h i]: (33) Lack of Palatalization in foreign vocabulary ticit h al ‘digital’ t hi ‘tea’ t h ina ‘Tina’ mot h ip ‘motive’ pati ‘body’ t h ipi ‘TV’

478

Young-mee Yu Cho

The two sets of surface exceptions to Palatalization for which there is no concrete evidence of an underlying diphthong are responsible for the NDEB e¤ect for which Palatalization is renowned, illustrated again in (34). (34) Stage 6: Contemporary Korean /maci/ a. mati

Faith-ROOT

Pal

Faith

*!

*

*

+ b. maci /mati/ + a. mati b. maci

* *!

*

/mat-i/ a. mati + b. maci

*! *

In lieu of the tautomorphemic example mati ‘joint’ in (32), we can easily substitute an example from the loan vocabulary such as t h i (‘tea’) with the same result, with no special ranking provision for recent loans. In both cases, a root containing an internal TI sequence is subject to Faith-ROOT in the Cyclic Stratum. We have seen clear historical reasons for the morpheme-internal nature of synchronic exceptions to Palatalization in Korean: morpheme-internal /ti, t h i/ sequences are immune to Palatalization (because they were not eligible to undergo the acrossthe-board Palatalization sound change), while morphologically derived environments regularly undergo it.8 Although careful investigations are needed, other NDEB cases seem to be analogous to Korean Palatalization in their provenance. Chamorro Vowel Lowering, another derived-environment rule, is reported to apply to ‘‘virtually all native Chamorro words’’ except in a handful of native words and a large class of loans (Chung 1983). This is the exact configuration in the Korean lexicon. The Korean pattern reflects the historical facts that all dentals at one point in history changed to palatals before [i] and that those synchronically available TI sequences are products of another independent process or recent borrowing. The idea that words showing NDEB are ‘‘marked’’ or ‘‘marginal’’ is also supported by other languages. Exceptions to Finnish Vowel Coalescence (Anttila, chap. 18, this volume) and Polish First Velar Palatalization (Lubowicz 1998) consist exclusively of recently borrowed words. These languages have a more restricted set of

A Historical Perspective on Nonderived Environment Blocking

479

exceptions than do Korean and Chamorro, in that the original sound change, undergone by all native forms, was never obscured by a later process in the way that Korean Monophthongization rendered Palatalization opaque. Kiparsky (1982, 77, b) makes a similar observation that NDEB exceptions are restricted about Sanskrit, noting, of the retroflexing ‘‘Ruki’’ rule in Sanskrit, that ‘‘/»/ is the highly normal case in the Ruki environment even in nonderived environments. Words like /bisa/ ‘lotus’ and /kusuma/ ‘flower’ are marginal in the Sanskrit lexicon. In contrast, words such as /vi»a/ ‘position’ and /kar»/ ‘drag’ are quite common.’’ Korean Palatalization confirms the hypothesis that NDEB is not an inalienable property of a particular phonological alternation per se but is derivable from constraint interaction. Korean Palatalization was at one point an across-the-board process that admitted no exceptions. Even when the subsequent introduction of exceptions turned Palatalization into a ‘‘derived-environment’’ rule, CI sequences still remained unmarked relative to TI sequences, just as [i»] in Sanskrit was unmarked, relative to [is], even in nonderived environments. NDEB e¤ects are not intrinsically necessary; they result from the introduction of an external source of opaque exceptional forms, either through additional language change or through borrowing.9 19.7

Postlexical Palatalization

In addition to the lexical Palatalization of dental stops discussed up to this point, Korean also has an automatic postlexical palatalization process. Before a high front vowel, all coronals are palatalized phonetically, resulting in prepalatalized segments (35a). As shown in (35a) versus (35b), the postlexical Palatalization process is, predictably, insensitive to morphological boundaries. The palatalization of /s, n, l/ (35c) is also generally attributed to the postlexical manifestation of the lexical Palatalization rule: (35) Postlexical Palatalization t¸i a. ti t hi t¸ h i t’i t¸’i b. os-i osˇi nun-i nunˇi kot¸il na kot il na c. si s˘i su¨psˇu¨ps’inun sˇinun k’ini k’inˇi p’alli pa··i ariarˇi-

‘where’ ‘blemish’ ‘belt’ ‘clothes (nom)’ ‘eye (nom)’ ‘soon get up’ ‘poem’ ‘to be easy’ ‘germ’ ‘meal’ ‘fast’ ‘to hurt’

e

e

e

e

480

Young-mee Yu Cho

Iverson and Wheeler (1988) argue explicitly that lexical and postlexical palatalization result from one and the same rule. When the result of applying the rule lexically would be neutralizing, the RAC limits the rule’s application to derived environments. By contrast, when the result of applying Palatalization is a purely allophonic alternation, the RAC plays no role, thus accounting for the across-the-board application of Palatalization in the postlexical stratum. There is, however, a fundamental di¤erence between lexical and postlexical Palatalization that imperils a single-rule account. When dentals undergo Palatalization at the Cyclic stratum, they become alveopalatal a¤ricates ([c], [c h ]); when they undergo postlexical Palatalization, they become prepalatal, as illustrated in (35a). The di¤erence between the two Palatalization processes can be accounted for by ranking faithfulness di¤erently in the Cyclic and postlexical strata. In the Cyclic stratum, as we have seen, Pal outranks both Ident-[ant] and Ident-[del-rel], giving rise to a dental ! alveopalatal alternation. In the postlexical stratum, however, dentals become prepalatal. Their value for [anterior] changes, but their value for [delayed release] does not. We can attribute this to a reranking of the markedness constraint Affricate and the faithfulness constraint Ident-[del-rel]. The rankings at the two strata are illustrated in (36): (36) a. Cyclic stratum: Dentals become alveopalatal a¤ricates before high front vocoids 

Pal  Ident-[ant]  Ident-[del-rel] Affricate b. Postlexical stratum: Dentals become prepalatal before high front vocoids Pal  Ident-[ant], Affricate Ident-[del-rel]  Affricate An illustration of allophonic palatalization at the postlexical stratum is provided below: (37)

/ti/ a. ti

Pal

Ident-[ant]

Ident-[del-rel]

Affricate

*!

+ b. t j i

*

c. ci

*

* *!

All sequences in Korean obey Pal at the postlexical stratum, where Faith-LEX is ranked too low to interfere. This analysis relates the two palatalization processes, attributing both to Pal, but captures their di¤erences through di¤erent rankings of the constraints a¤ecting the optimal means of satisfying Pal.

A Historical Perspective on Nonderived Environment Blocking

481

One more fact remains to be discussed before we leave the section on Korean Palatalization. Oh (1995) claims that Palatalization depends on morphological information only, and that application of a phonological rule does not create a derived environment, as evidenced by the fact that Glide Formation in /te- / ! [ty ] ‘to burn’ does not trigger Palatalization of [t] even though the phonological environment is met. e

e

(38) Lack of Palatalization in a phonologically derived environment /te- / ! ty (Vowel Raising) ! *cy (Lack of Palatalization) e

e

e

This argument presupposes that the /e/ of /te/ is being directly converted to /y/, potentially feeding Palatalization. However, in fact Glide Formation (GF) only applies to /i/; the relationship between /e/ and /y/ in (38) is mediated by lexical variation between /e/ and /i/ in this particular morpheme. As shown by the data below, GF is possible only for morphemes that show this variation; for morphemes that do not show the variation, GF is not possible. (39) Variations in Glide Formation and Vowel Raising a. Lexical variation in vowel height: GF possible /te-/ @ /ti-/ ! /ty/ ‘to burn’ /pe-/ @ /pi-/ ! /py/ ‘to cut’ b. Lack of lexical variation in vowel height: No GF /he- / *hi *hy ‘to count’ /k nne- / *k nni *k nny ‘to hand over’ /e- / *i *y ‘to scrap out’ e

e

e e

e

e e e

e

e

e e

GF merely reflects di¤erent syllabic organization between [i] and [y]; its application should not be expected to a¤ect the applicability of Palatalization. The reason that Palatalization does not apply to the morphemes in (39a) is that the relevant TI sequences are morpheme-internal, protected by Faith-LEX. 19.8

Sanskrit RUKI

In this section, I discuss a slightly more complex case of NDEB that involves two kinds of derived environments—one morphologically derived and the other phonologically derived. Because of evidence that these two notions of derivedness do not always pattern together, some formulations NDEB have attempted to state the principle behind NDEB e¤ects in purely morphological terms. Most notably, Hammond (1992) proposes to derive NDEB e¤ects from a constraint on the acquisition of structurechanging morphologically conditioned rules. Hammond assumes that phonological rules cannot create a derived environment.

482

Young-mee Yu Cho

In the present framework, however, there is no reason to arbitrarily limit the scope of NDEB by distinguishing phonologically derived environments from morphologically derived environments since both kinds can be derived from the same ranking schema. Although Korean Palatalization only presents positive evidence of being conditioned by morphological derivedness, the basic ranking schema used for Korean NDEB e¤ects extends in principle to both phonologically and morphologically derived environments. Faith-LEX takes care of morphologically derived environments by giving special status to a lexeme. Phonologically derived environments are handled by introducing another markedness constraint (M1 , in (40b)), which, if ranked high enough, potentially interacts with the process in question (40b). (40) a. Morphologically derived Faith-LEX  M  Faith b. Phonologically and morphologically derived M1  Faith-LEX  M2  Faith The scenario in (40b) is clearly attested in Sanskrit. According to the well-known Ruki rule, /s/ retroflexes following any of the four segments /r, u, k, i/: (41) Sanskrit RUKI rule (Kiparsky 1982d) s ! » / {r, u, k, i} The rule is blocked in morphologically simple cases (42a) but applies in a number of morphologically derived contexts (42b): (42) a. Morphologically underived contexts barse ‘tip’ kusuma kisalaya ‘sprout’ pis b. Morphologically derived cases da-da-si ‘you give’ bi-bhar-»i kram-sya-ti ‘he will go’ vak-»ya-ti sena-su ‘armies’ agni-»u

‘flower’ ‘move’ ‘you carry’ ‘he will say’ ‘fires’

These familiar e¤ects are easily modeled using the framework developed in this chapter: at the stratum in which the a‰xes in (42b) are attached, the ranking Faith-LEX  Ruki  Faith obtains; ‘‘Ruki’’ represents the markedness constraints responsible for /s/ retroflexion. Ruki is triggered not only in morphologically derived environments, however, but also in phonologically derived environments. Ablaut is a morphophonological process that results in vowel change (43a) and vowel loss (43b), among other e¤ects, in morphological Ablaut environments. As shown in (43), the application of Ablaut can feed the Ruki rule:

A Historical Perspective on Nonderived Environment Blocking

483

(43) Zero Grade Ablaut and RUKI (Kiparsky 1982d, 76) a. /a/ ! /i/ /sas-ta/ ! si»a ‘taught’ b. /a/-loss /ja-ghas-anti/ ! jak»anti ‘eat (reduplicated 3. pl.)’ c. /vas/ ! u» ‘shine, dwell’ /va-vas-us/ ! u»us ‘shone, dwelt’ d. /-vas/ ! -u» (gen. sg.) /vid-vas-as-/ ! vidu»as ‘knowing’ In an analysis like Hammond’s, which draws inherent distinctions between ‘‘phonologically derived’’ and ‘‘morphologically derived’’ environments, it matters a great deal whether Ablaut in Sanskrit is a phonological process or not. For our purposes, this question need not be raised at all. Ablaut simply involves markedness constraints that figure in the computation of optimal outputs. The following tableaux illustrate the interaction of Ablaut and the Ruki rule. I assume a constraint, or constraint family, ‘‘Ablaut,’’ containing the markedness constraint(s) responsible for triggering the range of e¤ects shown in (43). Since Ablaut applies morpheme-internally, Ablaut must outrank Faith-ROOT, yielding the following overall ranking: (44) Ablaut  Faith-ROOT  Ruki  Faith In Ablaut environments, satisfaction of Ablaut neutralizes Faith-ROOT by disrupting lexeme-output identity. Once Faith-ROOT is violated, there is nothing to prevent satisfaction of the Ruki constraints. In the tableau below, (45a) is a case where the context for Ablaut is not met; Faith-ROOT is satisfied at the expense of the lowerranked Ruki constraint. By contrast, (45b) is an Ablaut environment. Satisfying Ablaut means violating Faith-ROOT; the candidate satisfying both Ablaut and Ruki is therefore the winner: NDEB in Ruki (Ablaut  Faith-ROOT  Ruki  Faith)

(45) a.

/bi-bhar-si/[Ablaut]

Ablaut

bibharsi

Faith-ROOT

Ruki *!

+ bibhar»i b.

Faith

*

/ja-ghas-anti/[þAblaut] jagha»anti jaksanti + jak»anti

*! *!

* **

484

Young-mee Yu Cho

The fact that both morpheme concatenation (42b) and the application of Ablaut (43) create a ‘‘derived environment’’ for Ruki is given a unified account in terms of constraint ranking. Both are environments in which Faith-ROOT is violated, removing the impediment to satisfaction of the lower-ranked markedness constraint, in this case, Ruki. Turning briefly to a diachronic perspective, we note that Ruki was a word stratum process in Rigvedic Sanskrit but became restricted to being a cyclic process in Classical Sanskrit (Kiparsky 1997a). (46) summarizes the change in terms of Faithfulness promotion, a very common trajectory of historical change. (46) a. Rigvedic Ruki applies across a‰x, compounding, and clitic boundaries, but not phrasally across word boundaries. Cyclic stratum: Faith-ROOT  Ruki  Faith Word stratum: Faith-STEM  Ruki  Faith Phrasal stratum: Faith-PWD, Faith  Ruki (no Ruki) b. Classical Ruki applies across a‰x boundaries but not across compound, clitic, or word boundaries. Cyclic stratum: Faith-ROOT  Ruki  Faith Word stratum: Faith-STEM  Faith  Ruki (no Ruki) Phrasal stratum: Faith-PWD, Faith  Ruki (no Ruki) The present study can also naturally be extended to the so-called SCC e¤ects in phrasal phonology (Rice 1990). For instance, Sanskrit and Catalan final voicing, triggered by a sonorant, applies between words in both compounds and phrases, but it does not apply within words (47a, b). (47) The ‘‘SCC’’ e¤ects in voicing assimilation a. Catalan (Mascaro´ 1987) to[t] [r]ic ! to[dr]ic ‘all rich person’ se[t] [ma]ans ! se[dm]ans ‘seven hands’ (cf. temple [-pl-] ‘temple’, submari [-pm-] ‘submarine’) b. Sanskrit (Whitney 1889) sa[t] [a]ha ! sa[da]ha ‘good day’ ta[t] [n]amas ! ta[dn]amas ‘that homage’ (cf. putra [-tr] ‘son’ marutþi ! maru[ti] ‘wind-locative’) c. Faith-PWD  Voicing Assim  Faith The SCC makes word-internal structure unavailable at the phrasal stratum just as the NDEB makes root-internal structure unavailable at the cyclic stratum. The ranking in (47c) accounts for the integrity of an isolated prosodic word, in contrast to the variation in phrasal concatenation.

A Historical Perspective on Nonderived Environment Blocking

19.9

485

Conclusion

In the chapter, I have shown that the development of Korean Palatalization follows the change in constraint ranking shown in (48). History reveals no inherent relationship between Palatalization and NDEB. NDEB e¤ects developed well after the onset of Palatalization, modeled here by a reranking of Faith-LEX above the triggering Pal constraint: (48) Historical development of Korean Palatalization a. Across-the-board Palatalization: Pal  Faith-ROOT, Faith b. NDEB: Faith-ROOT  Pal  Faith Two additional constraint-ranking schemas, involving markedness and faithfulness constraints, have also been validated by data from Korean and Sanskrit: (49) NDEB rankings a. Faith-LEX  M  Faith b. M1  Faith-LEX  M2  Faith

(M satisfied only in morphologically derived environments) (M2 satisfied in both morphologically and phonologically derived environments)

The factorial typology based on the four-constraint system in (49b) predicts three patterns: (i) application in all environments (M  Faith-LEX, Faith), (ii) application only in derived environments either phonologically or morphologically (M1  Faith-LEX  M2  Faith), and (iii) application only in morphologically derived environments (Faith-LEX  M  Faith). Application only in phonologically derived environments cannot be described in this system, which I believe to be a desirable consequence. Although there are a number of cases in which only morphologically derived environments are considered derived, there are few or no cases where only phonologically derived environments, but not morpheme combinations, count in NDEB.10 Notes Earlier versions of this chapter were presented at colloquia in 2001–2002 at University of Massachusetts, Rutgers University, and Cornell University. Over the course of working on this topic I have benefited from discussions with Arto Anttila, John Alderete, Gregory Iverson, John McCarthy, Joe Pater, Alan Prince, John Whitman, and Draga Zec. I am particularly grateful to the editors of this book. Abbreviations: adv ¼ adverbializer, caus ¼ causative, com ¼ comitative, cop ¼ copula, gen. sg ¼ genitive singular, nml ¼ nominalizer, nom ¼ nominative, pass ¼ passive, 3. pl ¼ third-person plural.

486

Young-mee Yu Cho

1. Korean Palatalization can be decomposed into two independent processes, lexical a¤rication and postlexical palatalization. Section 19.7 deals with the di¤erences between the two. I retain the term Palatalization because of its wide usage. 2. The idea of Faith-ROOT constraints was introduced in McCarthy and Prince 1995, and has been taken up in many subsequent works in OT. 3. Whether TSS represents a group of ranked constraints or one monolithic constraint will not be discussed here. 4. Kim (1999) argues that the a¤ricates assumed to be alveopalatal or palatal are, in fact, alveolars and that Palatalization is an a¤rication process. Clearly there seem to be a range of variations in place of articulation for these a¤ricates. However, even if it turns out that the process in question is not of change of place but that of manner, the overall point made in this chapter will still remain valid. 5. There is only one liquid phoneme in the language, represented here by ‘‘L.’’ It surfaces as /l/ in the coda position or as part of a geminate and as [r] in onset position. See Cho 1997 for a detailed discussion on the distribution of this liquid. 6. Korean has a productive phrasal process of obstruent voicing between two sonorants that render [mati] and [maci] as [madi] and [maji]. I ignore this process for the sake of simplicity of representation here. 7. In particular, Kang (1992) proposes an end-based account in which left-edge alignment creates proper morphological bracketing. 8. I also predict that loanwords that end with a dental would undergo Palatalization when they are combined with an i-initial su‰x. This is in line with the loss of exceptionality under derivation demonstrated in several other languages—for example, certain words in Catalan do not undergo vowel reduction when una‰xed but do undergo reduction when a su‰x is attached (Kiparsky 1993a; Inkelas 2000). One potential example in Korean is the English word pot. All borrowed nouns ending in a dental stop are optionally realized as [s]-final before the vowel-initial nominative, as in [pas-i]. The alternative is to apply vowel epenthesis, as in [p h at h -ka], which results in the selection of the /ka/ nominative su‰x. 9. A remaining outstanding question is: What caused the change from across-the-board application to an NDEB process? In OT terms, the question can be rephrased as ‘‘What caused the rise of Faith-ROOT?’’ 10. See, however, McCarthy 2003b, which presents a system for generating precisely this kind of NDEB e¤ect.

20

Lexical Storage and Phonological Change

Geert Booij

20.1

Introduction

Empirical investigations of the division of labor between storage and computation in language behavior can be executed in a number of ways. An obvious one is that of psycholinguistic experimentation. Another way, which will be focused on in this paper, is language change. Language change is a psycholinguistic laboratory of nature, a window on how speakers produce and interpret language. For example, if a language loses a phonological rule while the e¤ects of that rule are preserved in a number of words, a possible explanation is that the outputs of that rule must have been stored at the stage when the phonological rule was still active and thus survived after the loss of that rule. The question that I will address in this paper is how far phonological change provides evidence for the kinds of phonological information about lexical items stored in lexical memory. That is, whereas Kiparsky in his early work on phonological change since 1965 (compiled in Kiparsky 1982a) focused on phonological change as evidence for the structure of the grammar, I will take a di¤erent, extragrammatical perspective and ask what we can learn from phonological change about lexical memory. Asking this question is also in line with Kiparsky’s conclusion at the end of his book Explanation in Phonology that linguistic change does not provide a window on the structure of the grammar as directly as was hypothesized in Kiparsky 1968c: ‘‘Before we can exploit historical evidence for synchronic purposes we need a firm theory of the intervening factors’’ (Kiparsky 1982a, 234). Lexical storage is certainly one of these intervening factors that deserve more detailed investigation. A necessary preliminary remark is that the issue of storage versus computation with respect to a specific regularity of a language is not a matter of ‘‘either . . . or.’’ The conclusion that a particular linguistic form must be stored in the lexicon does not preclude the existence of a rule that accounts for most or all of the properties of that form. This position has not been a standard one in Generative Grammar, which has always been strongly influenced by the Bloomfieldian view of the lexicon as the basic list of irregularities. For instance, Kenstowicz (1994, 60) motivates the claim that predictable information is not stored lexically as follows.

488

Geert Booij

(1) ‘‘Generative grammar’s answer to this question is based on the hypothesis that the human capacity for language is designed in such a way as to minimize the amount of information that must be stored in the speaker’s mental lexicon.’’ However, a few pages later in the same book, Kenstowicz (1994, 69–70) points out that this point of view is no longer self-evident: ‘‘with the advent of neural science and more accurate estimates of the capacity of the human brain, this ‘‘economy of storage’’ argument is not compelling in and of itself.’’ What I suspect is that the storage argument given in (1) has never been a serious consideration. Rather, the aim has always been to give an elegant and formally as simple as possible analysis of the distribution and alternation patterns of a language. In the area of Generative Morphology, linguists have always been aware of the possible and even necessary simultaneity of rules and stored outputs. Many complex words, once formed, must be lexically stored because they have unpredictable formal and/or semantic properties. Nevertheless, the rules that created these complex words may still be productive. Jackendo¤ (1975) therefore advocated a view of morphological rules in which these rules function as redundancy rules with respect to existing complex lexical items, but this does not exclude their creative use for the coining of new complex words. That is, predictable information can be lexically stored, and redundancy rules tell us which part of that information is predictable, that is, does not count as independent information. Similar views were defended by Arono¤ (1976) and Booij (1977a, 1977b). In sum, we should avoid what has been called the ‘‘rule/ list fallacy’’ by Langacker (1987, 29), namely, the reasoning in which lists and rules are mutually exclusive. In fact, the relation between lexical storage and morphological rules is even stronger and of a more principled nature: a morphological rule does not exist without a set of listed words instantiating that rule. The native speaker first acquires complex words. It is on the basis of recurring patterns in sets of similar complex words with a systematic pairing of form and meaning that the speaker may conclude the existence of a morphological rule, which then may result in extension of the set of words of that particular form.1 This view of morphology presupposes that words (and idioms), not morphemes, are the units of lexical storage. The role of morphemes in the analysis of word structure is of a secondary nature: they play a role in establishing the relations between words but are not the primary building blocks of complex words.2 As mentioned above, in Generative Grammar the prevailing tendency has been to reduce the lexical storage of allomorphy as much as possible. Allomorphy is accounted for in terms of one underlying form for each morpheme and a set of rules for the computation of the actual surface allomorphs. In addition, predictable phonetic details are omitted from lexical representations: it is only phonemic distinctions,

Lexical Storage and Phonological Change

489

that is, contrastive phonetic properties, that are encoded in lexical representations. The adage is what can be computed, should not be stored.3 This relates to the wellknown abstractness controversy: the more abstract our phonology is, the more we can derive di¤erent surface forms from the same underlying representation. Linguists have tried to tackle this issue by looking at phonological change (Kiparsky 1968c), whereas psycholinguists have been trying to solve this problem by means of psycholinguistic evidence (cf. Lahiri and Marslen-Wilson 1992). In the beginning of the 1970s, Bybee and Vennemann argued in favor of a concrete view of phonological representations, partially on the basis of facts of language change (Hooper 1974b; Vennemann 1974). Generative phonologists, however, di¤er in the degree to which they strive for redundancy-free representations, as shown by the debate on underspecification. For instance, in Dutch the velarity of a nasal is contrastive word finally but not before a velar obstruent. Should we therefore omit the specification for place in the lexical representation of velar nasals before velar stops? This is not obvious, and Anderson (1985, 136f ) has warned us that the idea that linguistic representations should be redundancy free is by no means self-evident, and probably wrong. The principle of ‘‘Lexicon Optimization’’ advocated by phonologists working within Optimality Theory (Archangeli and Langendoen 1997) also implies that redundant, nondistinctive phonological information will be stored in lexical representations: the faithfulness condition implies that inputs should di¤er minimally from the corresponding outputs (see Archangeli and Langendoen 1997, 201¤ ).4 Although there are di¤erences in the degree of abstractness that generative phonologists allow for, most generative phonologists assume that at least the e¤ect of automatic phonological rules should not be encoded in the lexical representations of morphemes and words, because they are always computable on the basis of purely phonological information. However, we should not take this position for granted given the storage capacity of human memory. Even if one is willing to accept the storage of the e¤ects of phonological rules, the information stored may still be abstract in the sense that it is phonemes that are stored and not the actual details of the phonetic realization of these phonemes. This is what most generative phonologists assume, even those who advocate a concrete kind of phonology: the phonetic details, such as the acoustic parameters for a particular vowel, are not specified as such in lexical representations but accounted for by the set of language-specific rules of phonetic implementation. However, this position has recently been attacked, for instance in Flemming (1995) and Bybee (2000b). This issue will be returned to in section 20.5. In sum, there are three kinds of phonetically relevant information with respect to which the lexical representation is a point of discussion: (i) predictable phonetic properties that have a contrastive function in some contexts, such as the place of

490

Geert Booij

articulation of nasals; (ii) the e¤ects of phonological processes that create alternations; and (iii) details of phonetic realization. The point of departure for our discussion is that we know that the vastness of the lexical memory allows us to store much of what can be computed, even though it is completely regular, since it will speed up processing. In the area of morphology, there is solid evidence for the storage of frequent plural nouns in a number of languages (Baayen et al. 1997). The same applies to the syntactic level. There are thousands of noun phrases of the type adjective þ noun, such as yellow pages, red tape, green card, black hole, hard disk, little toe, that are listed in our lexical memory because they function as names for things (see Jackendo¤ 1997, chapter 7); yet this does not imply that we do not have a productive rule for the construction of such noun phrases. Given these general considerations about the possibility of storage, and given that there is no conflict between storage and regularity/productivity, it is important to ask ourselves how we can find out about the realities of storage without a priori positions as to what is memorized in the form of lexical representations and what in the form of rules. In this chapter, I will use facts of phonological change as one possible type of evidence for storage and provide evidence for surface forms of words being stored. Another general issue that should be addressed before going into the details of phonological change is that it is often taken for granted that the notions ‘‘lexical representation’’ in the sense of ‘‘the form of a word that is the starting point of morphological operations’’ and ‘‘lexical representation’’ in the sense of ‘‘stored information about a lexical item’’ coincide. This is not a logical necessity, however. Take a simple case like the following pair of nouns from Dutch. (2) hoed ‘hat’ [hut], plural hoeden [hud n] e

The standard assumption is to assume /hud/ as the lexical representation for hoed and derive the phonetic form of the singular from this underlying form by means of the rule of Auslautverha¨rtung. However, it is also logically possible that we store [hut] as the singular form and in addition the basic form /hud/, based on the alternation [hut]/[hud n]. It is even possible that we do not store an underlying form like /hud/ at all and only compute it from the stored phonetic form if we need to do so because we want to apply a morphological operation to that word. In other words, we then reverse the classical analysis: phonetic forms are stored, and underlying forms are computed. This option will be discussed in more detail below. e

20.2

Phonemicization of Allophonic Properties

A number of Germanic languages have seen the loss of the stop in word-final clusters of velar nasal þ velar stop. For instance, Middle Dutch coninc [ko:nIÐk] ‘king’ be-

Lexical Storage and Phonological Change

491

came koning [ko:nIÐ] in Modern Dutch. Thus, the property of the nasal consonant that it is velar before a velar stop lost its allophonic nature and became phonemicized: words ending in [Ð] now contrast with words ending in [m] or [n]. Does this phonological change tell us something about the lexical storage of the place of articulation of the nasal before the following stop got lost and thus about the storage of redundant properties? In a classical rule-based analysis of phonological change, the answer is negative: the nasal can be underspecified at the underlying level (usually graphically represented as N). Two rules apply: the rule that spreads place of articulation from the stop to the nasal and, subsequently, a rule of cluster simplification that deletes the stop (the addition of this rule is the change in the grammar of the adult). The surface form of coninc is therefore [ko:nIÐ]. The child will then store this surface form as the underlying form, thus giving phonemic status to the velar nasal. This account of phonological change makes a sharp distinction between the two generations involved in a change: the adult generation and the next generation that has to acquire the language system. It is doubtful, however, whether such a sharp distinction is justified. It is quite clear that adult speakers also continuously subject the outputs of their language system to reanalysis, that is, there is a continuous inspection of output forms (see Bynon 1977; Hopper and Traugott 1993). The possibility of inspection and reanalysis presupposes that these output forms are stored: have a certain degree of permanence in memory. Moreover, this also enables us to explain phonological change. In the present example, the loss of a final obstruent, the speaker can conclude that the final obstruent can be omitted without the loss of distinction between words, because the place of articulation of the nasal will su‰ce to distinguish a word like zing [zIÐ] ‘to sing’ from zin [zIn] ‘sense’. Specification of place of articulation of the nasal in the word-final consonant cluster is also necessary for theory-internal reasons if we take a constraint-based approach to phonology instead of the traditional rule-based approach. The rulebased analysis makes crucial use of rule ordering of a particular type, nonbleeding order, and of the assumption that a phonological change is to be seen as a rule that is always added at the end of the grammar. If we take an Optimality-theoretical constraint-based approach, it is immediately clear that the velarity of the nasal should be part of the lexical representation even before the loss of the final obstruent: an underspecified representation will induce an extra violation of faithfulness compared to a fully specified lexical representation, which goes against the principle of Lexicon Optimization. The surfacing of the velar nasal is then straightforward: the change involved is that the constraint *CC-Velar (a constraint belonging to the family of constraints on consonant clustering) that forbids velar consonant clusters is ranked higher than Max-IO (a faithfulness constraint that requires identity of input and output).

492

(3)

Geert Booij

ko:nIÐk

*CC-Velar

ko:nIÐk

*!

Max-IO

+ ko:nIÐ

*

If an underspecified lexical representation (forbidden by the principle of Lexicon Optimization, at least if no alternation is involved, see Inkelas 1995) had been assumed for the velar nasal, we would predict the underspecified nasal to surface as coronal [n] since this is the default value for nasal consonants. (I assume that default values are expressed by markedness constraints such as Coronal: the default place of articulation of consonants is coronal, and the place of articulation must be specified, the constraint of Full Specification). (4)

/ko:nINk/

Full Specification

ko:nIÐk ko:nIN

*!

ko:nIÐ

*CC-Velar

Coronal

*!

*

Max-IO

*

*

*!

*

+ ko:nIn

*

In conclusion, given a constraint-based analysis, the only way the velar nasal will surface after the diachronic process of velar cluster simplification is by specifying the place of articulation of this nasal consonant at the level of lexical representation. We should note, however, that this kind of evidence for lexical storage of predictable properties is theory dependent, since, as we saw above, a rule-based analysis did not imply lexical storage of the place of articulation of the nasal. 20.3

Phonologization and Lexicalization: Vowel Lengthening in Dutch

Another relevant case of phonologization of allophonic properties is that of vowel lengthening in open syllables in Early Middle Dutch. This process a¤ected both simplex nouns and complex nouns and even wordþclitic combinations. (5) a. Simplex nouns name ‘name’ smake ‘taste’ stave ‘sta¤ ’

n[a:]me sm[a:]ke st[a:]ve

Lexical Storage and Phonological Change

b. Singular sch[I]p ‘ship’ h[ ]l ‘hole’ oorl[ ]g ‘war’ d[þ]g ‘day’ c. Word þ clitic saetic /sþt Ik/

493

Plural sch[e:]p-en h[o:]l-en oorl[o:]gen d[a:]g-en [sa:tIk] ‘sat I’

c

c

The singular-plural pairs in (5b) have been preserved in modern Dutch (see Booij 1995; Kager, this volume). This process of vowel lengthening is a manifestation of a much more general tendency in Germanic languages toward a requirement that stressed syllables must be heavy, that is, they cannot end in a short vowel but must contain minimally either a long vowel or a short vowel followed by a consonant (also known as Prokosch’s Law).5 In the first stage of this process in Early Middle Dutch (see Van Loon 1986, 89), the lengthening of the vowel was an allophonic process. In the course of time (probably in the 12th century), however, the lengthened vowel was phonemicized, that is, it became an underlying long vowel. How do we know this? The first evidence concerns short vowels followed by geminate consonants. These were protected from lengthening because the first half of the geminate closes the syllable. In the 12th century, degemination took place. Yet, the short vowels that thus ended up in open syllables did not lengthen anymore, and so we find many words with short vowels followed by only one consonant, such as wikke [VIk ] ‘vetch’. That is, the following development took place. e

(6) VCV > V:CV (predictably long vowel) > V:CV (phonemic long vowel) VCCV > VCCV (predictably short vowel) > VCV (phonemic short vowel) These changes are illustrated by the following minimal pair (Van Loon 1986, 89). Modern Dutch week [we:k] wikke [wIk ] e

Early Middle Dutch weke [we:k ] wikke [wIkk ] e

e

(7) Early Germanic wika [wIka] ‘week’ wikkia [wIkkia] ‘vetch’

The classical generative interpretation of this kind of phonemicization is as follows (Kiparsky 1968b): The initial change is the addition of a rule of degemination to the phonological system of the adult speakers. This rule of degemination made the alternation between long and short vowels in the output forms opaque; vowel lengthening is ordered before degemination, and this is a nonbleeding order. The next generation of language users therefore interpreted the length contrast as a phonemic contrast, and hence the length contrast became part of the lexical representation. In addition, the process of vowel lengthening in open syllables disappeared.

494

Geert Booij

A second kind of evidence for the phonemicization of these initially allophonic contrasts for simplex nouns is provided by the later process of schwa apocope (thirteenth century). This phonological change did not a¤ect the length of the vowel in the simplex nouns ending in schwa: after apocope the vowels of words such as naam ‘name’ (