R E D U P L I C AT I O N

This groundbreaking new study takes a novel approach to reduplication, a

phenomenon whereby languages use repetition to create new words. Sharon
Inkelas and Cheryl Zoll argue that the driving force in reduplication is identity
at the morphosyntactic, not the phonological, level and present a new model of
reduplication – Morphological Doubling Theory – that derives the full range of
reduplication patterns. This approach shifts the focus away from the relatively
small number of cases of phonological overapplication and underapplication,
which have played a major role in earlier studies, to the larger class of cases
where base and reduplicant diverge phonologically. The authors conclude by
arguing for a theoretical shift in phonology, which entails more attention to
word structure. As well as presenting the authors’ pioneering work, this book
also provides a much-needed overview of reduplication, the study of which
has become one of the most contentious in modern phonological theory.

s h a r o n i n k e l a s is Professor in the Department of Linguistics, University

of California at Berkeley. She has over fifteen years’ research and teaching
experience and has been in the Department of Linguistics at the University of
California, Berkeley since 1992. She is the author of a wide variety of articles
in phonology and morphology.

c h e r y l z o l l is Associate Professor in the Department of Linguistics and

Philosophy, Massachusetts Institute of Technology, with research interests in
phonology and morphology, particularly in relation to African languages. She
is the author of Parsing Below the Segment (1998) as well as numerous articles
on a variety of topics in phonology and morphology.
 In this series
65 e v e v. c l a r k: The lexicon in acquisition
66 a n t h o n y r . wa r n e r: English auxiliaries: structure and history
67 p. h . m at t h e w s: Grammatical theory in the United States from Bloomfield
to Chomsky
68 l j i l j a n a p r o g o va c: Negative and positive polarity: a binding approach
69 r . m . w. d i x o n: Ergativity
70 ya n h ua n g: The syntax and pragmatics of anaphora
71 k n u d l a m b r e c h t: Information structure and sentence form: topic, focus, and the
mental representation of discourse referents
72 l u i g i bu r z i o: Principles of English stress
73 j o h n a . h aw k i n s: A performance theory of order and constituency
74 a l i c e c . h a r r i s and ly l e c a m p b e l l: Historical syntax in cross-linguistic
perspective
75 l i l i a n e h a e g e m a n: The syntax of negation
76 p a u l g o r r e l: Syntax and parsing
77 g u g l i e l m o c i n q u e: Italian syntax and universal grammar
78 h e n r y s m i t h: Restrictiveness in case theory
79 d . r o b e r t l a d d: Intonational morphology
80 a n d r e a m o r o: The raising of predicates: predicative noun phrases and the theory
of clause structure
81 r o g e r l a s s: Historical linguistics and language change
82 j o h n m . a n d e r s o n: A notional theory of syntactic categories
83 b e r n d h e i n e: Possession: cognitive sources, forces and grammaticalization
84 n o m t e r t e s c h i k - s h i r: The dynamics of focus structure
85 j o h n c o l e m a n: Phonological representations: their names, forms and powers
86 c h r i s t i n a y. b e t h i n: Slavic prosody: language change and phonological theory
87 b a r b a r a d a n c y g i e r: Conditionals and prediction
88 c l a i r e l e f e b v r e: Creole genesis and the acquisition of grammar: the case of
Haitian creole
89 h e i n z g i e g e r i c h: Lexical strata in English
90 k e r e n r i c e: Morpheme order and semantic scope
91 a p r i l m c m a h o n: Lexical phonology and the history of English
92 m at t h e w y. c h e n: Tone Sandhi: patterns across Chinese dialects
93 g r e g o r y t. s t u m p: Inflectional morphology: a theory of paradigm structure
94 j o a n b y b e e: Phonology and language use
95 l a u r i e b a u e r: Morphological productivity
96 t h o m a s e r n s t: The syntax of adjuncts
97 e l i z a b e t h c l o s s t r a u g o t t and r i c h a r d b. d a s h e r: Regularity in
semantic change
98 m aya h i c k m a n n: Children’s discourse: person, space and time across languages
99 d i a n e b l a k e m o r e: Relevance and linguistic meaning: the semantics and
pragmatics of discourse markers
100 i a n r o b e r t s and a n n a r o u s s o u: Syntactic change: a minimalist approach to
grammaticalization
101 d o n k a m i n k o va: Alliteration and sound change in early English
102 m a r k c . b a k e r: Lexical categories: verbs, nouns and adjectives
103 c a r l o ta s . s m i t h: Modes of discourse: the local structure of texts
104 r o c h e l l e l i e b e r: Morphology and lexical semantics
105 h o l g e r d i e s s e l: The acquisition of complex sentences
106 s h a r o n i n k e l a s and c h e r y l z o l l: Reduplication: doubling in morphology
Earlier issues not listed are also available
CAMBRIDGE STUDIES IN LINGUISTICS
General editors: p. a u s t i n, j . b r e s n a n, b. c o m r i e ,
s . c r a i n, w. d r e s s l e r , c . j . e w e n, r . l a s s ,
d . l i g h t f o o t, k . r i c e , i . r o b e r t s , s . r o m a i n e
n. v. s m i t h

Reduplication
R EDUP L I CAT I ON
DOU BL I N G I N M O RPHO LO G Y

SHA RO N IN K E LA S
University of California, Berkeley

and

CH ERYL Z O L L
Massachusetts Institute of Technology
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo, Delhi

Cambridge University Press

The Edinburgh Building, Cambridge CB2 8RU, UK

Published in the United States of America by Cambridge University Press, New York

www.cambridge.org
Information on this title: www.cambridge.org/9780521114509

© Sharon Inkelas and Cheryl Zoll 2005

This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.

First published 2005

Reprinted 2005
This digitally printed version 2009

A catalogue record for this publication is available from the British Library

ISBN 978-0-521-80649-7 hardback

ISBN 978-0-521-11450-9 paperback

Cambridge University Press has no responsibility for the persistence or

accuracy of URLs for external or third-party Internet websites referred to in
this publication, and does not guarantee that any content on such websites is,
or will remain, accurate or appropriate. Information regarding prices, travel
timetables and other factual information given in this work are correct at
the time of first printing but Cambridge University Press does not guarantee
the accuracy of such information thereafter.
 Contents

Acknowledgments page xi
Table of languages xiii
Abbreviations used in morpheme glosses xxi

1 Introduction 1
1.1 Two approaches to duplication 2
1.2 Morphological Doubling Theory 6
1.2.1 The morphology of reduplication 7
1.2.2 Constructions in morphology 11
1.2.3 Constructional semantics 13
1.2.4 Constructional phonology 16
1.2.5 The phonology of reduplication 18
1.3 Phonological copying 20
1.4 Distinguishing the two types of duplication 22
1.5 Wrapup and outline of book 23

2 Evidence for morphological doubling 25

2.1 Morphological targets: affix reduplication 27
2.1.1 Preverb reduplication: Hungarian 28
2.1.2 Reduplication within the derivational stem 28
2.1.3 Further implications 30
2.2 Morphotactic asymmetries: empty morphs 31
2.2.1 Empty morphs in morphology 31
2.2.2 Phonologically beneficial empty morphs 33
2.2.3 Empty morphs in reduplication 36
2.2.4 Simple melodic overwriting 41
2.2.5 Double melodic overwriting 43
2.2.6 Tier replacement 45
2.3 Synonym and antonym constructions 47
2.3.1 Root allomorphy 47
2.3.2 Synonym constructions 59

vii
viii Contents

2.3.3 Beyond synonyms 61

2.3.4 Wrapup 65
2.4 Comparison of MDT with OO correspondence 65
2.5 Conclusion 66

3 Morphologically conditioned phonology in reduplication:

the daughters 67
3.1 Cophonologies 70
3.1.1 Cophonologies vs. indexed constraints 74
3.1.2 Cophonologies in reduplication 75
3.2 Typical daughter modifications 77
3.3 Divergent modification 82
3.3.1 Hua 82
3.3.2 Hausa tonal modification 83
3.3.3 Tarok: divergent TETU 84
3.3.4 Parallel modification 86
3.3.5 Double modification outside of reduplication 91
3.4 Daughter independence vs. base dependence 92
3.4.1 Reduplicant shape 92
3.5 Conclusion 97

4 Morphologically conditioned phonology in reduplication:

the mother node 98
4.1 General approach to junctural phonology 99
4.2 Reduplication-specific alternations 100
4.3 Reduplication-specific non-alternation 103
4.3.1 BR-Faith is insufficient 104
4.3.2 Underapplication all over 105
4.3.3 Non-identity-enhancing underapplication in
reduplication 106
4.3.4 Layering and underapplication 108
4.3.5 Klamath 113
4.4 ∃-Faith 118
4.4.1 Predictions of ∃-Faith 121
4.4.2 Overapplication of reduplication-specific
phonology 122
4.4.3 Construction-specific insertion 124
4.4.4 Reduplication-internal variation 125
4.4.5 Parallels between reduplicative and nonreduplicative
phonology 128
4.4.6 Wrapup 133
4.5 Conclusion 134
 Contents ix

5 Morphologically driven opacity in reduplication 135

5.1 Daughter-based opacity: overapplication and underapplication
in Javanese 136
5.1.1 /a/-raising: underapplication by truncation 137
5.1.2 Suffix-triggered ablaut: overapplication by truncation 138
5.1.3 Opacity in suffixation and reduplication: wrapup 143
5.1.4 Active prefix 144
5.1.5 /h/-deletion: overapplication 145
5.1.6 Laxing: underapplication 147
5.1.7 Summary 150
5.2 Mother-based opacity: infixation 151
5.2.1 Chamorro 151
5.2.2 Eastern Kadazan 152
5.2.3 Infixation in MDT 155
5.3 Morphological opacity outside of reduplication 156
5.3.1 Opacity by truncation 156
5.3.2 Opacity by infixation 157
5.4 MDT vs. Coerced Identity theories 158
5.4.1 Opacity does not always increase identity 161
5.4.2 Distribution of opacity 161
5.5 Case study: Fox 165
5.5.1 Stem-internal alternations: opaque overapplication
by truncation 168
5.5.2 Junctural alternations: normal application 169
5.5.3 Summary of Fox 172
5.6 Conclusion: morphology underlies opacity 173
5.7 The question of backcopying 174
5.7.1 Lack of evidence for backcopying in morphological
reduplication 175
5.7.2 Backcopying as phonological assimilation 177
5.7.3 Implications 180

6 Case studies 181

6.1 Tagalog 181
6.1.1 Alternatives to backcopying 182
6.1.2 Prefixation vs. infixation 183
6.2 Chumash 185
6.2.1 Alternatives to backcopying 187
6.2.2 Chumash verb morphology 189
6.2.3 Which prefixes can contribute an overcopying final C? 190
6.2.4 Inner (Level 2) prefixes in the reduplication domain 190
6.2.5 A split among Inner prefixes in Ineseño 191
x Contents

6.2.6 Implications of the split among Inner prefixes 193

6.2.7 Summary 195

7 Final issues 197

7.1 Criteria distinguishing phonological copying from
morphological reduplication 197
7.2 The purpose and nature of phonological copying 197
7.3 The morphological purpose of reduplication 199
7.4 CV reduplication 201
7.5 The question of rhyme 203
7.6 The question of anti-identity 210
7.7 Beyond reduplication 212

Notes 213
References 225
Index of languages 245
Index of names 247
Index of subjects 251
 Acknowledgments

The broad focus of this book has given us the opportunity to benefit from
the expertise of a multitude of linguists whose interest and insightful input
have contributed immeasurably to the construction of a morphological the-
ory of reduplication. Orhan Orgun and Larry Hyman challenged and inspired
us throughout. Juliette Blevins, Andrew Garrett, Michael Kenstowicz, Teresa
McFarland, Anne Pycha, Richard Rhodes, Ronald Sprouse, and Donca Steri-
ade all read the manuscript in various stages and facilitated the development of
our ideas with their penetrating critiques. We owe a debt of gratitude as well
to Terry Crowley, Laura Downing, Danny Fox, Claire Lefebvre, Frank Licht-
enberk, Johanna Nichols, Nick Sherrard, Galen Sibanda, Rajendra Singh, Ken
VanBik, and many others who generously provided us with useful materials
and shared their knowledge about languages, phenomena, and theories once
unfamiliar to us. We also thank our students, east and west, for their stimulating
interest and willingness to participate in the development of MDT. The work
benefited significantly from the feedback of audiences at a number of colloquia
and conferences, including Phonology 2000 (MIT/Harvard), NAPhC (Concor-
dia University, Montreal), the 1999 Linguistics Society of America Meeting,
and the 2002 Graz Reduplication Conference (Graz, Austria), and from discus-
sions at UC Berkeley and MIT during various less formal presentations. Much
of this book was written with the generous support of the Radcliffe Institute for
Advanced Study at Harvard University, which assisted one of the authors with
a fellowship in 2001–2002. We are grateful also to the Linguistics Department
at UC Davis for providing office space and a library card during one crucial
summer. Anne Nesbet, Nancy Katz, Klara Moricz, and David Schneider pro-
vided unending encouragement. Vineeta Chand sustained us with her expert
and cool-headed copy-editing, and Anne Pycha took time in a busy semester
to research the language table. We thank, finally, our families: the children we

xi
xii Acknowledgments

raise on our laps while we type, Jem, Eli, and Lydia, and our spouses, Orhan
Orgun and Eric Sawyer, who remove them at crucial moments for their occa-
sional bath or hot meal, and who provide boundless moral and logistical support.
This book is dedicated to our children, and to Curry Sawyer, whose inspiration,
practical assistance, and unparalleled ability to get things done make everything
possible.
 Table of languages
This table of languages is a compilation of information from the sources consulted in this work and the SIL Ethnologue,
available from SIL International at http://www.ethnologue.com.

Language Primary Regions within Secondary Special

country primary country countries note

Abkhaz Georgia Abkhazia Turkey, Ukraine

Amele Papua New Madang Province
Guinea
Amuesha Peru Central and eastern Pasco region
Apma Vanuatu Central Pentecost (Raga)
Arapesh Papua New North coast
Guinea
Armenian Armenia 29 others
Arosi Solomon Islands Northwest Makira (San Cristobal) Island
Arrernte Australia Northern Territory, Alice Springs area
Axininca Peru Pichis and Sheshea tributaries of the
Campa Pachitea River
Babine Canada West central British Columbia
Banoni Papua New North Solomons Province, southwestern
Guinea Bougainville
Bella Coola Canada Central British Columbia coast
Bellonese Solomon Islands Rennell and Bellona Islands
Bierebo Vanuatu West Epi
(cont.)
Language Primary Regions within Secondary Special
country primary country countries note

Boumaa Fijian Fiji Nauru, New Zealand, Vanuatu

Burmese Myanmar South, central, and adjacent areas Bangladesh, Malaysia, Thailand, USA
Chaha Ethiopia West Gurage Region
Chamorro Guam Northern Mariana Islands
Chechen Russia Chechnya, north Caucasus Georgia, Germany, Jordan, Kazakhstan,
Kyrgyzstan, Syria, Turkey, Uzbekistan
Chichêwa Malawi West central and southwestern Botswana, Mozambique, Zambia,
Zimbabwe
Chinese, China Northern and southwestern China Brunei, Cambodia, Indonesia, Laos,
Mandarin Malaysia, Mauritius, Mongolia,
Philippines, Russia, Singapore, Taiwan,
Thailand, United Kingdom, USA,
Vietnam
Chukchee Russia Northeastern Siberia
Chumash USA Southern California coast Extinct
Dakota USA Northern Nebraska, southern Minnesota, Canada
North and South Dakota, northeastern
Montana
Diyari Australia South Australia, Leigh Creek Extinct
Dyirbal Australia Northeast Queensland
Eastern Malaysia Northeast Sabah, Sandakan,
Kadazan Labuk-Sugut, and Kinabatangan districts

Emai Vanuatu Emae

English United 104 others
Kingdom, USA
Fongbe Benin South central Togo
Fox USA Eastern Kansas-Nebraska border and
central Oklahoma
Fula Guinea Northwest, Fouta Djallon area Gambia, Guinea-Bissau, Mali, Senegal,
Sierra Leone
Gapapaiwa Papua New Milne Bay Province
Guinea
German Germany 40 others
Hausa Nigeria Throughout northern Nigeria Benin, Burkina Faso, Cameroon, CAR,
Chad, Congo, Eritrea, Germany, Ghana,
Niger, Sudan, Togo
(Modern) Israel Australia, Canada, Germany, Palestinian
Hebrew West Bank and Gaza, Panama, United
Kingdom, USA
Hindi India Throughout northern India Bangladesh, Belize, Botswana, Germany,
Kenya, Nepal, New Zealand, Philippines,
Singapore, South Africa, Uganda, UAE,
United Kingdom, USA, Yemen, Zambia
Hopi USA Northeastern Arizona, southeastern
Utah, northwestern New Mexico
Hua Papua New Eastern Highlands Province, Goroka
Guinea District
Hungarian Hungary Australia, Austria, Canada, Israel,
Romania, Slovakia, Slovenia, Ukraine,
USA, Yugoslavia
Ilokano Philippines Northern Luzon USA
Indonesian Indonesia Netherlands, Philippines, Saudi Arabia,
Singapore, USA
(cont.)
Language Primary Regions within Secondary Special
country primary country countries note

Japanese Japan 26 others

Jaqaru Peru Lima Department, Yauyos Province
Javanese Indonesia Java and resettlements in Irian Jaya, Malaysia (Sabah), Netherlands,
Sulawesi, Maluku, Kalimantan, and Singapore
Sumatra
Kashaya USA Northern California Nearly
extinct
Kawaiisu USA California; Mojave Desert Possibly
extinct
Khasi India Assam; Meghalaya, Khasi-Jaintia hills; Bangladesh
Manipur; West Benga
Khmer Cambodia China, France, Laos, USA, Vietnam
Kikerewe Tanzania Northwestern Ukerewe Island, southern
Lake Victoria, Kibara
Kinande Democratic Nord-Kivu Province
Republic of
Congo
Klamath USA South central Oregon Nearly
extinct
Kolami India Northwestern Kolami: Maharashtra,
Yavatmal, Wardha, and Nanded districts;
Andhra Pradesh; Madhya Pradesh
Southeastern Kolami: Andhra Pradesh,
Adilabad District; Maharashtra,
Chandrapur, and Nanded districts
Lango Uganda Lango province, north of Lake Kyoga
Lushootseed USA Washington, Puget Sound area Nearly
extinct
Madurese Indonesia Island of Madura, Sapudi Islands, Singapore
northern coastal area of eastern Java
Malay Malaysia Peninsular Malaysia, Sabah, and
Sarawak
Johore Malay Malaysia Johor state
Ulu Muar Malay Malaysia Southeast of Kuala Lumpur, Ulu Muar
District
Malayalam India Kerala, Laccadive Islands, and Bahrain, Fiji, Israel, Malaysia, Qatar,
neighboring state Singapore, UAE, United Kingdom
Mangarayi Australia Mataranka and Elsey stations, Northern
Territory
Marathi India Maharashtra and adjacent states Mauritius
Marquesan French Marquesas Islands: Hatutu, Nuku
Polynesia Hiva, Ua Huka, Ua Pou islands
South Marquesan: Marquesas Islands:
Hiva Oa, Tahuta, Fatu Hiva islands
Mende Sierra Leone South central Liberia
Miya Nigeria Bauchi State, Ganjuwa LGA, Miya town
Mongolian Mongolia Buryat ASSR of Russia and Issyk-Kul Kyrgyzstan, Russia, Taiwan
Oblast of Kyrgyzstan
Nadrogā Fiji Fiji Islands, western half of Viti Levu,
Waya Islands
Nakanamanga Vanuatu Efate, Shepherd islands
Namakir Vanuatu Efate, Shepherd islands
(cont.)
Language Primary Regions within Secondary Special
country primary country countries note

Nāti Vanuatu Malakula

Navajo USA Northeastern Arizona, northwestern New
Mexico, southeastern Utah,
southwestern Colorado
Ndebele Zimbabwe Matabeleland, around Bulawayo Botswana
Nez Perce USA Northern Idaho
Niuafo’ou Tonga Niuafo’ou and ’Eua islands
Orokaiva Papua New Oro Province Popondetta District
Guinea
Oykangand Australia Wrotham Park, Kowanyama, Edward
River, Queensland
Paamese Vanuatu Paama, east Epi, Vila
Pacoh Vietnam Bl’nh Tri Thien Province Laos
Pali India (literary language of Buddhist scriptures) Myanmar, Sri Lanka Extinct
Piro (Yine) Peru East central Urubamba River area
Ponapean Micronesia Pohnpei Island, Caroline Islands
Quechua Peru Northeast Huánuco Department
Raga Vanuatu Pentecost island
Rotuman Fiji Rotuma Island
Roviana Solomon Islands North central New Georgia, Roviana
Lagoon, Vonavona Lagoon; Western
Province
Sahaptin USA North central Oregon
Sanskrit India
Sekani Canada North central British Columbia
Serbo-Croatian Yugoslavia Albania, Australia, Austria,
Bosnia-Herzegovina, Bulgaria, Canada,
Croatia, Germany, Greece, Hungary,
Italy, Macedonia, Romania, Russia,
Slovakia, Slovenia, Sweden,
Switzerland, Turkey
Sereer Senegal West central Senegal and the Sine and Gambia
Saloum River valleys
Siroi Papua New Madang Province, Saidor District
Guinea
Siswati Swaziland Mozambique, South Africa
Slave Canada Northwest Territories, northern Alberta
Southeast Vanuatu Southeast Ambrym Island
Ambrym
Spokane USA Northeastern Washington
Sundanese Indonesia Western third of Java Island
Sye Vanuatu Erromango Island
Tagalog Philippines Manila, most of Luzon, and Mindoro Canada, Guam, Midway Islands, Saudi
Arabia, UAE, United Kingdom, USA
Tamil India Tamil Nadu and neighboring states Bahrain, Fiji, Germany, Malaysia
(Peninsular), Mauritius, Netherlands,
Qatar, Réunion, Singapore, South Africa,
Sri Lanka, Thailand, UAE, United
Kingdom
Tarok Nigeria Plateau State, Kanam, Wase, and
Langtang LGAs; Gongola State, Wukari
LGA
Tawala Papua New Milne Bay Province
Guinea
Telugu India Andhra Pradesh and neighboring states Bahrain, Fiji, Malaysia, Mauritius,
Singapore, UAE
(cont.)
Language Primary Regions within Secondary Special
country primary country countries note

Tohono USA South central Arizona Mexico

O’odham
Trukese Micronesia Chuuk Lagoon, Caroline Islands, some Guam
on Ponape
Turkish Turkey 35 other countries
Ulithian Micronesia Ulithi, Ngulu, Sorol, Fais islands, eastern
Caroline Islands
Umpila Australia Cape York Peninsula, north of Cairns
Urdu Pakistan
Vietnamese Vietnam Afghanistan, Bahrain, Bangladesh,
Botswana, Fiji, Germany, Guyana, India,
Malawi, Mauritius, Nepal, Norway,
Oman, Qatar, Saudi Arabia, South
Africa, Thailand, UAE, United
Kingdom, Zambia
Warlpiri Australia Northern Territory, Yuendumu, Ali
Curung Willowra, Alice Springs,
Katherine, Darwin, and Lajamanu
Yoruba Nigeria Oyo, Ogun, Ondo Osun, Kwara, and Benin, Togo, United Kingdom, USA
Lagos states; and western LGAs of Kogi
State
Yupik, Central USA Nunivak Island, Alaska coast from
Alaskan Bristol Bay to Unalakleet on Norton
Sound and inland along Nushagak,
Kuskokwim, and Yukon rivers
 Abbreviations used in
morpheme glosses

3sg 3rd person singular inf infinitive

abs absolutive intr intransitive
adj adjective loc locative
af actor focus mid middle
aor aorist mulrec multiple
ass.col associate reciprocal
collective mulser multiple serial
aug augmentative distribution
caus causative n. noun
cl classifier neg negative
comp completive n.inten non-intentive
compr comprehensive n.ser non-serious
cont continuative nml nominal(izer)
ctr control nonfut non-future
def definite nonpres non-present
des desiderative npast non-past
dim diminutive obj object
diss dissimulation obv obviative
dist distributive op operator
dom dominant perf perfective
du dual pl plural
dur durative poss possessive
dx deictic pp past participle
erg ergative pres/prs present
exas exasperation pro pronoun
excl exclusive rec recent completive
fut future (E. Kadazan)
imp imperative rec recessive
imperf/impf imperfective rep repetitive
incl inclusive rf referent focus

xxi
xxii Abbreviations used in morpheme glosses

s.o. someone trs transitive subject

s.t. something uf undergoer focus
sfx suffix un.dist unequal
sg singular distribution
sp species v verb
subj subject vbl verbalizer
tr/trans transitive wp witnessed past
1 Introduction

Repetition is encountered in every language and affects all types of linguistic

units. One finds reiteration of phrases, as in the Quechua example in (1a), of
words, as in the Amele example in (1b), and even of single segments, as in the
onomatopœtic ideophone from English (1c):

(1) a. Chawra mishi alpurhapita [horqorkur kutirkUchir] [horqorkur

kutirkUchir] huk umallantashi chunka ishkayta yupaykun.
‘Then the cat, repeatedly removing the head from the saddlebag and
returning it, counts the one head twelve times.’ (Weber 1989:323)1
b. Odeceb fojen. Rum oso eu fojen. Ihoc leceb haun rum oso na li fojen.
Ihoc leceb haun rum oso na li fojen. Ihoc leceb oso na ha li ihoc leceb
haun jo oso na toni nu lena. Fojen. Ihoc leceb jo oso na toni nu len eu na
fojen ihoc len. Rum cunug ca foji hedon. Odimei madon, “Quila qa ihoc,”
don.
‘Then she vomited. She vomited in that room. Then after she had filled
that room with vomit she went to another room and filled that with vomit
and then filled another room with vomit. Then she went down and went to
another house. She vomited there. She filled all those rooms with vomit.
Then she finished vomiting and said to him, “Now that is enough.”’
(Roberts 1987:255–56)
c. English: [ʃ-ʃ-ʃ-ʃ-ʃ-ʃ . . .] ‘be quiet!’

This book concerns itself with a select subtype of repetition, namely gram-
matical doubling or duplication effects within words, illustrated by the forms in
(2). Word-internal reduplication may be partial (a) or total (b, c); it may involve
perfect identity between copies (c) or exhibit imperfect identity (a, b).2

(2) a. Hausa kira ‘call’ kik-kira (pluractional)

b. Amele bala-doʔ ‘to tear’ bala-bulu-doʔ (irregular iterative)
c. Warlpiri kamina ‘girl’ kamina-kamina (plural)

The mechanism of reduplication and manner in which copies can differ from
each other have been a foundational concern in theoretical and descriptive

1
2 Introduction

linguistics over the past twenty-five years. They constitute the central interest
of this book.

1.1 Two approaches to duplication

Speaking broadly, two general approaches to duplication are possible: phono-
logical copying and morpho-semantic (MS) feature duplication. Phonological
copying is an essentially phonological process that duplicates features, seg-
ments, or metrical constituents, as in the example of ‘eat’ in (3a). Under MS
feature duplication, two identical sets of abstract syntactic/semantic features
(‘EAT’, in (3b)) are provided by the grammar and spelled out independently
(3b).

(3) a. [EAT] → [eat] → [eat-eat]

Spellout Phonological copying
b. [EAT] → [EAT] [EAT] → [eat-eat]
MS feature duplication Spellout

While theories of morphological reduplication have focused on the duplica-

tion mechanism of phonological copying, it is the thesis of the present work that
both mechanisms are needed and that their empirical domains of application
are nearly complementary. In introducing this vision it is useful first to consider
cases which clearly instantiate the two poles of duplication.
The clearest examples of phonological copying are those in which small
pieces of phonological structure are copied to satisfy a phonological well-
formedness constraint. In Hausa (Chadic), for example, the most productive
noun pluralization suffix is -oCi, whose medial consonant is a copy of the final
consonant of the noun stem (Newman 2000:431–32). As noted by Newman,
who calls this phenomenon “pseudoreduplication” (p. 511), copying is driven
by the need to flesh out an underspecified suffixal consonant (or, on another
view, to provide an onset to the final syllable of the suffix). The data shown
here are taken from Newman 2000:432:

(4) a. bindigà bindig-ogi ‘gun/guns’

b. fannı̀ fann-oni ‘category/categories’
c. hùkumà hukum-omi ‘governmental body/bodies’

The well-known Yoruba gerundive construction constitutes another instance

of phonologically driven copying (Akinlabi 1985, Pulleyblank 1988, Kawu
1998).3 Gerunds are formed by prefixation of a high front vowel marked with a
high tone (ı́-), preceded by a copy consonant whose presence Akinlabi and Kawu
 Two approaches to duplication 3

attribute to the need for Yoruba syllables to begin with a consonantal onset.
Although Alderete et al. (1999) treat Yoruba gerunds as resulting from mor-
phological reduplication, Akinlabi and Kawu instead view them as an instance
of what Newman (2000) calls “pseudoreduplication” or what Urbanczyk (1998)
calls “non-reduplicative copying.” The data below are taken from Kawu
1998:3:

(5) Verb Gerund

gbé gb-ı́-gbé ‘take; taking’

jε j-ı́-jε ‘eat; eating’
wɔ̀ w-ı́-wɔ̀ ‘enter; entering’
wã w-ı́-wã ‘measure; measuring’
bú b-ı́-bú ‘insult; insulting’

In contrast to phonologically driven local duplication of phonological mate-

rial are clear cases of MS feature duplication. Such cases typically have mor-
phological or syntactic, rather than phonological, motivation; they duplicate
more than a single phonological element; and they may not even result in
phonological identity. Consider, for example, Modern Hebrew VP-fronting,
a construction in which the verb is spelled out in two positions (Landau
2003:7):4

(6) lirkod, Gil lo yirkod ba-xayim

to-dance, Gil not will-dance in-the-life
‘Dance, Gil never will’

The source of MS feature duplication will vary across theoretical frameworks;

on Landau’s analysis, verb doubling results from the pronunciation of two links
of a chain (i.e. copies of identical feature bundles in the terms of Chomsky
1995; 2000). What is relevant is that the duplication in (6) cannot be analyzed
as phonological copying. It is not motivated by phonological well-formedness
and the two copies are not even phonologically (or morphologically) identical.
While there is full inflection on the lower copy of the verb, the higher verb is
an infinitive. Divergent spellout of this sort is a clear sign that what is being
copied is an abstract syntactic or semantic aspect of the representation, rather
than phonological material.
Despite the existence of these two very different mechanisms for duplicating
grammatical material, virtually no attention has been given in the reduplication
literature to arguing for one over the other. Instead, theoretical approaches to
morphological reduplication have focused nearly exclusively on the idea that
phonological copying occurs to flesh out a skeletal reduplicative morpheme (see,
4 Introduction

for example, Marantz & Wiltshire 2000 for a recent overview).5 The example
in (7), using the Warlpiri form for ‘girls’ (Nash 1986:130), illustrates total redu-
plication under the dominant derivational approach taken in the 1980s (Marantz
1982; Clements 1985; Kiparsky 1986; Mester 1986; Steriade 1988). Red, a
skeletal Prosodic Word affix marking the plural, is fleshed out by copying the
base segments and associating the copies by rule to the Red template.
(7) Morphological reduplication by phonological copying
Affixation Copy and Association
Red plural + PWd → Red PWd


kamina
 
kamina kamina
‘girl’ ‘girls’

The more recent Base-Reduplication Correspondence Theory (BRCT)

approach to reduplication in the Optimality Theory literature makes the same
assumptions, with the additional proposal that copying into the Red mor-
pheme is coerced by violable constraints that compel Red to be identical to the
base (McCarthy & Prince 1993; 1995a; for an overview, see Kager 1999). In
the Chumash example in (8) (Applegate 1976), Red is required, by the output
constraint Red = ␴ ␮␮ , to instantiate a bimoraic syllable and, by BR-Faith,
to correspond segmentally to the material in the base. IO-Faith » Red = ␴ ␮␮
prevents the base from truncating:
(8) Morphological reduplication by BR correspondence
...............................................

Red, čh umaš IO-Faith Red=␴ ␮␮ BR-Faith

) a. čh um-čh umaš aš

h h
b. č umaš-č umaš aš!

c. čh um-čh um aš!

A definitive rationale for providing a phonological copying analysis, rather

than a MS feature duplication analysis, of phenomena of this type has not been
provided in the literature.6 One likely motivation for the focus on phonologi-
cal copying is the variety of phonological modifications that often accompany
morphological reduplication.7 However, as is argued at length in Chapters 3
and 4 of this book, phonological modification is not restricted to morphological
reduplication and cannot be used as a criterion to determine the doubling mech-
anism. Truncation is found not only in morphological constructions which are
 Two approaches to duplication 5

not reduplicative (see, for example, Weeda 1992; McCarthy & Prince 1999b and
Chapter 3), but is found even in clearly “syntactic” reduplication constructions
where MS feature duplication is unequivocally at work.
Consider, for example, the phenomenon in Fongbe (Kwa) which Lefebvre
and Brousseau (2002) analyze as syntactic doubling of the verb. Verb dou-
bling occurs in four syntactic constructions: temporal adverbials (9a), causal
adverbials (9b), factives (9c), and predicate clefts (9d). In each case, an extra
copy of the verb appears initially in the verb phrase. The fronted copy of the
verb can be identical to the main verb; crucially, for some speakers, it can also
be truncated to its first syllable (Collins 1994, cited in Lefebvre & Brousseau
2002:505):

(9) a. sı́sɔ́ ∼ sı́ Kɔ́kú sı́sɔ́ tlóló bɔ̀ xὲsı́ ı̀ Bàyı́
tremble Koku tremble as.soon.as and fear get Bayi
‘As soon as Koku trembled, Bayi got frightened’
b. sı́sɔ́ ∼ sı́ Kɔ́kú sı́sɔ́ útú xὲsı́ ı̀ Bàyı́
tremble Koku tremble cause fear get Bayi
‘Because Koku trembled, Bayi got frightened’
c. sı́sɔ́ ∼ sı́ é-è Bàyı́ sı́sɔ́ ɔ́, vέ nú mi
tremble op-res Bayi tremble, def bother for me
‘The fact that Bayi trembled bothered me’
d. sı́sɔ́ ∼ sı́ wὲ, Kɔ́kú sı́sɔ́
tremble it.is Koku tremble
‘It is tremble that Koku did’

In Fongbe, the only difference between the main verb and its fronted copy is
the optional truncation; both copies can be assumed to have phonologically and
semantically identical inputs.
A second likely motivation for the lack of attention to the issue of whether
phonological copying or MS feature duplication underlies reduplication is
simply that so many cases of canonical morphological reduplication appear
amenable to either approach. By way of illustration, consider again the case of
total pluralizing reduplication in Warlpiri, for which a standard phonological
copying analysis was presented in (7). Suppose, instead, that plural formation
in Warlpiri is governed by a morphological construction, such as the one shown
in (10), which dictates double insertion of the singular form of the noun. In
this case, there would be no phonological copying. Such an analysis differs
from the account of Hebrew argued for by Landau only in that it is the mor-
phology rather than the syntax that provides multiple instantiation of identical
features.
6 Introduction

(10) Morphological reduplication by MS feature duplication:

Spellout
[girl +girl] plural → [girl +girl] plural

 
kamina kamina

While the analysis in (10) is logically possible, the prevailing intuition that
word reduplication should be treated as phonological copying (as in (7)) has
precluded the development of a detailed theory of word reduplication based on
MS feature duplication.
This book addresses this asymmetry in previous approaches to morphological
reduplication, surveying a wide range of duplication effects and developing
numerous arguments to support the use of MS feature duplication, formalized
as Morphological Doubling Theory, for morphological reduplication, while
reserving phonological copying as the correct analysis of purely phonologically
driven duplication.

1.2 Morphological Doubling Theory

The essential claim of Morphological Doubling Theory (MDT) is that redu-
plication results when the morphology calls twice for a constituent of a given
semantic description, with possible phonological modification of either or both
constituents. MDT has roots in proposals by Hyman, Inkelas, and Sibanda
(to appear) and also resonates in important respects with Yip’s (1997; 1998)
Repeat(Stem) constraint, with the large body of work on Bantu reduplica-
tion by Downing (1997; 1998a; 1998b; 1998c; 1998d; 1998e; 1999a; 1999b;
1999c; 2000a; 2000b), with the Reduplicative Blending Theory of Sherrard
(2001) and with the word-and-paradigm approach of Saperstein (1997), who
also argues for a type of double stem selection and eschews the use of a mor-
pheme “Red” which phonologically copies a base. MDT also has points of
contact with Steriade’s Lexical Conservatism approach to allomorphy; see, for
example, Steriade 1997; 1999. Arguments supporting elements of the MDT
approach to reduplication can be found in Pulleyblank (to appear).
Any morphological analysis requires an explicit morphological framework.
In this book, MDT is couched within Sign-Based Morphology (SBM; Orgun
1996; 1997; 1999; Orgun & Inkelas 2002), a flexible morphological frame-
work which can incorporate many different approaches to morphology. SBM,
discussed more fully in §1.2.2, is compatible both with item-based and with
realizational morphology; it is compatible with Optimality Theory and with
 Morphological Doubling Theory 7

rule-based theories of phonology. The SBM framework makes it easy to discuss

and depict morphological constructions, a centerpiece of the approach to redu-
plication developed here.

1.2.1 The morphology of reduplication

MDT assumes the basic structure in (11) for morphological reduplication.
A reduplicated stem (or “reduplication construction,” to use a theory-neutral
descriptive term) has two daughters that are featurally identical, i.e., mean the
same thing:
(11) [output][F + some added meaning]

/intput/[F] /input/[F] where [F] = semantic feature bundle

By requiring the two sisters to be identical only semantically, MDT makes

a prediction which sets it apart from all phonological copying theories: other
kinds of deviation, whether morphotactic or phonological, between the two
copies are expected to be possible.
A theory much like MDT is anticipated by Moravcsik (1978), who writes:
Constituents to be reduplicated may in principle be definable . . . either by
their meaning properties only, or by their sound properties only, or in reference
to both. They may, in other words, be either semantic-syntactic constituents,
such as one or more semantic-syntactic features, or morphemes, or words, or
phrases, or sentences, or discourses; or they may be phonetic-phonological
terms, such as one or more phonetic-phonological features, or segments, or
syllables; or they may be morphemes of a particular phonetic shape, or sen-
tences of a particular number of phonetic segments; etc. (pp. 303–304)

Moravcsik wrote this passage at a time when it was thought that the first type
of reduplication did not exist; she states (p. 305) that no language possesses
a reduplicative construction “which involves the reduplication of a syntactic
constituent regardless of its form . . . in reduplication reference is always made
both to the meaning and to the sound form of the constituent to be reduplicated.”
Similar statements are made on p. 315, fn. 8.
Our subsequent research has revealed some of the missing data that sup-
ports Moravcsik’s original hypothesis that reduplication does not necessarily
involve phonological identity. A number of morphological constructions require
semantic identity, semantic similarity or (in some cases) semantic dissimilarity
between their daughters. Among the cases of this sort, discussed in Chapter 2,
are languages exhibiting “synonym compounding,” in which the two members
of the compound are phonologically distinct, perhaps etymologically distinct
8 Introduction

synonyms (e.g. Khmer peel-weeliə ‘time,’ from Sanskrit peel ‘time’ + Pali
weeliə ‘time’; Ourn & Haiman 2000:485). In the Khmer and Vietnamese
examples below, the meanings of these constructions can be lexicalized but
frequently are the same as the meaning of the individual parts. Page numbers
for Khmer and Vietnamese refer to Ourn and Haiman 2000 and Nguyen 1997,
respectively:

(12) a. Khmer synonym compounds

cah-tum ‘old + mature’ ‘village elder’ 485
kee-mɔrdɔk ‘heritage + heritage’ ‘legacy’ 501
cɑmnəj-ʔahaa(r) ‘food + food’ ‘food’ 485
ʔaar-kɑmbaŋ ‘secret + secret’ ‘secret’ 500
cbah-prakɑt ‘exact + exact’ ‘exact’ 500
b. Vietnamese synonym compounds
ma.nh-kho
e ‘strong + strong’ ‘well in health’ 67
do’ b
ân ‘dirty + dirty’ ‘filthy’ 67
´
lu’ò’i-biêng ‘lazy + lazy’ ‘slothful’ 67
toi-l
.̌ ˜
ôi ‘offense + fault’ ‘sin’ 70
kêu-go.i ‘to call + to call’ ‘to call upon, appeal’ 70

It is argued in Chapter 2 that any theory with the ability to model these
constructions already has the ability to model reduplication and does not need
recourse to extra mechanisms like a Red morpheme or base-reduplication
correspondence. Some related arguments against a morphemic approach to
reduplication can be found in Saperstein 1997.
Another type of case discussed in Chapter 2 is divergent allomorphy, in which
the two copies – “base” and “reduplicant,” to use traditional terminology –
differ in their morphological makeup. Divergent allomorphy provides striking
evidence for the MS feature duplication approach because it clearly shows that
the two copies can have different morphological inputs, as long as they are
semantically matched. Recall that in Hebrew VP-fronting, alluded to in (6), the
two copies of the verb appear in different forms. Chechen (North Caucasian)
likewise illustrates the possibility of divergent allomorphy when a construction
calls for only semantic identity between independent copies. Chechen exhibits
syntactic reduplication to satisfy the requirements of a second position clitic
(Conathan & Good 2000; see also Peterson 2001 on the closely related language
Ingush). As shown in (13), from Conathan and Good (2000:50), chained clauses
are marked by an enclitic particle ʔa, which immediately precedes the inflected,
phrase-final, main verb. The enclitic must be preceded by another element in the
same clause. Two types of constituent may occur before the verb (and enclitic
particle) in the clause: an object (13a), or a deictic proclitic or preverb (13b).
 Morphological Doubling Theory 9

If neither of these elements is present in a chained clause, then the obligatory

pre-clitic position is filled by reduplicating the verb (13c):8

(13) a. Cickuo, [chʔaara =ʔa gina]VP , ʔi buʔu

cat.erg [fish =& see.pp] VP 3s.abs B.eat.prs
‘The cat, having seen a fish, eats it.’
b. Ah- mada, [kiekhat jaaz =ʔa dina]VP , zhejna dueshu
Ahmad.erg [letter write =& D.do.pp]VP book D.read.prs
‘Ahmad, having written a letter, reads a book.’
c. Ah- mad, [ʕa =ʔa ʕiina]VP , dʕa-vaghara
Ahmad [stay.inf =& stay.pp] VP dx.v.go.wp
‘Ahmad stayed (for a while) and left.’

The Chechen reduplicant occurs in infinitive form, while the main verb is
inflected. Inflected verbs require a different form of the verb stem from that used
in the infinitive; in some cases the stem allomorphy is clearly suppletive, e.g.
Dala ‘to give’ vs. lwo ‘gives,’ or Dagha ‘to go’ vs. Duedu ‘goes.’ As Conathan
and Good (2000:54) observe, the result is that Chechen can exhibit suppletive
allomorphy differences between base and reduplicant; they cite as one example
the reduplicated verb phrase Dagha ‘a Duedu, based on ‘go.’
What is going on in Chechen is the use of two verbs with (almost) the same
meaning. MS feature duplication allows for divergent allomorphy of this sort,
since there is no requirement that multiple tokens of concurring feature bundles
be expressed identically. Divergent allomorphy in reduplication is, however,
impossible to generate with phonological copying, since the normal base-to-
reduplicant copying process cannot introduce an allomorph into the reduplicant
that is not present in the base. Therefore, evidence that morphological redupli-
cation exhibited divergent allomorphy would provide strong support that MS
feature duplication, rather than phonological copying, is the driving force in
reduplication.
Divergent allomorphy does indeed occur in morphological reduplication.
Consider, for example, a fragment of data from Sye (Central-Eastern Oceanic;
Crowley 1998; 2002d).9 Sye presents the type of morphological divergence in
which reduplicant and base contain different suppletive allomorphs of the same
morpheme. The main points of Sye reduplication are these:

(14) a. Most verb roots in Sye appear in two different shapes: Stem1 and Stem2
b. Each affixation construction selects for one of the two stem shapes
c. Reduplication in morphological contexts calling for Stem1 yields two
copies of Stem1
d. Reduplication in contexts that call for Stem2 surfaces as Stem2-Stem1
10 Introduction

Examples of some stems showing this allomorphy are shown below:

(15) Stem1 Stem2 Gloss

evcah ampcah ‘defecate’ Crowley 2002d:704

ocep agkep ‘fly’ Crowley 1998:84
omol amol ‘fall’ Crowley 1998:79

Reduplication in Sye, which is total and has an intensifying meaning, is

illustrated in (16). As seen, the verb ‘fall’ is reduplicated and combined with
the third person future prefix, which conditions the Stem2 form of a verb. The
two copies of ‘fall’ assume different stem shapes: amol is Stem2 and omol is
Stem1:

(16) cw-amol-omol ‘3.fut-fall2 -fall1 = they will fall all over’

Because the phonological relationship between Stem1 and Stem2 is not fully
predictable, the “reduplicant” in a Sye reduplicated verb cannot always be
described as a phonological copy of the “base,” as phonological theories of
reduplication would require. Rather, in at least some cases the reduplicant and
base consist of different suppletive allomorphs of the same morpheme.
Ndebele (Nguni; Bantu) presents a different kind of divergent allomorphy in
reduplication: the reduplicant contains semantically empty morphs not present
in the base (Downing 1999a; 2001; Sibanda 2004; Hyman, Inkelas & Sibanda
to appear). One such morph is the stem-forming -a (see Chapter 2 for a fuller
discussion). Reduplication, which targets the verb stem and contributes the
meaning that “the action is done for a short while before it stops or is done
from time to time, perhaps not very well” (Sibanda 2004:282), truncates the
first copy of the verb stem to two syllables. When the root is itself disyllabic
or longer, the reduplicant consists of its initial two syllables. The outputs of
reduplication here and throughout are shown with the final vowel in place, the
standard citation form for verb stems. Data are taken from Hyman, Inkelas, and
Sibanda to appear (HIS) and Sibanda 2004 (S):

(17) nambith-a nambi+nambith-a ‘taste’ HIS

thembuz-a thembu+thembuz-a ‘go from wife to wife’ HIS
hlikihl-a hliki+hlikihl-a ‘wipe’ S:289
dlubulund-a dlubu+dlubulund-a ‘break free of control S:289
tshombuluk-a tshombu-tshombuluk-a ‘become unrolled’ S:289

There are some conditions, however, under which empty morphs can appear
in the reduplicant which are not present, because nothing motivates their pres-
ence, in the base. One condition is when the verb root is monosyllabic. As
 Morphological Doubling Theory 11

illustrated by the cases in (18), based on example (6) from Hyman, Inkelas,
and Sibanda to appear (see also Downing 2001; Sibanda 2004:284), subjunc-
tive and perfective suffixes are not in the domain of reduplication; they cannot
be duplicated. The disyllabicity requirement on the reduplicant must be satis-
fied, however, and in Ndebele compels the insertion of the semantically empty
stem-forming morpheme -a:

(18) lim-e lima-lime ‘cultivate (subjunctive)’

lim-ile lima-limile ‘cultivate (perfective)’
thum-e thuma-thume ‘send (subjunctive)’
thum-ile thuma-thumile ‘send (perfective)’

Example (19) illustrates the MDT approach to reduplication of the stem

lim- in Ndebele. Both the left-hand stem (“reduplicant”) and the right-hand
stem (“base”) have the same semantic description, i.e. ‘cultivate.’ This
morphosyntactic agreement is not disrupted by the fact that the reduplicant has
an input morph, semantically empty -a, that the base does not.

(19) [lima-lim-][F + ‘here and there’]

/ lim-a/[F] /lim/[F] where F = [CULTIVATE]

Suppletive allomorphy in Sye reduplication is illustrated in (20). Both stems

have the meaning of ‘fall’ and are thus semantically identical, even though they
differ in their phonology.

(20) [amol-omol] [F + ‘random’]

/amol/[F] /omol/[F] where F = [FALL]

In summary, morphological reduplication in MDT is double selection (inser-

tion) of a morphological constituent such as stem or root. There is no inher-
ent morphological asymmetry between the daughters; there is no morpheme
Red.The terms “base” and “reduplicant” in fact have no formal status, despite
being descriptively handy in some cases.

1.2.2 Constructions in morphology

MDT makes heavy use of the concept of a morphological construction to handle
reduplicative semantics and phonology, the latter being the primary focus of
Chapters 3–6. A few words of explanation are in order here, especially in light
of the great variety of morphological frameworks in use in the literature.
12 Introduction

A “construction,” broadly speaking, is any morphological rule or pattern that

combines sisters into a single constituent. Each individual affix, compounding
rule, truncation construction, and/or reduplication process is a unique mor-
phological construction. Constructions can be related to each other under the
rubric of more general “meta-constructions,” which capture commonalities in
the morphological component of the grammar.
The formalization of constructions and meta-constructions varies across mor-
phological theories, some of which explicitly use the term “construction” and
some of which do not. For Bochner (1992), the equivalent of a morphologi-
cal construction would be a “pattern” relating words in the same inflectional
or derivational paradigm, e.g. for English nouns, the rule in (21a). In a real-
izational morphological theory like that of Anderson (1992), the comparable
affixation construction would take the form of a phonological rule conditioned
by the features [Noun, plural] (21b). In an item-based morphological theory
like that of Lieber (1980) or Selkirk (1982), affixation constructions take the
form of subcategorization frames specifying features of each affix and the stem
it combines with (21c):

(21) a. <[X]Noun, singular ↔ [Xz]Noun, plural >

b. [Noun, plural]
/X/→/Xz/
c. [[ ]Noun, singular z]Noun, plural

The implementation of morphological construction used here draws heavily

on Sign-Based Morphology (SBM; Orgun 1996; 1997; 1999; 2002; Orgun &
Inkelas 2002), a versatile framework capable of capturing the insights of all
three frameworks sketched in (21). Readers with attachments to particular mor-
phological frameworks may freely translate the MDT constructions into the
framework of their choice, as long as two crucial aspects of constructions are
respected: the ability to encode idiomatic semantics and the ability to encode
morphologically conditioned phonology.
In SBM constructions (and meta-constructions) are grammatical primitives,
elaborated versions of phrase-structure rules which encode the semantic, syn-
tactic, and phonological mappings between daughters and mothers. The mor-
phological grammar of any language consists of a set of constructions, which
combine with roots or with each other to form complex words. The English
noun plural construction is sketched in (22), temporarily omitting phonologi-
cal details. Like all of the statements in (21), it specifies that the noun plural
construction has two daughters – a nonplural noun and the suffix /z/ – and that
its mother is a plural noun.
 Morphological Doubling Theory 13

(22) Affixation construction: Example:

⎡ Syntax = N ⎤ ⎡ Syntax = N ⎤
⎢Semantics = plural(X)⎥ ⎢Semantics = 'books' ⎥
⎢ ⎥ ⎢ ⎥
⎣ Phonology = g(Y) ⎦ ⎣ Phonology = [bυks] ⎦

⎡ Syntax = N ⎤ ⎡ Syntax = N ⎤
⎢Semantics = X ⎥ /z/ ⎢Semantics = 'book' ⎥
⎢ ⎥ ⎢ ⎥ /z/
⎣ Phonology = Y ⎦ ⎣ Phonology = [bυk] ⎦

As mentioned earlier, the constructional approach to morphology permits a

unified approach to the wide range of morphological construction types. Shown
below are SBM constructions for compounding (using noun-noun compounding
in English as an example) and for truncation (using nickname formation in
English as an example):

(23) Compounding schema: Example:

⎡ Syntax = N ⎤ ⎡ Syntax = N ⎤
⎢Semantics = 'Ny used for Nx'⎥ ⎢ Semantics = 'book case' ⎥
⎢ ⎥ ⎢ ⎥
⎣ Phonology = g(Px, Py) ⎦ ⎣ Phonology = [' bυk keis] ⎦
'

⎡ Syntax = N ⎤ ⎡ Syntax = N ⎤ ⎡ Syn = N ⎤ ⎡ Syn = N ⎤

⎢Semantics = Semx⎥ ⎢Semantics = Semy⎥ ⎢ Sem = 'book' ⎥ ⎢ Sem = 'case' ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ Phonology = Px ⎦ X ⎣ Phonology = Py ⎦ Y ⎣ Phon = /bυk/⎦ X ⎣ Phon = /keis/⎦ Y

1.2.3 Constructional semantics

The constructional schemata in (22–23) differ in a couple of ways from the
subcategorization frames or realizational rules with which readers may be more
familiar. One difference is that the semantics of the mother node is specified, in
each construction, as a particular function of the semantics of the daughters. This
function could be a simple percolation function, as in Lieber (1980); percolation
is appropriate for constructions in which meaning is entirely compositional,
and no aspect of the meaning of the whole needs to be stipulated. However, the
semantic function could also involve some information not derivable from either
daughter. This is appropriate for cases of idiomaticity or exocentricity. The
semantics of reduplication, at least from a naı̈ve descriptive point of view, vary
from the iconic to the potentially quite idiomatic, supporting a constructional
approach to morphology.
A natural null hypothesis as to the meaning of reduplication constructions
might be that the meaning of a given reduplication construction is a purely
iconic function of the meaning of its daughters: for nouns, plurality; for verbs,
iterativity or pluractionality; for adjectives and adverbs, intensity, and so forth.
14 Introduction

Many examples of reduplication do indeed meet this description. In Rotuman

(Central-Eastern Oceanic), for example, partial verb reduplication marks
“repetition, frequency, continuance or spatial extension of a state or action”
(Schmidt 2002:825); “The progressive tense or continuous aspect is formed by
reduplicating the verb (which can also signify repetition or frequency) and/or
adding a pronominal suffix” (ibid. p. 827). The prevalence of semantically iconic
reduplication can readily be seen, for example, in the brief survey conducted
by Moravcsik (1978) and the more extensive ones by Niepokuj (1997) and
Kiyomi (1993); the latter catalogs a very wide range of reduplication construc-
tions in the Bantu, Papuan, Austroasiatic, and Malayo-Polynesian language
families.
Iconic semantics is not, however, the general rule. Reduplication, especially
partial reduplication, is associated cross-linguistically with all sorts of mean-
ings, both inflectional and derivational, whose degree of iconicity is often neg-
ligible. In Tarok (Benue-Congo), for example, partial reduplication marks third
person singular possessive (Robinson 1976). In Arosi (Central-Eastern Oceanic;
Lynch & Horoi 2002), partial reduplication serves as one of four possessive clas-
sifiers, used for different semantic subclasses of possessive constructions. For
those possessive constructions involving “food or drink,” “things done to or
intended for the possessor,” or “things closely attached to the possessor” when
the possessor is first or second person, the possessor pronoun, which precedes
the possessed noun, is subject to CV prefixing reduplication, e.g. mu-murua
bwaa ‘cl-2du taro = your (dual) taro’ (Lynch & Horoi 2002:567). (In apparent
free variation with reduplication, a prefix ʔa- can be used instead, with different
word order, e.g. bwaa ʔa-murua ‘taro cl-2du = your (dual) taro’ [p. 567].)
Though no longer productive, reduplication in another Central-Eastern Oceanic
language, Marquesan, “marks dual and plural subject with some verbs,” e.g.
ʔua moe ia ‘perf sleep 3sg=he slept’, ʔu mo-moe ʔaua ‘perf redup-sleep
3du=the two of them slept together’ (Lynch 2002a:872). Triplication likewise
illustrates the non-iconic nature of reduplicative semantics. In Emai (Edoid;
Benue-Congo), for example, the number of iterations of an ideophone is a
function of the size of the base: monosyllabic ideophones triplicate (rı́-rı́-rı́
‘red’), disyllabic ideophones reduplicate (kútú-kútú ‘boiling (water),” trisyl-
labic ideophones occur as singletons (Egbokhare 2001:88). The number of
iterations is, in principle, orthogonal to the semantics of the construction.10
Reduplication can also serve seemingly arbitrary derivational functions,
marking changes in syntactic category or verbal argument structure. In Rotu-
man, “adjectives are formed from nouns by reduplication,” e.g. rosi ‘fraud,’
ros-rosi ‘cunning,’ ha.fu ‘stone,’ ha.f-ha.fu ‘stony, rocky’ (Schmidt 2002:822).
In Banoni (Western Oceanic), partial reduplication derives instrumental nouns
 Morphological Doubling Theory 15

from verbs (e.g. resi ‘grate coconut,’ re-resi ‘coconut grater’; Lynch & Ross
2002:442). In Ulithian (Micronesian; Lynch 2002b), reduplication derives
intransitive verbs from nouns, e.g. sifu ‘grass skirt,’ sif-sifu ‘wear a grass skirt’;
yaŋi ‘wind,’ yaŋi-yaŋi ‘blow’ (p. 799). The same is true in the Central-Eastern
Oceanic language Niuafo’ou, e.g. maka ‘stone,’ maka-maka ‘to be stony’ (Early
2002:856). In Marquesan (Lynch 2002a), “[s]ome verbs are derived from nouns
by reduplication,” e.g. ivi ‘bone,’ ivi-ivi ‘be thin,’ niho ‘tooth,’ niho-niho ‘be
notched’ (p. 872). In Siroi (Trans-New Guinea; Wells 1979), noun reduplication
yields an intransitive verb meaning ‘act like N’, in the -k- class (all Siroi verb
stems fall into one of four classes), e.g. zon ‘John,’ zon-zon-k-ate ‘John-John-
k-3sg.pres = he is acting (like) John’; ragitap ‘turtle,’ ragitap-ragitap-k-ate
‘he is acting like a turtle,’ gua ‘young child,’ gua-gua-k-ina ‘young child-
young child-k-3s.past = she acted childishly’ (p. 35; hyphenation has been
modified from the original). A similar pattern in which the first copy of the verb
also possesses a class marker connotes pretence, e.g. malmbi-k-et-malmbi-k-et-
ng-ate ‘cry-k-1sg.pres-cry-k-1sg.pres-ng-3sg.pres = he is pretending to
cry’ (p. 36). In Nadrogā (Central-Eastern Oceanic; Geraghty 2002:841), redu-
plication is used “to form intransitives of patient-oriented verbs,” thus vuli ‘[to
be] turned over,’ vuli-vuli ‘turn over’. (Agent-oriented verbs form frequentatives
when reduplicated, e.g. tola-vi-a ‘look at it,’ tola-tola-vi-a ‘look repeatedly at
it.’)
Idiomatic reduplicative constructions are similar to exocentric compounds,
like pick-pocket, blue-hair, etc. The meaning of the whole is not the result
of adding together the meanings of the parts; rather, it must be stipulated as a
property of the construction as a whole. The difference between pick-pocket and
e.g. the Banoni form re-resi ‘coconut grater’ is that the meaning of pick-pocket
is lexicalized, whereas in Banoni there is a semantic generalization over all
instances of this reduplication construction. The instrumental noun semantics
of Banoni verb reduplication is listed, in MDT, as a property of the mother node
of the schematic reduplication construction, as shown below:

(24) Banoni reduplication construction

⎡ Syntax: N ⎤
⎢ Semantics: used while y-ing ⎥
⎢ ⎥
⎣Phonology = concatenate daughters⎦

⎡ Syntax: V ⎤ ⎡ Syntax: V ⎤
⎢ Semantics: y ⎥ ⎢ Semantics: y ⎥
⎢ ⎥ ⎢ ⎥
⎣Phonology: truncation⎦ ⎣Phonology: identity⎦
16 Introduction

Example (resi → re-resi)

⎡ Syntax: N ⎤
⎢Semantics: ‘coconut grater’⎥
⎢ ⎥
⎣ Phonology: re-resi ⎦

⎡ Syntax: V ⎤ ⎡ Syntax: V ⎤
⎢Semantics: ‘grate coconut’⎥ ⎢Semantics: ‘grate coconut’⎥
⎢ ⎥ ⎢ ⎥
⎣ Phonology: resi → re ⎦ ⎣ Phonology: resi → resi ⎦

The advantage of a constructional approach is that semantically regular, par-

tially idiomatic, and totally idiomatic, constructions are all handled in the same
way, the difference being in how general or specific a particular construction is.
The same motivation drives the use of consructions in syntax (see, for example,
Fillmore, Kay & O’Connor 1988; Koenig 1992; Nunberg, Sag & Wasow 1994;
Goldberg 1995; Sag & Wasow 1999; Riehemann 2001; Kay 2002 for extensive
discussion).
This section is intended only as an illustration of how semantics would be
handled in MDT; reduplicative semantics is not used in this work to argue for
MDT over other theories. Certainly theories in which reduplication results from
the presence of a red morpheme would be equally well-equipped to handle
the range of observed effects.

1.2.4 Constructional phonology

Of greater import than semantics in the choice of a constructional approach to
reduplication is reduplicative phonology, one of the main focus areas of this
book. Constructions in SBM are equipped with phonological functions comput-
ing the phonology of the mother node (the “output”) from the phonology of its
daughters (the “inputs”). These mappings, termed “cophonologies” in Inkelas,
Orgun, and Zoll 1997, are necessary in order to describe morphologically con-
ditioned phonology. A language in which all cophonologies are identical has
no morphological conditioning; a language in which many constructions have
different cophonologies has a lot of morphologically conditioned phonology.
As argued extensively in work by Orgun (1996, 1997; see also Inkelas 1998; Yu
2000), cophonologies also intrinsically give rise to cyclic, or layering, effects.
Cophonologies figure prominently in Chapters 3–5, where they are used to
account for the attested range of phonological effects in reduplication.
In the affixation and compounding constructions illustrated thus far,
cophonologies do only moderate work. The cophonology of the plural
construction includes the voicing assimilation and epenthesis alternations,
 Morphological Doubling Theory 17

general to the language, that drive pural suffix allomorphy; the cophonology of
the compounding construction assigns compound stress.
Cophonologies play a highly important role in nonconcatenative morphol-
ogy, such as truncation, zero-derivation and ablaut constructions. These phe-
nomena have traditionally fallen outside the scope of phrase-structure rules or
subcategorization frames (see, for example, Lieber 1980), yet they are centrally
important in the morphology of many languages. Constructions for trunca-
tion, zero-derivation and ablaut are illustrated below. English nickname for-
mation exemplifies truncation, in (25a), whose cophonology (␾k ) deletes all
but the initial syllable (loosely speaking) of the input name. Zero-derivation is
exemplified in (25b) by English noun-to-verb conversion of the sort Kiparsky
(1982c) located in Level 1 of the morphology (e.g. pérmit (n.), permı́t (v.)).
Since English noun-to-verb conversion potentially results in stress shift, stress
assignment must be part of Cophonology ␾q . Ablaut is illustrated by the German
umlaut-only plural formation (e.g. Vater ‘father.sg’, Väter ‘father.pl’). In
German, the relevant cophonology (␾r ) requires back stressed vowels to front:

(25) a. ⎡ Truncation Schema: ⎤ ⎡ Example (Daniel → Dan): ⎤

Syntax = g(Synx ) Syn = masculine N
⎣Semantics = f (Semx ) ⎦ ⎣Sem = ‘nickname of Daniel’ ⎦
Phonology = ␾k (Px ) Phonology = /dæn/
⎡ | ⎤ ⎡ | ⎤
Syntax = Synx Syn = masculine N
⎣Semantics = Semx ⎦ ⎣ Sem = ‘Daniel’ ⎦
Phonology = Px X
Phonology = /dænjəl/ X

b. ⎡ Zero-derivation Schema: ⎤ Example (pèrmit → permı̀t):

Syntax = V ⎡ ⎤
⎢Semantics = do something ⎥ Syn = N
⎢ ⎥ ⎣ Sem = ‘permit’ ⎦
⎣ involving X ⎦
Phonology = [‘pm’]
Phonology = ␾q (Px )
⎡ | ⎤ ⎡ | ⎤
Syntax = Synx Syn = V
⎣Semantics = Semx ⎦ ⎣ Sem = ‘permit’ ⎦
Phonology = Px X
Phonology = /pərmit/ X

c. ⎡ Ablaut schema: ⎤ ⎡ Example (Vater → Vāter): ⎤

Syntax = Plural N Syn = Plural N
⎣Semantics = Plural of (Semx ) ⎦ ⎣ Sem = ‘fathers’ ⎦
Phonology = ␾r (Px ) Phonology = ␾q (/fatεi/) = [fetεi]
⎡ | ⎤ ⎡ | ⎤
Syntax = Singular N Syn = N
⎣ Semantics = Semx ⎦ ⎣ Sem = ‘father’ ⎦
Phonology = Px X
Phonology = /fatεi/ X
18 Introduction

As discussed by Orgun (1996), an interesting consequence of using construc-

tional schemata like these, rather than phrase-structure rules or subcategoriza-
tion frames, is the free choice it allows between an item-based approach to
affixation, as shown in (22), or the realizational approach argued for by mor-
phologists such as Anderson (1992). In a realizational constructional approach
the cophonology would be responsible not only for any morphologically condi-
tioned phonology associated with a particular affix, but also for realizing overt
affixes themselves. A realizational approach to the English noun plural suffix,
comparable to that in (21b), is provided below:

(26) Noun
⎡ pluralization in English
⎤⎡ Example: ⎤
Syntax = N Syntax = N
⎣ Semantics = plural of X ⎦ ⎣ Semantics = ‘books’ ⎦
Phonology = ␾2 (Px , /z/) Phonology = ␾2 (/bυk/, /z/) = [bυks]
⎡ | ⎤ ⎡ | ⎤
Syntax = N Syntax = N
⎣Semantics = Semx ⎦ ⎣Semantics = ‘book’ ⎦
Phonology = Px X
Phonology = /bυk/ X

The cophonology of this construction adds the plural /z/ ending as well as
performing voicing assimilation.

1.2.5 The phonology of reduplication

The primary phonological issues arising in reduplication are these: (a) how, if
at all, are the copies in reduplication phonologically modified relative to how
they would appear in isolation? (b) is surface phonological identity an extrinsic
requirement on reduplication?
In MDT, the essential identity in reduplication is semantic. This sets MDT
apart from what can be termed Coerced Identity theories (Wilbur 1973;
McCarthy & Prince 1995a), in which phonological identity between the copies
in reduplication is an explicit surface requirement of the grammar. MDT is, by
contrast, a Native Identity theory, in the sense that surface phonological identity
between the two copies occurs only as a side effect of semantic identity; often
the simplest, or only, way to assure semantic identity is to select exactly the
same morphological entity for the two daughters. Surface phonological identity
is not, however – and cannot be – required.
Beyond not requiring phonological identity, MDT also makes strong, distin-
guishing predictions about the potential for phonological modification of the
morphological elements involved in reduplication. The principal way in which
MDT differs from other theories of reduplication is in allowing the daughters
 Morphological Doubling Theory 19

of reduplication, as well as the mother, to be associated with potentially distinct

cophonologies.
Much of the reduplication literature has focused on certain kinds of redupli-
cation constructions in developing theories of what is phonologically possible.
Of primary focus have been partial reduplication constructions in which one
copy assumes the same form that it would in isolation, while the other is simpli-
fied in some manner – truncated, subjected to contrast neutralization, or both.
Our extensive surveys of reduplicative phonology, presented in Chapters 3 to
6, have shown that the range of phenomena goes far beyond this sort of case.
Reduplicative phonology does not always privilege one copy (the “base”) while
shrinking the other (the “reduplicant”); in some constructions, both daughters
are modified, even in different ways.
In MDT, phonological modification in reduplication results from the inter-
actions of three cophonologies: one for each daughter, and one for the mother.

(27) [zzz]
⇐ Cophonology Z

[xxx] [yyy]
Cophonology X ⇒ | | ⇐ Cophonology Y
/Stemi/ /Stemi/

Some reduplication constructions have active phonological effects only in one

or the other daughter; some have active effects only at the mother node; others
have active alternations in all three cophonologies.
In Hausa pluractional reduplication, for example, the first stem is reduced to
its initial CVC string, while the second is maintained intact (28). In addition,
the final C of the first copy geminates with the base initial consonant, a pro-
cess whose obligatoriness is specific to this construction (see Newman 1989a;
2000:424–25, as well as Chapter 4 of the present work). Examples are taken
from Newman 2000:424:

(28) a. kira kik-kira ‘call’

b. bugà bub-bùga ‘beat’
c. kawo kak-kawo ‘bring’

The construction in (29) illustrates the essential elements of an MDT analysis

of the pluractional data in (28). The construction calls twice for two seman-
tically agreeing stems, in this case kira: ‘call.’ Each daughter is subject to an
independent cophonology, an approach to the phonology of reduplication pio-
neered by Steriade (1988). In this case, the cophonology of the first daughter is a
20 Introduction

truncating one, and outputs a CVC syllable. The second daughter is subject to an
identity-preserving cophonology whose output matches its input. Together the
two pieces are subject to the cophonology of the mother node, which imposes
gemination of the consonant cluster at the juncture between the two copies.

(29) [kikkira] ‘call-pluractional’

⇐ Assimilation in consonant cluster

[kir] [kiraa]
Truncates ⇒ | | ⇐ Preserves identity
/kira/i /kira/i

‘call’ ‘call’

The precise formalization of each cophonology depends largely upon one’s

choice of phonological model. The framework used in this work is Optimality
Theory (OT; McCarthy & Prince 1993; Prince & Smolensky 1993), in which
each cophonology consists of a hierarchy of ranked and violable markedness,
faithfulness, and alignment constraints.11
A crucial aspect of the use of cophonologies in handling reduplicative phonol-
ogy is the layering intrinsic to the construction. Because Cophonologies X and
Y are associated with the daughters and Cophonology Z is associated with the
mother, there is an intrinsic cyclic effect: the outputs of Cophonologies X and Y
are the inputs to Cophonology Z. This chaining of cophonologies is, as argued
in Chapter 5, a crucial component of the phonological opacity which has put
reduplication in the forefront of phonological theory construction.

1.3 Phonological copying

As discussed above, the primary motivation for MDT comes from cases (e.g.
Sye and Ndebele) in which phonological copying cannot explain the different
morphotactics of the two copies or their morphological complexity. MS feature
duplication is clearly necessary in these cases, as will be argued throughout the
book. The question is whether there is still a role for phonological copying.
It is argued in Chapters 5 and 7 that MS feature duplication cannot replace
phonological copying, but that the scope of phonological copying is limited to
a narrow set of contexts. These include some phenomena that previously have
been classified as reduplication but which are not amenable to a morphological
doubling analysis, in part because the doubled element is something very small,
 Phonological copying 21

like a single consonant or vowel, and in part because the doubling has a purely
phonological purpose, rather than being associated with a change in meaning.
Restricting the use of phonological constituent copying to cases motivated by
phonological necessity, resonates with recent arguments by Gafos (1998b),
Hendricks (1999) and, to some extent, Zuraw (2002) that a theory should pro-
vide only one type of analysis to single-segment copying within a word, not
two. It is argued in this work that the proper analysis of such cases is phonologi-
cal copying, not morphological reduplication; although Gafos and Zuraw adopt
the seemingly opposite view that the unified analysis should be reduplication
(though without templates), their analyses, which use reduplicative morphemes
(Gafos) or the constraint reduplicate (Zuraw), are conceptually more con-
sistent with phonological copying or assimilation than with morphological
reduplication.
The Hausa and Yoruba onset copying cases discussed above, for example,
are unequivocal examples of phonological copying. In both Hausa and Yoruba,
the duplication of the consonant is driven purely phonologically, by the need for
a syllable onset. Spokane (Interior Salish) provides another such example, dif-
fering from Hausa and Yoruba in that phonological copying occurs in only one
allomorph of an otherwise nonreduplicative affix. The repetitive form of a verb
is formed by infixing /e/ into what would (due to unstressed schwa deletion)
be an initial consonant cluster (30a). For stems beginning with only a single
consonant, however, the stem-initial consonant is doubled, with /e/ appear-
ing between the two copies (30b) (Black 1996:210ff., Bates & Carlson 1998:
655).

(30) a. Repetitive e- infixes into initial consonant cluster:12

i. /-e-, šl’-n’-t-ən’/ → š-e-l’n’tén’
rep, chop-ctr-tr-IsgTrS ‘I cut it up repeatedly’
ii. /-e-, lč’-n’-t-ən’/ → l’-e-č’n’tén’
rep, tie-ctr-tr-IsgTrS ‘I tied it over and over’
b. Phonological copying provides onset for repetitive prefix e-
i. /-e-, šəl’/ → še- šil’
/rep, chop/ ‘I cut it up repeatedly’
ii. /-e-, nič’-n’-t-əxw / → n’-e-n’ı́č’n’txw
rep, cut-ctr-tr-2sgTrS ‘you kept cutting’

In all such cases, phonological copying is at work. Autosegmental phonol-

ogy would spread a consonant to the onset position; in Optimality Theory
the Onset constraint compels the insertion of a consonant that agrees featu-
rally with a nearby consonant (on string-internal segmental agreement, see, for
example, Walker 2000a; Hansson 2001; Rose & Walker 2001).
22 Introduction

1.4 Distinguishing the two types of duplication

This book focuses on morphological reduplication and the arguments for ana-
lyzing it as morphological doubling. In order to assess this argument, it is
important to have criteria for classifying a given duplication phenomenon as
morphological, in which case MS feature doubling is the correct analysis, or
phonological, in which phonological copying is called for.
One criterion distinguishing the two duplication effects is that phonologi-
cal copying serves a phonological purpose, while morphological reduplication
serves a morphological purpose, either by being a word-formation process itself
or by enabling another word-formation process to take place. Chapter 7 dis-
cusses several cases (from Chukchee, Kinande, and Nancowry) in which an
affix or stem-forming construction imposes a prosodic requirement that can
be satisfied only through semantically null morphological reduplication. All
three languages independently possess the relevant morphological reduplica-
tion construction, which is simply recruited, minus its semantics, to serve in
these cases.
The second criterion is proximity. Phonological duplication is proximal,
meaning that it targets the closest eligible element (a copied consonant is a
copy of the closest appropriate consonant, for example), while this is not neces-
sarily true of morphological reduplication. Sections 1.1 and 12.1 of this chapter
discuss several cases of syntactic reduplication in which the two copies are sep-
arated by other words; many parallel examples, in which base and reduplicant
are nonadjacent, exist in morphology as well, e.g. the opposite-edge reduplica-
tion in Chukchee nute-nut ‘earth (absolutive singular)’ (Krause 1980), Umpila
maka ‘die, go out’ → maka-l-ma ‘die, go out (progressive)’ (Harris & O’Grady
1976; Levin 1985b), or Madurese wã-mõw̃ã ‘faces’ (McCarthy & Prince 1995a
and references therein), all discussed in Chapter 7.
The third criterion is structural. Each instance of phonological copying
involves a single phonological segment, as in Hausa, Yoruba, or Spokane
onset-driven consonant copying; morphological reduplication involves an entire
morphological constituent (affix, root, stem, word), potentially truncated to
a prosodic constituent (mora, syllable, foot). This property of morphologi-
cal reduplication is discussed in Chapter 2, under the rubric of the Thesis of
Morphological Targets.
The fourth criterion involves phonological identity. Phonological copying
by definition involves phonological identity, while morphological reduplication
involves semantic identity. Phonological identity may occur in morphological
reduplication, but as a side effect of the fact that semantic identity requirements
 Wrapup and outline of book 23

result in identical inputs. Chapters 3 to 6 all discuss, in various ways, the evi-
dence that morphological reduplication does not presuppose or enforce phono-
logical identity between its daughters. Chapter 7 presents an argument that
phonological copying not only presupposes identity but actively enforces it,
solidly distinguishing the two types of process.

1.5 Wrapup and outline of book

MDT is a novel theory of reduplication which situates reduplication within
a solid morphological and phonological context. MDT differs from previous
phonological copying theories of reduplication in a number of important ways,
listed in (31):

(31) a. Most phonological theories of reduplication posit a Red morpheme,

while in MDT the reduplicant is a potentially morphologically
complex stem.
b. MDT posits separate, and potentially distinct, inputs for base and
reduplicant, while phonological copying theories posit a single,
shared input for the two output strings.
c. MDT draws no fundamental asymmetry between base and
reduplicant; neither is logically prior to the other. In most
phonological copying models, the base and reduplicant have an
asymmetrical relationship to the input.
d. MDT accepts phonological identity, insofar as it exists, between its
daughters as an epiphenomenon resulting from their semantic
identity but does not actively enforce it; by contrast, all phonological
copying theories minimally presuppose phonological identity as a
starting point in reduplication, and Coerced Identity theories like
BRCT actively require it in output.

The various chapters of this book demonstrate that MDT is more descriptively
adequate and at the same time more constrained in its predictions than existing
phonological approaches to reduplication. Chapter 2 presents the basic morpho-
logical evidence for MDT, focusing on data new to the theoretical literature on
reduplication that demonstrate the need for a theory in which the two daughters
in a reduplication construction are morphologically independent. Chapters 3,
4, and 5 closely investigate the phonology of reduplication. Chapter 3 covers
languages which motivate the phonological independence of the two daughters
and begins to make the case that reduplication-specific phonology is part of a
general phenomenon of morphologically conditioned phonology. This broader
context reveals that Coerced Identity theories are too narrow; by giving redu-
plicative phonology a special status, such theories are unable to account for the
24 Introduction

disposition of reduplicative phonology in the context of a language’s full mor-

phological system. Chapter 4 continues the argument that reduplication must
be understood in context, and presents evidence for the layered cophonologies
central to MDT’s conception of morphophonology. Case studies of underappli-
cation and sandhi phenomena in reduplication demonstrate the futility of try-
ing to account for reduplicative phonology with reduplication-specific devices
such as BR correspondence (McCarthy & Prince 1995a) or Existential Faith-
fulness (Struijke 2000a). Chapter 5 tackles the issue of phonological opacity
in reduplication, focusing on the so-called “overapplication” and “underap-
plication” effects in reduplication which have long been of special interest to
phonologists. Case studies of Fox, Javanese, and a number of other languages
show that reduplicative opacity effects result from the input–output chaining of
the cophonologies associated with the nodes in the morphological constituent
structure of a reduplication construction. One important result of this organic
approach to morphologically conditioned phonology, in which no reduplication-
specific technology is used, is that MDT can correctly account for the grouping
of alternations in a language with respect to overapplication and normal appli-
cation, patterns that appear random or arbitrary in Coerced Identity accounts
of reduplicative opacity. Building on the results of Chapter 5, Chapter 6 looks
at two cases of ostensible backcopying, in Tagalog and Chumash, which have
been cited in the literature in support of Coerced Identity theories. For both
cases, however, morphological structure and cophonologies are sufficient not
only to describe but also to explain the effects. Opacity in particular, and redu-
plicative phonology in general, can properly be understood only in the context
of the entire morphological system of a language. Finally, Chapter 7 reviews
the results of our investigation and stresses again the distinction between mor-
phological reduplication and phonological copying.
2 Evidence for morphological
doubling

Morphological Doubling Theory views reduplication as a morphological con-

struction containing some number of daughters – prototypically two – which
are identical in their semantic and syntactic features.
(1) Mother: [ ]g([F])

Daughters: [ ][F] [ ][F]

One or both of the daughters in a given reduplication construction may be the

product of a stem-forming construction which modifies them phonologically
by the phonological rules or constraints associated with the construction. In
addition, the reduplication construction itself may be associated with phono-
logical rules or constraints which apply to the construction as a whole. The
nature of reduplicative phonology is the particular subject of Chapters 3 and
4, and will not be discussed further in this chapter. It is, however, essential
to recognize that the two daughters have morphologically and phonologically
independent inputs. This is what allows them to diverge in the morphotactic
and phonological ways demonstrated in this chapter.
This chapter focuses on two essential morphological insights of MDT. One
is that the targets of reduplication – the types of thing that reduplication con-
structions propagate – are morphological constituents. The other is that the
foundational identity in reduplication is semantic. In MDT, reduplication cou-
ples morphological constituents which agree in their semantic (and syntactic)
specifications. These constituents are not required to match phonologically or
even in their internal morphotactic makeup; and as shown later in this chapter,
there are cases of reduplication in which the two constituents do differ in both
respects. The two central theses of this chapter are formulated below:
(2) The Thesis of Morphological Targets: a reduplication construction
calls for morphological constituents (affix, root, stem, or word), not
phonological constituents (mora, syllable, or foot)
The Thesis of Semantic Identity: reduplication calls for semantic
identity of its daughters, not phonological identity

25
26 Evidence for morphological doubling

This chapter marshals a variety of evidence supporting these two compo-

nents of MDT. The theses in (2) contrast fundamentally with the precepts of
the Phonological Copying view which has dominated the literature for thirty
years. The Phonological Copying perspective on reduplication, as represented
here by Base-Reduplicant Correspondence Theory (BRCT; McCarthy & Prince
1995a), is very different: reduplication is driven by the presence of an affixal
morpheme, “red,” which is associated with the grammatical requirement that
it phonologically copy material in the phonological string to which it is adjacent.

(3) Counterparts of (2) in BRCT:

The target of reduplication is phonological (e.g. mora, syllable, foot)
The strings that reduplication doubles are identical phonologically

This chapter shows that a theory that hews to the theses in (2) is the right way
to approach reduplication; the main insights of MDT correspond to the data
better than do the main insights of phonological copying theories like BRCT.
The additions to phonological theory which BRCT embodies are, therefore,
unnecessary.
We begin with the demonstration that there are clear cases of reduplica-
tion in which the essential identity between the copies is semantic, rather than
phonological. These cases support the Thesis of Semantic Identity.
It is common for the two copies in reduplication to differ phonologically.
This asymmetry could have two causes: (a) the copies are identical in input but
differ in output because of normal or special reduplicative phonology, or (b)
the copies are different in input. These two scenarios are contrasted, below:

(4) a. Copies are identical in morphological input [a-b]

Phonologically asymmetric outputs: [a] [b]

| |
Identical inputs: /x/ /x/

b. Copies differ in morphological input [a-b]

Phonologically asymmetric outputs: [a] [b]

| |
Different inputs: /x/ /y/

Any existing theory – MDT, Full Copy, BRCT – can describe scenario (4a).
Scenario (4b) is the one that is unique to MDT: base and reduplicant have
 Morphological targets: affix reduplication 27

different inputs, although the aggregate semantics of the inputs is identical.

This situation can arise in several ways. As discussed in §2.1, semantically
empty morphs occur in one copy but not the other, such that semantic identity
occurs without phonological identity. A special case of this is the phenomenon
famously known, since McCarthy and Prince 1999b [1986], as “Melodic Over-
writing,” discussed in §2.2.4. An alternative source of scenario (4b) occurs when
the two daughters in reduplication contain morphemes or morphological con-
stituents which are identical in their semantic content but differ phonologically –
i.e., they are synonyms of each other. Synonym constructions are discussed
in §2.3.
Before turning to evidence for the Thesis of Semantic Identity, however, we
address evidence bearing more directly on the Thesis of Morphological Targets,
namely affix reduplication.

2.1 Morphological targets: affix reduplication

The literature is full of examples of reduplication which are sensitive to the
morphological root or stem. This is not surprising on any view of reduplication.
Under most existing phonological copying theories, for example, reduplicants
are treated as affixes (Marantz 1982), and it is well known that affixes select for
certain types of roots or stems and can be sensitive to where their edges are.
Data which better show how the Thesis of Morphological Targets sets
MDT apart from other theories come from affix reduplication, the situation
in which reduplication targets a specific affix or type of affix, regardless of
its phonological shape. For example, in Dyirbal (Pama-Nyungan), the nominal
suffix -ŋaŋ ay ‘without’ can reduplicate, intensifying the semantic contribution
of ‘really/absolutely without’ that it normally carries (Dixon 1972:242):

(5) bana ‘water’

bana-ŋaŋgay ‘without water’
bana-ŋaŋgay-ŋaŋgay ‘with absolutely no water at all’

This construction is semantically iconic: the affix appears twice, and its
semantic contribution is reflected twice (by increasing the degree of absence of
the property designated by the noun with which the affix combines), and could
even be represented as multiple affixation rather than as reduplication per se.
In the following cases of affix reduplication, however, the meaning of redupli-
cation is not at all related to the meanings of the individual affixes being doubled.
Rather, the meaning is an arbitrary property of the reduplication construction
itself.
28 Evidence for morphological doubling

2.1.1 Preverb reduplication: Hungarian

Tauli (1966):182 states that reduplication of the Hungarian preverb – analyzed
variously in the literature as a prefix or proclitic – occurs “to express iterative-
ness or frequentativeness,” citing these examples: el-el-fújita ‘blow out many
times,’ vs. el-fúj ‘blow out’; vissza-vissza-vágyom ‘I frequently long to be back.’
Moravcsik (1978:306) also adds the form bele-bele-néz ‘he occasionally looks
into it.’ Many more examples are provided by Piñon (1991), in a more extensive
discussion based in part on Soltész 1959.
Although the roughly 60–70 preverbs that Piñon estimates exist in Hungarian
do tend to have their own meanings (e.g. el- means ‘away’), preverb redupli-
cation does not intensify this particular component of the meaning of the con-
stituent consisting of verb plus preverb. Rather, preverb reduplication adds a
fixed meaning component to any combination of preverb plus verb. Piñon char-
acterizes preverb reduplication as denoting “an irregular iteration of the event
denoted by the verb.” The fact that the meaning of preverb reduplication is not
derivable from the meaning of the preverb itself supports the constructional
approach Morphological Doubling Theory takes to reduplication, in which any
semantic idiosyncracies or noncompositionality can be listed as properties of
the mother node of the construction.1

2.1.2 Reduplication within the derivational stem

In Dyirbal, nominals reduplicate to mark plurality (Dixon 1972). A bare nominal
simply undergoes total reduplication, as shown in (6a). If, however, the nominal
is derived by a stem-forming nominal suffix, then there are two options for
marking plurality: root reduplication or suffix reduplication. Dixon is quite
clear that the two options are semantically equivalent (p. 272):

(6) a. midi-midi ‘lots of little ones’

gulgi
i-gulgi
i ‘lots of prettily
painted men’
b. midi-midi-baun ∼ midi-baun-baun ‘lots of very small
ones’
bayi ya
a-ya
a-gabun ∼ bayi yaa-gabun-gabun ‘lots of other men/
strangers’

From the perspective of Morphological Doubling Theory, what unites affix

and root doubling is the generalization that some morphological element in the
affixed stem occurs twice. Meaning is associated with the act of reduplicating
something, not with the reduplicated thing itself. This same generalization holds
in several other languages that permit affix reduplication.
 Morphological targets: affix reduplication 29

In Gapapaiwa (Papuan Tip cluster, Western Oceanic), reduplication within

the (potentially complex) verb stem marks imperfective aspect (McGuckin
2002). According to McGuckin (p. 308), if the verb stem contains a deriva-
tional prefix, such as causative vi- or ‘with the hands’ vo-, the derivational
prefix reduplicates, as shown in (7a). Otherwise, the verb root reduplicates, as
shown in (7b); according to McGuckin, root reduplication follows phonologi-
cally conditioned patterns, which unfortunately are not described.

(7) a. a-vi-vi-sisiya ‘1sg-impf-caus.past-speak = I was speaking’ 308

i-vi-vi-tete ‘3.nonpres-impf-caus.past-come.and.go’ 309
b. i-kam-kam ‘3.nonpres-impf-eat’ 318
a-ta-kita-kita ‘1sg-subj-impf-see = that I was seeing’ 309
ko-na-yeba- ‘2pl-fut-impf-fish = you [will be] fishing’ 318
yebagha
i-ne-nae ‘3.nonpres-impf-go’ 320
a-na-vo-yu-yuna ‘1sg-fut-with.hands-impf-gather’ 319

Amele (Gum, Trans-New Guinea) has several constructions reduplicating

inflectional affixes within the verb. To express iterative aspect in Amele, “the
whole stem is normally reduplicated if the verb does not have an object marker,
otherwise the object marker is reduplicated either in place of or in addition to
the reduplication of the verb stem” (Roberts 1991 [R91], pp. 130–31; see also
Roberts 1987 [R87], p. 252).2 Example ((8a) illustrates root reduplication, and
(8b) illustrates object reduplication:3

(8) a. qu-qu ‘hit’ R87:252

ji-ji ‘eat’ R87:253
budu-budu-eʔ ‘to thud repeatedly’ R91:131
gb atan-gb
 atan-eʔ ‘split-inf’ R91:131
b. hawa-du-du ‘ignore-3s-3s’ R87:254
gobil-du-du ‘stir-3s-3s = stir and stir it’ R87:254
guduc-du-du ‘run-3s-3s’ R87:254

Simultaneous action reduplication, which is partial, exhibits almost the same

pattern; if there is an object suffix, it reduplicates, e.g. mele-do-do-n ‘as
he examined’ (Roberts 1991:129); otherwise, the root partially reduplicates,
e.g. gb a-gb atan-en ‘as he split’ (Roberts 1991:128).

2.1.2.1 Double reduplication within the stem

In Boumaa Fijian (Central-Eastern Oceanic), stems formed by spontaneous or
adversative prefixes reduplicate both the prefix and the root in order to mark
plurality (Dixon 1988:226):
30 Evidence for morphological doubling

(9) ta-lo’i ‘bent’ ta-ta-lo’i-lo’i ‘bent in many places’

ca-lidi ‘explode’ ca-ca-lidi-lidi ‘many things explode’
’a-musu ‘broken’ ’a-’a-musu-musu ‘broken in many places’

In some of the other affix reduplicating languages, as seen, double redupli-

cation is an option, though not required as it seems to be in Boumaa Fijian.
Gapapaiwa and Amele both permit root doubling in stems that also have
affix doubling. In Gapapaiwa (10a), doubling of both results in an intensi-
fied semantic contribution; double reduplication indicates “a long time-frame
for the imperfective action” (McGuckin 2002:308). In Amele (10b), root dou-
bling in words with object suffixes simply appears to be an option, without any
particular additional semantic contribution:

(10) a. i-vo-vo-koi-koi ‘3.nonpres-with. (McGuckin 2002:306)

hands-weed’
b. bala-bala-du-du-eʔ ‘tear-3s-inf=to tear (Roberts 1991:131)
it repeatedly’

In Dyirbal, by contrast, double root-affix reduplication is prohibited; Dixon

(1972) says explicitly that words like *midi-midi-baun-baun are ungram-
matical (p. 242).4

2.1.2.2 Discussion of reduplication within the stem

If one assumes that root + stem-forming suffix in Dyirbal, root + object marker
in Amele, root + derivational prefix in Gapapaiwa, and root + nominalizing suf-
fix in Boumaa Fijian form morphological constituents that we can call “stems,”
then the generalization is the same in all four languages: reduplication doubles
some morphological constituent within the stem, irrespective of phonological
size or linear position. Without going into the details of the morphological con-
struction that handles affix doubling, it is clear that these languages strongly
support the Thesis of Morphological Targets (2).

2.1.3 Further implications

The existence of affix reduplication, as distinct from root reduplication, requires
reconsideration of the root vs. affix distinction among reduplicants postulated
in McCarthy and Prince’s Generalized Template Theory (GTT; McCarthy &
Prince 1994c; d; Urbanczyk 1996). According to GTT, there are two kinds
of reduplicative morphemes: roots and affixes. The only consequence of this
distinction is phonological.5 Roots are subject to constraints requiring them to be
minimally foot-sized, while affixes are subject to constraints requiring them to
be less than a foot (i.e. no larger than a syllable). The existence of affix doubling
 Morphotactic asymmetries: empty morphs 31

shows, however, that the relevant distinction between reduplicated roots and
reduplicated affixes is not phonological, but morphological. In Dyirbal, Amele,
and the other languages discussed in §2.1, reduplicated affixes can be of any
size, as can reduplicated roots.6
Although somewhat of a side issue, it is worth noting that the existence of
affix doubling, especially in conjunction with root doubling, appears to support
theories of morphology which treat affixes as morphological constituents. In
realizational theories in which affixed stems lack internal constituent structure,
e.g. the A-morphous theory of morphology developed by Anderson (1992), it
is difficult to express affix doubling and root doubling as instances of the same
phenomenon.

2.2 Morphotactic asymmetries: empty morphs

A major prediction of MDT, following from the Thesis of Semantic Identity,
is that there should be reduplication constructions in which the two daughters
are identical in their meanings but differ phonologically because their internal
structural makeup is different. This section examines a variety of cases in which
phonological identity is negated by the existence of empty morphs in one of
the copies. For example, in Ndebele (Nguni; Downing 2001; Sibanda 2004;
Hyman, Inkelas, and Sibanda to appear), to be discussed in greater detail below,
reduplication of a verb stem involves a preposed, disyllabic version of the verb
stem, as shown in (11a).7 When the base itself is not disyllabic, however, the
reduplicant must contain a morph not present in the base, simply in order to
satisfy disyllabicity. The only way to do this and still maintain semantic identity
is to use a semantically vacuous morph. This is illustrated in (11b), where the
reduplicant, dlayi, contains the root ‘eat’ (dl-) plus the semantically empty
morph -yi. Data are taken from examples (2) and (27) of Hyman, Inkelas, and
Sibanda to appear; see also Sibanda 2004:289–300:

(11) a. ‘taste’ /nambith-a/ nambitha nambi-nambitha

b. ‘eat’ /dl-a/ dla dlayi-dla

The presence of yi in the reduplicant is tolerated because it does not affect

the semantic identity between dlayi and dla. Cases of this sort prove that the
identity in reduplication is essentially semantic, not phonological.

2.2.1 Empty morphs in morphology

To have a proper context for understanding the role of empty morphs in redu-
plication, it is useful first to survey the use of empty morphs in morphology
32 Evidence for morphological doubling

generally. There is, in fact, nothing unique morphologically about the empty
morphs that occur in reduplication. The factors that motivate the presence of
empty morphs in reduplication are the same factors that motivate their presence
in nonreduplicative constructions. These factors can be either phonological or
morphological.
Many morphological constructions require the presence of a morphological
formative which has no clearly isolable meaning but plays a structural role. Final
vowels in Bantu are a good example. Verb stems in many Bantu languages are
required to end in one of a small set of “Final Vowel” (FV) suffixes, so called
because they close off a verb stem to further suffixation. Some FV suffixes do
make a semantic contribution; for example, in Ndebele, the FV suffix -e marks
subjunctive mood, and the FV suffix -ile markes perfective aspect (12). But
there is usually one FV suffix which is semantically vacuous, serving only the
structural role of closing off the verb. In Ndebele the FV suffix -a meets this
description. It is simply the “elsewhere” FV (see, e.g., Sibanda 2004):

(12) lim- ‘cultivate’ Subjunctive: lim-e

Perfective: lim-ile
Elsewhere: lim-a

Wise (1986) proposes empty morph status for the /t/ formative which
appears in the PreAndine languages of Peru (which include the Campa,
Amuesha, and Piro-Apurinã languages). Although in Axininca Campa and other
Campa languages the distribution of /t/ appears wholly governed by phonology
(Wise 1986:579; see also, for example, Payne 1981; Spring 1990; McCarthy &
Prince 1993), Wise cites examples from Piro and Amuesha (p. 578) where the
cognate formative “is not necessary on phonological grounds; it occurs simply
because certain verb roots and suffixes . . . cannot occur word or stem finally
and no other potentially closing suffix . . . occurs” (p. 581). In this description,
/t/ has a function similar to the final vowel in Bantu; it is a semantically empty
element which serves to close off a stem morphologically.
The elements Hockett (1954) termed “markers” of constructions are also pro-
totypical empty morphs. Hockett’s example was the English conjunction and,
but many examples occur in morphology proper. Aronoff (1994) points to theme
vowels in European languages, particularly those of Latin, as being “[a]mong
the best-known examples of empty morphs” (p. 45); the bindungs-elements
commonly found in compounding constructions constitute another. English
has this to only a limited degree, e.g. huntsman, batsman, swordsman, clans-
man; but many compounding constructions in other languages illustrate the phe-
nomenon more clearly, e.g. Serbo-Croatian vod-o-paad ‘water-marker-fall =
 Morphotactic asymmetries: empty morphs 33

waterfall,’ plaav-o-zelen ‘blue-marker-green = blue green,’ etc. Turkish

(see, for example, Lewis 1967; Kornfilt 1997) marks one of its compounding
constructions with a final suffix, which, though homophonous with the third
person possessive suffix, does not contribute any independent meaning to the
construction, e.g. bebek-hastane-si ‘baby-hospital-marker = baby hospital,’
okul kitab-ı ‘school-book-marker.’8

2.2.2 Phonologically beneficial empty morphs

Some empty morphs are not part of particular constructions in the sense dis-
cussed above but are used when they serve a phonological purpose, e.g. aid-
ing in the satisfaction of prosodic minimality. Many Bantu languages impose
minimality requirements on verbs or verb stems. For example, Chichêwa and
Ndebele require their verbs to be minimally disyllabic (on Chichêwa, see, for
example, Mtenje 1988; Kanerva 1989; Hyman & Mtenje 1999; on Ndebele, see
Downing 2001; Sibanda 2004; Hyman, Inkelas & Sibanda to appear). When
verbal prefixes are present, disyllabicity is guaranteed; no stem is smaller than a
syllable and all prefixes are monosyllabic. But in the imperative, which consists
of a bare stem, a verb whose stem is monosyllabic falls short of minimal size.
Chichêwa augments monosyllabic verbs with the formative i- (13a) (see, for
example, Hyman & Mtenje 1999:108), while Ndebele augments with yi (13b)
(example (24) from Hyman, Inkelas & Sibanda to appear; see also Downing
2001; Sibanda 2004):

(13) Infinitive Imperative Gloss

a. i. ku-fótókozera fotokozera ‘explain’

ii. ku-dyá i-dya ‘eat’
b. i. uku-lima lima ‘cultivate’
uku-bamba bamba ‘catch’
uku-nambitha nambitha ‘taste’
ii. uku-dla yi-dla ‘eat’
uku-lwa yi-lwa ‘fight’
uku-ma yi-ma ‘stand’
uku-za yi-za ‘come’

To take one other example, the Athapaskan languages Slave and Navajo use
(h)e- (Rice 1989:149, 132–35, 426) and yi- (McDonough 1990:138ff.), respec-
tively, to augment verb roots which lack prefixes. According to McDonough,
verb prefixes belong to a constituent (the I-stem) which is obligatory in every
verb and which cannot be empty.9
34 Evidence for morphological doubling

The analysis of empty morphs adopted here is as follows. Some empty morphs
are associated with particular constructions, in which they are obligatory. These
are the kind seen in §2.2.1. Others, however, are listed in the lexicon without
being mentioned in any particular constructions; these include the phonolog-
ically beneficial morphs in §2.2.2. Their distribution in inputs is regulated by
the grammar, in a way that we model here by building on existing proposals
in Optimality Theory for determining the lexical representation of morphemes
(Lexicon Optimization) and, in cases of allomorphy, for selecting among the
different lexical allomorphs of a morpheme.10 Lexicon Optimization (Prince &
Smolensky 1993; see also Inkelas 1995; Itô, Mester & Padgett 1995; Yip 1996,
among others) is a deterministic method for establishing underlying repre-
sentations by considering all of the possible inputs that would generate a given
output, and selecting the most harmonic. In some cases, more than one underly-
ing allomorph for a given morpheme must be memorized, in a form determined
by Lexicon Optimization. We follow Dolbey 1996; Kager 1996; Sprouse 1997;
Steriade 1997; 1999; Sprouse in preparation in assuming that these listed allo-
morphs compete for position in words calling for the semantic content that they
share. The grammar selects that allomorph which phonologically optimizes the
word containing it.
Combining these two proposals yields the following approach to determin-
ing morphological inputs: the input candidates considered for a word with
meaning (M) are those sets of lexically listed morphological formatives whose
meanings add up to M. These combinations can differ from each other in sev-
eral ways. They can vary by lexical allomorph (as in the cases considered by
Dolbey and Kager); they can vary by including, or not including, semanti-
cally empty morphemes, the case in question here. The presence or absence of
a semantically empty morph has no effect on the meaning (M) of any given
candidate.
The tableaus in (14) evaluate competing possible morphological inputs for
the Chichêwa words meaning ‘eat (imperative)’ and ‘explain (imperative).’
These tableaus mirror those used for Lexicon Optimization by Inkelas 1995; Itô,
Mester & Padgett 1995 and for allomorph selection by Dolbey 1996; Sprouse
1997; in preparation. In each case the optimal input is the one judged most har-
monic by standards of output well-formedness and input–output faithfulness;
each input is compared, separately, to the full range of possible output strings. In
the first tableau, for ‘eat (imperative),’ the competing inputs are /dy-/, dy-a/, and
/i-dy-a/. By virtue of containing the same root, /dy/, all mean ‘eat’; the absence
of other semantically contentful morphemes results (by means that cannot be
fully explored here) in the default interpretation of imperative. The input /dy-a/
 Morphotactic asymmetries: empty morphs 35

has one empty morph; the input /i-dy-a/ has two. Each input is compared with
three reasonable outputs, [dy], [dya], [idya], which are evaluated for well-
formedness. Constraints on syllabification (not formalized here, but abbreviated
as syll) eliminate the output [dy] from consideration, as it is not a well-formed
syllable. Disyll favors the disyllabic output [idya], while *struc (arbitrar-
ily evaluated in terms of syllable count) favors the more compact outputs.
IO-Faith favors those input–output pairs which are faithful over those which
exhibit epenthesis or deletion. Because Disyll outranks *struc, the output
[idya] is favored over the other two outputs; because IO-Faith is high-ranked,
the input /i-dy-a/ is optimal for the output [idya], and this input–output pairing
is therefore selected as optimal overall. The presence of empty morphs in the
input for ‘eat (imperative)’ is optimal because the morphs contribute to the
well-formedness of the output without disrupting faithfulness (as epenthesis,
for example, would). The second tableau, for ‘explain (imperative),’ exhibits a
parallel set of candidate inputs and outputs. The difference between ‘eat’ and
‘explain’ is phonological size; the root – fotokozer- is already polysyllabic,
meaning that Disyll is satisfied by all of the input–output pairs under consid-
eration. The input /i-fotokozer-a/, with two empty morphs, violates *struc to
a greater degree than does the input without the /i-/ empty morph, which wins
the overall competition.

(14)

‘eat (imperative)’ IO-Faith syll Disyll *struc

input output
candidates candidates

a. i. /dy-/ [dy] *! * *

ii. [dya] *! * *

iii. [idya] *!* **

b. i. /dy-a/ [dy] *! * * *

ii. [dya] *! *

iii. [idya] *! **

c. i. /i-dy-a/ [dy] *!* * * *

ii. [dya] *! * *

E iii. [idya] **
36 Evidence for morphological doubling

‘explain (imperative)’ IO-Faith syll Disyll *struc

input output
candidates candidates

a. i. /fotokozer/ [fotokozer] *! ****

[fotokozera] *! *****

[ifotokozera] *!* ******

b. i. /fotokozer-a/ [fotokozer] *! * ****

) ii. [fotokozera] *****

[ifotokozera] *! ******

c. i. /i-fotokozer-a/ [fotokozer] *!* * ****

ii. [fotokozera] *! *****

[ifotokozera] ******!

From here forward we will assume that an approach of this kind regulates
the distribution of empty morphs in inputs, without providing similarly lengthy
justifications for every case.

2.2.3 Empty morphs in reduplication

The presence of empty morphs in one or both daughters in reduplication can
disrupt phonological identity while still preserving semantic identity between
them.
One type of empty morph commonly found in reduplication constructions
is the so-called “linker morph,” which comes between the two copies in the
same way that bindungs-elements join the two members of compounds. For
example, Khasi iterative verb reduplication separates the two copies of the verb
by the linker ši (Abbi 1991:130), e.g. iaid-ši-iaid ‘to go on walking,’ leh-ši-leh
‘keep repeating,’ kren-ši-kren ‘keep talking’ (Abbi 1991:128). Tamil “discon-
tinuous” reduplication partially reduplicates a numeral (X) to mean ‘X and only
X;’ separating the two copies is a linker element -e-, as in: o-e-on.n.u ‘only one’
or na:l-e-na:lu ‘only four’ (Abbi 1991:65). In Huallago (Huánuco) Quechua
(Weber 1989), the first of two reduplicated verbs is followed by -r. Many lan-
guages with more than one reduplication construction use different linkers for
each; in Arrernte (Arandic; Central Australia), frequentative reduplication uses
 Morphotactic asymmetries: empty morphs 37

the linker -ep, e.g. akemir-em ‘is getting up’ vs. akemir-ep-ir-em ‘keeps getting
up,’ while attenuative reduplication uses the linker -elp, e.g. itir-em ‘thinking’
vs. it-elp-itir-em ‘half-thinking’ (Breen & Pensalfini 1999:6–7). In Jaqaru, a
Jaqi language of Peru (Hardman 2000), two total reduplication constructions
are distinguished only by the linker consonant separating the two copies. Inten-
sive adjective reduplication separates the two copies with the palatal glide y;
intensive verbs use the palatal affricate linker ch. Hardman presents the fol-
lowing minimal pair, based on t’usqi ‘dust’: t’usqi-y-t’usqi ‘very like smoke’
(the intensive adjective) vs. t’usqi-ch-t’usqi ‘to be causing a lot of dust’ (the
intensive verb) (p. 54). For some additional examples of linker constructions,
see Moravcsik 1978. Linker morphs may be assumed to be semantically empty
when there is no positive evidence that they, as opposed to the reduplication
they co-occur with, distinctively contribute any specific meaning to the con-
struction. What is clear is that they are concomitants of reduplication whose
presence could not be predicted based on other knowledge of the language.
The constituent structure of reduplication constructions with linker morphs
is not always determinate. If the so-called “linker” forms a morphological con-
stituent with one of the daughters (15b), then its presence disrupts phonological
identity. If, on the other hand, the linker is a third daughter of the construction
(15a), then identity is not disturbed.

(15) (a) t’usqicht’usqi (b) t’usqicht’usqi

t’usqi ch t’usqi t’usqi ch t’usqi

Only in the presence of evidence for a structure like (15b) (or its mirror image)
do linker morphs support the claims of MDT. Unfortunately, the morphological
constituency of linker morphs is difficult to determine in many cases; a similar
problem is presented by bindungs-elements in compounding. The claims of
MDT are more clearly illustrated by cases in which the empty morph unam-
biguously belongs to one daughter or the other.
In the Mon-Khmer language Pacoh, for example, an empty morph can
occur, under certain conditions, in the base of reduplication. Verb reduplication
expresses the meaning of simulation; reduplicating ‘laugh’ yields a meaning of
‘pretend to laugh’ (Watson 1966:95). Both copies – base and reduplicant – are
subject to prosodic size conditions. The second copy must be monosyllabic;
the first must be disyllabic. If the input root is disyllabic, then the first copy is
intact, while the second is truncated to the final (main) syllable of the input verb
38 Evidence for morphological doubling

(16a). If the input root is monosyllabic, then the second copy is intact, while
the first is augmented with the semantically empty morph qâN- (16b). All of
the reduplicated forms cited in (16) are preceded by the verb táq ‘to do’:
(16) a. kacháng ‘to laugh’ táq kacháng-cháng ‘to pretend to laugh’
qaqay ‘sick’ táq qaqay-qay ‘to act sick’
b. cha ‘to eat’ táq qân-cha-cha ‘to pretend to eat’
bı́q ‘to sleep’ táq qâm-bı́q-bı́q ‘to pretend to sleep’
pôk ‘to go’ táq qâm-pôk-pôk ‘to pretend to go’

We may assume, building on the discussion in §2.2.2, that the Pacoh lexicon
includes the empty morph qâN-. Though normally absent from inputs because
its presence would gratuitously violate *struc, qâN is available to inputs that
would not otherwise produce stems meeting the prosodic minimality condi-
tions. It is needed in the input to the first copy in reduplication to satisfy the
disyllabicity requirement; it is not needed in the second copy.
(17) [qâmpôkpôk] [‘pretend to go’] ⇐ Phonological output of mother

[qâmpôk] [pôk] ⇐ Phonological outputs of daughters

/qâm-pôk/ [‘go’] /pôk/ [‘go’] ⇐ Morphological inputs to daughters

(regulated by input optimization, not shown)

On this analysis of Pacoh, the inputs of the daughters can differ morphotacti-
cally but are still identical semantically. Cases of this sort, in which the base is
more complex than the reduplicant, do not necessarily distinguish among theo-
ries of reduplication; however, cases in which the reduplicant is more complex
than the base provide strong support for MDT.
Empty morphs in the reduplicant play a significant role in Bantu reduplica-
tion, as discussed, for example, in the significant bodies of insightful work
on Kinande (Mutaka & Hyman 1990; Downing 1998b; 2000a), Chichêwa
(Hyman & Mtenje 1999), Siswati and Kikerewe (Downing 1997a; b; 1999a;
c). Here the focus will be on Ndebele, drawing on work by Downing 1999a;
2001, Sibanda 2004, and Hyman, Inkelas & Sibanda to appear. Ndebele has
two empty morphs: yi, which augments subminimal words, and -a, which is
the default final suffix in verb stems (see below). Both also aid in achieving
reduplicant disyllabicity in verb stem reduplication.
As noted above, Ndebele exhibits a common Bantu verb stem reduplication
pattern: the first copy of the verb stem is restricted to the CVCV prosodic shape,
while the second is not phonologically restricted. Standard practice is to refer
 Morphotactic asymmetries: empty morphs 39

to the first copy as the “reduplicant” and the second as the “base”; of course,
in MDT the two copies have equal morphological status. Reduplication adds
the meaning of ‘do here and there, a little bit.’ As shown below, a cursory look
at the phenomenon suggests that the reduplicant consists simply of the initial
CVCV portion of the base. Verbs are shown in the infinitive, marked by the
prefix uku-:
(18) Unreduplicated stem Reduplicated counterpart

‘taste’ /uku-nambith-a/ uku-nambitha uku-nambi-nambitha

‘go from wife /uku-thembuz-a/ uku-thembuza uku-thembu-thembuza
to wife’
When the verb stem is sufficiently short, however, empty morphs can appear
in the reduplicant which are not present in the base. If the verb stem is disyllabic
only by virtue of containing an inflectional suffix, then the reduplicant must end
in the empty morph -a:
(19) a. ‘cultivate’ subjunctive /lim-e/ lime lima-lime
perfective /lim-ile/ limile lima-limile
b. ‘send’ subjunctive /thum-e/ thume thuma-thume
perfective /thum-ile/ thumile thuma-thumile
According to Downing (see, for example, Downing 1999a for an overview), a-
final reduplicants are common in the Bantu family; the pattern arises when
morphological factors prevent the morphological material making the base
disyllabic (or larger) from appearing in the reduplicant.
As observed by Hyman, Inkelas, and Sibanda (to appear) and Sibanda (2004),
the domain of reduplication in Ndebele is what Downing has called, for Bantu
languages generally, the Derivational Stem (DStem), consisting of root plus
any derivational suffixes (see, for example, Downing 1997; 1998a; 1999a; c;
2000a; 2001). The DStem combines with an inflectional suffix, which could be
the semantically empty final vowel (FV) -a, to produce the Inflectional Stem
(IStem), which is normally what Bantuists mean by the term “verb stem.”
Many Bantu languages, including Kinande, reduplicate the IStem. However,
as proposed in Hyman, Inkelas, and Sibanda (to appear) and Sibanda 2004,
Ndebele reduplication targets Dstems. It produces another Dstem, with the
added meaning of ‘do here and there, a little bit,’ which then must obligatorily
combine with a FV (not shown).
(20) [ ]Dstem; [F + “here and there, a little bit”]

Disyllabicity ⇒ [ ]Dstem;[F] [ ]DStem; [F]

40 Evidence for morphological doubling

If the Dstem is smaller than two syllables, the disyllabicity requirement on

the reduplicant can be satisifed only if a semantically empty morph is present
in input. The example below shows what the construction generating the redu-
plicative constituent [lima-lim], within the subjunctive verb stem lima-lime,
would look like:

(21) [lima-lim]Dstem; [‘cultivate here and there, a little bit’]

Disyllabicity ⇒ [lim-a]Dstem;[‘cultivate’] [lim]DStem; [‘cultivate’]

| |
/lim-a/ /lim-/

Empty a is needed in the reduplicant, and hence the input containing it is

preferred to the input not containing it; a is not needed in the base, and hence
the more economical input is preferred.
What is important here is that the morphotactic discrepancy observed in
Ndebele clearly requires the kind of semantic equivalence central to MDT and
cannot be accounted for in theories where reduplication consists exclusively of
phonological copying.
Some Ndebele reduplicants are so far from achieving disyllabic minimality
that they need more help than empty -a can provide. As shown below, Dstems
consisting only of a consonantal root need help not only from -a but also
from the other semantically empty morph in Ndebele, namely -yi, introduced
earlier in a discussion of subminimal verb augmentation. Both -a and -yi are
needed to make the reduplicants of consonantal DStems disyllabic. -a alone
would be insufficient (dl-a, for example, would still be only monosyllabic), as
would -yi; yi-dl is still only monosyllabic and has an illegal coda, while dl-yi
is monosyllabic and has an impermissible onset cluster. Data are taken from
example (29) of Hyman, Inkelas, and Sibanda to appear:

(22) Infinitive (no reduplication) Reduplicated counterpart

a. ‘eat’ /uku-dl-a/ uku-dla uku+dl-a-yi+dla

b. ‘stand’ /uku-m-a/ uku-ma uku+m-a-yi+ma
c. ‘fight’ /uku-lw-a/ uku-lwa uku+lw-a-yi+lwa
d. ‘come’ /uku-z-a/ uku-za uku+z-a-yi+za

For the Dstem meaning ‘eat,’ based on the consonantal root dl-, the optimal
input for the reduplicant is /dl-a-yi/, which can faithfully achieve a disyllabic
output; the optimal input for the base is /dl-/, which is both faithful and struc-
turally minimal (since no size conditions constrain it).
 Morphotactic asymmetries: empty morphs 41

(23) [dlayi-dl-]Dstem; [‘eat here and there, a little bit’]

Disyllabicity ⇒ [dl-a-yi]Dstem;[‘eat’] [dl-]DStem; [‘eat’]

| |
/dl-a-yi/ /dl-/

Ndebele reduplication is striking in the degree of phonological discrepancy

that can result between reduplicant and base. Forms like dlayi-dla or lima-limile
are clearly more consistent with the precepts of MDT than with the precepts of
BRCT; lima cannot be derived from limile, nor dlayi from dla, by phonological
copying. The equivalence is not phonological. It is semantic.
These facts are amply discussed by Sibanda (2004) and Hyman, Inkelas, and
Sibanda (to appear), who adopt a morphological analysis of the sort taken here;
suffice it to say here that the data in (22) confirm the degree of morphotactic
autonomy of the two copies in reduplication.

2.2.4 Simple melodic overwriting

In Ndebele, the empty morphs a and yi supplement consonantal roots. Some-
times, however, empty morphs compete for surface position with some of the
material from the semantically contentful morphemes in the input. Particularly
interesting is the fact that semantically empty -a and -yi are in free variation,
in the reduplicant, with material from derivational suffixes, as illustrated below
(data from example (3) of Hyman, Inkelas, & Sibanda to appear):

(24) a. lim-el-a lim-e+lim-el-a ‘cultivate for/at’ (applicative -el-)

lim-a+lim-el-a
b. lim-is-a lim-i+lim-is-a ‘make cultivate’ (causative -is-)
lim-a+lim-is-a

The variation is even more extreme with consonantal roots, as discussed by

Sibanda (2004) and Hyman, Inkelas and Sibanda (to appear), who take an MDT
approach to the data. The important point is, however, sufficiently illustrated
by the CVC roots in (24). The grammar has a choice between using the empty
morph -a or the input derivational suffix vowel in the second syllable in the
reduplicant; the two choices are equally preferred, resulting in the observed
variation.
The interested reader is encouraged to consult Downing 2001, Sibanda 2004,
and Hyman, Inkelas, and Sibanda (to appear) for further information on the
rich reduplication system of Ndebele. The main point here is that an empty
input morph can, under conditions like the disyllabic size condition in Ndebele
42 Evidence for morphological doubling

reduplication, surface at the expense of (part of) a semantically contentful input

morph.
It is competition of this sort that underlies the phenomenon commonly known
as Melodic Overwriting. “Melodic Overwriting” is the term given by McCarthy
and Prince 1986, Alderete et al. 1999, and others to the situation in which one
of the two copies in reduplication exhibits phonological material not present
in the other, material which appears to be replacing segments that would oth-
erwise be expected to be present in a way that cannot be accounted for by
natural phonological alternations. In parade examples of Melodic Overwriting,
like the Yiddish-English ironic construction exemplified by fancy-shmancy, the
replacive material is a segmental string which targets a syllable constituent at a
fixed edge of the copy designated as the reduplicant. Initial onset replacement,
as in fancy-shmancy, is probably the most common type of case discussed in the
literature; it also occurs, for instance, in the well-known, virtually pan-Asian
“X and the like” construction manifested as m- replacement in Turkish kitap-
mitap ‘books and the like’ (see Lewis 1967 for discussion), Armenian pətuʁ -
mətuʁ ‘fruit and stuff’ (Vaux 1998:246), and Abkhaz ga ák’-ma ák’ ‘fool and
the like’ (Vaux 1996, Bruening 1997; Vaux forthcoming), and as gi-replacement
in Kolami kota-gita ‘bring it if you want to’ and Telugu a ku giku ‘leaf, etc.’
(Emeneau 1955:101–102; Bhaskararao 1977:7ff.).
Just as initial onset replacement can be seen as affixation, final rime or
coda replacement can be seen as suffixation. One example of final rime
replacement occurs in Vietnamese; the “i ê´ c-hoá” construction, which “supplies
some emotional coloring (disinterest, irony, etc.) to the meaning of any base,”
reduplicates a base of any size, and replaces the rime of the final syllable
of the second copy the formative -i ê´ c, e.g. hát → hát.hi ê´ c ‘to sing,’ cà.phê
‘coffee and the like’ → cà.phê cà.phi ê´ c, câu.la.c.bô. ‘clubs and the like’ →
câu.la.c.bô. câu.la.c.bi ê´ c (Nguyen 1997:53–54).11 Apparent coda overwriting
occurs, for example, in the emphatic adjective reduplication construction, found
throughout Turkic (see, for example, Johanson & Csato 1998) as well as in
Armenian (Vaux 1998:242–45), which reduplicates the first CV of the adjec-
tive and interposes a fixed consonant between reduplicant and base. This inter-
posed consonant is normally interpreted in descriptions as part of the redu-
plicant; it syllabifies as a coda except when the adjective is vowel-initial,
in which case it syllabifies rightward as an onset. Turkic languages vary in
the precise details of this construction; one of the simpler manifestations
occurs in Mongolian, where the fixed consonant is always /v/, e.g. khav-khar
‘coal black,’ chiv-chimeegüi ‘completely silent,’ uv-ulaan ‘bright red’ (Bosson
1964:110).12
 Morphotactic asymmetries: empty morphs 43

Alderete et al. (1999) have argued for treating Melodic Overwriting as affixa-
tion; in MDT, the affix in question is generated within one of the copies, the one
traditionally termed the reduplicant. The affix is a semantically empty morph
required by the construction, parallel to the empty linker morphs discussed in
§2.2.1 except in the one respect that it supplants, instead of coexisting with,
input material. (Thus, in retrospect, the Ndebele data discussed in §2.2.1 could
be analyzed in terms of Melodic Overwriting, insofar as the empty -yi and -a in
the reduplicant potentially supplant input affixes.) Based on a proposal made
for comparable data in Inkelas 1998, we assume that the competition between
empty affix and input material is driven by faithfulness: the output must be
faithful prosodically, e.g. in syllable count, to the input size of the base of affix-
ation. In a case like Vietnamese càphê+càph-i ê´ c, the input base of the affixed
copy is càphê; faithfulness to its syllable count forces the suffix i ê´ c to replace ê:

(25) càphêcàphiêc

[càphê] [càph-iêc] ⇐ output has 2 syllables

| /\
/càphê/ /càphê-iêc/ ⇐ input root has 2 syllables

Melodic Overwriting is thus distinguished from ordinary cases of morpholog-

ically required linker morphs, discussed in §2.2.3, in just one respect: in cases of
Melodic Overwriting, a prosodic size faithfulness condition is imposed on the
subconstituent containing the linker morph. The ensuing competition between
affix and base of affixation is what provides evidence that the linker morph
belongs to one of the two daughters, and this fact is what makes Melodic Over-
writing particularly relevant to MDT. Any case of Melodic Overwriting is evi-
dence that one daughter in reduplication contains an empty morph that the other
does not, with the result that the two daughters are different phonologically –
but identical semantically. Melodic Overwriting is a prototypical case of what
MDT predicts in reduplication.

2.2.5 Double melodic overwriting

By the same token that one daughter can contain an empty morph not found in
the other, the MDT approach to reduplication leads one to expect constructions
in which both daughters contain empty morphs not found in the other. This
pattern is indeed attested. One well-known case, described in Yip 1982, occurs
in Chinese secret languages. Another, from Hua, is presented here.
In the Papuan language Hua (Haiman 1980), verbs are intensified through
total reduplication, followed (in most caes) by the helping verb hu. One
44 Evidence for morphological doubling

subpattern of reduplication replaces the final V with /u/ in the first copy and
final V with /e/ in the second copy: this is Melodic Overwriting in both copies
(data from Haiman 1980:126):

(26) kveki ‘crumple’ kveku kveke hu ‘crumple’

ebsgi ‘twist’ ebsgu ebsge hu ‘twist and turn’
ftgegi ‘coil’ ftgegu ftgege hu ‘all coiled up’
ha-vari ‘grow tall’ ha-varu ha-vare hu ‘grow up’

Sticking with the affixation approach to Melodic Overwriting, the Hua redu-
plication construction can be analyzed as calling for two different stem types:
a u-stem, formed by suffixing empy -u, and an e-stem, formed by replacing the
final stem vowel with -e.13

(27) kveku kveke

[kvek-u]u-stem [kvek-e]e-stem

The daughters of the Hua reduplication construction are formed by the stem-
forming constructions illustrated below.

(28) u-stem construction e-stem construction

[ ][F] [ ][F]

[ ][F] u [ ][F] e

[kveku] [kveke]

/kveki -u/ /kveki -e/

Both constructions in (28) are associated with a phonological requirement that

the output and input contain the same number of syllables, which results in
root-vowel deletion by standard means which it is not necessary to spell out
here.
The Hua reduplication construction illustrated in (27) is like all other MDT
reduplication constructions in calling for two daughters which are semantically
identical; it adds the further specification that the daughters must be of different
lexical types. Each has a semantically empty morph not present in the other,
and as a result the two are not phonologically identical.
If Hua had called for two daughters of the same lexical type, the expected
result would be phonological identity. Such a case occurs in Siroi, a non-
Austronesian language of Papua New Guinea, in which adjective reduplication
 Morphotactic asymmetries: empty morphs 45

signifies plurality. According to Wells (1979:37), “[t]he infix -g- replaces the
central consonant in two-syllable words and is added in one-syllable words,”
as shown below.14 The digraph “ng” represents the prenasalized stop [ŋg]:
(29) a. tango maye → tango mage-mage
‘man’ ‘good’
‘a mature man’ ‘mature men’
b. tango sungo → tango sugo-sugo
‘man’ ‘big’
‘a ruler’ ‘rulers’
c. tango kuen → tango kugen-kugen
‘man’ ‘tall’
‘a tall man’ ‘tall men’

This pattern can be described in terms of g-infixation to the beginning of the

final syllable; the intrusion of g causes the deletion of an input onset, plausibly
due to a prohibition on the consonant clusters that would otherwise result. This
analysis of Siroi thus parallels the analysis of Hua point for point, with the
exception that only one stem-forming construction (g-infixation) is involved,
rather than two.
This discussion of stem-forming constructions, and lexical stem types, draws
on a growing literature, not about reduplication, which includes Aronoff 1994;
Koenig & Jurafsky 1995; Orgun 1996; Blevins 2003. The reader is referred
to these works for further discussion of the role of stem types in morphology
generally.

2.2.6 Tier replacement

In the constructions examined thus far, a particular syllable constituent is tar-
geted for replacement by an empty morph, making affixation (whether adfix-
ation or infixation) a straightforward analysis. But reduplication constructions
also admit the possibility of modifications that fall in the fuzzy area between
affixation and morphologically conditioned phonology.
The Papuan language Amele has a construction which Roberts (1987; 1991)
calls the “irregular iterative”; it reduplicates the verb stem and adds a meaning
of haphazardness, spasmodicity, intermittency, etc. (p. 133). Morphologically
the construction doubles the verb root (with suffixes being added at the right
edge); phonologically, the vowels in the second copy are replaced. As shown
in (30a), input /a/ vowels are replaced with /u/; the same is true of (b), whose
stems have an input [-high]-i vocalism. In (c) are inputs whose input vowels
are all /i/. In this case the vowels of the second copy are replaced by /o/.15 The
data in (a–b) are from Roberts 1991:135; those in (c) are from p. 136.
46 Evidence for morphological doubling

(30) a. bala-doʔ ‘to tear’ bala-bulu-doʔ ‘to tear and scatter’

ʔaʔagan-eʔ ‘to talk in sleep’ ʔaʔagan-ʔuʔugun-eʔ ‘to talk sporadically
in sleep’
fag-doʔ ‘to pierce’ fag-fug-doʔ ‘to stick all over’
b. faliʔ-doʔ ‘to turn’ faliʔ-fuluʔ-doʔ ‘to revolve’
lahi-doʔ ‘to shake s.t.’ lahi-luhu-doʔ ‘to shake s.t.
all over’
c. ʔifiliʔ-doʔ ‘to open s.t. out’ ʔifiliʔ-ʔofoloʔ-doʔ ‘to open s.t. out
all over’
gili-doʔ ‘to move’ gili-golo-doʔ ‘to move from side
to side’

A similar situation obtains in Siroi; when verb roots reduplicate to indicate

“plurality and/or intensification and variableness of an action . . . all vowels in the
first form change to i,” e.g. kare ‘be hard’, kiri-kare ‘wither’; kutuŋ ‘move’, kitiŋ-
kutuŋ ‘wobble’ (Wells 1979:36).16 Tier replacement is not limited to vowels;
it also commonly affects tone. In Hausa adverbial reduplication, for example,
discussed in more detail in Chapter 3, input tones are completely replaced
in both copies by a distinct melody specific to the reduplication construction
(Newman 1989b). In Hausa, complete tonal tier replacement is also common in
affixation; it is not specific to reduplication (see, for example, Newman 1986;
2000).
Tier replacement is difficult to describe in the standard terms of item-based
morphology. Affixing /i/ to the Siroi stem kare is not going to produce kiri in
any automatic way. The analysis offered here of tier replacement effects like
these is completely parallel to the analysis of affixational Melodic Overwriting:
kiri is derived from kare, in Siroi, by a stem-forming construction whose only
effect is to replace all vowels with /i/. The replacement itself adds no semantics
to the stem:

(31) ⎡ Semantics = [F] ⎤

⎢Phonology: vowels → /i/ ⎥ [kiri] ‘be
⎢ ⎥ hard’, i-stem
⎣ Stem type = i-stem ⎦
|
[Semantics = [F]] /kare/ ‘be hard’

Although its manifestation is different phonologically, vowel replacement

in Amele and Siroi is morphologically equivalent to the cases traditionally
called Melodic Overwriting, which involve affixation of a more canonical,
segmentally overt affix. In both types of cases, reduplication calls for daugh-
ters which are products of different stem-forming constructions. Since the
 Synonym and antonym constructions 47

stem-forming constructions in question are semantically null, the essential

semantic identity is maintained. However, phonological identity is disrupted.
Chapter 3 goes into much more detail about how phonology is associated
with stem-forming constructions. For now, discussion will focus on the fact
that the phonology of the daughters can diverge, as a result of semantically null
morphology, while their meanings stay the same. This is strong support for the
Thesis of Semantic Identity.

2.3 Synonym and antonym constructions

This section presents a second major source of supporting evidence for the
Thesis of Semantic Identity, namely cases in which the sets of semantically
contentful morphemes in each daughter in reduplication are different. One type
of case is typified by Sye, in which the two daughters in the reduplication
construction may under certain specific circumstances be suppletive allomorphs
of the same morpheme. In meaning and function the two are identical, but in
phonological form they are lexically distinct.
Once the potential for selecting different lexical allomorphs of the same mor-
pheme is recognized, it becomes clear that MDT correctly predicts the existence
of a second, actually quite voluminous set of constructions which have been
observed to be very similar in character to reduplication, namely juxtaposition
constructions in which the two daughters are synonyms, or near-synonyms, or
in some cases antonyms, of each other. These constructions have played little
role in the development of phonological reduplication theories of reduplication
(though are well known in the repetition literature; see, for example, Wierzbicka
1986; Haiman 1998; Ourn & Haiman 2000). Singh (1982) makes, to our knowl-
edge, the first argument in the literature that these constructions are intimately
related to reduplication and should be accounted for in the same way.

2.3.1 Root allomorphy

We begin with a discussion of root reduplication in languages from two different
language families, both of which exhibit root allomorphy which is at least to
some degree lexicalized, i.e. suppletive. In both families there are conditions
under which root reduplication manifests both allomorphs, such that the two
copies in reduplication are based on different lexical allomorphs of the same
morpheme. Such constructions clearly revolve around semantic identity; the
construction is drawing twice from the lexicon, with different results in each
daughter.17
48 Evidence for morphological doubling

2.3.1.1 Kawaiisu
Root allomorphy interacts with reduplication in Kawaiisu (Numic, Uto-
Aztecan; Zigmond, Booth & Munro 1990). Kawaiisu retains the remnants of
historic alternations, primarily between p∼v, p∼b, t∼r, t∼d, k∼g, kw∼gw,
which show up synchronically as allomorphy, e.g. for ‘head,’ toci ∼ roci, or
for ‘taste,’ kama ∼ gama. Only some roots show allomorphy. This lack of pre-
dictability, coupled with the fact that for roots in /p/ and /t/ the nature of the
allomorphy is unpredictable, requires the allomorphy to be treated as lexically
suppletive.
For those roots exhibiting allomorphy, the voiced-initial allomorph is used
in second position in compounding (32a) and in repetitive/inceptive aspectual
reduplication (32b) (Zigmond, Booth & Munro 1990:8, 97):

(32) a. Compounding
i. kama ∼ gama ‘taste (intr.)’ ʔowa-gama- 80, 211
‘to taste salty’
ii. kar - ∼ gar - ‘sit’ kwinuur -gar-d 221
(name of mountain
near Tehachapi)
cf. kwinuur -d 221
‘to prepare yucca’
iii. toci ∼ roci ‘head’ c ga-roci 199
‘rough-head = tangle-
haired’
b. Reduplication
i. kaʔa ∼ gaʔa ‘eat’ ka-gaʔa-na=ina ʔ vi 8, 97
‘he’s starting to eat now’
ii. tono ∼ dono ‘hit, pierce, etc.’ to-dono-kwee-d =ina 97, 284
‘he stabbed him repeatedly’
iii. kiya ∼ giya ‘play, laugh’ ki-giya 214
iv. tahna ∼ rahna ‘put down/away’ ta-rahna 272
v. t niya ∼ d niya ‘to tell’ t-dniya 280

The fact that the two daughters in reduplication can consist of different lexi-
cally listed allomorphs (potentially truncated) of the same morpheme strongly
supports the prediction of MDT that the essential identity in reduplication is
semantic but not phonological. No phonological copying theory of reduplica-
tion can account for this sort of effect.
The productivity of this particular construction in Kawaiisu is somewhat
questionable; not all roots exhibiting allomorphy in general show different
allomorphs in reduplication. However, Kawaiisu is by no means the only
example of its kind.
 Synonym and antonym constructions 49

2.3.1.2 Central Vanuatu

Verb root allomorphy, largely centering on the root-initial consonant, is
widespread in the languages of Central Vanuatu, which also tend to have verb
root reduplication. Useful discussions of Vanuatu verb root allomorphy can be
found in Walsh 1982, Crowley 2002b:31ff., Lynch, Ross and Crowley 2002:44,
and, especially, Crowley 1991. In numerous languages, represented by those
in the table below, roots have two allomorphs, conditioned by morphological
context. One, termed “primary” or “basic” in the literature and referred to here
as Stem1, is generally the elsewhere case; the other, termed “secondary” or
“modified” and referred to here as Stem2, is found in a more restricted set of
morphological contexts. Root allomorphy is widely assumed to trace back his-
torically to one or more prefixes which fused with the verb roots to form the
Stem2 allomorphs. Crowley (1991) argues that this fusion may have occurred
independently several times, rather than being a reconstructable property of
Proto-Central Vanuatu. The following chart shows the differences in the ini-
tial consonant across root allomorphs in six languages representing the simplest
type of root allomorphy in which there are only two variants. The Nāti, Nakana-
manga, Namakir and Bierebo data come from Crowley 1991:198ff.; the Raga
and Apma data come from Walsh 1982:
(33)
Nāti Nakanamanga Namakir Bierebo Raga Apma

Stem1 Stem2 Stem2 Stem2 Stem2 Stem2 Stem2

v- mp- p- b- p- b- b-
w- mpw- pw - b- pw - bw-
vw- bw-
r- ntr- t- d-
t- nt- t- d- nd- d-
k- ŋk- ŋ- ŋ- ŋk- g-
g- ŋg -
ʔ- ŋk-
c- nj-

The morphological contexts calling for Stem2 vary across languages. In

the Nāti language of Malakula, for example, Stem2 is used “when there is a
preceding future tense prefix . . . or when the verb carries the negative prefix
sa-” (Crowley 1991:198).18 In Raga, Stem2 is used when the root “is directly
preceded by the action-in-progress marker . . . by any verb-aspect marker ±
hav ‘negation’ + mom ‘still, yet’ . . . [or when] preceded by ba ‘verb ligature’
when the verb preceding ba is in a context marked for action-in-progress”
(Walsh 1982:237). In Nakanamanga, Stem1 is used when any of the following
50 Evidence for morphological doubling

prefixes directly precede: conditional pe-, intentional ŋa, imperative pw a; in

nominalizations, marked by the prefix na- and suffix -ana; and when the verb
serves “as a postmodifying adjective or adverb, or as the second part of a
compound” (Crowley 1991:201); Stem2 is used elsewhere.
Though data showing reduplication in a context calling for Stem2 is not
plentiful in the published sources, several examples suggest that when a redu-
plicated verb root appears in morphological contexts requiring Stem2, the first
copy of the verb root assumes its Stem2 form while the second copy is Stem1.
The reduplicated verb thus consists of two different allomorphs.
In Raga (Walsh 1982; Crowley 2002a), for example, one of the morphological
contexts conditioning Stem2 is an immediately preceding “action-in-progress”
prefix (m- ∼ mwa- ∼Ø) (Walsh 1982:237). Alternations in root shape driven
by the presence or absence of this prefix are shown in (34a), below. Root
allomorphy in reduplication is illustrated in (b):
(34) Stem1 context Stem2 context
ŋ
a. no- gu vano-ana na-m bano
‘my going’ ‘I go’
b. i. ra-n tu-tunu ra-m du-tunu
‘they cooked on hot stones’ ‘they are cooking on hot stones’
ii. na-n van-vano na-m ban-vano
‘I used to keep on going’ ‘I keep on going’

Similar data is reported for Apma, spoken on the same island (Pentecost) as
Raga; Walsh (1982:238) cites the following examples of reduplicated roots in
Stem1 and Stem2 contexts:

(35) Stem1 context Stem2 context

a. te wiri-wiri na-mwa bwiri-wiri
‘he/she splashed water on face ‘I splash water on face repeatedly’
repeatedly’
b. te kaha-kahaabe mwa gaha-kahaabe
‘it (=bird) went from branch to ‘it (=bird) goes from branch to
branch’ branch’
c. te val-valtoo na-m bal-valtoo
‘he/she/it walked with legs apart’ ‘I walk with legs apart’

In Apma, as in Raga, the “action-in-progress” prefix conditions Stem2 (Walsh

1982:238).
Crowley (1991) does not explicitly mention reduplication in his brief sketch
of Nāti verb root allomorphy (pp. 198–200), but one example suggests behavior
parallel to that of Raga and Apma. In (36a), the Nāti verb ‘hold’ appears to be
 Synonym and antonym constructions 51

reduplicated; both halves are identical. The reduplicated verb root occurs in a
context where Stem1 is called for. In (b), however, where the future prefix a-
calls for a Stem2 base, the first copy of the root assumes its Stem2 shape, while
the second is Stem1 (Crowley 1991:198):

(36) a. Reduplicated verb root in Stem1 context: both copies = Stem1

ntar-vur-vur
1du.nonfut-hold1 -hold1 ‘we hold/held’
b. Reduplicated verb root in Stem2 context: first copy = Stem2,
second copy = Stem1
ntar-a-mpur-vur
2du.incl-fut-hold2 -hold1 ‘we will hold’

In the cases seen thus far, root allomorphy is generally phonologically trans-
parent. In Nāti, for example, Stem2 can easily be derived phonologically from
Stem1 through the addition of an initial nasal element, which in turn stops the
initial portion of a following continuant. Different analyses have been given
to phonologically analyzable mutations of this kind in the literature; see, e.g.,
the extensive literature on consonant mutation in Celtic and in West African
languages (Fula [Arnott 1970], Mende [Innes 1971; Conteh, Cowper & Rice
1985], Sereer [McLaughlin 2000], and many others). A morphological analy-
sis would attribute the nasal to a semantically bleached morpheme occurring in
the selected set of morphological contexts; alternatively, a more phonological
approach would posit a stem-initial nasalization rule triggered in the same set
of contexts. This issue need not be resolved here, since languages like Nāti do
not provide a test case for MDT in any event. Since the modification is derived
by the grammar, any theory of reduplication could give a cyclic account of the
mutation effects, on the assumption that the mutation-triggering prefix com-
bines with the output of verb root reduplication. As with any other stem in
a mutation context, only the consonant immediately following the prefix is
affected.
Although the languages in (33) do not distinguish MDT from other theories,
the languages that do are only a small step away. Recall that the mutations which
are apparently productive and clearly phonologically general in the languages
in (33) derive historically from prefix-root fusion. Any language in which the
reflexes of that fusion have been muddied by reanalysis or sound change will
not yield to a cyclic anlysis in which reduplication logically precedes mutation.
When a reduplicated verb root occurs in a mutation context, such languages
will require the direct selection, from the lexicon, of different lexically listed
allomorphs of the verb root in question.
52 Evidence for morphological doubling

The first such case to be examined here is Sye, which exhibits two-fold
allomorphy but has lost the phonological generality of the mutation process.
We then turn to another set of Vanuatu languages, Paamese and Southeast
Ambrym, whose more complex verb root allomorphy also precludes a simple
phonological account.

2.3.1.2.1 Sye Sye, spoken on the island of Erromango, resembles other lan-
guages of Southern and Central Vanuatu in exhibiting both verb root allomorphy
and reduplication. The discussion of Sye in this section draws heavily on the
detailed description and insightful analysis of Crowley 1998; 2002d. Verb root
reduplication, which is total, indicates that “an action takes place in a vari-
ety of locations at once, e.g. avan‘walk’, avan-avan ‘walk all over’” (Crowley
2002d:714). Sye has two series of verb allomorphs, termed here Stem1 and
Stem2. Stem2 forms appear in a collection of seemingly unrelated morpholog-
ical environments (e.g. in verbs inflected for present or future tense, or bearing
conditional or past habitual prefixes; Crowley 2002d:704). Stem1 forms appear
elsewhere and can be understood as the default stem type. The distribution of
stem alternants is illustrated in (37) for the verb ‘provoke’. Its Stem1 alternant,
arinova, is given in (a); in examples (b) and (c), ‘provoke’ co-occurs with future
tense prefixes, and thus appears in its Stem2 form, narinowa. Data are from
Crowley 2002d: 711:

(37) a. etw-arinowa-g
2sg.imp.neg-provoke1- 1sg ‘Don’t provoke me!’
b. co-narinowa-nt
3sg.fut-provoke2- 1pl.incl ‘(S)he will provoke us’
c. kokwo-narinowa-nd
1inc.du.fut-provoke2- 3pl ‘We will provoke them’

The verb ‘provoke’ is an example of what Crowley calls a weak verb. A weak
verb root exhibits a completely transparent relationship between its Stem1 and
Stem2 allomorphs: Stem2 consists of Stem1 with the addition of an initial nasal.
Data are taken from Crowley 2002d:

(38) Stem1 Stem2

tovop ntovop ‘laugh’ 704

avan navan ‘walk’ 704
arinowa narinowa ‘provoke’ 711

Strong verbs, on the other hand, exhibit a much less transparent relationship
between their allomorphs. While historically the Stem2 verb alternants appear
 Synonym and antonym constructions 53

to derive from a-accretion and nasalization, synchronically they are arbitrary

to a sufficient degree that at least some, if not all, root pairs must be lexically
listed (Crowley 1998:84).19

(39) Stem1 Stem2 Gloss Crowley source

ehvoh ahvoh ‘white’ 2002d:704

evcah ampcah ‘defecate’ 2002d:704
evinte avinte ‘look after’ 1998:83–84
evsor amsor ‘wake up’ 1998:84
evtit avtit ‘meet’ 1998:83–84
mah amah ‘die’ 2002d:704
ocep agkep ‘fly’ 1998:84
ochi aghi ‘see it’ 1998:84; 2002d:704
okili agkili ‘know’ 2002d:704
omol amol ‘fall’ 1998:79
oruc anduc ‘bathe’ 1998:84; 2002d:704
otvani etvani ‘split’ 2002d:720
ovoli ampoli ‘turn it’ 1998:84
ovyu- avyu- causative 1998:84
ovyu- ampyu- desiderative 1998:84
owi awi ‘leave’ 1998:84; 2002d:704
pat ampat ‘blocked’ 1998:84; 2002d:704
pelom ampelom ‘come’ 2002d:712
vag ampag ‘eat’ 1998:84; 2002d:704

Comparison of evcah → ampcah and evtit → avtit, for example, shows a lack
of complete phonological regularity in the mapping between root allomorphs.
Even where the mapping is systematic, e.g. replacement of absolute initial o
with a, but prefixation of am to consonant-initial words, there is no unitary
process that can be appealed to of the sort that makes a phonological account
of weak roots possible. At least some Sye root alternants must be lexically
listed; we make the simplifying assumption here, adopting Zuraw’s (2000)
approach to patterned exceptions, that the root alternants are all listed, although
the argument would not change if the phonology of Sye were allowed to derive
some of the allomorphy. The lexicon provides Stem1 and Stem2 alternants of
each root; affixal constructions specify the stem type they combine with. As
seen in (39), future tense prefixes select for a stem of type Stem2; the second
person singular negative imperative prefix selects for a stem of type Stem1
(or simply does not specify a stem type, Stem1 being the default).
We turn next to reduplication, an arena in which Sye patterns much like the
other Vanuatu languages seen thus far. In those morphological contexts calling
for Stem1, a reduplicated verb consists of two copies of Stem1. In contexts
54 Evidence for morphological doubling

calling for Stem2, by contrast, reduplication shows morphotactic divergence:

the first copy is Stem2, while the second is Stem1. This situation is illustrated
below using ‘fall’, whose Stem1 and Stem2 forms are omol and amol, respec-
tively (Crowley 1998:79). In (40), ‘fall’ is shown, reduplicated, in a prefixal
context calling for Stem2:
(40) cw-amol2 -omol1
3pl.fut-fall2 -fall1
‘they will fall all over’

As these Sye data show, reduplicants are not always phonological copies of
their bases, as phonological theories of reduplication would require. What is
instead occurring in (37) is semantic agreement between suppletive allomorphs:
it is morphological doubling, not phonological copying.

2.3.1.2.2 Paamese and Southeast Ambrym Both Paamese and Southeast

Ambrym exhibit a larger set of verb root alternants than what has been seen
thus far in Sye and the other Central Vanuatu languages discussed. In Southeast
Ambrym, roots have up to five allomorphs each, one default allomorph (Stem1)
and up to four other allomorphs (Stem2–Stem5), each restricted to a unique
subset of irrealis morphological contexts. Root allomorphs are distinguished in
the initial consonant position. As Crowley (1991:185) states in his summary
of the situation, “[v]erbs in Southeast Ambrym can be assigned to one of nine
classes (or conjugations) according to the nature of the mutation pattern of the
initial segment of the root.” The classes exhibiting alternations are presented
below (the ninth class, all consonant-initial roots, does not alternate):

(41)
Class Stem1 Stem2 Stem3 Stem4 Stem5

I t- t- d-

II x- x- g-

III Ø- x- m- v- g-

IV h- h- m- v- g-

V h- x- m- v- g-

VI h- h- m- v- b-

VII h- v- m- v- b-

VIII v- v- b-
 Synonym and antonym constructions 55

There is no immediate phonological generalization available. According to

Crowley (1991:185), Stem2 is used after the negative prefixes nā- and tā-;
Stem3, after the first person singular immediate irrealis prefix na-; Stem4, after
the second person singular immediate irrealis prefix o- and all nonsingular
immediate irrealis prefixes; and Stem5, after all realis prefixes unless followed
by the past tense marker te-. Most of these contexts are illustrated below for the
verb ‘dig’ (Crowley 2002c:665):

(42) a. va-hil ‘(s)he is about to dig’ (Stem1)

b. na-mil ‘I am about to dig’ (Stem3)
c. o-vil ‘you are about to dig’ (Stem4)
d. Ø-gil ‘(s)he digs/dug’ (Stem5)

Nothing in the overt segmental phonology of the prefixes could explain the
root allomorphy. Class III, whose Stem1 forms are apparently vowel-initial,
suggests a morphological account on which the basic pattern is to prefix x- to
Stem1 to form Stem2, prefix m- to form Stem3, prefix v- to form Stem4, and
prefix g- to form Stem5. However, the existence of four different patterns for
h-initial roots (classes IV-VIII) makes it impossible to derive Stem2 by any
regular means from Stem1.
Though Crowley does not provide any examples of reduplication, he does
state that Stem1, the default, is used “as the second of a pair of reduplicated
syllables,” strongly suggesting that Ambrym behaves like Sye and the other
languages discussed above in potentially using different allomorphs of the same
verb root in reduplication, the first conforming to the requirements of the higher
affix and the second consisting of the default stem allomorph. Parker (1968)
mentions reduplication only briefly, indicating that reduplication copies the first
CV of the stem (e.g. ka-kal ‘scratch,’ pi-pili ‘be red,’ mu-mun ‘drink,’ etc.). He
does not explicitly address the question of what happens when a reduplicated
verb occurs in a mutation environment (which, if our understanding is correct, he
calls “Aorist”). However, Parker does provide one suggestive example, namely
the reduplicated verb go-xoles ‘he exchanges.’ The first (truncated to CV) copy
of the root begins with g, while the second begins with x. This alternation would
put ‘exchange’ into Stem class V in the table in (41).
For Paamese, which has a similar system, we are fortunate to have more data.
The following discussion is based on Crowley 1982; 1991. The description in
Crowley 1991, covering all dialects, revises that of Crowley 1982, primarily
covering a southern dialect, in a number of minor respects, including the number
of verb stem classes posited. Presented below is the classification of verbs from
Crowley 1991:190, based on the types of allomorphy they show across different
56 Evidence for morphological doubling

morphological contexts. Verbs have up to five allomorphs each, depending on

class. The Northern (N), Central (C) and Southern (S) dialects of Paamese differ
in the verb root allomorphs occurring in contexts C and D:

(43) Primary Secondary

A B C D
(N/C/S) (N/C=S)

Stem1 Stem2 Stem3 Stem4 Stem5

I t- t- r- r-/d-/d- r-/d-
II t- t- r- mur-/d-/d- mur-/d-
III r- r- r- mur-/d-/d- mur-/r-
IV k- k- k- k-/g-/g- k-/g-
V Ø- k- k- k-/g-/g- k-/g-
VI Ø- k- k- k-/g-/ŋ- k-/ŋ-
VII Ø- k- k- muk-/g-/ŋ- k-/ŋ-
VIII h- h- v- v- h-
IX Ø- Ø- mu- mu- mu-/Ø-
X C- C- C- muC-/C-/C- C-
XI C- C- C- C- C-

The allomorphs labeled “Secondary” each have a tightly restricted morpho-

logical distribution; the allomorph labeled “Primary” occurs in the greatest vari-
ety of morphological contexts and is arguably the default. Stem2 is used in “the
second part of compound noun construction in which a noun is compounded
with a verbal stem”; Stem3 is used for serialized verbs with no inflectional
prefix; Stem 4 occurs after realis or negative prefixes; Stem5 is used in realis or
negative verbs which are inflected and initial in a serial verb construction – or
which are fully reduplicated (Crowley 1991:190–91). Stem1, or the “Primary”
allomorph, occurs following distant irrealis prefixes, immediate irrealis pre-
fixes, imperative prefixes, potential prefixes, the adjectival derivational prefix
ta-, “when there is no preceding morpheme and the verb is nominalized by -en,”
and – crucially – “in the second of a pair of reduplicated syllables” (p. 190).
Thus in reduplication there are apparently two possibilities. There are two
phonological patterns of verbal reduplication, CV- and CVCV-, both associated
with the same set of meanings: detransitivizing, habitual, random. Following
Crowley (1982), we will not attempt to distinguish the constructions here; they
behave alike for purposes of stem allomorphy.20 If the reduplicated verb stem
is realis or negative, both copies are expected to appear in their Stem5 form.
If, however, the reduplicated verb stem occurs in another context, the first copy
exhibits the allomorph expected in that context, while the second appears in
 Synonym and antonym constructions 57

its Primary, or default, form, i.e. Stem1. Examples taken from Crowley 1982
(C82) and Crowley 1991 (C91) illustrate these generalizations, below:

(44) a. na-ro-ngani-ani [naronganian] C82:164

1sg.real-neg-eat-eat In ‘I am not a big eater’
b. na-gulu-ulu [nagulūl] C82:209
1sg.real-shake-shake ‘I rattled’
c. ko-galo-kalo [kogalokal] C82:209
2sg.real-move-move ‘you moved’
d. ro-gu-kulu-tei [rogukulutei] C82:141
3sg.real.neg-swim-swim-part ‘(they) don’t swim’
e. Ø-vi-hiramu [vihiram] C82:155
3sg.real-flicker-flicker ‘flickered’
f. Ø-museh-seh (N) C91:195
∼ Ø-museh-a-seh (C/S)
3sg.real-breathe-breathe ‘he/she puffs/puffed’
g. vane-hane [vanehan] C82:153
3sg.real-copulate-copulate ‘copulates all over =
is promiscuous’
h. vutu-hutu [vutuhut] C82:154
3sg.real-abuse-abuse ‘swore’
i. gaa-kaa [gāka] C82:164
3sg.real-travel-travel ‘travels’
j. a-ga-kani-e [agekani] C82:152
3pl.real-eat-eat-3sg in ‘is this edible?’
k. a-muke-kaa [amukeka] C82:153
3pl.real-fly-fly ‘flew in all directions’
l. a-da-tangosaa [adetangosa] C82:153
3pl.real-rise-rise ‘went up’
m. a-laa-laapoo [alālāpo] C82:153
3pl.real-fall-fall ‘they fell down all
over the place’
n. re-demi-ene [redemien] C82:227
think-think-nom ‘ideas’

2.3.1.2.3 Analysis In Sye, Southeast Ambrym, and Paamese, selectional

restrictions on the part of affixes and other morphological constructions, includ-
ing serial verb constructions, compounding, and reduplication, determine which
allomorph of the verb stem is selected. In Paamese, where for a given verb there
may be as many as five stem types lexically listed, realis and negative prefixes,
for example, as seen in (44), select for Stem4. The verb reduplication con-
struction is less specific, lexically, about stem type; while the second daughter
always assumes its Stem1 form, the construction does not stipulate a particular
stem type for the first daughter which can be any of the stem types. Rather, the
58 Evidence for morphological doubling

construction stipulates only that the stem type of the first daughter determines
the stem type of the construction as a whole:

(45) Reduplication construction (Paamese)

[ ]Meaning = ‘F + habitual/random/etc.’; stem type = x

[ ]Meaning = [F]; stem type = x [ ]Meaning = [F]; stem type = 1

This permits Paamese reduplicated verbs to occur in a range of stem-selecting

contexts, as illustrated above.
An interesting exception to the generalization that the second stem in a redu-
plicated Paamese verb is always Stem1 is manifested in what Crowley terms
class V verbs (which in Crowley 1991 corresponds most closely to IX). In
this class, Stem1 and Stem2 have no prefix; Stem4 and Stem5 begin with mu-.
(Whether or not Stem3 begins with mu-is a little unclear; Crowley’s verb clas-
sification chart on p. 190 indicates that it does, but his prose description of
the class on p. 195 indicates that it does not.) According to Crowley (1982),
verbs in this class behave differently under reduplication according to whether
their Stem1 form begins with a consonant or a vowel. “[C]onsonant-initial
roots reduplicate on the basis of their [Stem1] roots, while vowel-initial roots
reduplicate on the basis of their [Stem4] root” (p. 126). Example (46) shows
a vowel-initial class IX stem appearing in its Stem4 form in both copies in
reduplication:

(46) Stem1 allomorph = uasi; Stem4 allomorph = muasi

na-ro-mu-muasi [naromumuas] Crowley 1982:141
1sg-real.neg-hit4 -hit4

The presumed reason that Stem4, rather than Stem1, appears as the second
daughter is that it is phonotactically superior, in possessing an onset. Crowley
(1982) points out (p. 126) that were a form like uasi to reduplicate as u-uasi,
“the regular phonological rules of the language would eliminate all trace of
[reduplication] having taken place.” The fact that Onset can outrank the pref-
erence for Stem1 supports the claim that the usual use of Stem1 form in the
second daughter position derives from its status as the default stem. In the terms
of Optimality Theory, the grammar would state a general preference for Stem1
allomorphs, which is overridable (a) by the specific requirements of higher
affixes or constructions or (b) by phonotactic constraints such as onset, as
illustrated schematically below:
 Synonym and antonym constructions 59

(47) Reduplication construction (Paamese, class IX)

[ ]Meaning = ‘F all over’; stem type = x

[ ]Meaning = [F]; stem type = x [ ]Meaning = [F]; ONSET » stem type = 1

2.3.1.3 Wrapup
This section has verified one of the most striking and unique predictions of
MDT, namely that reduplication constructions are able to use distinct allo-
morphs of the same morpheme in the two daughters. This prediction follows
from the thesis of MDT that the two copies in reduplication have independent
inputs. We turn in the next section to a larger-scale instantiation of this same
general phenomenon.

2.3.2 Synonym constructions

In exactly the same way that MDT predicts the existence of reduplication con-
structions in which the two daughters are different suppletive allomorphs of
the same morpheme, MDT also predicts the existence of constructions whose
two daughters are not the same morpheme at all but are semantically identi-
cal lexemes. Constructions of this sort abound; they are typically referred to
not as reduplication but by terms like “redundant compounds” (Singh 1982),
“semantically symmetrical compounds” (Ourn & Haiman 2000), or “synonym
compounds” (Nguyen 1997). Here we will adopt the general term “synonym
constructions” for these cases.
In what may be the first statement in the literature of the principal idea
underlying MDT, Singh 1982 notes parallels between reduplication and a con-
struction he refers to as “redundant compounds” in Hindi. These are noun-noun
compounds in which the nouns are synonymous but differ etymologically: the
first noun is native, while the second is of Perso-Arabic origin (p. 345). The
overall meaning is ‘noun, etc.’; the fact that echo-word reduplication with this
same meaning is prevalent through Asia (e.g. kota-gita in Kolami, kitap-mitap
in Turkish, etc.) supports Singh’s characterization of the Hindi constructions as
reduplicative. Examples from Singh 1982:346 are given below:

(48) tan-badan ‘body-body’ dharm-imān ‘religion-religion’

vivāh-šādi ‘marriage-marriage’ sneh-muhabbat ‘love-love’
dhan-daulat ‘money-money’ lāj-šarm ‘deference-deference’
šāk-sabji ‘vegetable-vegetable’ nāta-rista ‘relation-relation’
60 Evidence for morphological doubling

Abbi (1991:24) provides similar examples in which the first word is native and
the second comes from Urdu.
Synonym constructions are quite pervasive (for general discussion see Abbi
1991:24 and, especially, Ourn & Haiman 2000). To cite just two examples
outside of Hindi, Li and Thompson (1981:68ff.) describe Chinese as having
verbal compounds in which “[t]he two verbs that constitute a parallel verb
either are synonymous or signal the same type of predicative notions,” e.g. gòu
măi ‘buy-buy = buy,’ měi-lı̀ ‘beautiful-beautiful = beautiful.’ Ourn and Haiman
(2000) discuss a similar construction in Vietnamese, examples of which, taken
from Nguyen 1997, are provided below:

(49) ma.nh-kho
e ‘strong = strong = well in health’ 67

do’ bân ‘dirty = dirty = filthy’ 67

lu’ò’i-biêng ‘lazy = lazy = slothful’ 67
toi-l ˜
.̂ ôi ‘offense = fault = sin’ 70
kêu-go.i ‘to call = to call = to call upon, appeal’ 70

Clearly, these constructions cannot be derived by phonological copying. A

phonological copying theory would have to declare them unrelated to redupli-
cation. Yet this would be a mistake, as Singh argues; for one thing, they are
semantically identical to another type of construction in Hindi, a productive
process of echo-word reduplication in which the exact same stem is doubled.21
As is definitional of echo-word constructions, the second copy is required to
begin with a fixed segment, v, which replaces an existing onset. From p. 348:

(50) roti ‘bread’ roti-voti ‘bread, etc.’

namak ‘salt’ namak-vamak ‘salt, etc.’

Foreshadowing the theory of MDT, Singh argues for treating both constructions
as reduplication:

Reduplication is generally thought of as a morphological process that copies a

word either as is or in some phonologically modified form. In order to regard
our redundant “compounds” as outputs of a rule of . . . reduplication, we would
have to modify the notion of reduplication to include reduplication of lexical
items with or without certain morphological features such as [+Native]. We
will, in other words, have to make a three-way distinction: reduplication with
no modification whatsoever, reduplication with phonological modifications,
and reduplication with morphological modifications. (1982:349–350)

What Singh is suggesting here appears to be the same idea that MDT is
built on: reduplication can (or according to MDT must) double a morpho-
logically abstract entity, one defined essentially by semantic properties. The
 Synonym and antonym constructions 61

MDT schemas for the two structurally similar constructions in (48) and (50)
are given below:
(51)
Hindi synonym construction (= (48)) Hindi echo word construction (= (50))
⎡ Syntax = N ⎤ ⎡ Syntax = N ⎤
⎢Semantics = ‘x etc.’ ⎢Semantics = ‘x etc.’⎥
⎣ ⎦ ⎣ ⎦

⎡ Syntax = N ⎤ ⎡ Syntax = N ⎤ ⎡ Syntax = N ⎤ ⎡ Syntax = N ⎤

⎢ Semantics = ‘x’ ⎥ ⎢ Semantics = ‘x’ ⎥ ⎢ Semantics = ‘x’ ⎥ ⎢ Semantics = ‘x’ ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥⎢ ⎥
⎣Stratum = +Native⎦ ⎣Stratum = −Native⎦ ⎣Stratum = αNative⎦ ⎣Stratum = αNative⎦

2.3.3 Beyond synonyms

By the same token that constructions have the ability to require daughters to
be identical (or nearly so) semantically, we naturally expect other relationships
to be possible as well, i.e. subset and opposite relationships. Their existence
confirms the predictions of MDT.
Ourn and Haiman (2000) catalog four types of constructions which, like
reduplication, juxtapose two morphological constituents with a fixed semantic
relationship to one another: in addition to synonym constructions, which we
have just seen, they discuss constructions which juxtapose near synonyms, as
in “law and order”; members of the same semantic set, as in “field and stream”;
and antonyms, as in “life and death.” Similarly, Abbi (1991) describes a family
of reduplication-like constructions as “compris[ing] lexical items of the same
semantic field ranging from near-synonymous to polar terms” (p. 25).22 To these
arrays of existing juxtaposition constructions we can add semantic inclusion
or classifier relationships, of the type often found in noun incorporation (see,
for example, Mithun 1984; Gerdts 1998; Mithun 2000). Turkish, for example,
has a number of noun-noun compounds in which the second member denotes
a class of which the first is a member (often a prototypical member), e.g. martı
kuş ‘seagull bird = seagull,’ serçe kuş ‘sparrow bird = sparrow’; çam ağacı
‘pine tree,’ meşe ağacı ‘oak tree’; papatya çiçeği ‘daisy flower = daisy,’ gül
çiçeği ‘rose flower = rose’; etc. (Orhan Orgun, p.c.).
MDT situates reduplication squarely within this family of construction types,
whose members all obtain parallel analyses within MDT. In prototypical redu-
plication, the daughters agree in every morpho-semantic feature. In synonym
constructions, the daughters may vary etymologically; in the case of near-
syonym constructions, they must share most features but may differ in small
ways. In the case of antonym constructions, the relationship between the two
62 Evidence for morphological doubling

daughters is still defined semantically, but in terms of disagreement in certain

features, rather than agreement.
(52) Reduplication all features agree
Synonym constructions all features agree except stratum/register/etc.
Near-synonym constructions most features agree
Members of same semantic set basic semantic category features agree
Semantic inclusion the features of one are a proper subset of
the features of the other
Antonyms the two are specified with opposite values
for certain features
The goal here is not to provide a comprehensive survey of all of the types
of semantic relationship that constructions can mandate, or tolerate, among
their daughters, but rather to show that pure reduplication is simply one point
along the cline of semantic similarity. This is true cross-linguistically; it is also
true within languages. Many languages have some subset of these construction
types. Vietnamese, for example, has not only the near-synonym constructions
seen above, but also antonym constructions. For case studies of languages with
constructions at many points along the cline in (52), we turn to Khmer and
Acehnese. Both illustrate the benefits of a uniform approach to these six con-
struction types.

2.3.3.1 Khmer
Khmer is rife with coordinate or symmetrical compounds, including what
Ourn and Haiman call “near-synonym” compounds like the following (Ourn &
Haiman 2000:485, 500–502) (see also Gorgoniev 1976:318). With respect to
semantics, Ourn and Haiman state that, “They seem to be simply semantically
redundant and rhythmically elaborated ways of saying what could be said by
a single word” (p. 491). Ourn and Haiman clearly characterize these synonym
and near-synonym constructions as reduplication.
(53) a. cah + tum ‘village elder’ 485
‘old’ ‘mature’
b chap + rɔhah ‘fast’ 485
‘quick’ ‘fast’
c. clooh + prɑkaek ‘quarrel’ 485
‘squabble’ ‘argue’
d. ʔaar + kɑmbaŋ ‘secret’ 500
‘secret’ ‘secret’
e. cbah + prakɑt ‘exact’ 500
‘exact’ ‘exact’
f. kee + mɔrdɔk ‘legacy’ 501
‘heritage’ ‘heritage’
 Synonym and antonym constructions 63

Ourn and Haiman write that “[s]o strong is this tendency to compound near-
synonyms that sometimes a verb will be conjoined with a doublet” (p. 485).
The example in (54a), below, is one in which the two nouns are semantically
identical but have different etymological origins, one coming from Sanskrit
and the other from Pali. The example in (54b) exhibits not etymological but
morphological divergence; the second daughter is a simple verb meaning ‘step,’
while the first is a morphological construction, termed by Ourn and Haiman a
“cognate accusative,” literally meaning ‘take a step.’

(54) a. peel + weeliə ‘time’ 485

‘time’ (Sanskrit) ‘time’ (Pali)
b. baoh cumhiən + chiən ‘take steps’ 485
‘take steps’ ‘step’

Khmer also has the next point on the scale, constructions in which both
daughters are what Ourn and Haiman call “typical or representative members
of a group,” as follows:

(55) a. klaa + damrəj ‘big game’ 485

‘tiger’ ‘elephant’
b. cruuk + kdaan ‘hooved animals’ 485
‘pig’ ‘deer’
c. phuum + phiək ‘region’ 485
‘village’ ‘plot of land’
d. tək + dəj ‘country’ 485
‘water’ ‘earth’

Ourn and Haiman note that some of the (near)-synonym compounds are
semantically opaque, meaning (somewhat) exocentric (pp. 485, 503).

2.3.3.2 Acehnese
Acehnese provides phonological evidence for the structural parallel claimed
in MDT to hold between reduplication and synonym/antonym constructions.
Examples of total reduplication, which confers emphasis, include the following,
from Durie 1985:

(56) tambô-tambô ‘drum-drum’ 39

ma-ma ‘mother-mother’ 39
tuleueng-tuleueng ‘bone-bone’ 40
jamee-jamee ‘guest-guest’ 40

The two parts are usually identical in form but may vary slightly: sometimes
the second copy differs in its vowel or final consonant, and sometimes unstressed
64 Evidence for morphological doubling

syllables can differ (ibid. p. 43). Reduplication can also be partial, targeting the
final syllable (ibid. p. 42):

(57) singöh-ngöh ‘sometime indefinite in the future’ cf. singöh ‘tomorrow’

bubê-bê ‘as big as’ cf. bubê ‘size (classifier)’

Like Sye, Acehnese permits different, suppletive allomorphs of the same mor-
pheme to be used in what appears to be the same construction (p. 43):

(58) irang-irôt ‘zig-zag’ cf. irang ‘skew,’ irôt ‘skew’

kreh = kroh ‘rustling dry sound’ cf. kreh ‘rustling dry sound,’ kroh
‘rustling dry sound’

In what Durie calls “juxtaposition of opposites,” “two words of contrasting

meaning can be combined to give a meaning that encompasses both” (ibid.
p. 44):

(59) tuha-muda ‘old and young’

bloe-publoe ‘buy and sell’
uroe-malam ‘day and night’
beungöh-seupôt ‘morning and evening’

Durie remarks that this construction is “phonologically identical with redu-

plication,” meaning that it has the same unique double stress configuration.
Thus with respect to stress, reduplication and juxtaposition of opposites form
a natural class. Normally, in Acehnese, word stress falls on the final syllable,
regardless of the morphological makeup of the word. (Durie 1985:9). However,
precisely when a word is reduplicated – or when it is a juxtaposition of opposites
construction – it has double word stress, with an equal degree of stress on each
of the two salient parts. The examples below illustrate double stresses in total
reduplication (60a), partial reduplication (b), and juxtaposition of opposites (c):

(60) ureueng'-ureueng' ‘people’ (lit. ‘person-person’) 40, 42

singöh'-ngöh' ‘sometime indefinite in the future’ 42
(lit. ‘tomorrow-tomorrow’)
lakoe'-binoe' ‘men and women’ 44

Clearly, one would want to say about Acehnese that there is a meta-
construction whose daughters are either synonyms or antonyms, and which
assigns stress to each daughter. Treating double stress as the result of base-
reduplicant identity would obscure the relationship between reduplication and
juxtaposition of opposites; the allomorphy cases show that reduplication is not
a matter of BR identity to start with.
 Comparison of MDT with OO correspondence 65

2.3.4 Wrapup
Clearly there is a cline of semantic similarity on which total identity is an
endpoint. As attested by the fact that descriptive grammarians standardly use
the term “reduplication” for the kinds of constructions we have been looking
at, reduplication includes cases in which the essential relation between the two
daughters is one of semantic identity, similarity, or opposition.

2.4 Comparison of MDT with OO correspondence

In light of the facts discussed above, no theory could maintain that reduplication
is a purely phonological copying process in which the reduplicant is derived
entirely from the phonological material of the base. Indeed, recognizing some
of the kinds of data we have talked about in this chapter, researchers working
in BRCT have responded with some interesting theoretical modifications. One
proposal, found in a series of illuminating studies of Bantu verb reduplication
by Downing (see especially Downing 1998a; 1999a; 2000a), invokes output–
output (OO) correspondence, i.e. analogy to other (related) words, to generate
morphological material in the reduplicant which is not present in the base. In a
form like Ndebele lima-limile, for example, although lima does not correspond
perfectly with the base, it does correspond perfectly with the independently
occurring stem lima.
Output–output correspondence was originally founded on the assumption
that correspondence can occur between independently occurring words in the
same language (Kenstowicz 1996; Benua 1997; Kenstowicz 1997). However,
as Hyman, Inkelas, and Sibanda (to appear) point out, there are numerous
reduplicants in Ndebele, like dleyi- or dlayi-, two possible reduplicants of dl-el-
a‘eat-for/at,’ that have no counterpart in the set of actual or even possible words
in Ndebele. If OO correspondence theory is relaxed, as Downing proposes, so as
to allow correspondence between morphological subconstituents of a word, then
the theories become quite close; the interest in comparing OO correspondence
to MDT essentially comes down to how the sets of potentially corresponding
items are defined in the former approach.
In an independent development of OO correspondence, Steriade (1997; 1999)
allows correspondence between any part of the reduplicant and any other mor-
pholexical consitutent with listed or predictable status, even if it has a different
meaning. Thus far, however, we have not seen evidence of the need for this
added power in analyzing reduplication. The two copies in reduplication can
differ phonologically and morphotactically, but they always have the same
meaning.
66 Evidence for morphological doubling

2.5 Conclusion
This chapter has assembled support for two of the essential claims of MDT,
repeated below:
(61) The Thesis of Morphological Targets: reduplication doubles
morphological constituents (affix, root, stem, word), not phonological
constituents (mora, syllable, foot)

The Thesis of Semantic Identity: the constituents that reduplication

doubles are identical semantically, but not necessarily in any other way

Although phonological identity often follows, indirectly, from semantic iden-

tity, in that the usual means of ensuring the latter is to use the same morphemes in
each copy, phonological identity is epiphenomenal, not an extrinsic or intrinsic
requirement. The epiphenomenality of phonological identity, and the propensity
of reduplicative constructions to introduce phonological differences between
the two copies, are the subjects of Chapters 3 and 4.
3 Morphologically conditioned
phonology in reduplication:
the daughters

Phonology is often morphologically conditioned. It can be sensitive to part of

speech, as in Acehnese, where /p/ dissimilates to /s/ before another labial in
verbs, but not in nouns (Durie 1985:32, 79–81, 95–96); in Japanese (McCawley
1968; Poser 1984) or Kaulong (Ross 2002), where accent assignment works
differently in noun and verb roots, or in any of innumerable other examples.
Phonology is also commonly conditioned by morphological construction type.
For example, some phonological effects occur only in compounding construc-
tions. Rendaku voicing in Japanese applies at the internal juncture of com-
pounds, but not at stem-affix junctures; moreover, it applies only within a
particular subtype of compounding, namely compounds with head-modifier
semantics and Yamato vocabulary (e.g. Itô & Mester 1986; 1995b; 1996a). Sim-
ilarly, Malayalam exhibits junctural gemination in one but not the other of its
two compounding constructions (Mohanan 1995). Other phonological effects
are specific to certain affixation contexts. For example, vowel-initial suffixes
in Turkish divide into two sets: those which trigger intervocalic velar deletion
(e.g. the possessives: bebek ‘baby,’ bebe-in ‘baby-2sgposs’), and those which
do not (e.g. the aorist: gerek ‘be necessary,’ gerek-ir ‘be necessary-aor’) (see,
for example, Inkelas & Orgun 1995). Many phonological effects are associated
with only a single affix in a language. For useful overviews of morphologically
conditioned phonology, see, among others, Ford and Singh 1983; Dressler 1985;
Spencer 1998; Booij 2000; Comrie 2000. It is very likely that many languages
meet Ezard’s (1997:35) characterization of the Austronesian language Tawala:
“phonological processes are mostly applied to a subset of the data; rarely does
a rule apply consistently,” i.e. across all morphological contexts.
It is important to understand reduplicative phonology within this general con-
text of morphologically conditioned phonology. Reduplicative constructions
sometimes exhibit phonological effects that are different from the phonological
effects found in some other constructions within the same language. This should

67
68 Reduplicative phonology: daughters

not be surprising, given that reduplication is a morphological construction and

that phonology is often morphologically sensitive. Nonetheless, reduplicative
phonology is often discussed as a distinct phenomenon, with theories proposed
for it that have no connection to nonreduplicative morphologically conditioned
phonology.
The empirical question confronting scholars of reduplication is whether redu-
plicative phonology is qualitatively different, in any systematic way, from the
phonology of nonreduplicative constructions, or whether all types of phono-
logical patterns that can occur in conjunction with reduplication constructions
can also occur with nonreduplicative constructions. The answer to this ques-
tion ought to determine whether reduplicative phonology can be handled within
the theory of morphologically conditioned phonology that is needed anyway,
or whether there must be a special theory just to handle reduplicative phonol-
ogy. With respect to this question, theories of reduplication separate into two
classes: Native Identity and Coerced Identity theories. The difference lies in
the extent to which special reduplicative phonology is tied to the dimension
of base-reduplicant identity that makes reduplication unique, or whether redu-
plicative phonology falls out from purely general principles of morphologically
conditioned phonology.
MDT is a Native Identity theory. It seeks to understand reduplicative phonol-
ogy in particular within the larger context of morphologically conditioned
phonology in general and does not posit any special phonological status for
reduplicative constructions. Some early phonological copying theories, like
Marantz’s (1982) Copy and Association theory or Steriade’s (1988) Full Copy
theory in which reduplicants start out as total copies of their bases and are then
modified by morphologically conditioned phonological rules, are also Native
Identity theories in that they take identity as the starting point but do nothing
special to maintain it. Coercive Identity theories, from Wilbur’s (1973) Identity
Principle to McCarthy and Prince’s (1995a) Base-Reduplicant Correspondence
Theory (BRCT), have taken the opposite tack, arguing that the phonological
patterns that occur in reduplication are intrinsically tied to an identity require-
ment holding between reduplicant and base. Coercive Identity theories predict
that at least some reduplicative phonology will be unparalleled outside of redu-
plication; Native Identity theories predict all reduplicative phonology to follow
from the same principles generating morphologically conditioned phonology
outside of reduplication.
Both this chapter and the next argue, on the basis of a broad survey of redu-
plicative phonology, for the Native Identity approach, pointing to the parallels
between the phonology of reduplicative and nonreduplicative constructions to
 Reduplicative phonology: daughters 69

support the view that reduplicative phonology is not qualitatively different from
nonreduplicative phonology and, specifically, that reduplicative phonology is
not identity-driven. Rather, true insight into reduplicative phonology comes
from looking at reduplication in its full morphological context.
These two chapters are organized around several predictions about reduplica-
tive phonology. All follow from the basic architecture of MDT, specifically from
the MDT claim that each node in a reduplication construction – first daughter,
second daughter, mother – is associated with a cophonology, or subgrammar.
This architecture is explicated in §3.1.
The first phonological prediction of MDT, sketched below, is that the phono-
logical effects that one finds within individual daughters in reduplication con-
structions will be the same kinds of effects one finds applying outside of
reduplication:
(1) Generalized Phonology Prediction:
The set of phonological effects found applying within reduplication is
equivalent to the set of morphologically conditioned phonological
effects found outside of reduplication. There is nothing unique about the
phonology of reduplication constructions.

Evidence for this prediction, a version of which was originally introduced by

Steriade (1988), is presented in §3.2.
A second prediction of MDT is that that the phonological effects applying in
the two daughters in a reduplication construction can be divergent.

(2) Independent Daughter Prediction:

The phonological effects associated with the two copies in reduplication
are independent.

Two different factors contribute to this prediction, which distinguishes MDT

sharply from phonological copying theories in which one copy is dependent
phonologically on the other. The first is that, as seen in Chapter 2, the daugh-
ters in reduplication have morphologically separate inputs which may differ
phonologically; the second is that the cophonologies of the daughters in redu-
plication are independent of each other and may also differ. In both respects
MDT differs profoundly from theories that privilege one copy (“Base”) and use
it to derive the other copy. The prediction of phonological independence of the
two daughters is explored in §3.3.
The third distinctive prediction of MDT is that reduplication constructions as
a whole, not just the daughters within them, may be associated with distinctive
phonology.
70 Reduplicative phonology: daughters

(3) The Mother Node Prediction:

Reduplication constructions may be associated, as a whole, with
morphologically conditioned phonological effects.

These special effects are, of course, subject to the Generalized Phonology

Prediction: there is nothing unique to reduplication about them. The effects of
this sort that we will see are also capable of being associated with nonreduplica-
tive morphological constructions. Mother node effects are largely discussed in
Chapter 4.
MDT is a layered theory, in that it assumes that the output of the phonol-
ogy applying to daughters is the input to the phonology applying to mother
nodes in morphological constructions. The particular layered structure pro-
posed for reduplication accounts for a number of effects which require extra
statements in theories not assuming layering. These are discussed primarily in
Chapters 4–6.

3.1 Cophonologies
MDT uses cophonologies, or phonological grammars associated with particular
morphological constructions, to model morphologically conditioned phonology
in general. Cophonologies within a language have been motivated indepen-
dently of morphologically conditioned phonology to handle variation (Anttila
1997; Itô & Mester 1997; Anttila & Cho 1998) and what Zuraw calls pat-
terned lexical exceptions (Zuraw 2000, also Itô & Mester 1995a; b). The use
of cophonologies for the phonological patterns associated with different mor-
phological constructions was pioneered in Orgun 1994; 1996; Inkelas, Orgun
and Zoll 1997 and has been developed in much subsequent work (Itô & Mester
1996b; Orgun 1997; Inkelas 1998; Inkelas & Orgun 1998; Orgun 1999; Yu
2000; 2003).
The cophonology approach is a natural evolution of ideas put forward in the
theory of Lexical Morphology and Phonology (LMP; see, for example Pesetsky
1979; Kiparsky 1982a; b; Kaisse & Shaw 1985; Mohanan 1986; Pulleyblank
1986; Inkelas 1990), which handles morphologically conditioned phonology by
sorting the morphological constructions of a language into sequentially ordered
groups, each associated with its own distinct phonological grammar. In LMP,
phonological rules themselves are, in principle, quite general; it is the grouping
together of rules and morphological constructions within morphological levels
that produces morphological conditioning of phonology. Much research has
shown LMP in its original form to be both too strong and too weak. It is too
 Cophonologies 71

strong in claiming that levels are strictly ordered relative to one another, a
point argued in, for example, Mohanan 1986; Hargus 1988; Inkelas and Orgun
1995; 1998. It is too weak in the sense that it omits from consideration many
phonological effects which are too morphologically specific to merit positing an
entire level, and thus requires building very specific morphological information
back into some phonological rules or constraints (Inkelas 1998). For example,
stress shift is associated with Level 1 morphology in English (e.g. Kiparsky
1982b), yet only a subset of Level 1 suffixes actually induce stress shift. Level 1
suffixes not inducing stress shift have to be associated with an extrametricality
rule (e.g. Hayes 1981) that is stipulated to apply to some but not all Level 1
suffixes. This very specific kind of stipulation falls outside the scope of what
level ordering can do.
Cophonology theory shares with LMP the ambition of making phonologi-
cal alternations themselves fully general. Every morphological construction –
each compounding, affixation, zero-derivation, truncation construction, and so
on – is associated with a cophonology, which may potentially differ from the
cophonologies of other constructions. In cophonology theory phonological rules
or constraints are never themselves indexed to particular morphological con-
texts; it is the entire bundle of rules or the specific ranking of constraints consti-
tuting the cophonology itself which is morphologically indexed. Cophonology
theory differs from LMP in abandoning the failed claim that morphological
conditioning boils down to dividing the lexicon into a few ordered components
each with fully general phonology.
Building on Anttila 1997 and suggestions in Inkelas and Orgun 1998 and
Stump 1998, we assume that the cophonologies of any given language are
constrained by what we have called in other work (Inkelas & Zoll 2003) the
“Master Ranking,” a partial ranking of constraints that every cophonology in
the language conforms to. The Master Ranking is in some ways similar to
proposals such the Stratum Domain or Strong Domain hypotheses of LMP,
which were attempts to constrain the degree to which phonological strata in
the same language could differ. These last hypotheses, however, presuppose
that strata are extrinsically ordered; the arguments against stratum ordering
are voluminous (see, for example, Mohanan 1986; Hargus 1988; Hualde 1988;
Inkelas & Orgun 1995; Inkelas 1998; Inkelas & Orgun 1998). Cophonologies
are intrinsically unordered. Their relative ordering follows from the intrinsic
or extrinsic ordering of the morphological constructions with which they are
associated.
Below are schematic representations of compounding and affixation, illus-
trating the association of cophonologies with morphological mother nodes. As
72 Reduplicative phonology: daughters

discussed in Chapter 1, the formalism adopted here is based on Orgun’s (1996)

Sign-Based Morphology, though the relevant generalizations are available to
other morphological frameworks as well:

(4) a. Compounding schema:

⎡ Syntax = g(Synx, Syny) ⎤

⎢Semantics = f(Semx, Semy )⎥ φi”
Cophonology = “φ
⎢ ⎥
⎣ Phonology = φi(Px, Py) ⎦

⎡ Syntax = Synx ⎤ ⎡ Syntax = Syny ⎤

⎢Semantics = Semx⎥ ⎢Semantics = Semy⎥
⎢ ⎥ ⎢ ⎥
⎣ Phonology = Px ⎦ X ⎣ Phonology = Py ⎦ Y

b. Affixation schema:

⎡ Syntax = g(Synx) ⎤
⎢ Semantics = f(Semx) ⎥ φj”
Cophonology = “φ
⎢ ⎥
⎣ Phonology = φj(Px, Py)⎦

⎡ Syntax = Synx ⎤
⎢Semantics = Semx⎥
⎢ ⎥ [Py]Y
⎣ Phonology = Px ⎦ X

What is important for purposes of this chapter is simply the fact that distinct
cophonologies can be associated with the mother node of each distinct mor-
phological construction. It is when cophonologies differ across morphologi-
cal constructions that we say that phonology is morphologically conditioned.
We illustrate the use of cophonologies to model morphologically conditioned
phonology with the example of Turkish velar deletion, below.
Stem-final velar deletion is a typical morphologically conditioned phonolog-
ical alternation. Some suffixes trigger it and others do not. (On velar deletion,
see, for example, Zimmer & Abbott 1978; Sezer 1981; Inkelas & Orgun 1995;
Inkelas 2000.)

(5) a. Dative suffix: triggers velar deletion

Nominative Dative Gloss

bebek bebe-e ‘baby’

inek ine-e ‘cow’
 Cophonologies 73

b. Aorist suffix: does not trigger velar deletion

Past Aorist Gloss

gerek-ti gerek-ir ‘be necessary’

bırak-tı bırak-ır ‘leave’

To describe this split among affixes it is necessary to posit at least two

cophonologies for Turkish, shown below:
(6) a. Cophonology ␾1 : *VKV » Max-C delete velar consonants (“K”)
intervocalically
b. Cophonology ␾2 : Max-C » *VKV preserve velars even intervocalically

Affixation constructions triggering velar deletion, like the dative, are associ-
ated with Cophonology ␾1 ; affixation constructions not triggering velar dele-
tion, like the aorist, are associated with Cophonology ␾2 .

(7)
Dative suffix (triggers velar deletion) Example:

⎡ Syntax = Dative noun ⎤ ⎡ Syntax = Dative noun ⎤

⎢ ⎥ ⎢ ⎥
⎢ Semantics = Semx ⎥ ⎢ Semantics = ‘baby’ ⎥
⎣ Phonology = φ1(Px, /-E/)⎦ ⎣ Phonology = φ1(Px, /-E/) = [bebee] ⎦

⎡ Syntax = N ⎤ ⎡ Syntax = Nx ⎤
⎢ ⎥ ⎢ ⎥
⎢Semantics = Semx⎥ [/E/]Y ⎢ Semantics = ‘baby’ ⎥ [/E/]Y
⎣ Phonology = Px ⎦ X ⎣ Phonology = /bebek/⎦ X

Aorist suffix (doesn’t trigger velar deletion) Example:

⎡Syntax = Present tense verb⎤ ⎡ Syntax = Present tense verb ⎤

⎢ ⎥ ⎢ ⎥
⎢ Semantics = Semx ⎥ ⎢ Semantics = ‘look’ ⎥
⎣ Phonology = φ2(Px, /-r/) ⎦ ⎣ Phonology = φ2(/bak/, /-r/) = [bakɯr] ⎦

⎡ Syntax = Nx ⎤ ⎡ Syntax = Nx ⎤
⎢ ⎥ ⎢ ⎥
⎢Semantics = Semx⎥ [/r/]Y ⎢Semantics = ‘look’⎥ [/r/]Y
⎣ Phonology = Px ⎦ X ⎣Phonology = /bak/⎦ X

As observed in Chapter 1, constructions can treat affixation as realizational,

as well, although we generally opt throughout this work to represent affixation
as item-based, for reasons of graphical clarity.
Cophonologies are a theoretically economical method of describing mor-
phologically conditioned phonology, in that cophonologies are needed
74 Reduplicative phonology: daughters

independently to describe language-internal variation and, if desired, patterned

exceptions or stratal effects in the lexicon. The use of cophonologies also per-
mits phonological rules or constraints to be morphologically purely general;
it is the distribution of cophonologies, and their differences from one another,
that captures morphological conditioning of phonological patterns.
In comparing cophonology theory to alternatives, it is important to recog-
nize not only its theoretical economy but also two important predictions that
cophonology theory makes regarding the behavior of morphologically condi-
tioned phonology. The first is scope. Each cophonology has scope over the
daughters of the morphological construction with which it is associated, but
not over any other material. The second prediction, which is related, is the
inside-out nature of the phonology–morphology interface. It follows intrinsi-
cally from associating cophonologies with nodes in morphological structure
that the output of applying phonology to the base of a morphological con-
struction – e.g. to the base of affixation – will serve as input to the phonology
applying to the derived construction – e.g. to the stem produced as a result
of affixation. The output of mother cophonologies can never serve as input to
daughter cophonologies; “outside-in” effects are not describable. The inside-
out character of phonology–morphology interaction has long been recognized;
theories without layered cophonologies have had to stipulate base priority to
generate the effects (e.g. Kenstowicz 1996; Benua 1997 inter alia).1

3.1.1 Cophonologies vs. indexed constraints

Many researchers have developed approaches to the phonology–morphology
interface which do not utilize cophonologies. It is important to assess
these approaches for their ability to capture generalizations about the scope
and inside-out nature of the interface. The most widespread alternative to
cophonologies in use today is constraint indexation in Optimality Theory, which
maintains one single constraint ranking for the language as a whole but splits
constraints into sets whose members are indexed to specific morphemes or mor-
phological constructions and are ranked independently (see, for example, Benua
1997; Alderete 1999; Itô & Mester 1999; Alderete 2001). Constraint indexation
is capable of emulating many of the effects of cophonologies, though it does
not intrinsically generate the scopal and inside-out predictions of cophonology
theory.
Cophonologies and individual constraint indexation work similarly in simple
cases, such as in the accentual differences between nonderived nouns and verbs
in Tokyo Japanese. The presence and location of accent in nonderived nouns (8a)
is unpredictable and must be lexically specified; for verb roots, by contrast, while
the presence or absence of lexical accent itself is unpredictable, the location
 Cophonologies 75

of accent (if any) is predictable: accent falls on the syllable containing the
penultimate mora of the verb stem (8b) (Poser 1984:51–52, citing McCawley
1968). The data below are taken from Smith 1997:7–8, citing Haraguchi 1977
and Poser 1984:
(8) a. Nouns b. Verbs

ı́noti ‘life’ kák-u ‘write’

kokóro ‘heart’ nayám-u ‘worry’
atamá ‘head’ kumadór-u ‘graduate’

The cophonological approach to this contrast would posit a verb stem

cophonology in which accent placement constraints (FixLoc) outrank accen-
tual Faithfulness (FaithLoc) (9a), and a noun stem cophonology in which
the ranking is the reverse (9b).2
(9) Noun stem cophonology: FaithLoc(Accent) » FixLoc(Accent)
Verb stem cophonology: FixLoc(Accent) » FaithLoc(Accent)

The indexed constraint account, developed by Smith 1997, is conceptually

similar; it splits accentual Faithfulness into two versions, with the noun-specific
version ranked above, and the other version ranked below, FixLoc:
(10) FaithLoc Noun (Accent) » FixLoc(Accent) » FaithLoc(Accent)

Any given pair of cophonologies can always be unpacked in this manner

into a single ranking with multiple, indexed versions of the constraints whose
ranking differs in the cophonologies. Thus if one attends to the phonology
of only a single morphological constituent at a time, the cophonological and
indexed constraint approaches will always be exactly equivalent.
It is in morphologically complex words, more than one of whose morpho-
logical subconstituents is associated with distinctive phonology, that the two
approaches diverge. The indexed constraint approach does not automatically
make either the scopal or the inside-out predictions of cophonology theory
(although, as we noted earlier, it can stipulate the effects; see, for example,
Benua 1997; Alderete 1999; 2001). The virtue of making these predictions in
reduplication is discussed below, and more extensively in Chapter 4, under the
rubric of “layering,” and it is for this reason that we adopt cophonology theory
in this work.

3.1.2 Cophonologies in reduplication

Three morphological constituents play a role in the MDT approach to redupli-
cation: the two daughters and the mother. The daughters are the stem-forming
constructions that independently generate the two semantically identical stems
76 Reduplicative phonology: daughters

(as discussed in Chapter 2); the mother is the reduplication construction itself,
which puts those two stems together and associates the result with a par-
ticular meaning. Each constituent in the construction is associated with a
cophonology:3

(11) Mother node:

Cophonology Z

Daughter #1: Daughter #2:

Cophonology X Cophonology Y
| |
/Input #1/ /Input #2/

The three cophonologies may potentially differ from one another. Any of them
may also differ from the cophonologies of other constructions in the language,
resulting in the appearance of reduplication-specific phonological effects.
Consider, by way of illustration, intensive reduplication in Sanskrit, as dis-
cussed in Steriade 1988:108 (forms are prefinal, lacking the application of vowel
coalescence rules):4

(12) kan-i-krand- ‘cry out’

tai-tvais- ‘stir’
dau-dyaut- ‘shine’
sa:-svap- ‘sleep’
ga:-grabh- ‘seize’
va:-vyadh- ‘pierce’

The two copies are differentiated in two phonological respects: the first copy is
truncated down to a single syllable, and its initial consonant cluster (if any) is
truncated to its least sonorant consonant (kr → k, sv → s, etc.).
In MDT these modifications are performed by the cophonology of the first
daughter (Cophonology X). In Cophonology X, markedness constraints (e.g.
*Complex, which bans complex onsets) rank higher, relative to faithfulness,
than they do in the cophonology of the second daughter (Cophonology Y) or
the mother (Cophonology Z). The contrast in rankings is shown below.

(13) Cophonology X: *Complex, Dep-V » Max-C

Cophonologies Y, Z: *Dep-V, Max-C » Complex

Truncation is also accomplished in Cophonology X; a constraint limiting the

output to a single heavy syllable ranks above Max-seg in Cophonology X, but
is low-ranked in Cophonologies Y and Z:
 Typical daughter modifications 77

(14) Cophonology X: Output = ␴ ␮␮ » Max-seg

Cophonology Y, Z: Max-seg » Output = ␴ ␮␮

A third aspect of Sanskrit intensive reduplication is the i vowel appearing

between the two copies of the verb root. Steriade characterizes this vowel as a
morphological requirement of the construction as a whole; its presence is not
required by general syllable structure requirements of Sanskrit. Clusters arising
across other kinds of morpheme boundaries are not, as a rule, broken up with
i (for example, Steriade (1988) cites the reduplicated perfect form va-vrk-tám,
based on the root ‘cut up’ where -tám is a suffix, on p. 126). In MDT the
i vowel is inserted by the mother node cophonology of the intensive redupli-
cation construction. Cophonology Z breaks up consonant clusters across its
internal boundary, as shown below. Contiguity ensures that i does not break up
clusters internal to either daughter:

(15) Cophonology Z: Contiguity » *CC » Dep

Cophonologies X, Y: Contiguity, Dep » *CC

It is important to note that nothing in this MDT account is reduplication-specific.

The generality of all of the phonological statements predicts that the effects
that happen, in a given construction in some language, to be reduplicant-
specific can occur in a nonreduplicative construction in that or some other
language.
We demonstrate in the remainder of this chapter and in Chapter 4 that the three
cophonologies in (11) successfully generate the attested range of reduplicative
phonological effects without the need for any phonological statements that are
specific to reduplication, and that each of the predictions of MDT, sketched in
the introduction, is correct.

3.2 Typical daughter modifications

Theories of Coerced Identity take the independence of base-reduplicant faith-
fulness (BR-Faith) and input-output faithfulness (IO-Faith) as the driv-
ing force behind reduplication-specific phonological effects (Wilbur 1973;
McCarthy & Prince 1995a; Struijke 2000a). If this thesis is correct, then it
should be the case that reduplicative phonology differs from the phonology of
other morphological constructions not involving reduplication, since identity
plays no role in the latter. By contrast, Native Identity theories, like MDT,
treat all morphologically conditioned phonology in the same way and predict
no difference between the kinds of effects found in reduplication and those
78 Reduplicative phonology: daughters

found elsewhere. The Generalized Phonology Prediction of MDT says that

reduplicative phonological effects are neither a subset nor a superset of the types
of morphologically conditioned phonological effects outside reduplication. In
this prediction, MDT shares with Steriade (1988) the view that the mechanisms
(truncation, or segmental changes) “that frequently accompany reduplication
are operations independent of and unrelated to the copying process central to
reduplication” (p. 75).
In MDT, any phonological effect distinguishing the output form of the base or
reduplicant from its input is accomplished either in the daughter cophonologies
or in the mother cophonology. We focus in this section on the effects found in
daughter cophonologies. These consist of assimilation, dissimilation, deletion,
insertion, truncation, augmentation, lenition, fortition, and neutralization – in
short, the set of phonological effects observed in languages generally. None
is limited to reduplication. For example, syllable structure simplification –
cluster simplification, length neutralization, coda deletion – is often cited as
a hallmark of reduplicant phonology; onset cluster simplification occurs, for
example, in reduplicants in Sanskrit intensive reduplication (e.g. kan-i-krand-
‘cry out’; Steriade 1988:108) and Tagalog ‘just finished’ construction (e.g.
ka-ta-trabaho ‘just finished working’; McCarthy & Prince 1999b:252, citing
Bowen 1969; see also, for example, Ramos & Caena 1990:38, Schachter &
Otanes 1972:362). But onset cluster simplification is of course not specific
to reduplicants; many languages allow onset clusters only in morphologically
or phonologically prominent positions. In Lango (Nilotic), for instance, onset
clusters (of which the second consonant is necessarily a glide) occur only root-
initially (Noonan 1992:7); Babine (Athapaskan) permits complex onsets only
word-initially (Hargus 1995). Burmese, discussed in Chapter 4, simplifies syl-
lable onsets in compounds, also a nonreduplicative environment. Similar points
can be made for rime simplification and the neutralization of segmental con-
trasts. Comparison of Alderete et al. 1999 and Steriade 1988, who survey the
kinds of neutralizations or simplifications that occur in reduplicants, and Barnes
2002, who surveys the kinds of neutralizations or simplifications that can be
positionally conditioned and are independent of reduplication, shows complete
parallelism. We have found nothing reduplication-specific in this respect about
the contents of Cophonologies X and Y.
This observation extends as well to the more arbitrary, morphophonological
effects that are sometimes observed in reduplication, particularly truncation and
input–output dissimilation. Neither is a natural phonological rule, setting them
apart from the ordinary kinds of reductions of markedness discussed above,
 Typical daughter modifications 79

but like the markedness-reducing effects, neither is specific to reduplication,

either.
Consider, for example, truncation, which in MDT is what results in partial
reduplication: the stem-forming construction producing the relevant daughter is
associated with a truncating cophonology. In the case of Diyari reduplication, for
example, the relevant cophonology truncates its input down to a foot (McCarthy
& Prince 1999b:263, citing Austin 1981 [see also Poser 1989]):
(16) a. wila-wila ‘woman’
b. tj ilpa-tj ilparku ‘bird sp.’

Truncation is handled in Optimality Theory by ranking the constraints that

induce foot truncation (Lex≈PWd and PWd≈Foot) above IO-Faith.

(17) [tj ilpa]

| ⇐ Cophonology: PWd≈Foot » IO-Faith
/tj ilparku/

Because there is nothing specific to reduplication in the contents of daughter

cophonologies, MDT makes the prediction that a comparable truncation process
might occur in a nonreduplicative context as well. This is of course correct; and
pointing it out was a major contribution of McCarthy and Prince 1986 (published
in edited form as McCarthy & Prince 1999b). As shown by McCarthy and
Prince, Weeda 1992, and others, truncation comparable to what is found in
reduplication can also serve as the sole morphological marker of constructions,
e.g. the Central Alaskan Yupik hypocoristics in (18), from Weeda 1992 (based
on work by Anthony Woodbury):5

(18) iluʁaq iluq ‘male cross cousin of male’ 163

qəuqi:n (HBC) qəuq (name) 163
cupə:aq cupə (name) 163
aŋivʁan aŋif (name) 163
kalixtuq kalik (name) 163
qətunʁaq (HBC) qətun ‘son’ 163

The Generalized Phonology Prediction holds true even of reduplicative

dissimilation, an effect which has often been discussed in reduplication-
specific terms (see, for example, Yip 1997; 1998; Kelepir 2000, among others).
Dissimilation is quite common in cases of Melodic Overwriting (discussed in
Chapter 2) constructions, as discussed, for example, in Emeneau 1955; Apte
1968; Abbi 1991, and others. In Abkhaz, for example, an echo-word construc-
tion meaning ‘X etc.’ doubles a word and replaces the onset of the second copy
80 Reduplicative phonology: daughters

with /m/ (a) (Bruening 1997, based on a manuscript by Bert Vaux). If the word
already begins with /m/, however, /č’/ is used instead (b):

(19) a. čə́-k’ čə́k’-mə́k’ ‘horse’

gaá-k’ gaák’-maák’ ‘fool’
pəstəh◦ -k’ pəstəh◦ k’- məstəh◦ k’ ‘fog, mist’
b. maát maát č’aát ‘money’
maš◦ ə́r-k’ maš◦ ə́rk’-č’aš◦ ə́rk’ ‘miracle’
maá-k’ maák’ č’aák’ ‘secret’

Similar effects are found in cognate constructions throughout Asia (see Abbi
1991 for an overview), as well as in the emphatic adjective partial reduplication
construction, discussed in Chapter 2, found throughout Turkic (e.g. Johanson
& Csato 1998) and in Armenian (Vaux 1998).
Yip (1997; 1998) characterizes such cases in reduplication-specific terms.
Using a BRCT-like constraint, Repeat, to drive reduplicative identity, she
invokes a higher-ranking constraint, *Repeat, to drive (minimal) dissimila-
tion between base and reduplicant.
Of course, however, dissimilation that takes place across base and reduplicant
can generally be described equally well as dissimilation that takes place between
input and output, for one of the daughters in reduplication. Instead of assuming
that the reduplicant-initial m-prefix in Abkhaz is dissimilating (to č’-) with
respect to the initial m- of the base, we could equally well assume that it is
dissimilating with respect to the initial input consonant of the reduplicant:

(20) Input–output dissimilation (MDT) Base-reduplicant dissimilation (BRCT)

⇐ BR
maák’

’-aák’ maák’

’-aák’

⇐ IO ⇐ IO ⇐ IO

/maák’/ /maák’/
/maák’/

Input–output dissimilation is observed in nonreduplicative contexts where

IO correspondence is the only descriptive possibility, as in the famous “toggle”
cases of voicing or vowel length in Nilotic languages; see Kurisu 2001 for
a recent overview of these and other cases. Input–output dissimilation can
be triggered in overtly marked morphological contexts as well.6 For exam-
ple, input-output quantity dissimilation is triggered in stems by the location-
forming suffix -an in Tagalog (Schachter & Otanes 1972:98), as illustrated
below:
 Typical daughter modifications 81

(21) a. if base has penultimate long vowel, length shifts to preceding open
syllable, else vanishes
gu:lay ‘vegetables’ gulay-an ‘vegetable garden’
ta:goʔ ‘hide’ taguʔ-an ‘hiding place’
hala:man ‘plant’ ha:laman-an ‘garden’
lanso:nes ‘sp. of fruit’ lansones-an ‘place for growing lansones’
b. if base has penultimate short vowel, it lengthens
hiram ‘borrow’ hi:ra:m-an ‘place for borrowing’
ʔaklat ‘book’ ʔakla:t-an ‘library’
kumpisal ‘confess’ kumpi(:)sa:l-an ‘confessional’
taraŋkah ‘lock’ ta:raŋka:h-an ‘gate’

Length dissimilation is only part of the story; when the stem-final vowel
lengthens, as in (21b), so does the stem-initial vowel, as well, optionally, as
other stem-medial vowels in open syllables. Schachter and Otanes report sim-
ilar behavior for several other suffixes. According to Schachter and Otanes
1972, all syllables with long vowels are stressed; it is thus conceivable that
the alternations above are ultimately stress-related, but the pattern is clearly
dissimilatory, regardless of the precise prosodic parameter along which the
dissimilation occurs.
Given that input–output dissimilation is necessary in order to describe
nonreduplicative phonological effects, Occam’s Razor suggests it is also the
right analysis of reduplicative dissimilation as well.
The conclusion reached here, namely that none of the phonological modifica-
tions typically found in reduplication is reduplication-specific, finds interesting
echoes not only in the Native Identity approach of Steriade (1988) but also
in the following statement of Alderete et al. (1999), who argue for a Coerced
Identity approach to reduplication:

(22) Reduplicant/Inventory Relation III (Alderete et al. 1999)

Any phonological restriction on the reduplicant of one language is a possible
restriction on the whole of another language.

By this Alderete et al. mean that the constraints driving reduplicant-specific

phonotactics in one language are the same constraints that, ranked differ-
ently, drive general, nonreduplicative phonotactics in some other language(s).
However, the data surveyed in this work suggest that Alderete et al. are
too narrow in singling out reduplication. The statement in (22) would be
more accurate if ‘the reduplicant’ were replaced by “a morphological con-
struction,” recalling the original Generalized Phonology Prediction of MDT,
reworded below to refer more explicitly to Optimality Theory and constraint
ranking:
82 Reduplicative phonology: daughters

(23) Generalized Phonology Principle (Optimality Theory version):

A constraint ranking which in one language may be indexed to a particular
set of morphological constructions may in some other language be indexed
to a different set of constructions, possibly larger, possibly smaller.

It has often been noted that opaque or unnatural phonological alternations

are more likely to be restricted to a small set of morphological constructions
than to be fully general. One would not want to claim that any phonological
alternation which is found in reduplication in one language has the potential of
being a fully general process in another; rather, the claim put forth there is that
the alternations associated cross-linguistically with reduplication are neither a
proper subset nor a superset of the alternations associated with morphological
constructions of any kind, across languages.

3.3 Divergent modification

Textbooks describe two prototypical cases of reduplication: perfect reduplica-
tion, in which both copies surface in the shape they assume as independent
words, and reduplication in which one of the copies is modified phonolog-
ically in some way. BRCT is designed to capture these two kinds of cases.
Either the reduplicant matches the base exactly, satisfying BR identity, or it
deviates in some way, either by truncating or by improving its phonological
well-formedness (IO-Faith » Markedness » BR-Faith).
MDT predicts a third possibility: both copies can be modified, in potentially
independent ways (the Independent Daughter Prediction). The next sections
illustrate this prediction with cases in which both copies undergo distinct mor-
phologically conditioned phonology. We term this situation “divergent modifi-
cation.” MDT is alone in predicting it to occur.

3.3.1 Hua
As seen in Chapter 2, Hua intensifying reduplication (Haiman 1980) involves
total reduplication in which each copy is subject to final vowel modification. In
the first copy, the final vowel is replaced with u; in the second copy, the final
vowel is replaced with e. The data are repeated in (24); recall that hu is a support
verb.

(24) kveku kveke hu ‘crumple’ kveki ‘crumple’

ebsgu ebsge hu ‘twist and turn’ ebsgi ‘twist’
ftgegu ftgege hu ‘all coiled up’ ftgegi ‘coil’
ha-varu ha-vare hu ‘grow up’ ha-vari ‘grow tall’
 Divergent modification 83

In Chapter 2 this case was analyzed as an instance of double Melodic Over-

writing. Reduplication calls for daughters which are semantically identical but
consist of different lexical types; one is a u-stem, meaning it must be formed by
an existing semantically vacuous construction which replaces the final vowel
with u, and the other is an e-stem, formed by a comparable construction which
effects e-replacement. Those who, like Alderete et al. (1999), draw a sharp
distinction between Melodic Overwriting and phonological modification in the
direction of increased phonological well-formedness might not consider Hua
to represent divergent modification. We are not committed to the view that
Melodic Overwriting and cophonological modification are always distinguish-
able; certainly a realizational view of morphology would not necessarily want
to distinguish the two. From a larger perspective, it does not matter what is
causing the differentiation between base and reduplicant. What matters is that
they are both different, not only from each other, but also from their inputs.
Nonetheless, we present next several cases which fall more clearly into the
category of divergent cophonological modification.

3.3.2 Hausa tonal modification

As described by Newman (2000:74ff.), numerous augmentative Hausa adjec-
tives of the form X-e Cè coexist with a synonymous reduplicated construction
in which the same base, X, appears in both daughters with a suffix -à . The first
copy has all High tone; the second, all Low:7

(25) gansame mè ∼ gansama -gànsàmà ‘tall and stout’ 76

shar̃ta e è ∼ shar̃ta a -shàr̃tà à ‘long and sharp’ 76
zungure rè ∼ zungura -zùngùrà ‘long, tall’ 76

The base of these constructions is in most cases bound, not appearing (even
with a vowel suffix) as a free word; however, Newman notes a few forms
like gabzà ‘heap up a lot of something,’ presumably historically the base of
gabje jè ‘bulky’ and gabza -gàbzà ‘huge (people)’ (p. 76).8
A similar construction is reported by Newman (1989b:251); Niepokuj
(1997:56) and Downing (2000a) offer discussion and additional data. Neither
copy is tonally faithful to the isolation form:

(26) càkw àli ‘slush’ cakw al-càkw àl ‘slushy’

jini ‘blood’ jina -jı̀nà ‘bloody’
r̃àbè niya ‘dangling’ r̃abe -r̃àbè ‘pendulous (breasts)’

For an overview of tonal replacement patterns of this kind, see Downing

2004. In the present MDT analysis, the formation of the reduplicated H-L
84 Reduplicative phonology: daughters

adjectives in (26) calls for daughters of different types, in that the first requires
a cophonology assigning High tone, while the second requires a cophonology
assigning Low tone. Both daughters impose -a stem-finally.
(27) Daughter #1 cophonology: H tone imposed (and -a: suffixed)
Daughter #2 cophonology: L tone imposed (and -a suffixed)

3.3.3 Tarok: divergent TETU

In Tarok (Benue-Congo; Nigeria), nominal reduplication expresses third singu-
lar possession (Robinson 1976 [R]; Sibomana 1980 [S]; 1981; Longtau 1993;
Niepokuj 1997 [N]). In reduplicated forms, as illustrated by the reduplicated
polysyllabic nouns in (28), the tone of the second copy is neutralized to Mid:
(28) a-fini a-fini-fini ‘his/her yarn’ N23
a-górò a-górò-goro ‘his/her cola-nut’ N23
ı̀-gı̀sàr ı̀-gı̀sàr-gisar ‘his/her broom’ N23
a-rı́jı̀yá a-rı́jı̀yá-rijiya (alt a-rı́jı̀yá-jiya) ‘his/her spring’ N23
a-dànkálı̀ a-dànkálı̀-dankali (alt a-dànkálı̀-kali) ‘his/her potato’ N23

The interest of the Tarok facts lies in the divergent modification exhibited
by the two copies of a monosyllabic base. Neither copy is faithful to the input,
and each shows neutralization along a different dimension. As shown in (29),
reduplication of monosyllables manifests tonal reduction in the second copy,
and truncation and vowel neutralization in the first copy. In the first copy, the
final consonant deletes and all vowels raise to [+high]. In the second copy, tone
(which in Tarok can be High, Low, or Mid) is neutralized to Mid.
(29) ı̀-jáŋ ı̀-jı́-ja:ŋ ‘his twin’ R206
a:-sáŋ a:-sə́-sa:ŋ ‘his rope’ R203
ı̀-tòk ı̀-tù-tok ‘his chair’ S203
ı̀-gy el ı̀-gi-gy el ‘his/her chin’ N115

The Morphological Doubling Theory reduplication construction for Tarok

monosyllables selects two semantically identical daughters, each subject to its
own (distinct) cophonology:9
(30) [tù-tok]
/
/

[tù] [tok]
X: Vowel raising and truncation ⇒ | | ⇐ Y: Reduction to Mid tone
/tòk/ /tòk/

Cophonology X effects truncation and vowel raising by ranking the following

constraints above IO-Faith: Output=CV, which mandates truncation to a
 Divergent modification 85

light syllable, and *V[-high], which neutralizes the height of the reduplicant
vowel to the unmarked value, [+high]. Both effects are observed in tù-tok (31).
Failure to truncate causes a fatal violation of the templatic constraint (31a, b),
and failure to raise the vowel crucially violates the vowel markedness constraint
(31c). The optimal outcome is a light syllable with a high vowel (31d).
(31)
Cophonology X /tòk/ Output=CV *V[-high] IO-Faith

a. tòk *! *

b. tùk *! *

c. tò *! *

) d. tù **

In the second daughter in the reduplication construction, the only alternation

is neutralization of tone to Mid. Cophonology Y thus ranks the tonal markedness
constraints above tonal faithfulness:
(32)
Cophonology Y /tòk/ *High, *Low IO-Faith *Mid

a. tòk *!

) b. tok * *

The problem for the Basic Model of BRCT presented by Tarok arises with
tonal neutralization in the second copy, which, because it is not truncated, would
be analyzed as the base. With respect to tone, the reduplicant is more faithful
to the input than the base is. Accounting for this while maintaining that IO-
Faith is responsible for base faithfulness results in a ranking paradox.10 We
know independently that IO-Faith must outrank the tonal markedness con-
straints (*High, *Low) since High, Low, and Mid all occur in nonreduplicated
forms:
(33)
tòk IO-Faith-Tone *High, *Low

) a. tòk *

b. tok *!

In standard BRCT (McCarthy & Prince 1995a), IO-Faith is the same

as IB-Faith. Herein lies the divergent modification ranking paradox. The
ranking required for nonreduplicated forms, namely IO-Faith-Tone » tonal
86 Reduplicative phonology: daughters

markedness constraints, incorrectly predicts preservation of the base tone in

reduplication. Consider the candidates in (34). BR-Faith-Tone is shown low-
ranked here, but this is immaterial; the fact is that no ranking of the constraints
in (34) will correctly identify (34a) as the winner. Candidate (34d) violates a
proper subset of the constraints violated by candidate (34a), so candidate (34a)
can never win.

(34)
Red-tòk IO-Faith-Tone *High, *Low BR-Faith-Tone

()) a. tù-tok *! * *

b. tu-tok *!

c. tù-tòk **!

0 d. tu-tòk * *

Because MDT generates reduplicant and base independently, it handles both

mappings solely in terms of IO-Faith and markedness constraints. The fact
that constraints can be ranked independently in the two cophonologies means
there is no ranking paradox.11

3.3.4 Parallel modification

If reduplicant and base can be associated with divergent cophonologies, neither
of which is general in the language, it should also be possible to find cases
in which both copies are associated with the same cophonology. We call this
phenomenon “parallel modification.” It is of interest because, although pre-
dicted to be possible by MDT, parallel imposition of many kinds of phono-
logical modifications typical of reduplication (truncation, contrast reduction
within the reduplicant) had previously been thought not to occur. The founda-
tional logic of BRCT rules out parallel contrast reduction effects (McCarthy &
Prince 1994a; 1995a); Generalized Template Theory (McCarthy & Prince
1994c; d; Urbanczyk 1996) was added to BRCT in order to rule out parallel
truncation.
As demonstrated in this work, however, parallel modification does occur,
supporting MDT in its prediction that the phenomenon is possible. Newman
(2000) describes a Hausa construction forming adjectives from common nouns
via “full reduplication of the underlying noun, preserving the original tone.
This is accompanied by shortening of the final vowel of both parts” (Newman
2000:27):
 Divergent modification 87

(35) gishiri gishiri-gishiri ‘salt/salty’

bùhu bùhu-bùhu ‘sack/sacklike’
gà ri gà ri-gà ri ‘flour/powdery’
mà kubà mà kubà-mà kubà ‘mahogany tree/dark brownish color’
This is parallel modification. In a derivational approach, parallel modifica-
tion would be handled by first making the modification in question, and then
reduplicating the result:

(37)
Step 1 truncation
mà kubà → mà kubà

Step 2 reduplication
mà kubà → mà kubà mà kubà

In MDT, parallel modification occurs as a result of the fact that in the redupli-
cation construction in question, both daughters impose the same cophonology.
Recall that the daughters in reduplication are themselves constructions that take
a morpheme or collection of morphemes as input and map them, via a specific
cophonology, to an output which serves as one of the two inputs to the redu-
plication construction. If we assume, as in Chapter 2, that every morphological
construction in the language is identified by some sort of index, which we will
call type (following the type hierarchy literature), then to model a reduplication
construction in which both daughters are the product of the same construction,
we specify that the daughters agree not only semantically but also in type.
Type features are nonsemantic diacritic features of the sort that are needed any-
way to represent things like conjugation or declension class, or nonsemantic
gender.
Each daughter of the Hausa construction illustrated in (35) is the product of
the same stem-forming construction, shown below, which modifies noun stems
by shortening the final vowel and produces stems of type #47:

(38) Stem-forming construction: Example:

⎡ Stem type: #47

Semantics: X ⎤ ⎡Semantics: ‘sack’ ⎤
Stem type: #47

⎢ Syntax: N ⎥ ⎢ Syntax: N ⎥
⎣Phonology: truncate last vowel of Y⎦ ⎣ Phonology: bùhu ⎦
| |
⎡ Semantics: X⎤ ⎡ Semantics: ‘sack’ ⎤
⎢ Syntax: N ⎥ ⎢ Syntax: N ⎥
⎢ ⎥ ⎢ ⎥
⎣Phonology: Y⎦ ⎣ Phonology: /bùhu / ⎦
88 Reduplicative phonology: daughters

The reduplication construction responsible for generating the forms above

simply calls for two type #47 stems, which it concatenates, identifying the
output as an adjective, as below:

(39) Denominal adjective reduplication (for data in (35))

⎡ Semantics: ‘having to do with X’ ⎤
⎢ Syntax: adj ⎥
⎢ ⎥
⎣Phonology: concatenate daughters⎦

⎡Stem type: #47⎤ ⎡Stem type: #47⎤

⎢ Semantics: X ⎥ ⎢ Semantics: X ⎥
⎢ ⎥ ⎢ ⎥
⎣ Syntax: N ⎦ ⎣ Syntax: N ⎦

The MDT account of Hausa denominal adjectival reduplication retains the

insight behind the derivational approach to these data sketched in (37). In the
MDT account, one construction, i.e. the stem-forming construction (#47) which
produces each of the daughters, truncates nouns, while another construction
calls for two truncated nouns and groups them into a constituent labelled “adjec-
tive.” Thus the output of truncation is the input to the cophonology of the mother
node of the reduplication construction, which concatenates the two copies. The
resulting schema, compiled into one representation, is shown below:

(40) ⎡ Semantics: ‘having to do with X’ ⎤

⎢ Syntax: adj ⎥
⎢ ⎥
⎣Phonology: concatenate daughters⎦

⎡Stem type: #47⎤ ⎡Stem type: #47⎤

⎢ Semantics: X ⎥ ⎢ Semantics: X ⎥
⎢ ⎥ ⎢ ⎥
⎣ Syntax: N ⎦ ⎣ Syntax: N ⎦
| |
⎡⎢Semantics: X⎤ ⎡⎢ Semantics: X⎤
⎣ Syntax: N ⎥⎦ ⎣ Syntax: N ⎥⎦

While nonderivational, this maps closely to the idea of “first truncate, then
copy.”
Parallel application of reduplication-specific phonological modification
presents a challenge for standard BRCT. To illustrate this prediction, we again
use Hausa adjective formation, which shortens final long vowels in both copies
in reduplication (mà kubà → mà kubà-mà kubà). In order for the final long
vowel to shorten in the base, it has to be the case that the ban on final long
 Divergent modification 89

vowels (*V ]) outranks IO-Faith, which preserves input vowel length. No

matter where BR-Faith ranks, it will ensure that the reduplicant also has a
final short vowel:

(41)
long-Red BR-Faith *V ] IO-Faith

a. long-short * *

b. long-long **

()) c. short-short *

d. short-long * * *

The problem is that Hausa does not shorten final long vowels generally (not
even in other reduplication patterns), and, therefore, the ranking *V ] » IO-
Faith cannot be correct. The only way for BRCT to describe the Hausa
pattern would be to adopt cophonologies, such that for adjective formation,
*V ] » IO-Faith, but for other constructions, IO-Faith » *V ]. This move would
bring BRCT closer to MDT but would vitiate the central tenet of BRCT, which
is that it is the relative ranking of BR-Faith that distinguishes reduplicative
phonology.12
Interestingly, BRCT is capable of generating a superficially similar paral-
lel modification effect. McCarthy and Prince (1999a) credit René Kager and
Philip Hamilton for independently observing what has come to be known as
the “Kager–Hamilton problem”: given base-reduplicant correspondence (as in
BRCT), it is logically possible for a truncated reduplicant to cause a base to
become truncated as well. This can happen in case BR-Faith is ranked so
high that a reduplicant-specific constraint on prosodic size (Red = X) is indi-
rectly imposed on the base as well. Thus, for example, the ranking Red =
␴ ␮ , BR-Faith » IO-Faith would produce a situation where bases and redu-
plicants are both truncated to light syllables. The belief that such phenomena
do not occur inspired the development of Generalized Template Theory (GTT;
McCarthy & Prince 1994c; Urbanczyk 1996), a serious modification to BRCT
which prohibits the use of direct prosodic constraints on reduplicant size. How-
ever, the existence of doubled truncated nicknames like JoJo or CoCo (from
Josephine or Collette) suggests, as observed by Downing (2000a), that in fact
there may be no empirical basis to the “Hamilton–Kager” problem; any time
truncation and reduplication are in a morphological feeding relationship, the
appearance of double truncation can be generated, with or without GTT, and
GTT is therefore unnecessary.13
90 Reduplicative phonology: daughters

Parallel affixal modification is fairly common in reduplication, as discussed

in Chapter 2. For example, Hausa forms frequentatives, which Newman (2000)
describes as “pseudoplural deverbal nouns” (p. 196), from verbs via modifica-
tion of stem tone, truncation and suffixation. As is general throughout Hausa
verbal morphology, the vocalic suffix, here -e, replaces the verb’s inherent grade
suffix (if any). As shown below, stem tone is also replaced with a L*H melody
in both copies:
(42) gya rà ‘repair’ gyà re-gyà re ‘corrections’ 196
tàmbayà ‘ask’ tàmbàye-tàmbàye ‘repeated questioning’ 196
sàssa à ‘carve’ sàssà e-sàssà e ‘repeated carving’ 196
sha ‘drink’ shà ye-shà ye ‘drinks, repeated 196
drinking’

Newman notes (p. 196) that the form each copy takes in frequentative redu-
plication is identical to the stative form of the same verb, e.g. sàssà e ‘carved.’
However, Newman argues that the frequentative and stative constructions are
not synchronically related, either functionally or semantically. What this shows
is that Hausa has a stem-forming construction producing L*H verbs ending
in -e:
(43) Construction: Example:

⎡ Stem type: #29

Semantics: X ⎤ ⎡ Stem type: #29
Semantics: ‘ask’ ⎤
⎢ Syntax: V ⎥ ⎢ Syntax: V ⎥
⎣Phonology: φ29(Y) ⎦ ⎣ Phonology: tàmbàye ⎦

⎡ Semantics: X ⎤ ⎡ Semantics: ‘ask’ ⎤

⎢ Syntax: V ⎥ Phonology: /e/ ⎢ Syntax: V ⎥ Phonology: /e/
⎢ ⎥ ⎢ ⎥
⎣ Phonology: Y⎦ ⎣ Phonology: tàmbayà ⎦

Stem #29 is a semantically neutral stem-forming construction (on stem types,

see Aronoff 1994; Blevins 2003). It produces stems for use in other construc-
tions. The frequentative construction calls for two semantically identical stems
of type #29. The stative calls for one.
Note that the stem-forming construction itself involves melodic overwriting
of two kinds: the replacement of the verb’s final vowel by -e, and the replace-
ment of the verb’s tones by Low. The choice to model these two replacements
differently, one as a semantically empty morph and the other in the cophonology
of the construction, is arbitrary; both modifications, or neither, could have been
attributed solely to the cophonology. This ambiguity of analysis derives from
 Divergent modification 91

the more general issue of whether morphology is item-based or realizational,

an issue on which we would like to remain as neutral as possible, and which
does not affect our central claim.

3.3.5 Double modification outside of reduplication

Support for the MDT approach which uses independent daughter cophonologies
to perform the modifications found in double daughter modification examples
is that exactly the same mechanism is needed outside of reduplication. Two
familiar examples are Japanese loanword clipping, which applies to each mem-
ber of borrowed compounds (44a), and two processes of nickname formation in
English, illustrated in (44b, c), which clip each of two juxtaposed words down
to initial syllable-sized units.14

(44)
Double truncation outside of reduplication
a. Japanese clipped compound loanwords (Itô 1990:220)
waado purosessaa → waa puro ‘word processor’
paasonaru koNpyuutaa → paso koN ‘personal computer’
paNtii sutokkiNgu → paN suto ‘panty stockings’
sukeeto boodo → suke boo ‘skate board’

b. Stanford University compound clipping slang

Corner Pocket in the Tresidder Union → Co Po
Florence Moore Hall → Flo Mo
Frozen Yogurt → Fro Yo
Memorial Auditorium → Mem Aud
Memorial Church → Mem Chu
Residential Education → Res Ed
Stanford University Employees → SU Emps
c. English shorthand slang
slo-mo ‘slow motion’
ficto-facto documentary based on very little fact
SoHo neighborhood in New York south of Houston St.
hi-fi ‘high fidelity’
sci-fi ‘science fiction’
Bo-Jo ‘Bob Jones university’
fin-syn from ‘financial syndication’ = proposed FCC policy to ease up
on control of the networks

Clearly, the ability of both members of a compound to be truncated shows

that double truncation itself is not a property of reduplication alone. The con-
struction posited for Hausa double reduplicative truncation, above, could easily
be adapted to account for compound truncation in Japanese or English. It would
92 Reduplicative phonology: daughters

be necessary to remove the semantic identity requirement from the daughters,

and, of course, to alter as needed the semantic and syntactic description of the
mother node.

3.4 Daughter independence vs. base-dependence

The evidence assembled thus far clearly shows that daughter cophonologies
are independent of one another and that base and reduplicant have morpho-
logically independent (though semantically equivalent) inputs. Insofar as MDT
guarantees this independence, it predicts the absence of the phenomenon termed
“base-dependence,” in which the output form of one copy depends on the out-
put form of the other. Such a situation cannot be directly modeled in MDT,
since in MDT the outputs of the two daughters in reduplication are not in a
correspondence relationship. The following sections survey the evidence for
base-dependence and find it wanting.

3.4.1 Reduplicant shape

The strongest arguments for base-dependence have been made on the basis
of reduplicant shape. The most commonly described scenario is that in which
prefixing reduplicants exhibit CV vs. VC allomorphy, such that the CV allo-
morph is used when the base is C-initial and the VC allomorph is used when the
base is V-initial. McCarthy and Prince (1999b) analyze several such cases in
terms of base-dependence, subsequently formalized in the Optimality Theory
frameworks (e.g. McCarthy & Prince 1993; 1995a). The idea is simple. In such
languages, the reduplicant is constrained to be a light syllable. CV reduplicants
meet that description, but VC reduplicants (prefixed to V-initial bases) do not,
since the final C of a VC reduplicant is necessarily syllabified as the onset of
the surface syllable containing it. The only reason, according to McCarthy and
Prince, that the C in question finds its way into the reduplicant at all is that
it can supply an onset to the following base. An example from McCarthy and
Prince’s (1986) analysis of Orokaiva (published in McCarthy & Prince 1999b),
is presented below:
(45) From McCarthy and Prince 1999b (example (16), p. 250)
σ σ σ σ

uhuke uhuke

If reduplicant shape were determined solely within Cophonology X, VC ∼ CV

allomorphy would be very difficult for MDT to handle.
 Daughter independence vs. base-dependence 93

However, our survey of the literature shows that the evidence for base-
dependent reduplicant shape is very slim, if not nonexistent. The prediction
of MDT that base-dependence should not occur appears to be correct.

3.4.1.1 Orokaiva, Roviana, Tawala

We begin our discussion with McCarthy and Prince’s (1986) original analysis
of Orokaiva (Binandere; Papua New Guinea), published (with emendation) in
McCarthy and Prince 1999b. McCarthy and Prince’s discussion of Orokaiva
is based on Healey, Isoroembo and Chittleborough 1969 (HIC), from which
McCarthy and Prince cite the following examples of pluractional reduplication
(p. 249; see HIC, p. 35):15
(46) Bare stem Pluractional Shape of reduplicant Gloss
a. waeke wa-wa-eke CV ‘shut’
hirike hi-hi-rike CV ‘open’
tiuke ti-ti-uke CV ‘cut’
b. uhuke uh-uhuke VC ‘blow’

In their 1986 manuscript, McCarthy and Prince propose a base-dependent

analysis in which the reduplicant is a syllable; it copies a final base consonant
just in order to provide a hiatus-breaking onset consonant for the following base.
If this were the correct analysis, it would be difficult to implement straightfor-
wardly in MDT, since the reduplicant cophonology (X) does not have access
to the base. In the published, revised version of the (1986) manuscript, how-
ever, McCarthy and Prince (1999a:251) reject the base-dependent analysis,
proposing instead that Orokaiva repetitive reduplication is infixing, and that the
reduplicant has a fixed CV shape, as shown below:
(47) Bare stem Pluractional Shape of reduplicant Gloss

a. waeke wa-wa-eke CV ‘shut’

hirike hi-hi-rike CV ‘open’
tiuke ti-ti-uke CV ‘cut’
b. uhuke u-hu-huke CV ‘blow’

Consideration of all of the data and discussion in Healey, Isoroembo, and

Chittleborough 1969, strongly suggests that McCarthy and Prince’s revised
analysis is the correct one. Healey, Isoroembo, and Chittleborough (1969)
cite thirteen reduplicated pluractional forms. Only two are vowel-initial: the
uhuhuke form cited by McCarthy and Prince (1986), and another form, indidike
‘eat’ (Healey, Isoroembo & Chittleborough 1969:36), which is correctly han-
dled by the infixation analysis, but not by the base-dependent prefixation
analysis:
94 Reduplicative phonology: daughters

(48) Bare stem indi

Prediction of base dependence analysis: VC reduplicant *in-indi-ke
Prediction of infixation to first CV: CV reduplicant in-di-dike

Since the infixation account works neatly for all the data, and since CV
infixing reduplication is common not only in the world’s languages in gen-
eral but in Oceanic languages in particular, we may safely conclude with
McCarthy and Prince (1999b) that Orokaiva does not present evidence for base
dependence.16
We have not found other robust examples of the kind of CV ∼ VC reduplica-
tive allomorphy, conditioned by whether the following base begins with a C or a
V, that would signal base dependence.17 There are certainly cases, like Roviana
(Corston-Oliver 2000), where phonological alternations applying at the mother
node level give rise to phonologically conditioned CV ∼ VC allomorphy. How-
ever, Roviana is not a case of base-dependence; all of the surface reduplicative
allomorphs can be derived from the same basic (C)VC reduplicant structure.
In Roviana, the general method of forming imperfectives is to reduplicate the
first two syllables of the verb root. As shown below, when the first two syllables
are both CV, the second V is often deleted from the reduplicant (49a, b). If
V-deletion creates a geminate, simplification occurs (b):

(49) Imperfective V-deletion,

Verb root reduplication degemination

peka ‘dance’ peka-peka pek-peka 470

raro ‘cook’ rara-raro rar-raro → ra-raro 470

In the case of a vowel-initial verb, the reduplicant would be VCV. The final
vowel of the reduplicant fuses with the initial vowel of the base, giving the
appearance of VC reduplication.

(50) Imperfective V-deletion,

Verb root reduplication degemination

ene ene-ene en-ene ‘walk’ 470

The result is superficial CV ∼ VC reduplicant allomorphy. However, because

the C-final reduplicative allomorphs can be derived from a V deletion rule
at the mother node, there is no need to attribute en-ene to base-dependent
reduplication.
A similar analysis is probably correct for another apparent case of base-
dependence in Tawala, an Austronesian language of Papua New Guinea (Milne
Bay Area; Ezard 1980; 1997). Progressive aspect is marked by verb root
 Daughter independence vs. base-dependence 95

reduplication. Roots that begin with a CVV sequence reduplicate as CV, as

shown in (51a) (Ezard 1980:147, Ezard 1997:43).18 Vowel-initial roots system-
atically reduplicate as VC (51b), as shown by these data from Ezard 1980:147:

(51) a. gae ge-gae ‘to go up’

tou tu-tou ‘to weep’
beiha bi-beiha ‘to search’
houni hu-houni ‘to put it’
b. apu ap-apu ‘to bake’
atuna at-atuna ‘to rain’
otowi ot-otowi ‘to make an appointment’

Before concluding that Tawala exhibits base-dependent CV ∼ VC reduplica-

tive allomorphy, however, it is important to take note of what happens with
CVCV-initial bases (see Ezard 1997:41–42); the data below are taken from
Ezard 1980:147:

(52) hopu hopu-hopu ‘to go down’

geleta gele-geleta ‘to arrive’
hune-ya hune-huneya ‘to praise (tr.)’
bahanae baha-bahanae ‘to speak (lit. talk-go)’
hanahaya hana-hanahaya ‘to bite’

The data in (52) suggest that the target output shape for the reduplicant is
not CV but CVCV, which affects the interpretation of the data in (51). The data
with apparent VC reduplicants may very well instead be instantiating VCV
reduplication and experiencing a V-V → V simplification like that of Roviana.
Indeed, there is independent evidence for vowel elision in Tawala. Ezard (1997)
describes a regular process by which the initial vowel of the derivational pre-
fix om- is elided following a vowel-final prefix, thus, for example, /hi-om-
hoe/ → himhoe ‘they left’ (p. 36). Ezard also describes a process by which
morphemes containing a lexical vowel sequence undergo the loss of the first
vowel when another morpheme follows in the same word or phrase: i-mae
‘3sg-stay = he stayed,’ vs. i-me duma ‘3sg-stay very = he stayed a long time’
(p. 37).
If we accept vowel elision in (52), then the data in (51b) and (52) fall together.
In each case the output of the reduplication is two moras. What requires expla-
nation is not the VC allomorph in (52) but the CV allomorph in (51a). Although
we cannot be sure, Ezard’s observation about VV reduction in non-final mor-
phemes may well be relevant here. If indeed VV → V simplification is taking
place at Cophonology Z (or even at the word level or phrasally), the output of
the reduplicant may in these cases be bimoraic (CVV) as well.
96 Reduplicative phonology: daughters

3.4.1.2 Other possible cases

There are more claimed cases of base-dependent reduplicative shape than we
can analyze here. Mokilese progressive verbs, for example, exhibit CVC redu-
plication for consonant-intial stems (e.g. pɔdok ‘plant’ → pɔd-pɔdok) and VCC
reduplication for V-initial stems (e.g. andip ‘spit’ → and-andip, alu ‘walk’ →
all-alu) (Harrison 1973; 1976, analyzed in Levin 1985a; Blevins 1996; see
also discussion in McCarthy & Prince 1999b, McCarthy & Prince 1995a:308).
Reduplicant-final gemination, as in all-alu, could easily be accomplished at
Cophonology Z and need not be stipulated as a property of the reduplicant.
Double consonant copying, as in and-andip, is more of a concern, since in
MDT the d would have to be preserved in Cophonology X. Without pretending
to resolve this issue here, we do note that all of the VCi Cj reduplication exam-
ples provided by Harrison involve homorganic consonant clusters, as the nd in
andip. It is possible that the homorganicity of these clusters inhibits simplifica-
tion in Cophonology X. Ulu Muar Malay, analyzed by Kroeger 1990 and Wee
1994, is a case in which the reduplicant (the first copy) consists of material from
the left and right edge of the input; whether the right edge material surfaces
depends on the nature of the initial consonant in the second copy. However, it
is possible to assume that Cophonology X always includes the final consonant
in its output, and that Cophonology Z simplifies impossible clusters, just as we
assume it simplifies vowel sequences in Tawala and Roviana.
McCarthy and Prince 1999b cite several examples of VC* reduplication as
candidates for a base-dependence analysis, e.g. Oykangand prefixing redupli-
cation (e.g. eder → ed-eder ‘rain’) and Mangarayi pluralizing infixing redu-
plication (e.g. gabuji → g-ab-abuji ‘old person’). Because the reduplicants in
these languages have a fixed shape, base dependence is not strictly necessary in
their analysis; the reduplicant can be derived in a context-free manner, as MDT
requires. McCarthy and Prince argue, however, that only a base-dependence
analysis can explain the VC* shape of the reduplicant: it ends in a consonant
because it always precedes a vowel with which that consonant can syllabify.
(McCarthy & Prince dispute Sommer’s [1981] claim that the canonical syl-
lable in Oykangand is VC*.) Their argument, subsequently formalized in the
Optimality Theory framework, is typological; they point to cross-linguistic gen-
eralizations about the shape and position of reduplicants, particularly infixed
ones. McCarthy and Prince (1999b) state, for example, that fixed VC pre-
fixing reduplication is unattested (except, apparently, in the case of strictly
V-initial languages like Oykangand; p. 251), and that no language will exhibit
CV reduplication which is infixed after the first consonant (the mirror image
 Conclusion 97

of Mangarayi) (McCarthy & Prince 1994b). However, more recent work by Yu

(2003) and McCarthy (2002a) has cast doubt on statements of this kind. Yu
(2003) cites two languages, for example, which infix a consonant after the first
consonant of the stem (the Austronesian language Leti, discussed in Blevins
1999, Yu 2003:89, and references cited therein; and the Pingding dialect of
Mandarin Chinese [Yu 2003:91 and references therein]). In the absence of com-
pelling individual cases requiring base-dependence, and in the absence of strong
typological universals connecting reduplicant shape to segmental context, we
conclude that MDT is correct in generating the daughters of reduplication inde-
pendently of one another.

3.5 Conclusion
In this chapter we have documented a number of cases in which both daughters
in reduplication are modified, and we have provided analyses making use of
the principle in MDT that requires only semantic identity between the two
daughters in reduplication and allows the daughters to diverge phonologically
and morphotactically.
By accounting for reduplicative phonology using the same method –
cophonologies – which we use to account for nonreduplicative phonology,
we correctly predict that all of the types of phonological effects that occur in
reduplication occur outside of reduplication as well. Morphological Doubling
Theory is economical, in not using any reduplication-specific phonological
rules or constraints; it also makes the right predictions about the phonology
of reduplication. Theories that focus solely on reduplication fail to capture the
fundamental parallels between reduplicative and nonreduplicative phonology.
4 Morphologically conditioned
phonology in reduplication:
the mother node

Chapter 3 demonstrated the need for independent daughter cophonologies in the

reduplication construction that is the centerpiece of Morphological Doubling
Theory (MDT). This chapter shows that, in addition, a cophonology associated
with the mother node, i.e. indexed to the construction as a whole, is required
in order to handle junctural alternations conditioned by the juxtaposition of the
two daughters in a reduplication construction. To illustrate briefly what we call
junctural effects, consider the case of vowel reduction/syncope in Lushootseed
(Hess 1977; Hess & Hilbert 1978; Bates 1986; Hess 1993; Bates, Hess &
Hilbert 1994; Hess 1995; Urbanczyk 1996). In the Lushootseed diminutive
reduplication construction, an unstressed /a/ either reduces to [ə] (1a) or, when
the resulting consonant cluster is legitimate, deletes altogether (1b) (Urbanczyk
1996). Since the diminutive prefix is usually stressed, in most cases the base
vowel reduces. However, when the base vowel is stressed, the vowel of the
reduplicant alternates. This is a classic junctural effect: the environment for the
alternation is provided by the concatenation of the two pieces of the construc-
tion, and the process is agnostic with respect to the affiliation of the unstressed
vowel it strikes.1
(1) a. s-túləkw s-tú-tələkw ‘river’/‘creek’ 167
b. wális wá-w’lis ‘type of frog’/‘Little Frog 167
(wife of p’ic’ikw )’
c. q’aXa-cut q’ə-ʔq’áXa-cut ‘uncover, bring to light’/‘show 169
self a little’
The application of reduction in Lushootseed is taken up in more detail in
§4.4 (see particularly §4.4.4).
Other languages exhibit junctural alternations that apply in some morpho-
logical constructions but fail to apply in reduplication. One such example
involves stress assignment in Indonesian (Cohn 1989). As shown in (2a), in
compounds, where there are potentially two primary stresses, the stress of the
first word is subordinated. The behavior of compounds contrasts with that of

98
 General approach to junctural phonology 99

reduplication, however, in which primary stress persists in both copies (2b).

The underapplication of junctural effects of this sort is taken up in §4.3. Data
are taken from Cohn 1989:185–88:
(2) a. tùkaŋ cát ‘artisan-print = printer’ *túkaŋ cát
b. búku búku ‘books’ *bùku búku
Our surveys have found that the range of junctural alternations occurring
in reduplication is as broad as the range of junctural alternations generally,
including epenthesis, lenition, metathesis, coda sonorization, assimilation, dis-
similation, and syncope. Carrying on with the argument developed in Chapter 3,
this chapter demonstrates that reduplication-specific phonology requires the
same approach as other morphologically specific phonology, and provides evi-
dence for the mother node cophonology provided in the MDT reduplication
construction. We first illustrate our general approach to junctural alternations,
arguing that the existence of reduplication-specific junctural effects that target
the base require any descriptively adequate theory of reduplication to include
layered phonologies, i.e. cophonologies or constraints indexed to constructions
as well as to terminals. Cases discussed include so-called underapplication;
we demonstrate that MDT, rather than a theory built around base-reduplicant
correspondence, such as BRCT (Base-Reduplicant Correspondence Theory;
McCarthy & Prince 1995a), is the right way to account for them. The final
piece of this argument considers, and then rejects, a potential alternative to lay-
ered phonologies, namely Existential Faithfulness (∃-Faith) (Struijke 1998;
2000a; b), on the grounds that it is both insufficient and unnecessary.

4.1 General approach to junctural phonology

To illustrate our general approach to junctural effects, consider the behavior
of stems affixed with dominant suffixes in Japanese (Poser 1984), recently
discussed from an Optimality Theory perspective by Alderete (1999; 2001).2
In Japanese, morphemes may be accented or unaccented. When a dominant
accented suffix attaches to a stem, it deletes the stem accent (if there is one),
and the word is realized with the suffixal accent (3a). The loss of stem accent
triggered by dominant affixes contrasts with the behavior of stems concatenated
with a recessive suffix (3b), in which case the stem accent, if present, prevails
(Alderete 2001:202):
(3) a. Accented stem + Dominant accented suffix: stem accent deletes
/adá + ppóDom + i/ ada-ppó-i ‘coquettish’
b. Accented stem + Recessive accented suffix: stem accent prevails
/yóm + táraRec / yón-dara ‘if he reads’
100 Reduplicative phonology: mothers

This case dramatically illustrates an important feature of construction-

specific junctural phonology: the alternation conditioned by the presence of the
affix (in this case accent deletion) is a property of the whole affixed stem, not just
of the triggering affix itself.3 Such effects can only be modeled by a theory that
allows constraints or cophonologies to be indexed to particular constructions,
rather than to hold only over terminals. Following the approach developed by
Inkelas (1998) for similar data, we propose that dominant and recessive affixes
are associated with different cophonologies. As shown schematically in (4), in
the cophonology associated with dominant affixes, *Accent outranks stem
faithfulness, which results in deletion of stem accent. In the cophonology asso-
ciated with recessive affixes, the ranking of these constraints is reversed, thus
preserving the lexical accent of the stem.

(4) Dominant suffix schema Recessive suffix schema

FAITH SUFFIX »*ACCENT » FAITH STEM FAITH STEM » *ACCENT » FAITH SUFFIX

stem suffix stem suffix

The need to reference constructions in stating subphonological patterns tran-

scends the extensive debate about the use of cophonologies vs. indexed con-
straints.4 Any adequate account of subphonological junctural patterns must
refer to constructions. Poser (1984), who indexes tone deletion to domi-
nant affixes, specifies that deletion has scope over the base of affixation;
Alderete (1999; 2001) indexes his accent deletion constraint not to the domi-
nant affix itself but to the affixed stem. As Inkelas (1998) argues, cophonolo-
gies do a better job of correctly predicting what all approaches recognize:
the scope of deletion is the affixed stem – the mother node, in constructional
terms.

4.2 Reduplication-specific alternations

We return to reduplication-specific junctural effects by examining the case of
Dakota coronal dissimilation. Dakota morphology uses reduplication to mark
the plurality of inanimate subjects as well as the iterative, distributive, and inten-
sive forms of verbs (Patterson 1990:90). Reduplication in Dakota is described
as a process that occurs to the right of the root, copying the final syllable (Shaw
1985; see also Boas & Deloria 1941 and, more recently, Steyaert 1976; Marantz
 Reduplication-specific alternations 101

1982; Sietsema 1988; Patterson 1990).5 The data in (5) illustrate the pattern for
vowel final roots; the reduplicant is boldfaced (Shaw 1985:184):

(5) ixá ixá-xa ‘to smile’ / ‘to grin’ (reduplicated)

skokpá o-skókpa-kpa ‘to be scooped out’ / ‘puddle’

Consonant-final roots always have a final epenthetic stem-forming /a/. In

reduplication, however, this vowel is not copied (Shaw 1985:184):6

(6) /ček/ čék-a ček-čék-a ‘stagger’

/čap/ čáp-a čap-čáp-a ‘trot’
/t’is/ t’iz-a t’is-t’iz-a ‘draw tight’
/khuš/ khúž-a khuš-khúž-a ‘lazy’

If the first coronal in a cluster of coronals is {t, č, n, d}, it becomes dorsal,
but this happens only if the cluster straddles the juncture between reduplicant
and base (Shaw 1985:184):

(7) /žat/ žag-žát-a ‘curved’

/theč/ thek-théč-a ‘be new’
/čheč/ čhek-čhéč-a ‘to look like’
/nin/ nig-nı́n-a ‘very’

Dissimilation does not occur word-internally or in lexical compounds (Shaw

1985:184):

(8) sdod + čhı́-ya ‘know.I + you-cause’ ‘I know you’

phed + nákpa-kpa ‘fire + crackle (reduplicated)’ ‘sparks’

The relevant constraints governing this alternation are shown in (9):

(9) IO-Ident(Place) penalizes a change of place features between input

and output
*Cor-Cor assesses a violation for a cluster of two coronals

In reduplication, *Cor-Cor must outrank IO-Ident in order to force

coronal dissimilation. Outside of reduplication, by contrast, the opposite rank-
ing must hold.7 With only a single hierarchy and one IO-Ident constraint, the
result is an apparent ranking paradox:

(10) a. Ranking for coronal dissimilation in reduplication:

*Cor-Cor » IO-Ident(Place)
b. Ranking for no coronal dissimilation outside reduplication:
IO-Ident(Place) » *Cor-Cor
102 Reduplicative phonology: mothers

MDT, which permits different cophonologies within a language, encounters

no paradox; it associates the different rankings in (10) with different construc-
tions (11). In each case, the relevant cophonology must be a property of the
construction as a whole, since the triggering environment for dissimilation is
junctural. In MDT terms, this translates into mother node cophonologies. The
necessity for such cophonologies provides strong support for the layered struc-
ture of the MDT model of reduplication.

(11) Reduplication Compounding

[Dissimilation cophonology] [Non-dissimilation cophonology]

[ ] [ ] [ ] [ ]

The standard BRCT indexing solution to potential ranking paradoxes that

arise in reduplication – faithfulness specific to the reduplicant – cannot work
here, because it is a consonant of the base rather than the reduplicant that dissim-
ilates. The problem is identical to the Divergent Modification Ranking Paradox
from Chapter 3. First, since coronal dissimilation does not occur in unredupli-
cated words, it must be the case that IO-Ident(Place) outranks *Cor-Cor
in Dakota. The tableau in (12) shows how this ranking blocks dissimilation
in unreduplicated forms. High-ranking IO-Ident(Place) penalizes dissimila-
tion of the coronal to velar (12b), rendering the more marked sequence of two
coronals intact in the optimal form (12a).

(12)
/phet – nákpakpa/ IO-Ident(Place) *Cor-Cor

a.) phed – nákpakpa *

b. pheg – nákpakpa *!

The problem Dakota raises, however, is that reduplication-specific coronal

dissimilation targets a consonant in the base. Because in BRCT, IO-Faithfulness
constraints are responsible for bases in and out of reduplication, the ranking
of IO-Ident(Place) over *Cor-Cor should prevent the alternation from
targeting any base, even in reduplication. This is shown in the following tableau:
IR-Ident(Place) is necessary to guarantee faithfulness of the reduplicant to the
input (13c), but it cannot subvert the protection of the base by the high-ranking
IO-Ident(Place).
 Reduplication-specific non-alternation 103

(13)
/sut-Red-a/ IO-Ident(Place) IR-Ident(Place) *Cor-Cor

0 a. sut-sút-a *

()) b. suk-sút-a *!

c. suk-súk-a *!

It is clear that the solution for Dakota lies in having a cophonology specific
to the reduplication construction, as we propose above.8

4.3 Reduplication-specific non-alternation

Further support for layered phonologies comes from junctural alternations that
fail to apply. Indonesian stress assignment is a familiar example of junctural
underapplication in reduplication. In the variety of Indonesian analyzed in Cohn
1989, words carry a single primary stress, which is usually penultimate. This
culminativity is enforced in compounds by the demotion of the main stress of
the first member of a compound to secondary stress (14) (Cohn 1989:188).

(14) tùkaŋ cát ‘artisan + print = printer’

anèka ráam ‘various + way = varied’
polùsi udára ‘pollution + air = air pollution’
ı̀kut sərtá ‘go along + part of = included’
lùar nəərı́ ‘outside + country = abroad’

Stress subordination fails to apply in unaffixed reduplication, however, where

two primary stresses persist (Cohn 1989:185):

(15) hák-hák ‘rights’

búku-búku ‘books’
kərá-kərá ‘monkeys’
kəcı́l-kəcı́l ‘small-dist’
wanı́ta-wanı́ta ‘women’
minúman-minúman ‘drinks’ (n.)
màšarákat-màšarákat ‘societies’
kəkuráŋan-kəkuráŋan ‘lacks’ (n.)
kəmàšarakátan-kəmàšarakátan ‘societies (abstract)’

The MDT analysis of Indonesian associates different cophonologies with

the compounding and reduplication constructions. They differ crucially in the
ranking of Culminativity, which permits only one primary stress in the
output, and Ident io -stress, which preserves primary stress:
104 Reduplicative phonology: mothers

(16) Compounding cophonology: Culminativity » Ident io -stress

Reduplication cophonology: Ident io -Stress » Culminativity

As this demonstration illustrates, MDT treats underapplication as nonap-

plication; it is analyzed with garden-variety input–output faithfulness con-
straints. In BRCT, underapplication is analyzed as purposeful, blocking an
alternation that would, if it occurred, destroy the identity of reduplicant and
base. Consider the sketch of Cohn and McCarthy’s (1998) BRCT analysis
of Indonesian in (17), which appears as a small part of their comprehen-
sive account of the complex stress patterns in the language and parallels the
independently arrived at Coerced Identity account proposed in Kenstowicz
1995. Cohn and McCarthy argue that Culminativity 9 is outranked by con-
straints enforcing metrical identity between the base and the reduplicant, which
we abbreviate here as BR-MetricalFaith.10 Crucially, the constraint is
violated when corresponding vowels disagree in degree of stress. Therefore,
the reduplicated form with two primary stresses (17a) is optimal, since the
stress subordination in (17b) introduces a fatal difference between the two
copies:11

(17)
[RED-buku] BR-MetricalFaith Culminativity

) a. búku-búku *

b. bùku-búku *!

4.3.1 BR-Faith is insufficient

A Native Identity theory such as MDT, where identity in reduplication is a by-
product of MS feature duplication but is not actively maintained by the phonol-
ogy, does not attribute underapplication to constraints on base-reduplicant iden-
tity. With regard to underapplication, the MDT approach makes at least two
predictions that distinguish it from BRCT. First, MDT predicts that underap-
plication is possible under any kind of morphological conditioning, regardless
of identity. Second, MDT predicts that there will be underapplication effects
that occur in reduplication that are not identity enhancing. Both of these predic-
tions are correct, with the result that the Coerced Identity approach to underap-
plication in BRCT is insufficient as it stands. To make BRCT descriptively
adequate requires including the kind of indexed cophonologies/constraints
that are at the heart of MDT. In doing so, BR-Faith constraints become
redundant.
 Reduplication-specific non-alternation 105

4.3.2 Underapplication all over

Recall the Generalized Phonology Prediction from Chapter 3:

(18) Generalized Phonology Prediction:

The set of phonological effects found applying within reduplication is
equivalent to the set of morphologically conditioned phonological effects
found outside of reduplication. There is nothing unique about the phonology
of reduplication constructions.

With respect to underapplication, the Generalized Phonology Prediction

leads us to expect that underapplication is a potential property of all mor-
phologically conditioned phonology, not just of reduplication. This prediction
is certainly true. For example, alternations that occur in derived environments
often underapply in nonderived environments (see, for example, Kiparsky 1993,
as well as Inkelas 2000; Anttila in press for recent overviews). Moreover, in
many languages the phonological alternations of one stratum fail to apply in a
different stratum, as has been well documented, for example, in English, Malay-
alam, Kashaya, Sekani, and Turkish among others (see, for example, Orgun
1996; Inkelas 1998; Anttila 2002 for recent overviews). It is also common to
find underapplication in words of a particular morphosyntactic category. Smith
(1997; 1998) documents cases in which alternations applying in verbs in some
languages underapply in nouns. In other cases an apparently arbitrary class of
morphemes fails to trigger an alternation found in other constructions. Consider,
for example, a morphologically conditioned syncope process in Hopi, described
by Jeanne (1982). For roots subject to syncope, their final vowel deletes when
a suffix is added (19a). In all other roots, syncope underapplies (19b).

(19) Non-future Future

a. Syncope applies: soma som-ni ‘tie’ 246

yŋ y a yŋ-ni ‘enter (pl)’ 246
mooki mok-ni ‘die’ 246
hiiko hikw -ni ‘drink’ 246
b. Syncope underapplies: maqa maqa-ni ‘give’ 248
pana pana-ni ‘put in’ 248
qat qat-ni ‘sit’ 248
ʔöki ʔöki-ni ‘arrive (pl)’ 248

Along the same lines, we find a morphologically conditioned syncope alter-

nation in Piro (Arawakan) (Matteson 1965; Kenstowicz & Kisseberth 1979).
According to Matteson (1965:36), in Piro, “Arbitrary vowel loss occurs preced-
ing certain suffixes and incorporated postpositives.” The data in (20), based on
106 Reduplicative phonology: mothers

the stems neta ‘I see’ and yopnuha ‘he wards off there,’ serve as illustration.
For any given stem, some suffixes arbitrarily trigger deletion of the final vowel
(20a), while others do not (20b):

(20) a. Syncope applies:

locative net-ya yopnuh-ha
3sg.obj net-lu yopnuh-lu
b. Syncope does not apply:
anticipatory neta-nu yopnuha-nu
yet neta-wa-lu yopnuha-wa-lu

The cophonologies at the heart of MDT are general enough to encom-

pass the full range of underapplication effects found cross-linguistically.
These cases clearly require indexation of cophonologies or constraints to
syntactic categories (noun/verb), arbitrary affixation constructions (e.g. in
Piro), and so on. BRCT’s identity-based account of underapplication, on the
other hand, is insufficient. With the recognition that underapplication (non-
application) is not a special feature of reduplicative phonology, and that indexed
constraints/cophonologies are necessary, BR-Faith constraints are rendered
superfluous in the analysis of underapplication.

4.3.3 Non-identity-enhancing underapplication in reduplication

Because the MDT account of underapplication does not rely on phonological
identity, it predicts that there will be cases where underapplication occurs in
reduplication even when it is not identity enhancing. This is in contrast with
the expectation in BRCT, where base-reduplicant identity is at the heart of
reduplication-specific effect. Again, the prediction of MDT is the correct one.
Chamorro (Topping 1973), for example, exhibits underapplication of a stress
shift, independent of any plausible identity considerations. Consider the data
in (21), taken from Topping 1973. Default stress in Chamorro is penultimate.
When a suffix attaches to a root, it shifts the stress over to the right to maintain
penultimate stress:

(21) a. karéta ‘car’ 108

karetá-hu ‘my car’ 108
karetá-ta ‘our (incl.) car’ 108
kareta-n-mámi ‘our (excl.) car’ 108
kareta-n-ñı́ha ‘their car’ 108
b. hásso ‘think’ 42
h-in-ásso ‘thought’ 42
h-in-assó-mu ‘your (sg.) thought’ 42
h-in-asso-n-mámi ‘our (excl.) thought’ 42
 Reduplication-specific non-alternation 107

While the penultimate stress pattern is preserved under suffixation (21a)

and infixation (21b), it is disrupted by both of the CV reduplication patterns
in Chamorro. Final CV reduplication, which serves to intensify adjectives,
negatives, and directional adverbs (Topping 1973:183), repeats the final CV of
the stem. As shown in (22), this reduplication pattern results in antepenultimate
stress.

(22) ñálang ‘hungry’ ñálalang ‘very hungry’

métgot ‘strong’ métgogot ‘very strong’
bunı́ta ‘pretty’ bunı́tata ‘very pretty’
guátu ‘there, in that direction’ guátutu ‘furthest’
mági ‘here, in this direction’ mágigi ‘nearest’
ni háyiyi ‘no one else’
ni guáhuhu ‘not even me’
hi taimánunu ‘no matter how’

Klein (1997) presents a Coerced Identity account of the underapplication of

default stress in the intensive reduplication construction. He argues that stress
shift is inhibited in reduplication because it would ruin the otherwise perfect
identity between the reduplicant and the base:

(23)
/Red-bunı́ta/ BR-Ident(Stress) Penult

a. normal application bunitá-ta *!

) b. underapplication bunı́ta-ta *

A more comprehensive look at the language reveals, however, that Coerced

Identity cannot be at the root of underapplication in Chamorro. A second redu-
plicative construction, which copies the first CV of the stressed syllable, also
fails to shift stress, despite the fact that there would be no consequences for
identity. Stressed CV reduplication serves a variety of functions, including nom-
inalization (24a) and the marking of continuative aspect (24b). In the result-
ing forms stress surfaces on the antepenultimate, rather than the penultimate,
syllable. Data are taken from Topping 1973:

(24) a. hátsa ‘lift’ háhatsa ‘one that was lifting’ 102

b. saga ‘stay’ sásaga ‘staying’ 192, 259
s-um-ága ‘lived’ s-um-ásaga ‘living’ 191–192
c. átan ‘look at’ á’atan ‘looking at’ 171
d. táitai ‘read’ tátaitai ‘reading’ 259
e. hugándo ‘play’ hugágando ‘playing’ 259
108 Reduplicative phonology: mothers

In this case, assuming a high-ranked Non-Finality constraint, either the

reduplicant or the base must carry the stress. Klein’s account incorrectly pre-
dicts that for this construction, stress shift should apply normally, since nor-
mal application of stress shift (25a) does not increase the severity of the
BR-Ident(Stress) violation:

(25)

/Red- sága / BR-Ident(Stress) Penult

0 a. normal application sa-sá-ga *

b. underapplication sá-sa-ga * *!

Coerced Identity cannot be the right analysis of underapplication in

Chamorro. The pattern bears out the prediction of MDT that underapplica-
tion in reduplication is not tied to identity enhancement. In this case, as we
will also argue below for Indonesian and Klamath, apparent underapplication
is simply non-application: the cophonology in question does not enforce the
pattern in question.12

4.3.4 Layering and underapplication

Klein’s analysis of Chamorro highlights the danger of theorizing on the basis of
a reduplicative construction without taking into account the broader picture of
the morphology of a language. In that case, while one reduplication construction
appears to be consistent with the Coerced Identity view of underapplication, it
is clear that Coerced Identity cannot motivate underapplication in the whole set
of constructions (including other reduplication constructions) taken together.
To make the point that underapplication is not in general motivated by identity
requirements, we return to Indonesian reduplicative underapplication, another
case which has been analyzed from the Coerced Identity perspective, as dis-
cussed above. We show, by looking at reduplication in the context of affixation
generally, that the apparent identity affect is ephemeral; instead, Indonesian
shows a cophonological difference between application and non-application of
stress reduction which happens to give the appearance of identity preservation
in some cases.
Indonesian stress forms the basis of an intriguing argument for the Coerced
Identity account of underapplication, presented by Kenstowicz (1995) and Cohn
and McCarthy (1998). The argument hinges on the stress patterning of redupli-
cated forms.
 Reduplication-specific non-alternation 109

The default location of main stress in Indonesian is penultimate. Penulti-

mate main stress is reassigned under suffixation; stresses further to the left are
subordinated (Cohn 1989:176):
(26) a. bicára ‘speak’
məm-bicará-kan ‘speak about (something)’
məm-bicàra-kán-ña ‘speak about it’
b. acára ‘plan’
məŋ-acará-kan ‘plan (something)’
məŋ-acàra-kán-ña ‘plan it’
Reduplication does not conform perfectly to the generalization that the right-
most stress is primary and others are secondary. A focus of Cohn (1989) is the
fact that the copies in reduplication bear equal stress, as shown on the left in
(27). Intriguingly, however, stress subordination does apply when a reduplicated
stem is suffixed; as shown on the right in (27), such forms pattern exactly like
the suffixed stems in (26) with regard to stress shift and subordination (Cohn
1989:185):
(27) búku-búku ‘books’ [bùku-bukú]ña ‘the books’
wanı́ta-wanı́ta ‘women’ kə[wanı̀ta-wanitá]an ‘womanly’
màšarákat-màšarákat ‘societies’ [màšaràkat-màšarakát]ña ‘the societies’
minúman-minúman ‘drinks’ n. [minùman-minumán]ña ‘the drinks’

Stress shift and subordination under affixation decrease identity between the
two copies. Kenstowicz (1995) and Cohn and McCarthy (1998) independently
argue that this is a kind of TETU effect; once identity is disrupted for other rea-
sons, unmarkedness emerges.13 As modeled in (28), the unmarkedness effect
in this case is stress reduction; the disruption in identity is caused by stress
shift upon suffixation. As seen, the addition of a suffix to a reduplicated form
forces the main stress on the second copy to shift rightward (by high-ranking
Penult). This shift induces non-identity between the two copies, since it
results in differently stressed syllables in each. Given that perfect identity is
impossible, the effects of Culminativity emerge, triggering stress subor-
dination in the first copy in typical TETU fashion:14
(28)

[RED-buku-ña] “Penult” BR-MetricalFaith Culminativity

a. búku-búku-ña *!
b. búku-bukú-ña * *!
) c. bùku-bukú-ña *
110 Reduplicative phonology: mothers

Note that in Cohn and McCarthy’s analysis it is crucial that BR-Metrical

Faith is assessed categorically. Although the optimal candidate, (28c)
obscures metrical faithfulness in two ways (location and degree of stress), it can
win only if this leads to a single violation of the BR-Faith constraint. Other-
wise, it would be bested by the candidate in (28b), whose base and reduplicant
are more faithful to each other but which violates Culminativity.
The characterization of normal application of stress subordination as a TETU
effect contrasts with the treatment of this phenomenon in a Native Identity the-
ory such as MDT, where BR-Faith constraints play no role. In the Native
Identity view, it is an accidental (although functionally understandable) fact
about the reduplication construction that its cophonology happens not to trig-
ger stress subordination. Adding a suffix to a reduplicated form introduces a
new layer of structure and a new cophonology, one that includes regular stress
subordination:

(29) Native Identity view:

[bùku bukú-ña] ⇐ Stress subordination cophonology

No subordination cophonology ⇐ [búku búku]

[búku] [búku] /-ña/

The BRCT view is appealing in its apparent explanation of a pattern that in

MDT is essentially arbitrary. However, further facts show that stress subordina-
tion in affixed reduplicated words is in fact a consequence not of TETU, but of
cophonological layering, as predicted by MDT. Consider first the data in (30),
drawn from Cohn (1989:185). Bases with only a single stressable vowel, which
include both monosyllabic bases and disyllabic bases where one of the vowels is
a schwa, exhibit the same pattern of stress subordination as do the polysyllabic
forms seen in (27). The downgrading of primary to secondary stress does not
apply in reduplicated forms – except where the reduplicated word is followed
by a suffix. What distinguishes the data in (30) from those in (27) is that stress
never shifts. Because the words have only one stressable vowel, stress is final;
under suffixation, it stays put, due to the penultimate stress generalization. No
stress shift occurs – but stress subordination takes place regardless:

(30)
pás ‘try’ di [pàs-pás]kan ‘tried on, repeatedly’
kərá ‘monkey’ kərá-kərá ‘monkeys’ [kərà-kərá]an ‘toy monkey’
kəcı́l ‘small’ kəcı́l-kəcı́l ‘small (dist.)’ məŋ[əcı̀l-ŋəcı́l]kan ‘to belittle s.t.’
 Reduplication-specific non-alternation 111

Cohn and McCarthy (1998) recognize stress subordination in these forms

as a potential problem for their analysis of underapplication. In the Coerced
Identity scenario, stress subordination should occur in reduplication only as
a TETU effect when other factors make perfect identity impossible. How-
ever, there is no obvious cancellation of metrical identity in these short words:
stress does not shift. The prediction would thus appear to be that stress sub-
ordination should not occur either; BR-identity in stress degree should be
preserved:

(31)
[RED-pas-kan] BR-MetricalFaith culminativity

a. pàs-pás-kan *!

0 b. pás-pás-kan *

Cohn and McCarthy solve this problem by proposing that although stress does
not shift, metrical structure changes under suffixation by virtue of enlarging the
second metrical foot to two syllables. Footing differences are argued to trigger
likewise a violation of BR-MetricalFaith; the decision among candidates
is thus again passed down to culminativity, with the result that only the
second stress retains its primary status:

(32)
[RED-pas-kan] BR-MetricalFaith culminativity

) a. (pàs)-(pás-kan) *

b. (pás)-(pás-kan) * *!

As noted earlier, this analysis relies heavily on the unorthodox assumption

that BR-MetricalFaith is evaluated categorically. This mode of violation
constitutes a significant departure from standard assessment of faithfulness con-
straints in Optimality Theory (Prince & Smolensky 1993) and Correspondence
Theory (e.g. McCarthy & Prince 1995c; d). In general, it is crucial for a faith-
fulness constraint to be violated for every infraction. Consider, for example, a
language where high-ranking *Complex penalizes complex codas and sim-
plifies them by deleting all but the postvocalic consonant (e.g. hand → han).
As shown by the tableau in (33), the pattern would be impossible to generate
if faithfulness is evaluated categorically, since IO-Max-C could not serve to
minimize deletion. A NoCoda constraint, ranked anywhere, would always
favor the candidate with maximal deletion, i.e. (c).
112 Reduplicative phonology: mothers

(33)
/hand/ *complex IO-Max-C NoCoda

a. hand *!

b. han * *!

0 c. ha *

In the Indonesian stress case, this problem is exacerbated further by the fact
that, in a fully developed analysis, the single metrical BR-MetricalFaith
constraint would have to be broken down into a family of metrical faithfulness
constraints, as is standard in any analysis of stress (see, for example, Kager
1999 for an overview). In that case, we would not expect to see the TETU
effect at all. Since suffixed output candidates violate foot identity regardless of
whether or not stress subordination occurs, the candidate that at least preserves
the same degree of stress would always be optimal:

(34)
[RED-pas-kan] Br-Ident BR-Ident
(stress degree) (foot membership)

a. (pàs)-(pás-kan) *! *

) b. (pás)-(pás-kan) *

Even if it were possible to work out the technical details of Cohn and
McCarthy’s Indonesian analysis, the deeper issue is whether or not the intuition
Cohn and McCarthy, and Kenstowicz, pursue for Indonesian is the right one,
cross-linguistically. The intuition behind their analysis is that if a language has
a phonological alternation that fails to apply in reduplication, the inhibition is
due to BR identity; only when perfect identity is impossible for independent
reasons is the alternation predicted to apply, in the manner of TETU.
Our investigations, detailed below, fail to support this intuition. Lack of appli-
cation of a phonological alternation in reduplication is due to its absence from
the cophonology associated with that reduplication construction; application of
the alternation when the construction is embedded in further morphology is due
to the presence of the alternation in the cophonology associated with the higher
morphological constructions. Interestingly, this is the analysis originally given
to Indonesian by Cohn (1989); we have found its counterpart to be the correct
analysis in numerous other cases as well.
 Reduplication-specific non-alternation 113

4.3.5 Klamath
We illustrate our argument against the equation of phonological underapplica-
tion with maintenance of base-reduplicant identity with Klamath intensive redu-
plication, a well-known case for which that equation has been made (McCarthy
& Prince 1995a).15 The relevant phonological alternation in Klamath is vowel
reduction/syncope. Both prefixes and root reduplication trigger vowel reduction
in the second syllable of the derived stem (Barker 1963; 1964).16 If the second
syllable is closed, its vowel reduces to schwa (35a, b). If the second syllable
would be open, the vowel deletes (35c, d) altogether. Henceforth, page numbers
for data taken from Barker (1963) are prefixed with “B63”; those for data taken
from Barker (1964) are prefixed with “B64.”17

(35) Plain stem Prefixed stem

a. domna so-d[ə]mna B63:121
‘hears, obeys, understands’ ‘hear each other’
b. čonwa hos-č[ə]nwa B63:166
‘vomits’ ‘cause to vomit’
c. qabata has-q[Ø]ba:ta B63:310
‘sets something heavy up ‘causes s.o. to put a heavy or
against’ pronged object up against’
d. wp’eq’a ʔi-p[Ø]q’a B63:308
‘hits in the face with a long ‘puts plural objects on the face’
instrument’

Barker (1964) considers Intensive reduplication to be prefixal (though, since

reduplication is total, no evidence identifies either copy as the reduplicant as
opposed to the base).18 As shown in (36), the Intensive does not trigger vowel
reduction.

(36) Plain stem Intensive

a. /beq’-l’i/ ‘be bay colored, dawn’ beq-beq-l’i B63:61
b. /Liw’-a/ ‘shiver (from fatigue, hunger)’ Liw-Liw’-a B63:229

McCarthy and Prince (1995a), who adopt Barker’s prefixing analysis of

Intensive reduplication, argue that the failure of vowel reduction to apply in
the Intensive is a BR-identity effect. Their analysis relies not only on BR cor-
respondence but also on backcopying, a possibility unique to BRCT, in which
the surface form of the reduplicant can influence, via BR correspondence, the
surface form of the base. McCarthy and Prince’s (1995a) backcopying analysis
is illustrated by the tableau in (37). The constraint called Reduce stands for
114 Reduplicative phonology: mothers

the set of constraints that motivate vowel reduction and deletion; it is outranked
by IO-Faith-␴ 1 , which protects the vowel in the initial syllable from
reduction/deletion, with the result that only the second syllable is normally
affected (as in (35)). In the Intensive form shown in (37), normal second syllable
reduction, which would affect only the base (37a), is blocked because it would
render the base and the reduplicant non-identical. Reducing both copies, which
would also maintain identity, fatally violates the high-ranking IO-Faith-␴ 1
(37b) that protects the initial syllable from reduction. The optimal solution is
to reduce neither (37c):

(37)

..............................................
/INT-Wič-l’i/ IO-Faith-␴ 1 BR-Faith Reduce

a. W[ə]č-W[ə]č-l’i *!

b. Wič-W[ə]č-l’i *! *

) c. Wič-Wič-l’i **

If this is the right analysis, then the failure of reduction in Intensive redu-
plication requires some version of BR faithfulness, and would be incompatible
with MDT. However, the backcopying analysis has two major drawbacks. First,
it does not take into account the fact that vowel reduction also fails to apply in
certain nonreduplicative morphological contexts; second, it does not take into
account the fact that reduction does apply in some reduplicative contexts. The
BR identity analysis presupposes an implicational connection between redupli-
cation and lack of vowel reduction which simply does not exist, a point made
by Zoll (2002), on which this discussion draws; Park (2000) reaches similar
conclusions as well. To support this view, we present some background on
Klamath morphology.
The verb in Klamath consists of a number of prefix classes, a root, and a
variety of suffixes (Barker 1964; Delancey 1991; 1999):

(38) Distributive Causative, Causative, Classifiers Intensive Root Suffixes

Reflexive- Transitive
reciprocal

Vowel reduction is morphologically conditioned, applying only in certain

zones of the verb. As shown below, vowel reduction/deletion of the second
syllable in the morphological domain is not triggered by suffixes (39a, b),
nor does it apply root-internally (39c) (White 1973; Thomas 1974). Roots are
underlined:
 Reduplication-specific non-alternation 115

(39) a. se-tpek-bli-bg-a se-tp[ə]k-bli-pg-a *se-tp[ə]k-bl[ə]-pg-a B64:155

‘reaches back of oneself’
b. bo:s-dgi-di:l-a → bo:s-tgi-di:l-a *bo:s-tg[Ø]-di:l-a B64:141
‘turns black underneath’

Bare roots and root-suffix sequences are contexts where BR identity is not at
stake; thus some other mechanism is needed to prevent vowel reduction in these
domains, while allowing it in prefixed stems like those in (35). MDT assumes,
as for all morphologically conditioned phonology, a cophonological differ-
ence. Stems, consisting of root plus suffixes, are associated with a cophonol-
ogy that does not enforce vowel reduction (IO-Faith » Reduce). By con-
trast, prefixes like those in (35) are associated with a cophonology that does
enforce vowel deletion in noninitial syllables (IO-Faith-Initial » Reduce »
IO-Faith).
The same cophonological differences found outside of reduplication also
obtain among reduplication constructions. Vowel reduction fails to apply in the
Intensive reduplication construction (36), as we have seen; however, it does
apply in another reduplication construction: the Distributive (40).

(40) qlin ‘chokes . . .’ qli-ql[ə]n ‘choke (Distributive)’ B63:321

domna ‘hears, so-d[ə]mna ‘hear each other (Distributive)’ B63:121
obeys . . .’
čonwa ‘vomits’ hos-č[ə]nwa ‘cause to vomit’ (Distributive)’ B63:166
pag-a ‘barks’ pa-p[Ø]g-a ‘bark (Distributive)’ B63:296

McCarthy and Prince (1995a) distinguish Distributive reduplication from

Intensive reduplication by splitting BR-Faith into two constraints, one spe-
cific to the Distributive reduplicant and one specific to the Intensive redu-
plicant. Reduce ranks below BR-Faith-Intensive and above BR-Faith-
Distributive. This is essentially equivalent to the cophonology approach that
MDT takes, without invoking BR-Faith at all: Distributive stems are asso-
ciated in MDT with the cophonology Reduce» IO-Faith, while Intensive
stems are associated with the cophonology IO-Faith » Reduce. Because
construction-specific constraint reranking – cophonology differentiation – is
needed anyway in Klamath, IO-Faith is sufficient to account for the observed
variation. BR-Faith in general, and backcopying in particular, are superfluous.
A concern even greater than superfluity for the backcopying analysis is an
incorrect prediction that it generates. Recall that on the backcopying analysis
of simple reduplicated stems, Reduction/Deletion is blocked in the redupli-
cant because it is word-initial; BR identity is responsible for its blockage in the
Base (37). The backcopying analysis therefore clearly predicts that when Red
116 Reduplicative phonology: mothers

is not word-initial, and thus not protected by positional faithfulness, reduction

will occur in both Red and Base (McCarthy & Prince 1995a:349). However,
as illustrated by the forms in (41), when an Intensive is preceded by a pre-
fix that normally triggers second syllable vowel reduction, only the syllable
immediately following the prefix reduces.
(41)
a. /Wič/ ‘be stiff’
Wič-Wič-l’i ‘stiff’ B63: 458
Wi-W[ə]č-Wičl’i ‘stiff (Dist)’ * Wi-W[ə]č-W[ə]č-l’i B64:121
b. /Leq’/ ‘whisper’
Leq-Leq’-a ‘whispers’ B63:228
Le-L[ə]q-Leq’-a ‘whisper (Dist)’ *Le-L[ə]q-L[ə]q’-a
se-L[ə]q-Leq’-a ‘whispers to oneself, *se-L[ə]q-L[ə]q’-a
to each other’
c. /Wit’/ ‘flop’
Wit-Wit’ -a ‘flops’ B63:458
Wi-W[ə]t-Wit-kang-a ‘flop around (Dist)’ *Wi-W[ə]t-W[ə]t-kang-a
sni-W[ə]t-Wit-kang-a ‘make s.t. flop around’ *sni-W[ə]t-W[ə]t-kang-a

The prediction that reduction will be backcopied to the base, yielding the
starred forms in (41), is illustrated below. Only Reduce violations for the
second syllable are shown:
(42)

se-Red-Leq’-a IO-Faith-Initial BR-Faith Reduce IO-Faith

a. se-Leq’-Leq’-a *!

()) b. se-L[ə]q’-Leq’-a *!

0 c. se-L[ə]q’-L[ə]q’-a *

McCarthy and Prince (1995a) cite a different form which they interpret as
showing that prefixed Intensive forms do show reduction in both base and redu-
plicant: /sw’V-Red-ciq’-a/ → sw’i-c[ə]q-c[Ø]q’a ‘shake the head’ (1995a:349),
which has reduction in the reduplicant and deletion in the base. This form is
assumed to be representative of prefixed reduplicated forms in Klamath gen-
erally. (McCarthy & Prince observe that the form is problematic in that the
reduplicant and base reduce differently: schwa in the reduplicant, and deletion
in the base.)
An exhaustive search of Barker (1964) reveals, however, that this form is
not representative and that double reduction of prefixed intensive stems is not
the norm. Crucially, there are no examples in which an unprefixed intensive
 Reduplication-specific non-alternation 117

stem, without reduction, coexists with a prefixed counterpart showing reduc-

tion in both reduplicant and base. As the examples below illustrate, either the
unprefixed form shows a regular Reduction/Deletion pattern targeting the sec-
ond syllable (43a), or there is no unprefixed form given (43b, d). The predicted
alternation between a doubly reduced prefixed form and an unreduced plain
form is unattested.

(43) Root Reduplicated Prefixed reduplicated

a. /čiq’/ čiq-cq’a sw’i-č[ə]q-č[Ø]q’-a B63:76
‘shakes’ ‘shakes the head (like a horse)’
b. /k’ač’/ tga-k’[ə]č-k[Ø]č’-ač’ B63:197
‘from earth to ‘stands from earth to sky
sky’ (said of a character in a myth)’
c. /peL’/ sw’e-p[ə]L-p[Ø]l’-a B63:300
‘shake all over’ ‘shakes all over, shakes oneself
(as a dog)’
d. /qt’ aq’/ sa-qt’ [ə]q-qt[Ø]q’-a B63:324
‘clap the hands’ ‘claps the hands together’

Under any account, forms like č[ə]q-č[Ø]q’-a, which exhibit double reduc-
tion even when unprefixed, are lexical exceptions and must be listed as such.
Given this, the patterning of the rest of the data reflects normal application of
vowel reduction.
In summary, the failure of the Intensive to trigger reduction is not directly
attributable to BR identity. Instead, it must be attributed to the association of
the Intensive construction with a non-reducing cophonology. The side-by-side
structures in (44) of an Intensive verb and a prefixed Intensive verb show how the
differential application of reduction in the two cases follows directly from the
association of the appropriate cophonologies with the constructions involved.
Each arrow represents the effect of the cophonology associated with the mother
node just above:

(44) Unprefixed Intensive Prefixed Intensive

se-L[ə]q-Leq’
⇐ 2nd syllable reduction
Leq-Leq’ Leq-Leq’
⇐ No reduction ⇐ No reduction
Leq’ Leq’ se- Leq’ Leq’

Zoll (2002) argues that the association of Intensive reduplication with a

non-reducing cophonology is not arbitrary but follows from the fact that the
Intensive construction is stem-internal, thus subsumed under the generalization
118 Reduplicative phonology: mothers

that reduction does not apply stem-internally. This analysis accords with the
insights of Barker (1964) and White (1973), who treat Intensive reduplication
essentially as root-root compounding.
By contrast, the Distributive, a late morphological process, patterns with
stem-external prefixes in triggering vowel reduction. The association of clusters
of morphological constructions, or zones in the word, with similar cophonolo-
gies is the kind of pattern that stratal ordering was originally developed to
handle; it can be captured in cophonology theory through meta constructions;
see, for example, Inkelas and Orgun 1998; Stump 1998; Anttila 2002 for
discussion.
In summary, the failure of vowel reduction in Intensive reduplication in
Klamath is a case of normal non-application of Reduction/Deletion stem-
internally, not backcopying underapplication. The proposed cophonology anal-
ysis is capable not only of distinguishing constructions triggering reduction
from those which do not, but also of correctly predicting the outcome in forms
containing both kinds of constructions. In the latter respect it is more descrip-
tively adequate than the backcopying analysis argued for in McCarthy and
Prince 1995a. Our reanalysis thus eliminates Klamath as support for BR cor-
respondence, while also making the more general point that morphological
investigation is essential to the understanding of the non-application of a phono-
logical alternation.19

4.4 ∃-Faith
The MDT view that reduplicative phonology is simply a subcase of morpho-
logically conditioned phonology, modeled in exactly the same way, contrasts
sharply with the BRCT view, stated most strongly by Spaelti (1997):

all deviations from complete identity between the two parts of the reduplication
are due to [TETU]. In contrast to this any unexpected identity between the redu-
plicant and the base [is] the result of what I will call Identity Induced
Failure of Alternation (IIFA). These two concepts constitute the
entirety of the theory of reduplication. (p. 43)

The difficulty with this view, as we have seen, is that it does not extend
to the broader range of morphologically conditioned phonology found across
languages, nor does it fully account for the phonology of reduplication itself.
One proposal within the BRCT framework that has made some inroads into
the latter problem is existential faithfulness (∃-Faith), a proposal by Struijke
(1998; 2000a; 2000b), building on earlier work by Raimy and Idsardi 1997;
 ∃-Faith 119

Spaelti 1997; Fitzgerald 2000 (see also Rose 1999) which addresses the issue
of ranking paradoxes arising when alternations which are not fully general
in the language target the base of reduplication. One such paradox arises in
Dakota (§4.2), where coronal dissimilation applies across the internal juncture
only in reduplication; Tarok, in Chapter 3, represents another ranking paradox
in which reduplication-specific neutralization applies to the base. Klamath, in
which vowel reduction applies to the base in some cases of reduplication but
not generally in the language, is yet another such example. In each case, the
paradox is the same. For the alternation in question to apply to the base of
reduplication, it must be the case that the relevant markedness constraint (set)
outranks IO-Faith. That predicts, however, that the alternation will also occur
outside reduplication, which is not true.
Perhaps the best-known example in the literature of a reduplication-induced
ranking paradox comes from diminutive reduplication in Lushootseed, dis-
cussed earlier in this chapter, and one of the languages highlighted in the work
of Struijke (2000a; b). Lushootseed exhibits a process of unstressed vowel
syncope/reduction. Earlier work on Lushootsheed (Hess 1977; Hess & Hilbert
1978; Broselow 1983; Bates 1986; Hess 1993; 1995) observes that reduction fre-
quently occurs between voiceless obstruents; Urbanczyk (1996) demonstrates
that the alternation targets unstressed vowels, primarily the low vowel /a/. An
overview of the morphological conditioning of syncope, oversimplified in a cru-
cial way to be clarified shortly, is as follows. In unreduplicated words, reduction
is optional, leading to free variation (a), while in reduplicated words, reduction
is obligatory (b). Which vowel reduces – reduplicant or base – depends on
the location of stress, as discussed in Urbanczyk 1996 and references therein.
Whether an unstressed /a/ syncopates or reduces is predictable from segmental
environment; we subsume both processes under the term “reduction.” The data
below are taken from Urbanczyk 1996:

(45)
a. caq’(a) ‘spear, jab’ cá-cq’ ‘act of spearing big game on water’ 167
w w
šaX -il ‘grass, hay’ sá-ʔsX -il ‘(short) grass, lawn’ 167
wális ‘type of frog’ wá-w’lis ‘Little Frog (wife of p’ic’ikw )’ 167
laq-il ‘late’ lá-ʔlq-il ‘be a little late’ 167
q’aXa-cut ‘uncover’ q’ə-ʔq’áXa-cut ‘show self a little’ 169
b. sčúsad ∼ sčúsəd ‘(established) star’ 166
hı́qab ∼ hı́qəb ‘too, excessively’ 166
qacágw=ac ∼ qəcágw=ac ∼ qcágw=ac ‘ironwood, ocean spray, spiraea’ 166
Xá␭’=al=ap ∼ Xə␭’=ál=ap ‘steer a canoe with a paddle held 166
over the stern (like a rudder)’
120 Reduplicative phonology: mothers

The ranking paradox lies in the fact that the constraint rankings driving the
pattern in bases of reduplication are clearly different from those driving the
pattern in unreduplicated bases. In particular, the ranking that entails obliga-
tory reduction in diminutive reduplication, Reduce » IO-Faith, incorrectly
predicts obligatory reduction outside of reduplication as well.20
The cophonology approach would resolve this paradox by positing two
cophonologies, one associated with unreduplicated stems, and one associated
with the reduplication construction.

(46) Reducing cophonology: Reduce » IO-Faith

Optionally reducing cophonology: IO-Faith ∼ Reduce

Existential faithfulness, or ∃-Faith, addresses the paradox without appeal-

ing to cophonologies by introducing a modified type of IO faithfulness which
calls for every segment and feature of the input to have some correspondent
in the output. The only difference between ∃-Faith and IO-Faith, which
it is intended to replace, is that, in reduplication, ∃-Faith is satisfied by the
appearance of input material in either the reduplicant or the base. The follow-
ing definitions of ∃-Max and ∃-Ident, members of the ∃-Faith family, are
taken from Struijke 2000a:

(47) a. ∃-Max io (p. 21)

Let seg ∈ input: then there is some seg’ ∈ output, such that segseg’
Every segment in the input has some correspondent in the output
a. ∃-Ident io (p. 24)
Let seg ∈ input be in the domain of , and seg is [␣F]: then there is
some seg’ ∈ output, such that segseg’ and seg’ is [␣F].
Some output segment corresponding to an input segment preserves
the feature specification [aF] of that input segment.

∃-Faith gives BRCT some purchase on the reduplication-specific junctural

effects discussed in previous sections by potentially substituting for constraint
indexation to the reduplication construction. ∃-Faith limits certain types of
alternations to just the context where one correspondent of an input segment is
preserved. Struijke’s analysis of Lushootseed appeals to ∃-Faith to distinguish
reduplication from nonreduplicative contexts. The insight is intuitive: reduction
is, in principle, optional in both constructions. In nonreduplicated constructions,
reduction is in a tug of war with ∃-Faith, which is violated whenever reduction
is satisified. In reduplicative constructions, however, ∃-Faith is satisfied even
when reduction applies in the base, because the input vowel is still preserved
intact in the reduplicant.
 ∃-Faith 121

(48)
/sčúsad/ ∃-Faith Reduce

) a. no reduction sčúsad *

b. reduction sčúsəd *!

/Red-cáq’/

c. no reduction cá-caq’ *!

) d. reduction cá-cq’

If ∃-Faith can not only resolve the ranking paradox which MDT resolves
using different cophonologies but also predict which construction – redupli-
cation or nonreduplication – will be associated with which pattern, ∃-Faith
would be an appealing alternative to cophonologies.
To evaluate the potential of ∃-Faith to obviate cophonologies, we turn
now to an in-depth exploration of the intuitions underlying ∃-Faith and the
predictions it makes. We ultimately conclude that ∃-Faith is neither nec-
essary nor sufficient in accounting for reduplicative phonology, nor can it
extend, as cophonologies do, to nonreduplicative morphologically conditioned
phonology.

4.4.1 Predictions of ∃-Faith

∃-Faith potentially offers a way of understanding reduplicative phonology
without recourse to cophonologies. The claim is that reduplicative phonology
differs from nonreduplicative phonology for a functional reason: the existence
of two correspondents for each output permits neutralization to occur once in
reduplication even in languages that do not permit the corresponding neutral-
ization process outside of reduplication.
This is not, however, the right generalization. As we will see, reduplicative
phonology does not differ from nonreduplicative phonology in precisely this
way. ∃-Faith is therefore not sufficient to account for the differences between
reduplicative and nonreduplicative constructions in the same language. More-
over, even within reduplicative constructions in the same language we find
variation for which ∃-Faith cannot account. The mechanisms required to
describe this variation render ∃-Faith unnecessary. ∃-Faith is therefore nei-
ther necessary nor sufficient to account for reduplication-specific phonology.
The intuition it embodies is not correct in any general way.
122 Reduplicative phonology: mothers

In this section we explore four aspects of ∃-Faith that lead to the conclusion
we have just stated.

(49) a. ∃-Faith undergenerates, predicting no overapplication of

reduplication-specific phonology
b. ∃-Faith undergenerates, being unable to derive construction-
specific insertion in reduplication
c. ∃-Faith is superfluous in handling reduplication-internal
differences in the same language
d. ∃-Faith does not capture parallels between reduplicative and
non-reduplicative construction-specific phonology

4.4.2 Overapplication of reduplication-specific phonology

∃-Faith predicts that if a neutralization occurs in either the reduplicant or the
base which does not occur generally in the language, it should affect only seg-
ments in one copy and, in particular, it should affect only those segments which
have intact correspondents in the other copy. The reason is that if the neutraliza-
tion does not occur generally, ∃-Faith must outrank the relevant Markedness
constraint; from that ranking it follows that the neutralization can affect only
one of the two output correspondents of the input structure in question.
This prediction is, however, incorrect. It is possible to find double applica-
tion of neutralization in reduplication. We saw cases in Chapter 2 of double
Melodic Overwriting in which neither reduplicant nor base preserved an input
segment that, outside of reduplication, is always preserved. Of course, Melodic
Overwriting presents extra complications, and it could be that ∃-Faith is sim-
ply outranked in that case by the need for the Melodic Overwriting affix(es) to
surface. However, the case of Hausa reduplicative adjective formation is a real
counterexample to the predictions of ∃-Faith.
In Hausa, “[a]djectives are formed from common nouns by full reduplication
of the underlying noun, preserving the original tone. This is accompanied by
shortening of the final vowel of both parts” (Newman 2000:27):21
(50) gishiri gishiri-gishiri ‘salt/salty’
bùhu bùhu-bùhu ‘sack/sacklike’
gàri gàri-gàri ‘flour/powdery’
màkubà màkubà-màkubà ‘mahogany/dark brownish color’

The process of vowel shortening that occurs in both copies in reduplication

does not occur in the unreduplicated forms, nor is word-final vowel shortening
a general process in the language. The fact that vowel length contrasts are
generally preserved word-finally in Hausa means that ∃-Faith must outrank
the ban on long vowels: ∃-Faith » *Final-V:
 ∃-Faith 123

(51)
/ bùhu/ ∃-Faith *Final-V:

) bùhu

bùhu *!

Following the logic of ∃-Faith, it is possible, in a language with the general

ranking ∃-Faith » *Final-V, to get final vowel shortening in reduplication –
but only once (52b), and only if BR-Faith is ranked very low:
(52)

.....................................................
/Red-bùhu/ ∃-Faith *Final-V: BR-Faith

possible a. bùhu-bùhu **

possible b. bùhu-bùhu (or * *

bùhu-bùhu)

impossible c. bùhu-bùhu *!

Given the ranking ∃-Faith » *Final V:, there is no ranking of BR-Faith

that can achieve overapplication of final vowel shortening (52c). The only out-
come ∃-Faith does not predict possible is the actual outcome in Hausa.
A theory with ∃-Faith could, of course, describe this Hausa construction
if we abandon the general ranking of the language and rerank ∃-Faith below
*Final-V: for this particular construction:

(53) Hausa generally: ∃-Faith » *Final V:

Hausa adjective reduplication: *Final-V: » ∃-Faith

But this move undermines the primary motivation behind ∃-Faith, which
was to resolve ranking paradoxes without resorting to reranking. Once reranking
is introduced as a possibility, there is no longer any need for ∃-Faith; we can
simply use cophonologies, as MDT does, which utilize markedness constraints
and standard IO-Faith to generate the right outcomes. The cophonological
approach to Hausa is shown below:

(54) Hausa generally: IO-Faith » *Final V:

Cophonologies X and Y in adjective reduplication: *Final-V: » IO-Faith

Once we acknowledge the need to rerank Faithfulness (whether IO-

Faith or ∃-Faith) and markedness constraints across reduplicative and
nonreduplicative constructions, ∃-Faith becomes unnecessary as a tool for
distinguishing reduplicative and nonreduplicative phonology.
124 Reduplicative phonology: mothers

4.4.3 Construction-specific insertion

Another prediction resulting from reliance on ∃-Faith as the instrument of
reduplication-specific phonology is that reduplication-specific phonology will
be limited to alternations that alter material within the reduplicant or the base,
and will not include insertion of material between the reduplicant and base.
This prediction follows from the standard assumption that epenthetic material,
especially at morpheme junctures, does not belong to either morpheme (see, for
example, Prince & Smolensky 1993: 20–21 for a recent statement in Optimality
Theory).
This prediction is overly restrictive, as illustrated in (55) by data from
Fox. As described in Jones 1919; Dahlstrom 1997 (see also Burkhardt 2002)
and discussed more extensively in Chapter 5, vowel hiatus at the base-
reduplicant juncture is resolved via h-epenthesis. Data are taken from Dahlstrom
1997:

(55) amwe-h-amwe:wa ‘he eats him’ 219

ayo-h-ayo:ya:ni ‘I use it; conjunct’ 219
iwa-h-iwa ‘he says’ 219

While the process of epenthesis at VV junctures is completely general, the

quality of the epenthetic segment is construction specific. Vowel hiatus across
prefix-stem junctures is broken up by an epenthetic t (56a), rather than h. (In yet a
third morphological context, discussed in Dahlstrom 1997, n is used instead.) A
form with both epenthetic t and epenthetic h is shown in (56b); the reduplicated
stem is bracketed:

(56) a. ke-t-en-a:wa ‘you say to him’ 220

ne-t-amw-a:wa ‘I eat him’ 220
ke-t-a:čimo ‘you tell a story’ 220
ne-t-ekw a ‘he says to me’ 221
b. ne-t-[ena-h-ina:pi] ‘I look [there]’ 216

Following Optimality Theory approaches to markedness (e.g. Smolensky

1993; Kiparsky 1994; DeLacy 2002), we assume that the quality of epenthetic
segments is determined by the relative ranking of faithfulness and markedness
constraints. On this standard view, Fox therefore requires (at least) two different
junctural cophonologies, one for affix-stem constructions and the other for
reduplication. We use segment-specific IO-Dep here; other analyses are also
imaginable:

(57) Affix-Stem construction: IO-Dep(h) » IO-Dep(t)

Reduplication construction: IO-Dep(t) » IO-Dep(h)
 ∃-Faith 125

Since epenthesis in Fox involves insertion between reduplicant and base,

rather than modification internal to either constituent, the difference between the
two types of constructions cannot be reduced to a consequence of BR-Faith
constraints alone; the epenthetic segment is outside the scope of BR-Faith.
More to the point, junctural epenthesis also cannot be reduced to ∃-Faith. This
is because the intent of ∃-Faith is to make sure that everything present under-
lyingly appears somewhere on the surface. It will not be violated by the presence
of anything extra on the surface (unless it violates ∃-Contiguity, by insertion
within a morpheme). Recognizing this, Struijke uses M-seg (McCarthy 1993)
in place of Dep,in the context of a different discussion, to regulate epenthesis
in reduplication:

(58) M-seg ‘Every segment belongs to a morpheme.’

The substitution of M-seg for Dep in Fox is straightforward, but to achieve

the necessary construction-specificity of the epenthetic segment, an M-seg
analysis still requires constraint reranking across different morphological con-
structions. We therefore reach the same conclusion arrived at above: ∃-Faith
cannot take the place of constraint reranking, or cophonologies, in accounting
for reduplication-specific phonology. Again, given the existence of cophonolo-
gies, ∃-Faith is superfluous.

4.4.4 Reduplication-internal variation

Another crucial lacuna in ∃-Faith theory is its inability to capture varia-
tion across reduplicative constructions in the same language. We will illustrate
this point using data from Lushootseed. Recall, from our earlier discussion of
Lushootseed, the caveat that the picture of Lushootsheed reduction presented
in (45) was oversimplified in a crucial way.22 The information missing from
that picture is the following. As discussed in Urbanczyk 1996 and the literature
cited therein, Diminutive reduplication is not uniformly subject to reduction.
Rather, some stems reduce invariably when they reduplicate, including those
shown in (45a), while others invariably fail to reduce when they reduplicate
(Urbanczyk 1996:184ff.). The data below are taken from Urbanczyk 1996:186;
unreduced unstressed /a/ vowels in reduplicated stems are in boldface:

(59) Plain stem Diminutive stem

Xa␭’-il Xá-Xa␭-il ‘argue’/‘squabble’

␭ac’=ap-əb ␭’á-␭ac’=ap-əd ‘belt’/‘ant, small (lit. little cinched up one)’
talə táʔ-talə ‘dollar’/‘small amount of money’
čagw -is čá-čagw -is ‘become irritated’/?
126 Reduplicative phonology: mothers

According to Urbanczyk, stems which fail to reduce in reduplication repre-

sent a minority pattern; nonetheless, they must be accounted for. Urbancyzk
proposes a constraint reranking account (p. 184), which is equivalent to using
cophonologies. The behavior of non-reducing stems requires the ranking IO-
Faith » Reduce.
(60) Reducing cophonology: Reduce » IO-Faith
Non-reducing cophonology: IO-Faith » Reduce

The problem for ∃-Faith is that if ∃-Faith is substituted for IO-Faith, the
distinction between these two patterns collapses. ∃-Faith predicts reduction
in reduplication regardless of where it is ranked with respect to Reduce.
(61)

...................................
/Red-Xa␭’-il/ ∃-Faith (V) Reduce

()) a. Xá -Xa␭’-il *!

0 b. Xá -X␭’-il

By its very nature ∃-Faith is incapable of being reranked in such a way

as to account for the difference between reducing and non-reducing variants of
the diminutive reduplication construction. The ∃-Faith analysis must rerank
BR-Faith instead, as shown below:
(62) Non-reducing cophonology: ∃-Faith » BR-Faith » Reduce
Reducing cophonology: ∃-Faith, Reduce » BR-Faith

A tableau showing the correct result for a non-reducing stem is given below
(compare to the failed tableau, above):
(63)
/Red-Xa␭’-il/ ∃-Faith (V) BR-Faith Reduce

) a. Xá -Xa␭’-il *

b. Xá -X␭’-il *!

The only way to save the ∃-Faith approach from descriptive inadequacy is
to permit reranking of BR-Faith and Reduce across different reduplication
constructions, as shown below.
(64) Reducing reduplicative construction: ∃-Faith, Reduce » BR Faith
Non-reducing reduplication construction: ∃-Faith, BR Faith » Reduce

But as soon as this is done, ∃-Faith immediately becomes superflu-

ous. Once we accept reranking of BR-Faith with respect to Markedness
 ∃-Faith 127

constraints across different constructions, we have essentially emulated the

MDT cophonology analysis, which requires neither ∃-Faith nor BR-Faith.
As shown below, IO-Faith and Markedness, ranked suitably in the two rele-
vant cophonologies, are all that is needed:
(65) Reducing cophonology: Reduce » IO-Faith
Non-reducing cophonology: IO-Faith » Reduce

∃-Faith thus plays no role, in the end, in explaining the applicability of

reduction in reduplication in Lushootseed. While the intuition it is intended to
capture – that reduplication will exhibit more neutralization than is permitted
outside reduplication, because of the increased opportunity it has to preserve
input contrasts – is not falsified by the data, it is true only for a circumscribed
portion of the data. A fully general approach to reduplicative phonology requires
consideration of that data in a broader context.
Lushootseed is by no means unique; many other Salish languages exhibit
variation in the application of syncope/reduction inside and outside of redupli-
cation (on Bella Coola, for example, see Newman 1971; Broselow 1983; Nater
1984; 1990; Carlson 1997; Raimy & Idsardi 1997). Farther afield, we find vari-
ation of a different type in Hausa, where codas neutralize differently not only
inside and outside of reduplication but also within the set of reduplication con-
structions the language possesses. Hausa imposes fairly strict constraints on
codas throughout the language. Coronals are allowed the most leeway, exhibit-
ing the following variety of behaviors (data and generalizations from Newman
2000):
(66) Outside reduplication: plosives rhotacize; fricatives do not
alternate (a)
Most reduplication: all coronals rhotacize, including fricatives (b)
Pluractional reduplication: all coronals geminate (c)
a. a:tà: ‘to damage’ 233
àr̃na: ‘damage (n.)’
cf. kasko: ‘earthen bowl’ 233
fı̀zga: ‘grab, snatch’ 233
b. /gwado:/ gwàr̃-gwado: ‘proportion, 235
moderation’
/maza/ mar̃-maza ‘very quickly’ 235
/kwà:sa:/ kwar̃-kwà:sa: ‘driver ant’ 235
c. /fı̀ta/ fı̀f-fı̀ta (∼ fı̀r̃-fı̀ta) ‘go out’ 425
/gasà:/ gag-gàsa: (∼ gar̃-gàsa:) ‘roast’ 425

It is clear that reduplicative phonology is not monolithic in nature.

128 Reduplicative phonology: mothers

4.4.5 Parallels between reduplicative and nonreduplicative phonology

The final argument against relying on ∃-Faith to understand the differences
between reduplicative and nonreduplicative phonology is the body of evidence
in which the morphologically conditioned phonological effects that set some
reduplication constructions apart from other constructions in the language find
parallels in nonreduplicative constructions in the same language as well. Not
only is reduplicative phonology not internally monolithic; it is also not distinct,
in aggregate, from nonreduplicative phonology.
Like a blunt knife, ∃-Faith divides constructions into reduplicative vs.
nonreduplicative ones with respect to their predicted phonological behavior.
However, we have seen from Lushootseed and Hausa that reduplication con-
structions in the same language are not identical in their phonology. What we
will show now is that the same kinds of differences found among reduplicative
constructions are found among nonreduplicative constructions as well, moti-
vating a view like that in (c):
(67) a. ∃-FAITH prediction b. Lushootseed
phonology x
phonology x Some reduplication
Reduplication
constructions
constructions
phonology y

other reduplication
Nonreduplication
constructions
constructions
phonology z

phonology y
Nonreduplication

constructions

c. Still other languages

Some reduplication other reduplication

constructions and constructions and

some nonreduplication nonreduplication

phonology x phonology y

yet other phonologies

yet other

constructions
 ∃-Faith 129

The facts we discuss are of a familiar type; the novelty in our discussion lies
only in pointing out their implications for ∃-Faith.

4.4.5.1 Reduplication and other derived environments

In Tohono O’odham (Uto-Aztecan), the primary differentiation in phonolog-
ical behavior lines up with the distinction between morphologically derived
and morphologically underived words. Stress assignment in Tohono O’odham
places main stress on the initial syllable, followed by alternating secondary
stress. The interesting twist in Tohono O’odham is that nonderived words pro-
hibit final stress (69a), but derived ones permit it (69b) (Hill & Zepeda 1992;
Fitzgerald 1999; 2000; Yu 2000 (Y); Fitzgerald 2001 (F); on Tohono O’odham
phonology generally, see also Zepeda 1983; 1984; Hill & Zepeda 1998):

(68) a. Nonderived words: no stress on final syllable

kı́: ‘house’ Y118
pı́:ba ‘pipe’ Y118
ʔásugal ‘sugar’ Y118
sı́minˇul ‘cemetery’ Y118
b. Derived words: stress permitted on final syllable
cı́kpan-dàm ‘worker’ Y119
má:ginà-kam ‘one with a car’ Y119
pı́miàndo-màd ‘adding pepper’ Y119

As shown in (69), reduplicated words pattern with derived words in allowing

secondary stress to fall on the final syllable:

(69) pı́-pibà ‘pipes’ Y119

pá-padò ‘ducks’ F942
sı́-sminˇùl (cf. sı́minˇul ‘cemetery’) ‘cemeteries’ F945
tá-tablò ‘shawls’ F945

This is a species of derived environment effect.23 For our purposes, what

makes it interesting is that it shows one common classification of words into
subphonologies: derived vs. underived, where the derived words include both
reduplicated and unreduplicated words. Yu (2000) demonstrates that the only
approach to this derived environment effect that can succeed is one that includes
distinct cophonologies for derived and underived words, as shown in (70). Sim-
plifying Yu’s analysis somewhat, the difference between the two cophonologies
amounts to the relative ranking between Parse-␴, which favors metrification
of all syllables, and constraints that penalize final stress, represented here as
Non-f inality (Prince & Smolensky 1993):
130 Reduplicative phonology: mothers

(70) a. Underived: Non-Finality » Parse-␴

b. Derived: Parse-␴ » Non-f inality

Tohono O’odham manifestly requires indexation of cophonologies (or con-

straints) to broad construction types. ∃-Faith is irrelevant, since reduplication
patterns with nonreduplicative derivation. Fitzgerald (2001) proposes to derive
the different stress patterns using a monostratal ranking, arguing that what drives
final stress is a morpheme-to-stress constraint. In the case of suffixed words, this
principle acts transparently to attract stress to the suffix. As Yu shows, however,
this analysis is problematic for various reasons, the simplest being the fact that
imperfective truncation attracts stress to the final syllable despite providing no
overt morpheme on which the stress can surface:

(71) Imperfective Perfective

a. sı́kon sı́ko ‘hoe object (imperf/perf)’ Y130
b. wáčuwı̀-čud wáčuwı̀c ‘make someone bathe (imperf/perf)’ Y130
bathe-caus
c. wákon-am`d wákon-àm ‘go and wash (imperf/perf)’ Y130
wash-to go

Yu’s cophonology account works straightforwardly for these data. Imperfec-

tive truncation is a morphological construction whose output is morphologi-
cally derived; it therefore undergoes final stress assignment, a property of all
cophonologies associated with morphologically derived environments.24

4.4.5.2 Reduplication and compounding

In Roviana (New Georgia; Northwest Solomonic subgroup of Oceanic), redu-
plication and compounding pattern together to the exclusion of other morpho-
logical constructions in three phonological respects (Corston-Oliver 2002:470):
stress, vowel deletion, and consonant cluster reduction.25 As shown in (72a),
stress falls on the first syllable of both constituents in reduplication and in com-
pounding. In suffixed words, by contrast, there is just one stress allowed per
morphologically complex word (with the exception of the stressed transitive
suffix -i [p. 469]).26 As shown in (72b), both CVCV reduplication and com-
pounding also optionally delete the second V in the first constituent; this appar-
ently does not occur in other morphological constructions. Finally, as shown in
(72c), clusters resulting from vowel deletion are simplified in the same way in
reduplication (RED) and compounding (CPD): nasals assimilate in place to a
following voiceless stop, geminates degeminate, and oral consonants optionally
delete:
 ∃-Faith 131

(72)
a. RED 'ham bo-'ham botu-ana27 ‘sit-sit-nml = chair’ 469
CPD '␤etu-'lotu ‘house-pray = church’ 469
b. RED ham bo-ham botu-ana ∼ ‘chair’ 470
ham-ham botu-ana
peka-peka ∼ pek-peka ‘dance-dance = dancing’ 470
CPD ␤etu-lotu ∼ ␤et-lotu ‘church’ 470
c. RED heγe-heγere ∼ heγ-heγere ∼ he-heγere ‘laugh-laugh’ 470
ra-raro (< rar-raro < raro-raro) ‘cook’ 470
pino-pino ∼ pim-pino (< pin-pino) ‘star’ 470
CPD <no examples given>

Even though vowel and consonant reduction are neutralization effects that
could, in reduplication, be attributed to ∃-Faith, an ∃-Faith account would
do nothing to capture the parallelism between reduplication and compounding
with respect to these effects.
Parallelism, in particular languages, between compounding and reduplication
constructions has often been noticed; a common response has been to propose
that reduplication and compounding both operate on prosodic words. While this
may well be the right approach to Roviana reduplication, it still does not elim-
inate the need for cophonologies. Vowel deletion is not a general phenomenon
at the end of prosodic words in Roviana; thus, it is still necessary to specify,
cophonologically, its occurrence in reduplication and compounding.
A similar constellation of reduplication and compounding is manifested in
Burmese. As is typical of languages in the Tibeto-Burman family, Burmese
imposes strict constraints on syllable structure. Complex onsets are allowed
only in the “main,” or “full” syllable of the word; a preceding “minor” syllable
is permitted only a simplex onset. Green (2002) analyzes “full” syllables as
feet and treats complex onsets as being licensed only foot-initially. Synchronic
alternations show that this distributional generalization is actively enforced.
As analyzed in VanBik 2003, Burmese has a process of compounding marked
phonologically by voicing of the initial obstruent of the second member (e.g. pè
‘peanut’ + pouʔ ‘rot/putrid’ = pè-bouʔ ‘fermented soybean’). Compounds are
subject to reduction, whereby the rime of the first member reduces to [ə], and
its onset simplifies and, if the second member begins with a voiced obstruent,
voices:28

(73) a. pà + moú → pə-moú

‘cheek’ ‘mound’ ‘prominence of cheek’
b. nwà + má → nə-má
‘cow’ ‘female’ ‘female cow’
132 Reduplicative phonology: mothers

c. ␪wà + teʔ → ðə-deʔ

‘tooth’ ‘rise’ ‘canine tooth’
d. pyà + tu → bə-du
‘bee’ ‘hammer’ ‘hornet’

Even though the onset reduction in reduplication might normally call out for
an ∃-Faith account, the clear parallelism between compounding and redupli-
cation shows that ∃-Faith is irrelevant in this reduction process.

4.4.5.3 Random grouping of things with same cophonology

Hausa is known for its distinction between affixes which preserve base tone and
those which replace base tone with a fixed melody. Newman (1986; 2000) terms
these “non-tone-integrating” and “tone-integrating,” respectively. Whether or
not a given affix is tone-integrating (the majority pattern) is, at least synchron-
ically, an essentially arbitrary fact about that affix.
As illustrated below, reduplication constructions are no different from affixa-
tion constructions in subdividing, arbitrarily, into those that are tone-integrating
and those that are not. The number of reduplication and affixation constructions
in Hausa is very large; we have selected only four here, for illustration. The
interested reader is encouraged to consult Newman (2000), from whose work
this data is drawn, as well as Downing (2004) for a recent survey of such effects.

(74) a. Tone-integrating, nonreduplicative: plural -ai; L*H (p. 434)

alı̀bi (HLH) → àlı̀b-ai (LL-H) ‘student(s)’
àlmùbazzàr̃i (LLHLH) → àlmùbàzzàr̃-ai (LLLL-H) ‘spendthrift(s)’
màkanikè (LHHL) → màkànı̀k-ai (LLL-H) ‘mechanic(s)’
b. Non-tone-integrating; nonreduplicative: -wa verbal noun former; (L)H
(p. 705)
kar̃ànta (HLH) → kar̃àntâ-wa (HLH-LH) ‘read(ing)’
tsoratar̃ (HHH) → tsoratâr̃-wa (HHH-LH) ‘frighten(ing)’
c. Tone integrating; reduplicative: Hypocoristic formation; L*HH output
tone (p. 348)
Àlhajı̀ (LHL) → Àlhàji-ji (LLH-H)
Ladı̀ (HL) → Làdi-di (LH-H)
Mù e (LH) → Mù e- e (LH-H)
d. Non-tone-integrating; reduplicative: pluractional CVG- (p. 425–26)29
dàgurà (LHL) → dàd-dàgurà (L-LHL) ‘gnaw at (pluractional)’
girma (HH) → gig-girma (H-HH) ‘grow up (pluractional)’
mutù (HL) → mur̃-mutù (H-HL) ‘die (pluractional)’

The relevance of these data to the theoretical treatment of morphologi-

cally conditioned phonology is discussed by Inkelas (1998), who applies the
 ∃-Faith 133

terms “dominant” and “recessive” to tone-integrating and non-tone-integrating

constructions. Inkelas argues on grounds of scope and layering that each
morphological construction must be associated with a cophonology which either
deletes or preserves input tone.
The distinctive tonal phonology of certain Hausa reduplication constructions
has nothing to do with special reduplicative faithfulness. First, tone-integrating
reduplication completely wipes out input stem tone; ∃-Faith is playing no
role. Second, tone-integrating affixation shows the same behavior. The differ-
ences between reduplication and nonreduplication constructions are identical
to the differences obtaining within reduplication, on the one hand, and within
nonreduplication constructions, on the other.

4.4.6 Wrapup
Our discussion of ∃-Faith has revealed that even in a theory with ∃-Faith
constraints, it is still necessary to appeal to cophonologies to capture the full
range of reduplicative phonology. Once cophonologies are invoked, ∃-Faith
becomes superfluous. ∃-Faith is neither necessary nor sufficient in the anal-
ysis of reduplicative phonology.
The main problem with ∃-Faith lies with its twin presuppositions that
reduplicative phonology is different from nonreduplicative phonology and that
the difference emanates from base-reduplicant identity. Both presuppositions,
however, derive from an overly narrow focus on reduplication. By situating
reduplication in the larger context of morphologically conditioned phonology,
we have found that the same cophonological approach needed outside redupli-
cation is necessary, and sufficient, to account for reduplicative phonology as
well.
Part of the appeal of ∃-Faith lies in its evocation of a commonly held view
that contrast preservation is a desideratum of grammar. Reduplication, offering
two chances to preserve each input contrast, might therefore be expected to
behave differently from other constructions. Syntagmatically, certainly, we have
seen that ∃-Faith is not substantiated. The paradigmatic function of contrast
preservation is harder to evaluate, and may have some truth. It is possible
that functional pressures to recover input information have created a situation
in which reduction is more likely, statistically, within reduplication; we do
not know, and it would take a massive diachronic and synchronic study to
be able to test this hypothesis. In any case, if one looks only at the types
of reductions that are possible, one finds that reduplicants have no special
status.
134 Reduplicative phonology: mothers

4.5 Conclusion
Reduplication constructions, as wholes, are often associated with phonologi-
cal effects not observed generally in the language. We have argued at length
that this phenomenon is exactly parallel to the phenomenon of morphologi-
cally conditioned phonology outside of reduplication. Theories that approach
reduplication out of morphological context and focus exclusively on effects
that appear related to the phonological identity between bases and reduplicant
miss the larger picture and unnecessarily burden the theory with reduplication-
specific analytical tools.
5 Morphologically driven opacity
in reduplication

One of the objectives driving theories of reduplication is accounting for opacity

effects, to which reduplication is perceived to be exceptionally prone. Opacity
is the situation in which an alternation which, elsewhere in the language, is nor-
mally conditioned in some particular environment either exceptionally fails to
apply in the expected environment or exceptionally applies in another environ-
ment where it is not supposed to (Kiparsky 1973a). In reduplication, the former
phenomenon is termed “underapplication”; the latter, “overapplication.”
Dakota (Shaw 1980:344–45) provides an example of the kind of opacity with
which reduplication is rife. In Dakota, velars palatalize after /i/; spirants voice
intervocalically. In the example of CVC reduplication, below, velar palataliza-
tion “overapplies” in the sense that it is conditioned transparently in the first
copy of the root but not in the second:

(1) wičh á-ki-čax-čax-ʔiyèya ‘he made it for them quickly’ (/kax/)

napé kı́-čos-čoz-a ‘he waved his hand to him’ (/kos/)

Because of languages in which reduplication appears to be the only mor-

phological context exhibiting a given pattern of overapplication or underap-
plication, opacity has emerged as a hallmark of reduplication. In a number of
theories (Clements 1985, Mester 1986, McCarthy & Prince 1995a), reduplica-
tive opacity is given special treatment that distinguishes it from other kinds of
opacity. Recent work on reduplication (McCarthy & Prince 1995a, building on
Wilbur 1973) has attributed reduplicative opacity effects to principles of base-
reduplicant identity. These approaches presuppose that opacity effects increase
identity between base and reduplicant.
This chapter, building on the results of Chapters 3 and 4, presents a different
view. In MDT reduplicative opacity follows from the morphology of reduplica-
tion constructions, which often conspires to obscure the conditions on the appli-
cation of phonological processes. The same approach that is taken in MDT to
opacity in nonreduplicative constructions carries over to reduplication as well,
without the need for reduplication-specific proposals like BR correspondence.

135
136 Morphologically driven opacity in reduplication

The perception that opacity is particularly prevalent in reduplication derives

from the fact that there are multiple sources for opacity in any given reduplicated
word. A principal factor is the morphological layering intrinsic to the MDT
reduplication construction, which gives rise to the cyclic, or stratal, interactional
effects to which Kiparsky (2000) attributes opacity in general.
In MDT, the output of the daughter cophonologies serves as input to the
mother cophonology. There are thus three cophonologies involved, and any of
them may produce opacity. For example, truncation within one of the daughters
may render opaque another phonological alternation that the daughter under-
goes. This type of daughter-based opacity occurs in Javanese, as discussed
in §5.1. Alternatively, infixation of one daughter into the other at the mother
node level may obscure phonological alternations conditioned within one of
the daughters. Mother-based opacity of this kind is discussed in §5.2.1 and in
the case studies in Chapter 6.
As we will show in a number of case studies presented in this chapter –
Javanese, in §5.1, Eastern Kadazan and Chamorro, in §5.2, and Fox, in §5.5,
among others – the cophonologies associated with reduplication in MDT are
sufficient to account for opacity effects in reduplication. Reduplication-specific
opacity-inducing mechanisms like Wilbur’s Identity Constraint, or McCarthy
and Prince’s (1995a) BR correspondence relations, which have been invoked
in phonological copying theories, turn out to be unnecessary once the role of
morphology in reduplication is properly exploited. BR identity is not enhanced
by all cases of reduplicative opacity; BR identity is therefore not a sufficient
explanation for these effects. Further, identity-based theories of opacity make
incorrect predictions about what kinds of reduplicative opacity are possible, as
argued in §5.4.
Our claim that cophonologies and layering drive the opacity effects that have
attracted attention in reduplication connects with the strong claim of Kiparsky
2000, according to whom all opacity effects, not just those in reduplication,
arise through stratal ordering: a process on stratum n obliterates the environment
which transparently conditions a process applying on stratum n-i. While we do
not agree with Kiparsky that all phonological opacity has a stratal (layering)
source, we do agree that much of the opacity found in reduplication is of this
kind.

5.1 Daughter-based opacity: overapplication and underapplication

in Javanese
Javanese is rife with instances of phonological opacity in reduplication, some of
which have featured prominently in the theoretical literature (e.g. McCarthy &
 Daughter-based opacity in Javanese 137

Prince 1995a). Opacity in Javanese is daughter-internal, arising when truncation

deletes input phonological structure that is crucial to determining the applica-
bility of an alternation whose effects are opaquely present – or in some cases
opaquely absent – on the material that survives truncation. Our understanding
of Javanese is based on the following sources: Horne 1961 (H); Sumukti 1971
(S); Dudas 1976 (D); see also Uhlenbeck 1953; 1954. Many parallel effects
occur in Johore Malay (Onn 1976).

5.1.1 /a/-raising: underapplication by truncation

Javanese has a total reduplication construction applying to adjectives and verbs;
the construction is characterized by Dudas (1976) as indicating “various sorts of
plurality – either of objects or actions – although there are certain cases where
it is used in conjunction with a specific affix with a slightly different semantic
import” (p. 202). According to Horne (1961), “Doubled adjectives . . . have a
plural meaning: either they modify more than one noun, or they modify a plural
noun . . . Some (not all) active verbs may be doubled . . . In general, the meaning
conveyed by a doubled verb is a less purposeful action than for the single verb;
perhaps it is performed without serious aim, or lazily. Or, it may be done over
an extended period of time, or repeatedly” (pp. 221–22). The examples below
are taken from Horne 1961:222–23. Because transcription systems for Javanese
vowels vary across sources, we have converted all transcriptions of vowels to
IPA here:1
(2) kirɔ ‘think’ kirɔ-kirɔ ‘about, approximately’
takɔn ‘ask’ takɔn-takɔn ‘keep asking’
wɔtjɔ ‘read’ di-wɔtjɔ-wɔtjɔ passive of ‘read without concentration’
turu ‘sleep’ turu-turu ‘fall asleep’

The reason Javanese has attracted interest in the reduplication literature is

illustrated by the /a/ ∼ /ɔ/ alternation in (3): /a/ surfaces before suffixes, as
shown in (3b), but raises to /ɔ/ word-finally (3a) and in both copies in bare stem
reduplication (3c). What has attracted attention are the data in (3d): when a
stem is both reduplicated and suffixed, both of its copies end in /a/, despite the
fact that the first is unsuffixed. Because /ɔ/ is expected when no suffix follows,
/a/-raising can be said to be underapplying in the first copy of the verb roots
in (d):2

(3) a. medjɔ ‘table’ H70, D206

b. medja-ku ‘my table’ H70
medja-mu ‘your table’ H70
medja-ne ‘his/her table’ H70
c. medjɔ-medjɔ ‘tables’ D206, H70
138 Morphologically driven opacity in reduplication

d. medja-medja-ku ‘my tables’ H70

medja-medja-mu ‘your tables’ H70
medja-medja-ne ‘his/her tables’ H70, D206

In a BR-identity approach (McCarthy & Prince 1995a), the opacity in (3d)

could be attributed to the requirement that the final vowels in reduplicant and
base must be identical, whether constrained in the second copy to be /ɔ/ (when
final) or /a/ (when suffixed).
Our perspective on these data is entirely different. Supported by evidence
from a number of suffixed reduplicated stems, we argue that /a/ raising is
transparently conditioned within each daughter of reduplication. In (3c), the
daughters are both bare stems; in (3d), they are both suffixed stems. The con-
ditions inhibiting raising are rendered opaque in the first daughter of (3d) by
truncation, which deletes the suffixal material entirely.
The two analyses are compared below:

(4) Opacity by BR identity Opacity by truncation

medja ↔ medja-ne medja-medja-ne

[medja] [medja-ne]

/RED-medja-ne/ /medja-ne/ /medja-ne/

Although the data in (3) are neutral between the two analyses in (4), abundant
evidence within Javanese supports the opacity-by-truncation analysis.

5.1.2 Suffix-triggered ablaut: overapplication by truncation

Initial support for holding truncation responsible for opacity effects in Javanese
reduplication comes from the co-occurrence of reduplication with causative and
locative suffixes. Javanese has three causative suffixes: the simple Causative
/-(q)akε/, the Causative Imperative /-(q)nɔ/, and the Causative Subjunctive
/-(q)nε/. There is a similar array of locative suffixes: the simple Locative /-(n)i/,
the Locative Imperative /-(n)ɔnɔ/, and the Locative Subjunctive /-(n)ane/. The
parenthesized consonants surface when the preceding root is vowel-final; recall
that “q” represents [ʔ]. Although our sources (Horne 1961; Sumukti 1971;
Dudas 1976) do not do this, it seems reasonable to treat /-(q)-/ and /-(n)-/ as
general causative and locative stem formers, respectively; these consonantal
suffixes, or their zero allomorphs, create bound stems to which one of the three
available extensions must attach.
 Daughter-based opacity in Javanese 139

These consonantal causative and locative stem-forming suffixes trigger par-

ticular specific effects on the preceding stem vowel (Horne 1961:176, 208),
which we will refer to by the cover term “Ablaut.” Causative /-(q)-/ lowers /ɔ/
to /a/, /u/ and /o/ to /ɔ/, and /i/ and /e/ to /ε/; the Locative /-(n)-/ triggers these
same effects as well as tensing /i/ and /υ/ to /i/ and /u/. Illustrative examples
are provided below. (Note that all of the verbs are cited in their full form and
bear either the ng- active prefix or the di- passive prefix. On the ng- prefix in
reduplication, see §5.1.4.)

(5) Plain stem Gloss of root Causative Locative Alternation

bisɔ ‘can, be able’ di-bisa-q-ake di-bisa-n-i ɔ→a H209, 178

laku ‘walk’ nge-lakɔ-q-ake nge-lakɔ-n-i u→ɔ H212, 179
(a)nggo ‘use, wear’ ng-anggɔ-q-ake ng-anggɔ-n-i o→ɔ H213, 180
bali ‘return’ m-balε-q-ake m-balε-n-i i→ε H209, 177
gawe ‘make’ di-gawε-q-ake ng-gawε-n-i e→ε H208, 178
apeq ‘good, nice’ ng-apeq-ake ng-apiq-i eC → iC H209, 177

When causative and locative stems are reduplicated, identical stem vowels
occur in the two copies, parallel to the phenomenon manifested in the suffixed
reduplicated stems in §5.1.1. In the case of causative and locative suffixation
there is positive evidence pointing to the source of this effect. As shown in
(6), the consonantal causative and locative stem formers surface on both copies
when the root in question is vowel-final (see, for example, Sumukti 1971:97,
Horne 1961:223, Dudas 1976:31). Double underlining identifies those vowels
and consonants in the reduplicant for whose quality (or existence) the suffix is
responsible:

(6) Root Suffixed stem Reduplicated suffixed stem

a. ambu ‘odor’ ng-ambɔ-q-akε (caus) ngambɔq-ngambɔqake D31

ng-ambɔ-n-i (loc) ngambɔn-ngambɔni D31
b. bali ‘return’ m-balε-q-ake (caus) mbalεq-mbalεqake D31
m-balε-n-i (loc) mbalεn-mbalεni D31
c. djərɔ ‘deep’ n-djərɔ-q-ake (caus) ndjərɔq-ndjərɔ-qake D31
n-djərɔ-n-i (loc) ndjərɔn-ndjərɔ-ni D31
d. uni ‘sound’ ng-unε-q-ake (caus) unεq-unε-q(a)ke H216, S97
‘to berate’
e. tibɔ ‘to fall’ tiba-q-(a)ke (caus) tibaq-tiba-q(a)ke S97
‘repeatedly drop’
140 Morphologically driven opacity in reduplication

Why do only the causative and locative stem-forming consonants, rather than
the entire causative and locative suffix complexes, surface on the first copy?
One possibility is morphological: perhaps only the causative and locative stem
formers are present in the input to reduplication, with the extensions being added
to the entire reduplicated stem. The other hypothesis, which we adopt here, is
phonological: the reduplicant, i.e. the output of Cophonology X, is subject to
a restriction that its syllable count cannot exceed that of the input root. We
formalize this by positing a constituent, the Proot (Prt), which contains all of
the syllables of the morphological root (Rt), plus any surrounding consonants
which can syllabify into the Proot without increasing its syllable count (see
(8) for formalization). Cophonology X truncates its input down to the Proot,
which consists of the material in the morphological root plus any incorporated
affixal consonants. In the schema below, representing the derivation of ‘berate’
in (6d), each daughter takes as input the Causative stem {uneq}ake, based on
the input /uni-q-ake/. Cophonology X truncates its input down to the Proot,
{uneq}; Cophonology Y is faithful.

(7) Causative stem formation: {uneq}ake Ablaut triggered by -q-

/uniRt-q-ake/

Reduplication of causative stem: {unεq}{unεq}ake

{unεq} {unεq}ake
Cophon. X: truncation to Proot ⇒ | | ⇐ Cophon. Y: Identity
{uneq}ake {uneq}ake

Ablaut, triggered by the Causative stem former /-(q)-/, applies on the

causative stem cycle; its effects are thus present in the input to the first daugh-
ter of reduplication. Because truncation removes most of the triggering suffix
material, the ablaut is nearly opaque on the surface.3
Constituents like the Proot, based on but not identical to a morphological
constituent, have a long history in phonological analysis; see, for example,
Booij 1984; Nespor & Vogel 1986; Sproat 1986; Inkelas 1990; Booij & Lieber
1993, among many others. The role of P-constituents in reduplication has been
explored by Cole (1994) and Downing (1998c; 1998d) (see also Pykllänen
1999), who document a number of onset overcopying cases in which a segment
from an adjacent morpheme is incorporated (or “conscripted,” to use Cole’s
term) into a following Proot, often to satisfy onset or minimality requirements
 Daughter-based opacity in Javanese 141

on the root. Our analysis of Javanese is firmly rooted within this tradition of
mismatches between morphological and prosodic constituents.
The Root-Proot (RP) correspondence constraints involved in truncating the
output to a Proot which is based on the input morphological Root but which
incorporates an adjacent non-Root consonant are given below:

(8) RP-Max-seg: All root segments should have correspondents in the Proot
RP-Max-syl: All root syllables should have correspondents in the Proot
RP-Dep-syl: All Proot syllables should have correspondents in the root
RP-Dep-seg: All Proot segments should have correspondents in the root
IP-Max-seg: All input segments should have correspondents in the Proot

In Javanese, RP-Max constraints are always satisfied; the Proot always includes
the Mroot. Of interest is the potential for the Proot to contain affix material as
well. The ranking illustrated in (9), RP-Dep-syl » IP-Max-seg » RP-Dep-seg,
results in inclusion of non-root input segments in the Proot when doing so does
not add syllables to the Proot which lack correspondents in the root:

(9) uni-q-ake RP-Dep-syl IP-Max-seg RP-Dep-seg

a. {unε}qake qake!

b. {unεqake} qa!, ke qake

) c. {unεq}ake ake q

d. {u}nεqake niqake

Semantic evidence is consistent with this analysis of reduplicated causative

and locative roots. The meaning of verb reduplication is inflectional, conveying
randomness or casualness. Causative and locative affixation are clearly deriva-
tional. Thus even without any phonological evidence we would automatically
expect the causative or locative to be present in the input to reduplication, as
the MDT analysis claims, rather than the opposite.
Further evidence that suffixes whose effects are felt in both copies are in
fact present in both inputs to reduplication, if not necessarily in both outputs,
comes from the nominalizing suffix -/(a)n/, which surfaces as /-an/ follow-
ing consonant-final stems and as /-n/ following vowel-final stems.4 Like the
causative and locative suffixes, the nominalizer triggers its own pattern of stem
vowel ablaut:
142 Morphologically driven opacity in reduplication

(10) a. tules ‘write’ tulis-an ‘handwriting, script’ H126

raop ‘wash face’ raup-an ‘water in which face H126
has been washed’
tutop ‘close’ tutup-an ‘a closed house with H126
nobody at home’
b. tuku ‘buy’ tukɔ-n ‘money for buying H126
something/things bought’
wɔtjɔ ‘read’ watja-n ‘reading matter’ H126
tampi ‘receive’ tampε-n ‘amount of money H136
received’

When vowel-final stems, i.e. those taking the /-n/ allomorph of the nomi-
nalizer, are reduplicated, the suffixal /-n/ surfaces, as expected under the MDT
analysis, in both copies; not surprisingly, the ablaut patterns triggered by /-n/
appear in both copies as well:5

(11) /uni/ ‘sound’ unε-n+unε-n ‘noise, saying’ S98

/aŋgo/ ‘to wear’ aŋgɔ-n+aŋgɔ-n ‘jewelry’ S98
/ombe/ ‘to drink’ ombε-n+ombε-n ‘beverage’ S98

When, however, the root is consonant-final, /-an/ surfaces only following the
second copy. As predicted in the MDT account, the ablaut effects associated
with the nominalizer are nonetheless found on both copies:

(12) apeq ‘good, nice’ apiq+apiq-an ‘something nice’ H127

εntoq ‘receive, get’ εntuq+εntuq-an ‘use, benefit’ H127
d.on ‘descend; get off’ d.un+d.un-an ‘things which have been H127
taken off or unloaded’

The analysis already developed for reduplicated causative and locative roots
extends straightforwardly to the nominalizer. Cophonology X truncates the suf-
fixed, nominalized stem input down to its Proot, which consists of the input mor-
phological root plus any adjacent material that can syllabify into the Proot with-
out increasing its syllable count. This analysis correctly predicts that only the /n/
allomorph of the nominalizer will survive truncation; vowel ablaut effects, how-
ever, will surface in the output of Cophonology X regardless of whether suffixal
material does. Example (13) shows nominalizer stem formation for vowel-final
and consonant-final roots, and (14) illustrates the fate of Nominalized stems in
reduplication:

(13) {unεn} {apiq}an

/\ /\ Cophonology ablauts final stem vowel
/uniRt -an/ /apeqRt -an/
 Daughter-based opacity in Javanese 143

(14) {unεn}{unεn} {apiq}{apiqan}

{unεn} {unεn} {apiq} {apiq}an

| | | |
{unεn} Nml {unεn} Nml {apiq}anNml {apiq}anNml

The MDT analysis in which reduplication operates on nominalized stems

is consistent with cross-linguistic expectations about the relative ordering of
morphological constructions in the same word; as a derivational process, nom-
inalization is expected to occur inside of an inflectional process like Javanese
reduplication.6

5.1.3 Opacity in suffixation and reduplication: wrapup

Our discussion of opacity in suffixed Javanese reduplicated stems began, in
§5.1.1, with forms like medjɔ-medjɔ ‘tables’ and medja-medja-ku ‘my tables’;
we proposed there that the opaque distribution of /a/ and /ɔ/ in the output of
reduplication is due to truncation. The daughters of the construction whose
output is medjɔ-medjɔ are both [medjɔ]; the daughters of medja-medja-ku are
both, on the MDT account, [medja-ku]. /a/-raising applies (or fails to apply)
transparently in the daughters in both cases; its failure to apply in the first stem
copy in medja-medja-ku is obscured by the fact that the suffix -ku is not part
of the Proot and therefore does not survive truncation in the output of the first
copy, as shown below:

(15) {medja} Output truncated to Proot

{medja}ku /a/ raising not conditioned

/medja Rt -ku/

The question arises at this point as to why the k of -ku cannot syllabify into
the Proot, yielding *medjak-medjaku, in the same manner as the -q Causative
stem former (as in (6)) or the -n allomorph of the Nominalizer (as in (11)).
We hypothesize that a principle of Morpheme Integrity, along the lines of that
proposed for Kinande by Mutaka and Hyman 1990, prevents the partial pars-
ing of Javanese affixes into the Proot. Either the whole suffix (if consonantal)
incorporates into the Proot, or none of it does. What distinguishes the Causative
-q and Nominalizer -n suffix allomorphs is that they are monoconsonantal, able
to syllabify in their entirety into a preceding vowel-final Proot.
144 Morphologically driven opacity in reduplication

5.1.4 Active prefix

The account developed thus far, which attributes opacity in suffixed reduplicated
stems to truncation of the first stem copy to its Proot, provides insight into an
interaction between prefixation and reduplication in Javanese that has been cited
in the literature as support for BR Identity. The prefix in question is the Active
prefix, one of two prefixes (the other being Passive -di) found on most transitive
verbs (see, for example, Dudas 1976:14).
As shown in (16), the Active prefix, typical of nasal-final prefixes through-
out Austronesian, fuses phonologically with the following root under certain
phonological conditions, on which see Dudas 1976:14–15 and Horne 1961:103.
If the root is vowel-initial, the Active prefix surfaces as [ŋ] (ng) or in some cases
[m] (Horne 1961:103). The Active prefix surfaces as prenasalization on a fol-
lowing voiced obstruent (16b); it fuses completely with a following voiceless
obstruent or sonorant, resulting in a fully nasal consonant (16c).7

(16) a. adjar ‘learn’ ngadjar D15

uleh ‘go home’ m-uleh H103
b. bajar ‘pay, salary’ m-bajar D15
golεq ‘get, seek’ ng-golεq D15
c. kεləm ‘become submerged’ ngεləm D15
tulis ‘write’ nulis D15
sijυng ‘canine tooth’ njijυng D15
wɔtjɔ ‘read’ mɔtjɔ D15

The behavior of the Active prefix in reduplication has attracted attention in

the literature (e.g. McCarthy & Prince 1995a) because of its propensity to occur
on both copies. According to Dudas, when the verb root begins with the type
of consonant with which the Active prefix does not fuse segmentally, the prefix
can appear either on both copies or only on the first (17a, b). However, when
the verb root begins with the kind of consonant with which the Active prefix
does fuse, the prefix must be doubled (17c):

(17) a. idaq ‘to step’ ŋidaq-idaq ‘keep stepping’ S89

b. baliq ‘turn’ mbaliq-mbaliq-ake ∼ mbaliq-baliq-ake D30
c. tulis ‘write’ nules-nules ‘write aimlessly, without H222
accomplishing much’
nulis-nulis-ake (*nulis-tulis-ake) D30

Horne, who does not mention the possibility of phonological variation of

the kind Dudas describes, presents the facts somewhat differently, stating that
reduplication can either precede or follow Active prefixation (1976: 222–23).
Horne offers forms like the following to illustrate the two options: First
 Daughter-based opacity in Javanese 145

prefix, then reduplicate: tules ‘write’ → nules → nules-nules, or first redu-

plicate and then prefix: isen ‘shy’ → isen-isen → ng-isen-isen. The five exam-
ples Horne cites (1976:223) of this latter type begin with vowels or /l/, falling
into the class of root types for which Dudas says Active prefix doubling is
optional.
Unless a clear semantic difference can be found between the two construction
types Horne describes (and see Uhlenbeck 1954 for suggestive discussion), we
assume Dudas is correct in treating the difference as essentially phonological;
however, the issue clearly requires further investigation by those better versed
than we are in Javanese morphology. It should also be borne in mind that,
according to Onn (1976), verb reduplication in Johore Malay patterns almost
identically with Dudas’s characterization of Javanese.
Assuming, with Dudas, that the distribution of the Active prefix is phono-
logically determined, the issue becomes one of under what conditions the nasal
prefix can be recruited into the Proot. Both Downing (1998c; 1998d) and Cole
(1994) have proposed to handle effects similar to these in other languages by
conscripting a preceding consonant into a following Pword.8 A prefix which
has fused phonologically with the stem-initial consonant must be conscripted:
/ng-tulis/ → {nulis}. Prefixes which retain their segmental integrity appear, on
Dudas’s description, to have the option of incorporating or not (e.g. m-baliq →
m{baliq} ∼ {mbaliq}).

5.1.5 /h/-deletion: overapplication

/h/-deletion is the third alternation type (in addition to vowel changes and
Active prefix nasal fusion) contributing to reduplicative opacity in Javanese;
like Active prefix nasal fusion, it has been invoked as support for BR corre-
spondence (McCarthy & Prince 1995a:285–87). However, BR correspondence
is not needed in the analysis of /h/-deletion, whose behavior follows straight-
forwardly from the MDT analysis we have already developed.
Intervocalic /h/- deletion is triggered by certain, though not all, vowel-initial
suffixes; (18) illustrates its application before Locative, Demonstrative, and
Hortatory suffixes:

(18) a. Locative Stem

ng-adɔ-i adɔh ‘far’ D164

ng-akε-i akεh ‘much, many’ D164
m-ili-i pilih ‘choose’ D164
nj-ala-i salah ‘mistaken’ D164
146 Morphologically driven opacity in reduplication

b. Demonstrative

ane-e aneh ‘strange’ D133

kɔkɔ-e (> kɔkɔwe) kokoh ‘rice mixed into soup’ D133
c. Hortatory

sarε-ɔ sarεh ‘be patient!’ S92

/h/-deletion is suffix-specific; as illustrated below, for example, vowel-initial

allomorphs of causative suffixes do not trigger /h/-deletion:9

(19) Causative Stem

ng-adɔh-ake adɔh ‘far’ D164

ng-akεh-ake akεh ‘much, many’ D164
m-ilih-ake pilih ‘choose’ D164
nj-alah-ake salah ‘mistaken’ D164

Because /h/-deletion is suffix-specific, it must be accomplished by the

cophonology associated with the triggering suffixes. (We will assume that the
relevant suffixes all trigger an /h/-deleting cophonology.) The following exam-
ple illustrates Demonstrative stem formation:

(20) {bəd

. a}e Cophonology deletes intervocalic /h/

bəd.ahRI -e/

What has drawn attention in the literature is the following: In reduplicated

suffixed /h/-final roots, /h/-deletion opaquely overapplies in the first copy when
a vowel-initial suffix follows the second copy (and causes deletion there):

(21) Reduplication Suffix + reduplication

a. bəd.ah-bəd.ah bəd.a-bəd.a-e cf. bəd.ah ‘broken’ D208

b. d.ajɔh-d.ajɔh d.ajɔ-d.ajɔ-e cf. d.ajɔh10 ‘guest’ D209
c. təngah-təngah tənga-tənga-e H71
‘middle’ ‘the middle’
d. kukυh-kukυh kuku-kuku-w-e cf. kukυh ‘solid’ D209
e. njala-njala-i cf. salah ‘make a D154–155
mistake’
f. oma-oma-an cf. omah ‘house’ S92
‘to play house’
g. saq-adɔ-adɔ-e cf. adɔh ‘far’ S92
‘as far as possible’
h. pili-pili-y-an cf. pilih ‘choose’ D154

The MDT analysis developed for Javanese reduplication handles these cases
straightforwardly. The suffix triggering /h/-deletion is present in the input to
 Daughter-based opacity in Javanese 147

both copies in reduplication; the first copy undergoes truncation to its Proot,
preserving the effects of /h/-deletion but obscuring its suffixal trigger. This is
illustrated below for the reduplicated demonstrative form of ‘broken’ (21a). The
inputs to reduplication are both {bəd.a}e, itself the output of the Demonstrative
stem-forming construction whose input is /bəd.ah-e/:

(22) bəda-b
. ədae
.

bəda . bədae
. Daughters of reduplication
| |
{bəda}e
. {bəda}e
. Demonstrative stems
| |
/bədah-e/
. /bədah-e/
.

As with /a/-raising and nasal prefix fusion, /h/-deletion is opaque in the first copy
because the environment which conditions its application within the suffixed
stem is obscured by morphological truncation in reduplication.

5.1.6 Laxing: underapplication

Thus far in our discussion of Javanese we have seen several examples of overap-
plication and one example of underapplication (/a/-raising). A significant plank
of support for the MDT analysis of opacity is that it predicts whether a given
alternation which is transparent in the daughters of the reduplication construc-
tion will, as a result of reduplicative truncation, appear to exhibit overappli-
cation vs. underapplication. If truncation obscures a trigger for the alternation
(as in /h/-deletion), the result will appear to be overapplication; if truncation
removes the environment inhibiting the alternation (as in /a/-raising), the result
will appear to be underapplication. By contrast, in BRCT whether a given alter-
nation overapplies or underapplies is an arbitrary effect of constraint ranking,
not predictable from the nature of the alternation itself. We illustrate this point
with our final case study from Javanese, a pattern of alternations in suffixed
stems that affects both vowels and consonants.
As shown in (23), High Vowel Laxing laxes high tense vowels in stem-final
closed syllables (Dudas 1976:55ff.).

(23) UR Plain Demonstrative Gloss

/apiq/ apiq apiq-e ‘good (-demonstrative)’

/kluwung/ kluwυng kluwung-e ‘rainbow (-demonstrative)’
/djupuq/ djupυq n-djupuq-ɔ ‘go get (-hortatory)’
/tulis/ tulis nulis-ɔ ‘write (-hortatory)’
/wiwit/ wiwit wiwit-an ‘beginning (-substantive)’
148 Morphologically driven opacity in reduplication

Consonants alternate as well. As described by Dudas (1976:118–19), certain

stem-final consonants alternate between “heavy” (voiced) and “light” (voice-
less): they are voiceless word-finally or before a C-initial suffix, but voiced
before V-initial suffixes, e.g. the Demonstrative, illustrated below:

(24) Plain Demonstrative

gaip gaib-e ‘secret’ D118
səbap səbab-e ‘because’ D118
bləbət bləbəd-e ‘protective cover’ D118
djogεt djogεd-e ‘the classical Javanese dance’ D118

Dudas terms this pattern Consonant Neutralization; we will refer to it instead

as Devoicing. Both High Vowel Laxing and Consonant Devoicing underapply in
reduplicated suffixed stems. Example (25) illustrates the failure of High Vowel
Laxing to apply to the first daughter of a reduplicated suffixed (Demonstrative)
stem; by contrast, it applies transparently in both daughters of a reduplicated
unsuffixed stem:

(25) a. d.ud.υq ‘place’ d.ud.uq-d.ud.uq-e d.ud.υq-d.ud.υq D207

b. abυr ‘flight’ abur-abur-e abυr-abυr D207
c. apiq ‘good, nice’ apiq-apiq-e apiq-apiq D207
d. gilik ‘cylindrical’ gilig-gilig-e gilik-gilik D207
e. murit ‘student’ murid-murid-e murit-murit D207

Similar behavior is exhibited by the stem-final consonants (25d, e); they

transparently devoice in both copies of bare reduplicated stems but opaquely
retain their underlying voicing in the first copy of a suffixed reduplicated
stem.
Both of these underapplication effects follow directly from the MDT analysis
developed thus far. Laxing and Consonant Devoicing apply transparently on the
stem cycle; their conditioning environments are obscured by suffix truncation in
the first copy of reduplicated suffixed stems. The Demonstrative stem-forming
construction generating (25e) is shown below:

(26) {murid}-e


/muritRt -e/

Neither Laxing nor Consonant Devoicing are conditioned in this stem; both
transparently fail to apply. The appearance of underapplication in reduplication
is a result of two factors: (a) the truncation, in Cophonology X, of the suffix
 Daughter-based opacity in Javanese 149

which, on the stem cycle, renders the stem-final syllable open, permitting the
vowel to remain tense and the consonant to remain voiced, and (b) the fact that
Laxing and Devoicing do not reapply in Cophonology Z.

(27) murid-muride

murid muride
| |
{murid}e {murid}e

MDT does not predict that Laxing and Consonant Devoicing have to be
opaque in the output of reduplicated suffixed stems. It would be possible, for
example, within MDT for Laxing and Consonant Devoicing to apply again in
Cophonology X, producing an output in which both copies – the first, trun-
cated copy, and the second, intact copy – exhibit transparent phonology: murit-
murid-e. The fact that they do not apply cyclically is an arbitrary fact about
Javanese. What MDT does clearly predict, however, is that the opacity within
each daughter is an effect of daughter-internal phonology; this is a prediction
in which MDT differs sharply from BRCT or other Coerced Identity theories.
In a BRCT account of Laxing, underapplication in murid-muride would follow
from the fact that Laxing would disrupt BR identity if it applied in the first copy
(where it is conditioned); therefore it is coerced to apply in neither. The problem
is that identity would be maintained equally well if Laxing overapplied, yielding
murid-muride with transparent application in the first copy by opaque applica-
tion in the second copy. In BRCT what determines whether underapplication
or overapplication of Laxing will be the outcome is relative markedness. If the
constraint against a tense vowel in a closed syllable (*Tense/Closed) outranks
the ban on a lax vowel in an open syllable (*Lax/Open), Laxing will overapply;
if the ranking is reversed, Laxing will underapply. In MDT, by contrast, there
is no means of deriving Laxing overapplication. Laxing is conditioned neither
in the input nor in the output of the second copy in forms like murid-muride.
If Laxing were a word-level alternation, rather than a stem-level alternation,
MDT could describe the situation in which Laxing applied in the first copy but
not in the second (the hypothetical murit-muride), but it could never generate
murit-muride.
In summary, in every case of opacity we have surveyed in Javanese, the
alternation is transparent in its domain of application (the stem); the opacity is
generated when the stem in question is embedded in a reduplication construction
which truncates that part of the stem orginally determining the applicability
of the alternation. MDT not only describes this opacity effortlessly; it also
150 Morphologically driven opacity in reduplication

correctly predicts whether the opacity will take the form of underapplication or
overapplication.

5.1.7 Summary
In surveying opaque reduplicative phonology in Javanese, we have concluded
that BR identity does not play a role. If anything, reduplication obscures identity,
by virtue of the concomitant truncation. It is not the case in Javanese that
reduplicative phonology creates or preserves identity that would not otherwise
have existed.
In finding that reduplication operates on morphologically complex stems, we
concur with the original insight of Dudas (1976), who concludes (p. 209) that
reduplication is ordered after the phonological alternations in question.
A strong prediction that our analysis makes is that alternations internal to
the daughters of reduplication may be rendered opaque by reduplication, but
that word-level phonology, which applies to the output of reduplication, cannot
be. This prediction appears to be correct. Glide Insertion (Dudas 1976:135–36)
applies between two non-identical vowels when the first is tense and the second
is lower or fronter than the first. In the examples below, Glide Insertion breaks
up vowel hiatus resulting from h-deletion. Notice that it applies transparently,
not opaquely: only the second copy in reduplication, not the first, is followed
by an inserted glide:

(28) Reduplicated Suffixed

Stem stem reduplicated stem

a. gadjih gadjih-gadhih11 gadji-gadji-y-e ‘salary’ D209

b. kukυh kukυh-kukυh kuku-kuku-w-e ‘solid’ D209
c. pilih ‘choose’ pili-pili-y-an D154

If Glide Insertion is a word-level rule, MDT has an explanation for why it

applies only after the second copy, and not the first. The input to Glide Insertion
in, say, (28b) is the output of reduplication, i.e. kuku-kukue. There is only one
context for Glide Insertion, and it applies only once: kukukukuwe.
Our claim that the kind of opacity found in reduplication is simply the result
of processes at a higher cophonology masking the triggers of processes in lower
cophonologies resonates with the view of Kiparsky (1997:2):

Insofar as templatic morphology functions as an operation on phonological

representations (copying, truncation, etc.), it must show “overapplication”
of phonological processes: processes triggered at the relevant level by the
phonological context in the base will appear in the copied or truncated version,
provided that the constraints on the reduplicant itself permit it (whether or not
they are “opaque” there).
 Mother-based opacity: infixation 151

5.2 Mother-based opacity: infixation

In this section we look at two cases of opacity driven by interactions between
the mother and the daughter. Both involve infixation.

5.2.1 Chamorro
In Chamorro, opacity results from the interaction between stem-level stress
assignment and subsequent infixing reduplication (Topping 1973). Recall from
our discussion of Chamorro in Chapter 4 that stress is normally penultimate,
shifting rightward under suffixation:

(29) karéta ‘car’ 108

karetá-hu ‘my car’ 108
karetá-ta ‘our (incl.) car’ 108
kareta-n-mámi ‘our (excl.) car’ 108
kareta-n-ñı́ha ‘their car’ 108

This penultimate stress pattern is perturbed by both CV reduplication pat-

terns in Chamorro. Stressed CV reduplication, which copies the first CV of
the stressed syllable, is illustrated in (30a). Final CV reduplication repeats the
stem-final CV (30b).12 Both patterns result in antepenultimate stress, rendering
the predictable assignment of penultimate stress opaque.

(30) a. sága ‘stay’ sásaga ‘staying’ 259

hugándo ‘play’ hugágando ‘playing’ 259
hátsa ‘lifted’ háhatsa ‘one that was lifting’ 102
b. métgot ‘strong’ métgogot ‘very strong’ 183
bunı́ta ‘pretty’ bunı́tata ‘very pretty’ 183
dánkolo ‘big’ dánkololo ‘very big’ 183

The MDT analysis of Chamorro parallels that developed for Javanese. Penul-
timate stress assignment applies transparently in the stems which are input to
reduplication, as shown below:

(31) Stem cycle: hugándo

| ⇐ Stem cophonology: PENULT-STRESS » *STRUC-STRESS
/hugando/

Reduplication calls for two copies of each stem, the first of which is truncated
by Cophonology X to (in the case of stressed CV reduplication) the initial CV
portion of its stressed syllable. The second copy in reduplication, associated
with Cophonology Y, preserves the stem intact. Cophonology Z infixes the
output of Cophonology X into the output of Cophonology Y:
152 Morphologically driven opacity in reduplication

(32) hugá-ga-ndo

Cophonology X ⇒ ga hugándo ⇐ Cophonology Y

| |
hugándo Stem[F] hugándo Stem[F]

The location of stress in the output of Cophonology Z is opaque in not follow-

ing the regular penultimate stress rule. This type of opacity is so familiar that it
is hard to recognize, the distinction between affixes that occur within vs. outside
of the domain of stress assignment being so commonplace. Yet the obscuring
effect that subsequent affixation (here, infixation) can have on the conditioning
environment of a phonological alternation (here, stress assignment) plays an
important role in more complex examples of reduplicative opacity which we
analyze below.13

5.2.2 Eastern Kadazan

Opacity caused by infixation occurs in the reduplication system of Eastern
Kadazan, a Bornean language of Malaysia described by Hurlbut (1988) as
having “two or three times as many affixes . . . as, for example, in Tagalog”
(p. 1). The following example shows words containing both root reduplication
and affixes. To avoid prejudging the issue of where in the morphological layering
reduplication occurs, we list, on the left, the underlying representation of all
of the morphemes contained in the word (sans reduplication), with roots in
boldface, and on the right, the reduplicated words, with the duplicated sequence
in boldface. Infixes are set off by commas in the left-hand column but linearized
in the right-hand column. Data are based on Hurlbut 1988:14

(33)
‘dim-go/be like’ /soN-kaa/ sangkakaa 77
‘rec-move-I’ /ko-puu-ku/ kopupuuku 34
‘ass.col-ignore-someone-rf’ /pog-baya-an/ pogbabayaan 60
‘af-du.rec-aug-obey’ /m-pi-ku-bojo/ mikubobojo 107
‘des-lie down’ /si-kili/ sikikili 110
‘comp-n.inten-un.dist-spouse’ /N-o-piN-savo/ nopinsasavo 92
‘af-cont-lift’ /m-iN-kakat/ mingkakakat 105
‘af-mid-poss-go around’ /m-pog-ki-liput/ mogililiput 87
‘exas-arrive/reach’ /to-rikot/ toririkot 72
‘comp, mul.rec-be side by side’ /-in-, poi-ganding/ pinoigaganding 109
‘af, come’ /-um-, ravung/ rurumavung 102
‘catch an illness-uf-you’ /ruvang-o-ko/ ruruvangoko 112
 Mother-based opacity: infixation 153

‘comp, af-dur-to sun’ /-in-, m-mogin-sidang/ minoginsisidang 105

‘compr-pock mark’ /so-gorontok/ sogogorontok 49, 106

The reason for reserving judgment as to the hierarchical position of redu-

plication in the word is the behavior of vowel-initial roots (34). These show
doubling of the initial V of the root as well as the preceding, prefix-final C:

(34)
‘af-du.rec-n.ser-paddle a boat’ /m-pi-siN-alud/ misingangalud 107
‘comp-n.inten-cont-eat’ /no-ko-iN-aran/ nokoingangaran 80
‘af-diss-sleep’ /m-posiN-odop/ mosingongodop 60
‘dim-hold in hand’ /soN-onggom/ songongonggom 77

The MDT account of the data in (33) and (34) is that Eastern Kadazan
reduplication is infixing. As in Javanese, the Proot plays a crucial role. In Eastern
Kadazan, reduplication, a late morphological process, truncates the first copy
of the word down to the initial CV of the Proot and infixes the result directly
preceding the Proot in the second copy. The Proot, as in Javanese, consists of
those syllables projected from the morphological root. In the case of consonant-
initial roots, as in (33), the morphological root and the Proot are identical, and
reduplication reproduces the initial syllable. In the case of vowel-initial roots,
as in (34), the Proot can include a consonant from a preceding prefix if the
morphological root is vowel-initial; in such a case, reduplication reproduces
the syllable composed of that consonant and the root-initial vowel.

(35) a. Consonant-initial root m-pi-ku-{bojo} → mpiku-bo-bojo

b. Vowel-initial root: m-posi{N-odop} → mposi-ngo-ngodop

There is independent evidence in Eastern Kadazan that reduplication is infix-

ing. Example (36) shows that when a vowel-initial root is not preceded by a
prefix, reduplication targets the initial CV portion of the root, skipping the initial
vowel:

(36) ‘run away-rf-I’ /idu-an-ku/ i+du+duanku VCV 103

‘wait-rf-n.comp.m’ /indad-an-po/ in+da+dadanpo VCCVC 103, 116

The Proot requires an initial onset, as modeled by the constraint ranking

Proot-Onset » Align-L(Mroot, Proot);this requirement can either
cause material from outside the Mroot to be pulled into the Proot, as in (34) or
(35b), or, as in (36), it can cause the initial vowel in the Mroot to be excluded
from the Proot. Proot formation for a prefixless vowel-initial root is modeled
below:
154 Morphologically driven opacity in reduplication

(37)
indad-an-po Proot-onset Align(Mroot,Proot)

) a. in{dad}anpo i

b. {indad}anpo *!

Cophonology X truncates the stem to the initial CV of the Proot, and

Cophonology Z positions the output of Cophonology X immediately preceding
the Proot in the output of Cophonology Y.
(38) in{da}{dad}anpo

{da} in{dad}anpo
| |
in{dad}anpo in{dad}anpo

Further support for the infixation analysis is evidence that, completely inde-
pendent of onset overcopying issues, the input to reduplication can be a mor-
phologically complex stem. This assumption is crucial to the claim that redu-
plication is a late process. The data in (39) show that the Proot can contain the
completive infix -in- (a) as well as what appear to be denominal verb stems
(b, c):
(39) a. -in- + vaal /v-in-aal/ vi+vinaal ‘It was made up’ 101
comp-make
b. in-sada /m-pog-iN- mo+gi+ginsadano ‘He catches fish 63
vbl-fish sada-no/ regularly’
c. to-ruol15 /m-posiN-to- mosin+to+toruol ‘She is pretending 60
?-? ruol/ to have an illness’

The infix -in- is one of the outermost affixes (Hurlbut 1988:23); its redu-
plicability confirms that reduplication is a late morphological process, looking
potentially deep into the word for the Proot.
In sum, what appears initially to be onset overcopying is actually internal
reduplication of a prosodic constituent (Proot), a phenomenon for which there
is strong precedent in the form of internal reduplication of stressed syllables (as
in Chamorro; see Yu 2003 for a recent survey). We have provided independent
evidence for the two crucial parts of the analysis: reduplication is infixing, and
reduplication operates on morphologically complex constituents.
The semantics of Eastern Kadazan reduplication are also consistent with
an analysis in which reduplication is a relatively late process. According to
Hurlbut (1988:100), reduplication is basically semantically iterative, though
 Mother-based opacity: infixation 155

its meaning can vary according to what other affixes are present in the word.
Together with the m-prefix, reduplication marks progressive iterative aspect;
together with completive aspect, reduplication gives the sense that the action
of the verb was undertaken without purpose (or, apparently, in bad faith). The
fact that the semantics of reduplication depends on the other affixes in the word
is consistent with its being a late morphological process.
Further support for this conclusion comes from two examples illustrating
whole-word reduplication (Hurlbut 1988). With what appears to be a purely
iterative meaning, reduplication targets all three syllables of the complex stems
mingguli in (40a) and pointaamon in (40b):

(40) a. tiap sodop mingguli-mingguli ka ino (102)

‘Every night that animal returned again and again’
b. pointaamon-pointaamon dii iri tapon (51)
‘And as he was fishing he kept throwing in the fishing-line’
c. /m-piN-guli/ AF-Un.Dist-return/repeat
/po-iN-taam-on/ DerivedState-Cont -throw up-UF

Clearly, the observed variation is more plausible if reduplication is a late

morphological process. If reduplication were the innermost morphological con-
struction, the expectation would be variation between partial and total redupli-
cation of the Proot, not the whole word; there would be no structural basis with
respect to which partial Proot reduplication and total Word reduplication would
form a natural morphological class.
Eastern Kadazan is just one of several languages in which infixation produces
an apparent overcopying effect. Two other languages for which such effects have
been of great interest to the theoretical literature are Chumash and Tagalog. They
are discussed in Chapter 6.

5.2.3 Infixation in MDT

Infixation in MDT must be accomplished by Cophonology Z, which is respon-
sible for collating the phonological outputs of its daughters. Infixation has been
the subject of much recent work. The kind of infixation seen in Chamorro
and Eastern Kadazan falls into the category of “positive prosodic circumscrip-
tion” (McCarthy & Prince 1990; Lombardi & McCarthy 1991; McCarthy &
Prince 1995b), alignment to a prosodic constituent (McCarthy & Prince 1994b),
or Phonological Subcategorization (Yu 2003). In Chamorro, the reduplicant,
which is truncated to the stressed syllable of its input, is positioned adjacent to
the stressed syllable of its sister; in Eastern Kadazan, the reduplicant, itself a
truncated Proot, is positioned adjacent to the Proot of its sister.
156 Morphologically driven opacity in reduplication

It is interesting that in both of these cases the reduplicant surfaces adjacent to

the portion of the base that it phonologically resembles. As far as we know this is
true generally of infixed reduplicants; see, for example, Tagalog and Chumash
(Chapter 6), and the over fifty cases of internal reduplication documented in
Yu 2003 (Appendix III).16 Adjacency is certainly not a general requirement in
reduplication; in languages as diverse as Chukchee (Krause 1980), Madurese
(Stevens 1985; McCarthy & Prince 1995a), and Umpila (Harris & O’Grady
1976; Levin 1985b), for example, the reduplicant is truncated to the initial por-
tion of the word being doubled, yet surfaces after, not before, the nontruncated
copy (thus, for example, from Chukchee, nute-nut ‘earth’ or, from Umpila,
maka-l-ma- ‘die, go out (fire)’).
It is beyond the scope of this study to determine why internal reduplication
is apparently more restricted in this way. One speculation, however, draw-
ing on work by Inkelas 1990, is that the generalization may follow from the
impossibility of discontinuous prosodic constituents. In a case like Eastern
Kadazan, where a doubled word has an internal Proot (e.g. in{dad}anpo),
an output in which the two Proots were not adjacent would violate continu-
ity: *{da}in{dad}anpo, cf. the actual output in{da}{dad}anpo (which perhaps
resolves to in{dadad}anpo). We leave this issue for future work on infixation
to illuminate.

5.3 Morphological opacity outside of reduplication

In containing no reduplication-specific elements, the MDT analysis of redu-
plicative opacity predicts parallel opacity effects outside of reduplication. It
has long been known that morphology–phonology layering, of the sort intrin-
sic to our cophonological construction model, is a potential source of opacity.
Indeed, Kiparsky (2000) has gone so far as to claim that all opacity results from
this type of layering. This claim is almost surely too strong, as there are clear
cases of opacity derived within a single cophonology (see, for example, the
case of Turkish velar deletion, discussed in Sprouse 1997). However, it is clear
that reduplication has no corner on the opacity market. We briefly examine two
cases here that resemble our reduplication examples in the type of opacity they
exhibit but do not involve reduplication themselves.

5.3.1 Opacity by truncation

Just as truncation in reduplication can render opaque otherwise transparent
phonology, we find truncation outside of reduplication producing the same
effects. A well-known example from the recent literature is English nickname
 Morphological opacity outside of reduplication 157

formation. As documented in Benua 1997, in a certain dialect of American

English, [æ] is prohibited before tautosyllabic /r/, though [æ] may precede /r/ if
it ends an open syllable.17 This distribution is violated in nickname formation;
names with heterosyllabic [ær] sequences can truncate to syllables ending in
[ær]: Larry [læri] → Lar [lær], etc. Benua treats the Larry-Lar relationship as
word-to-word (Output–Output) rather than input–output, though she does have
to stipulate a “Base-Identity” principle whereby Larry, as the morphological
“base,” is privileged over Lar. That is why markedness constraints are satisfied
in Larry but violated in Lar. We find it more plausible to set up, as does Weeda
1992 for parallel phenomena, an input–output relationship between Larry and
Lar such that the output of Larry [læri], a form obeying the distributional
restrictions on [æ], serves as input to truncation. The output of truncation, Lar
[lær], violates the ban on [. . . ær] syllables but is faithful to the input.

(41) Lar [lær]

⇐ truncation:[æ] opaque but faithful to input
Larry [læri]
⇐ [æ] transparently OK
/lAri/

In its details this analysis is almost exactly parallel to the analysis of Javanese
laxing. What can be thought of as an æ → ε rule seems to underapply in the
output of truncation, but that is simply because the output is faithful, in its vowel
quality, to the input, where the rule is transparently not conditioned. A similar
example involving place of articulation is Gram [græm]; for those speaker who
pronounce the word Grandma as [græmma], the place of articulation in the final
consonant of Gram can be explained only by assimilation to the following -ma
of Grandma, its morphological base. Numerous other examples can be found
in Weeda 1992; see especially pp. 78ff., where Weeda discusses cases of vowel
tensing in Madurese and New York English that closely parallel Benua’s data
and involve clearly productive alternations.

5.3.2 Opacity by infixation

Just as it does in reduplication, infixation can induce opacity outside of redu-
plication. Sundanese is a well-known case in which morphological infixation
(Robins 1959:368) renders an otherwise transparent phonological alternation
opaque (Cohn 1990; Benua 1997). Sundanese is subject to rightward nasal
harmony, triggered by a nasal consonant and affecting following vocoids. It is
blocked by supralaryngeal consonants. The singular forms in (42a) show nasal
harmony applying transparently. In the plural forms in (42b), the plural infix
158 Morphologically driven opacity in reduplication

-al- ∼ -ar-, which follows the first consonant, renders nasal harmony opaque.
Harmony appears to overapply to the right of the infix, where it is not condi-
tioned on the surface; the final liquid of the infix is the type of consonant that
blocks harmony.

(42) Singular Plural Gloss

a. ŋãũr ŋãlãũr ‘say’

b. ŋãı̃ãn ŋãrãı̃ãn ‘wet’
c. mı̃ãsih mãrı̃ãsih ‘love’
d. ŋũliat ŋãrũliat ‘stretch’
e. mãrios mãlãrios ‘examine’
f. ŋãluhuran ŋãrãluhuran ‘be in a high position’
g. mãwur mãlãwur ‘spread’

Standard accounts of Sundanese nasal harmony appeal to the cycle (Cohn

1990) or its equivalent (Benua 1997; see also Anderson 1974:150–51). The
idea, in MDT terms, is that Nasal Harmony is part of the cophonology of
the stem that the plural infix combines with as well as the cophonology of the
stem that the plural infix produces. Nasal Harmony therefore applies to the
input as well as to the output of plural infixation, producing the attested surface
forms. The layered analysis of Sundanese plural infixation in (43) combines two
constructions: the singular stem-forming construction in which Nasal Harmony
applies to the singular root, and the plural stem-forming construction in which
Nasal Harmony applies, again, to the output of plural infixation.

(43) 
[ŋãruliat]
⇐ Plural stem cophonology: Nasal harmony

[ŋuliat]
| ⇐ Singular stem cophonology: Nasal harmony
/-al-/ /ŋuliat/

5.4 MDT vs. Coerced Identity theories

Even though its occurrence may be more common – or simply more noticeable –
in reduplication, opacity is clearly not a reduplication-specific effect. Nonethe-
less, the literature has seen the development of a number of theories designed
precisely to handle opacity (over- and underapplication) in reduplication, all
of which treat opacity as resulting from phonological identity requirements on
base and reduplicant. Wilbur (1973) proposed the Identity Principle, which has
been given new life in the notion of BR correspondence in Optimality Theory.
Both Mester (1986) and Clements (1985) propose representations specific to
 MDT vs. Coerced Identity theories 159

reduplication which have the effect of requiring the phonological outputs of

base and reduplicant to be the same; their insights, too, can be seen in the BR
correspondence proposal. Other reduplication-specific proposals for handling
reduplicative opacity include those in Raimy 2000 and Frampton 2003.
What all of these proposals have in common is the claim that the redupli-
cant is derived, by some phonological duplication mechanism, from the base,
and that either structural or correspondence relations keep the base and redu-
plicant from diverging too far. The underlying empirical premise is that redu-
plicative opacity keeps base and reduplicant more similar than they otherwise
would be. In this section we show this premise to be unfounded. Our discus-
sion will make reference to the most influential of the theories based upon it,
namely BR correspondence theory, which enforces surface identity between
reduplicant and base even in face of pressures to the contrary from markedness
constraints.
Consider, for example, the case of Javanese suffix ablaut, seen in §5.1.2, and
exemplified by the pair apeq ‘prayer,’ with underlying /e/, vs. apiq + apiq-an
(reduplicated, nominalized), with ablauted [i] in both copies. In BRCT, the iden-
tity in root vowels across base and reduplicant in apiq-apiq-an is achieved using
the ranking BR-Faith, Ablaut » IO-Faith, IR-Faith, as shown below:

(44)
red-apeq-an BR-Faith Ablaut IO-Faith IR-Faith

a. apeq-apeq-an *!

b. apeq-apiq-an *! *

) c. apiq-apiq-an * *

d. apiq-apeq-an *! * *

On the BRCT analysis, the opaque application of Ablaut in the reduplicant,

which in MDT we attributed to the transparent application of Ablaut in the input
suffixed stem, is attributed to its transparent application in the base, which is
followed by an overt triggering suffix. The ranking Ablaut » IO-Faith forces
ablaut to apply in the base; BR-Faith forces the base and reduplicant to be
identical, thus resulting in the observed “overapplication” effect.
Underapplication has a similar provenance in BRCT. Consider the example
of underapplication from Javanese, in which closed syllable laxing of /i/ to
/i/ opaquely fails to apply in the reduplicant of murid-murid-e, though it does
apply to the same root (“student”) in isolation and in unsuffixed reduplication:
160 Morphologically driven opacity in reduplication

(45) murit murit+ murit murid+murid-e

Transparent Transparent Opaque Transparent

application application underapplication nonapplication

The transparent failure of Laxing to apply in the base of reduplication,

/murid-e/, is attributed to the ranking *Tense/Closed syllable » *Lax,
IO-Faith, which requires laxing in closed syllables but prevents it from
applying in open syllables. The opaque failure of Laxing to apply in the redu-
plicant is due to BR-Ident, which is ranked above *Lax, the ban on lax
vowels.

(46)
Red-murid-e *Tense/Closed BR-Ident *Lax IO-Faith

) a. mu.rid-mu.ri.d-e
b. mu.rid.-mu.ri.d-e *! * *

c. mu.rid.-mu.ri.d-e *!* *

d. mu.rid-mu.ri.d-e *! *

BRCT and MDT might appear to work equally well for these Javanese exam-
ples of over- and underapplication – although, as argued in §5.5.3, MDT makes
stronger predictions than BRCT regarding which alternations will show over-
application and which will show underapplication.
We add to this argument in the next several sections of this chapter, showing
other ways in which MDT provides a better account of reduplicative opacity
than does BRCT in general. We make three points:

Not all reduplicative opacity increases BR identity; therefore BRCT
is not sufficient

BRCT predicts kinds of identity effects that don’t occur; thus BRCT
overgenerates

MDT can handle the apparent identity effects that do occur; thus BRCT
is not necessary

Opacity in MDT derives from morphological structure; MDT is therefore

more explanatory than theories of opacity which do not take internal hierarchical
structure into account.
 MDT vs. Coerced Identity theories 161

5.4.1 Opacity does not always increase identity

Although the work of Wilbur (1973) and McCarthy and Prince (1995a) has
highlighted those instances of reduplicative opacity in which BR identity is
increased, not all cases of opacity are of this kind. To take one example, the
Chamorro infixing reduplication analyzed in §5.2.1 creates opacity that has
nothing to do with increasing identity. Recall examples like the following:

(47) Plain form: hugándo

Reduplicated form: hugágando

What is opaque about the reduplicated form is the location of stress, which is
normally penultimate in Chamorro. But this “displacement” of stress does not
result in the base and reduplicant being more identical than they otherwise would
be; if stress were penultimate in the reduplicated form, giving us hugagándo,
base and reduplicant would be no less similar.
MDT can handle both the effects in Chamorro, in which identity clearly plays
no role, and those in Javanese, in which identity could be said to play a role, in
the same manner. By contrast, BRCT has no light to shed on Chamorro. BRCT
would presumably also have to appeal to cyclicity in some guise and would
therefore derive the opacity the way MDT does; BR correspondence would
play no role.

5.4.2 Distribution of opacity

The predictions of Morphological Doubling Theory and Identity theories are
very clearly distinguished in the area of junctures. MDT limits the scope of redu-
plicative opacity to mother–daughter relationships. The daughter cophonologies
can render the phonology of their inputs opaque; the mother cophonology can
render the phonology of its daughters opaque. Overapplication effects occur
when conditioning environments triggering an alternation in a lower cophonol-
ogy are obscured by some operation in a higher cophonology; underapplication
effects occur when an alternation that would apply in a lower cophonology
cannot because the conditioning environment is absent, even though a higher
cophonology provides it. We have seen examples of both kinds already in this
chapter. (Klamath, discussed in Chapter 4, is another instance of the latter type.)
What a cophonology cannot do is render opaque phonological effects condi-
tioned in morphological constituents over which the cophonology does not have
scope. The cophonology of one daughter in reduplication cannot influence the
outcome of the cophonology of the other daughter; similarly, no cophonology
in reduplication can influence the phonological outcome of morphemes outside
the morphological constituent produced by reduplication.
162 Morphologically driven opacity in reduplication

The effects of the cophonological structure of MDT can be summed up in

the following prediction:

(48) Scopal Opacity Prediction (SOP): cophonologies can render only the
phonology of their inputs opaque
Mother ← can opacify inputs

can opacify input → Daughter Daughter ← can opacify input

| |
Transparent phonology → input input ← Transparent phonology

The SOP does not follow from BRCT. As a result, BRCT makes dramatically
different predictions regarding phonological opacity in reduplication. Below we
survey three types of phonological alternations which in BRCT are predicted
to participate in over- or underapplication. We will show that only one of these
types of effects actually does participate, namely stem-internal phonology, pre-
cisely as MDT predicts.

5.4.2.1 Stem-internal phonology

Both MDT and BRCT predict the possibility that a phonological alternation
whose applicability is determined solely in input should be able to over- or
underapply in reduplication, as below:

(49) a. Overapplication (h-deletion)

/bəd.ah-e/ bəd.a-bəd.ae (vs. *bəd.ah-bəd.ae)
b. Underapplication (closed-syllable laxing)
/gilik-e/ gilig-gilige (vs. *gilik-gilige)

We have already seen how MDT and BRCT handle such cases; (a) is discussed
in the section on Javanese h-deletion (§5.1.5), and (b) is discussed in the section
on Javanese laxing (§5.1.6).

5.4.2.2 Internal junctural alternations

While BRCT and MDT both predict the possible opacity of over- or under-
application of daughter-internal (for BRCT, base-internal) phonology in redu-
plication, BRCT predicts additional types of opacity that MDT does not. The
first of these involves phonology applying across the base-reduplicant juncture.
Consider the hypothetical opaque internal junctural alternation below. “NPA”
stands for nasal place assimilation, the hypothetical alternation at hand:

(50) bihan → biham-biham

tarem → taren-taren
 MDT vs. Coerced Identity theories 163

At the internal base-reduplicant juncture, the nasal ending the first copy assim-
ilates in place to the consonant beginning the second copy; this alternation is
then reflected in the second copy.
MDT cannot describe this effect. There is no cophonology that could assim-
ilate the final segment of the second copy to the first segment of itself or of
the preceding copy. MDT predicts (50) not to occur. By contrast, BRCT can
easily describe it. Let us assume for the sake of exposition that in (50) the
second copy is the reduplicant. Letting “NPA” in (51) stand for the constraint(s)
responsible for requiring nasal-stop clusters to agree in place, the ranking NPA »
IO-Ident-place ensures that NPA will take place generally in the language,
and therefore in bases of reduplication as well. Satisfaction of IO-Ident-
place compels the reduplicant-final consonant to be identical to the base,
yielding overapplication of a BR-juncture effect:

..............................................
(51) /bihan-Red/ NPA IO-Ident BR-Faith

a. biham-bihan * *!

b. underapplication bihan-bihan *!

) c. overapplication biham-biham *

The BRCT analysis of this case is actually formally parallel to the BRCT
treatment of the actual opaque stem-internal Javanese alternations analyzed in
§5.4.2.1, above. The problem with this is that cases like bihan → biham-biham
do not exist. McCarthy and Prince (1995a), discussing parallel hypothetical
but unattested examples, are able to offer no formal explanation for the gap
(pp. 327–28).18

5.4.2.3 External junctural alternations

A third possible type of opacity, which MDT rules out but BRCT predicts pos-
sible, is exemplified by the hypothetical alternation below, in which consonant
assimilation, triggered by a suffix external to the reduplicated stem, is propa-
gated long distance to both members of the reduplicated stem to which the suffix
attaches. Assume, for the sake of argument, that the second copy is the base:

(52) a. Total reduplication tapan → tapan-tapan

b. Suffixation to reduplicated stem: tapan-tapan + -la → tapal-tapal-la

In (52b), total assimilation of the base-final nasal to the following suffix-

initial /l/ is reflected in the reduplicant, where it is not conditioned. If such effects
occur, MDT is in trouble and BRCT is supported, since only BR correspondence
164 Morphologically driven opacity in reduplication

could produce this result. BRCT can easily describe the hypothetical situation
in (52b):
(53)
/[Red-tapan]-la/ BR-Faith assimilation IO-Faith

a. transparent tapan-tapal-la *! *

) b. overapplication tapal-tapal-la *

c. underapplication tapan-tapan-la *!

To our knowledge, effects like these do not occur. However, the analysis
BRCT offers of these hypothetical data is formally completely parallel to the
opaque stem-internal alternations above, predicting that they should be equally
common. BRCT misses a major cross-linguistic generalization in generat-
ing, with equal facility, overapplication effects for phonology applying stem-
internally (which occurs often), phonology applying across the BR-juncture,
and phonology applying across junctures external to reduplication (neither of
which occur reliably).
The statement that cases like those in (50) and (52) do not occur does not
mean that there are no data which superficially resemble these patterns, as seen
already in the discussion of Javanese in §5.1. To take another example, the data
from Tawala (Austronesian; Ezard 1997:38), below, might initially appear to
instantiate overapplication of an external junctural effect (as in (52)). All of
these words exhibit CVCV reduplication, which marks durative aspect. Data
from the reduplication of long roots shows that the first copy is the truncated one
(e.g. geleta ‘to arrive,’ gele-geleta (durative); Ezard 1997:41). As seen below,
suffix-induced vowel change in the second copy is replicated in the first copy
as well.
(54) a. dewa-dewa-ya /dewa-dewa-ya/ ‘. . . -3sg’
b. dewe-dewe-ya /dewa-dewa-e-ya/ ‘. . . trv-3sg’
c. dewi-dewi-yai /dewa-dewa-iyai/ ‘. . . -1pl.excl’

Whether this qualifies as overapplication of an external junctural alternation

depends entirely on one’s morphological analysis. If what is being redupli-
cated is the bare root, then the double vowel-raising effects require expla-
nation; if what is being reduplicated is a stem consisting of root plus suf-
fixes, then the double vowel-raising effects are unremarkable. The input to
reduplication in (54b) is deweya, which reduplicates, as expected, as dewe-
deweya. There is reason to believe this latter analysis is correct for Tawala.
 Case study: Fox 165

The two suffix types whose effect on a preceding root vowel is replicated in
reduplication are the transitivizing suffix “trv” shown in (54b), and the six
object agreement suffixes, one of which is illustrated in (54c). As a derivational
suffix, the transitivizer is expected to occur closer to the root, hierarchically
speaking, than inflection, which includes object agreement and aspect. Cross-
linguistic expectations about inflectional affix ordering, e.g. Bybee 1985, hold
that object agreement is generally closest to the verb root, followed by aspect;
tense and subject agreement are even farther from the verb root. Thus the null
hypothesis for Tawala would be that durative reduplication operates on a stem
including the verb root, any derivational morphology, and object agreement.
On this hypothesis the data in (54) are completely expected.19
Eastern Kadazan, seen earlier in §5.2.2, might also appear at first glance to
involve overapplication of an external junctural effect between prefix and redu-
plicant. As argued in §5.2.2, however, morphology again explains the pattern:
the opacity in Eastern Kadazan is due to infixation. The alternation in question –
adjunction of a prefix-final consonant into an otherwise vowel-initial Proot – is
stem-internal.
Two other examples like Eastern Kadazan, namely Tagalog and Chumash,
have in fact been analyzed in the BRCT literature as instantiating the overappli-
cation of an effect at the external juncture between prefix and reduplicant. It is
argued in Chapter 6 that the opacity in Tagalog and Chumash should be under-
stood in a parallel fashion to that of Eastern Kadazan: independent evidence
exists that both languages have internal reduplication, and that the juncture in
question is internal to the stem being reduplicated.

5.4.2.4 Summary
MDT makes restrictive predictions about what kinds of phonological effects
can over- or underapply in reduplication; BRCT predicts a much wider range of
reduplicative opacity. The predictions of MDT are borne out in the data, while
those of BRCT are not. Observations parallel to ours about the apparently false
predictions of BRCT are also made in Kiparsky (1997), in the context of a
discussion of Sanskrit reduplication.

5.5 Case study: Fox

We turn now to a case study of Fox which illustrates, more fully than any of
our previous studies, the different behavior of stem-internal vs. stem-external
alternations in reduplication, confirming the predictions of MDT in this regard.
Our discussion is based on Dahlstrom 1997. Reduplication in Fox targets a
166 Morphologically driven opacity in reduplication

well-defined potentially morphologically complex constituent which we will

call the stem. Fox has both stem-internal and stem-external phonology. The
former is subject to overapplication effects; the latter applies transparently in
reduplication. This is exactly as MDT (but not BRCT) predicts.
The more productive of Fox’s several reduplication patterns, disyllabic pre-
fixing reduplication marks what Dahlstrom characterizes as iterative aspect
(repeated action or action distributed over group of subjects or objects); see
Conathan and Wood 2003 for further elaboration of the semantics of the con-
struction. As can be seen in the data in (55b), vowel length is preserved in the
first syllable of the reduplicant but is lost in the second. Similarly, in (55c), coda
consonants are preserved in the first syllable of the reduplicant but not in the
second. Dahlstrom (p. 218) relates both these facts to the phonotactics of Fox
words, which must end in light (CV) syllables. The reduplicant, in boldface, is
thus best understood as a disyllabic (minimal) prosodic constituent; following
Dahlstrom, we will call it a minimal Pword.20

(55) a. pye:taw-e:wa pye:ta-pye:tawe:wa ‘he brings it for him’ 217

kw i:nom-e:wa kw i:no-kw inome:wa ‘he longs for him’ 217
ni:škesi-wa ni:ške-ni:škesiwa ‘he is overburdened’ 217
ki:hpoče:-wa ki:hpo-ki:hpoče:wa ‘he eats his fill’ 217
b. we:ne:haki we:ne-we:ne:haki ‘who (pl)?’ 218
*we:ne:-we:ne:haki
mayo:-wa mayo-mayo:wa ‘he cries’ 218
*mayo:-mayo:wa
po:swe:kesi-wa po:swe-po:swe:kesiwa ‘he cries louder’ 218
*po:swe:-po:swe:kesiwa
c. nenehke:nem-e:wa nene-nenehke:neme:wa ‘he thinks about him’ 218
*neneh-nenehke:neme:wa
nakiškaw-e:wa naki-nakiškawe:wa ‘he meets him’ 218
*nakiš-nakiškawe:wa
kokw a:ške:-wa kokw a-kokw a:ške:wa ‘he is jerked’ 218
*kokw aš-kokw a:ške:wa

Disyllabic reduplication is sandwiched morphologically between inflectional

suffixation, which is present in the input to reduplication, and inflectional pre-
fixation, which is not.
The data in (56a) illustrate that inflectional suffixal material can reduplicate
if it falls within the first two syllables of the stem. The forms in (56b) show
that inflectional prefixes are not in the scope of reduplication. The claim that
reduplication operates inside of prefixation is reinforced by (56c), nesi, a word
consisting only of prefixes because its root ‘say’ is phonetically null. The word
nesi, lacking any overt root or suffix material, cannot reduplicate at all:
 Case study: Fox 167

(56) a. mi:n-e:wa mi:ne-mi:ne:wa ‘he gives it to him’ 208

kot-aki kota-kotaki ‘he swallows it; conjunct’ 219
b. ne-mi:n-a:wa ne-mi:na-mi:na:wa ‘I give it to him’ 220
ke-nepa ke-nepa-nepa ‘you sleep’ 220
c. ne-s-i-Ø — ‘I say’ 220

Based on these facts, we follow Dahlstrom in positing the following structure

for a Fox word containing prefixes and suffixes:

(57) Word
|
Prefixed stem

Stem

Prefixes Root Suffixes

Reduplication operates on Stems – thus “outside” of suffixation – but “inside”

of prefixation. Layering reduplication in this manner accounts for why nesi, a
word in which the Stem is phonetically null, cannot overtly manifest redu-
plication. It also accounts for why suffixes but not prefixes can be doubled
in reduplication. A word containing reduplication, prefixation, and suffixation
will have the structure below; we assume that prefixes can combine either with
Stems or with Rstems, where ‘Rstem’ is the product of reduplication:

(58) Word
|
Prefixed Rstem

Rstem

Stem Stem

Prefixes Root Sfx Root Sfx

This layered morphological structure also makes predictions about the redu-
plicability of the effects of phonological alternations conditioned within Stems
versus those conditioned solely in Rstems or Words. Stem-internal alternations
are predicted on this Morphological Doubling Theory analysis to be reduplica-
ble, in the sense that they apply within both daughters of the Rstem. However,
alternations applying only within Rstems or Words are not expected to be dou-
bled. These predictions are completely supported by the data.
168 Morphologically driven opacity in reduplication

5.5.1 Stem-internal alternations: opaque overapplication by truncation

Dahlstrom (1997) describes a number of alternations which we know to be
stem-internal because they are conditioned by individual suffixes. Because Fox
disyllabic reduplication targets the (suffixed) stem, we expect all of these stem-
internal alternations to be reflected in both copies of the reduplication construc-
tion. This prediction is borne out. Opacity arises in just those cases in which
the triggering suffix is truncated but the affected segments are not.
The first example involves the root ‘say,’ which has two suppletive allo-
morphs: /en/ and /Ø/. The choice of root allomorph is determined by the imme-
diately following suffix. /Ø/ is used before the inflectional suffixes -ekw - “inverse
theme sign,” -ekw i- “inanimate subject,” and -en(e)- “2nd person object.” As
shown below, when stems formed from the root for ‘say’ are subject to disyl-
labic reduplication, the same suffix-triggered root allomorph is present in both
copies. Because the root is so short, the first syllable or two of the suffix always
surfaces in the reduplicant, making the root allomorphy transparent:21
(59) a. ne-t-en-a:wa ne-t-ena-h-ina:wa ‘I say to him’ 220
ke-t-en-a:wa ke-t-ena-h-in-a:wa ‘you say to him’ 220
in-e:wa ine-h-ine:wa ‘he says to him (obv)’ 220
b. ne-t-Ø-ekw a ne-t-ekw a-h-ikw a ‘he says to me’ 221
ke-t-Ø-ekw a ke-t-ekw a-h-ikw a ‘he says to you’ 221
Ø-ikw a ikw a-h-ikw a ‘he (obv) says to him’ 221

Below we sketch several other instances in which stem-internal phonolog-

ical alternations appear in each copy in reduplication. The alternations are all
clearly stem-internal; they are morphologically conditioned to affect a root-final
segment in the environment of suffixes.
(60) No reduplication Reduplication

a. Root-final /n/ → /š/ before /i/ (in some roots)

/ke-t-en-ipena/ ke-t-eš-ipena ke-t-eši-h-išipena 223
‘you say to us’
b. Root-final /Cw/-e → C-o
/amw-ekoči/ amokoči amo-h-amokoči 223
‘he [obv.] eats him’
c. Root-final /aw/-e . . . → o: (where e is part of certain inflectional suffixes)
/ke-na:taw-ene/ ke-na:to:ne ke-na:to-na:to:ne 224
‘I go after it for you’
d. Umlaut: root-final a: → e: before independent indicative 3rd person suffixes
/nepa:-wa/ nepe:-wa nepe-nepe:wa 224–225
‘he sleeps’
/na:kw a:-waki/ na:kw e:-waki na:kw e-na:kw e:waki 224
‘they leave’
 Case study: Fox 169

Not all suffixes trigger these root-final effects; compare, for example, the exam-
ples in (61) to those in (60d). The Umlaut alternation is triggered by indepen-
dent indicative third person suffixes (Dahlstrom 1997:224); the examples in
(61), with second person (a) or conjunct paradigm (b) suffixes, do not show any
alternation in the root-final vowel:
(61) /ke-nepa:-pwa/ ke-nepa:-pwa ke-nepa-nepa:pwa 224–25
‘you (pl) sleep’
/na:kw a:-waki/ na:kw a:-wa:či na:kw a- na:kw a:wa:či 224
‘they leave; conjunct’

As they are suffix-specific, the alternations in (60) are necessarily Stem-level

processes:
(62) Rstem

Stem Stem
Umlaut, truncation ⇒ ⇐ Umlaut, etc. (certain suffixes)
… Root Sfx Root Sfx

Alternations like Umlaut therefore apply internal to the stem that is doubled and
(in the first instance) truncated in reduplication. Insofar as reduplicative trun-
cation obscures the suffixal triggers for these processes, they apply opaquely.
As Dahlstrom puts it (p. 223), “bisyllabic reduplication copies the output of
the rule but not the environment which triggers the rule.” This is precisely the
description of daughter-internal opacity in MDT.

5.5.2 Junctural alternations: normal application

Morphological Doubling Theory predicts “overapplication” effects only in the
case of alternations triggered within both daughters of the reduplication con-
struction. Alternations triggered outside of one daughter, which happen to affect
that daughter, are not expected to overapply to the other daughter (unless inde-
pendently conditioned there). Fox provides clear illustration of the different
behavior of daughter-internal phonology and junctural phonology in regard to
overapplication. Having already seen cases of daughter-internal phonology in
Fox, we turn next to cases of junctural phonology, looking at phonological
alternations triggered by prefixes external to reduplication and at alternations
triggered at the base-reduplicant juncture.

5.5.2.1 Prefix-Rstem junctures

“Initial Change” is an ablaut process marking certain inflectional categories
(Dahlstrom 1997:221); when it applies to reduplicated words it affects only
170 Morphologically driven opacity in reduplication

the initial consonant, rather than both copies. Equating Initial Change formally
with prefixation, Dahlstrom observes that it is outside, not inside, the scope of
reduplication. Initial Change turns initial /a/, /e/, /i/ to /e:/, and initial /o/ to /we/.
The examples below illustrate Initial Change in its function of marking conjunct
participle inflection in reduplicated words. As seen, Initial Change affects the
initial vowel in the word, resulting in phonological divergence between the two
copies of the root:

(63) /amw-/ e:mwa-h-amw-a:čihi ‘the ones whom they (repeatedly) eat’ 222
/ašam-/ e:ša-h-ašam-a:či ‘that which he (repeatedly) feeds him’ 222
/kanawi-/ ke:na-kanawi-ta ‘the one who gives speeches’ 222

In Morphological Doubling Theory, Initial Change is handled in the

cophonology of the prefixal construction creating conjunct participles:

(64) Word
|
Prefixed stem
⇐ Initial change (Prefix Cophonology)
Rstem

Stem Stem

Prefixes Root Suffixes Root Suffixes

Initial Change applies external to the stem and is therefore not present in
the input to reduplication; we correctly predict no “overapplication” in those
cases where Initial Change happens to affect the first of the two daughters in
the reduplication construction.
A second alternation applying at the Prefix-Rstem juncture is /t/-epenthesis,
triggered by VV sequences across the Prefix-Stem or Prefix-Rstem juncture:

(65) ne-t-en-a:wa ‘I say to him’ 220

ke-t-en-a:wa ‘you say to him’ 220
ke-t-a:čimo ‘you tell a story’ 220
ne-t-ekw a ‘he says to me’ 221

As a Stem-external junctural rule, /t/-insertion must be handled by a

cophonology higher than the Stem, namely the Prefix cophonology:
 Case study: Fox 171

(66) Word
|
Prefixed stem /t/-insertion (Prefix Cophonology)

Rstem
Stem Stem

Prefixes Root Suffixes Root Suffixes

Morphological Doubling Theory predicts strongly that, like Initial Change,

/t/-insertion will apply transparently to the output of the Rstem created by
reduplication. This prediction is correct, as shown by the reduplicated form
below (Dahlstrom 1997:220), with only one instance of /t/-insertion.

(67) ne-t-amw-a:wa ‘I eat him’ ne-t-amwa-h-amw-a:wa (*ne-t-amwa-t-amw-a:wa)

5.5.2.2 Rstem internal juncture

The /h/ at the juncture between the two copies in (67) is due to a separate
pattern which inserts /h/ between vowels at the base-reduplicant juncture. This
is illustrated further below:

(68) aškw at-amwa aškw a-h-aškw at-amwa ‘he has [food] left over’ 217
a:mi:-wa a:mi-h-a:mi:wa ‘he moves camp’ 218
opye:ni opye-h-opye:ni ‘slowly’ (preverb) 218
amw-e:wa amwe-h-amwe:wa ‘he eats him’ 219
ay-o:ya:ni ayo-h-ayo:ya:ni ‘I use it; conjunct’ 219
i-wa iwa-h-iwa ‘he says’ 219

Many prefixed reduplicated forms show both /t/- and /h/-epenthesis (see, for
example, (59) and (67); neither alternation over- or underapplies in the service
of greater base-reduplicant identity. This is exactly as MDT predicts; neither
inserted /t/ nor inserted /h/ is present in the input to reduplication, i.e. within
either daughter of the Rstem. Because it is a junctural alternation, /h/-insertion
has to be handled higher up, either at the Rstem or, as argued below, at the Word
cophonology.

5.5.2.3 Word-level alternations

Short /e/ is raised to /i/ in absolute word-initial position; thus the root /ena:pi-/
‘look [there]’ surfaces with /e/ when prefixed (ne-t-ena:pi ‘I look [there]’) but
with /i/ when initial (ina:pi-wa ‘he looks [there]’ (Dahlstrom 1997:215–16)).
Because /e/ → /i/ is conditioned by word-initial position, we expect it to be
172 Morphologically driven opacity in reduplication

a Word-cophonology alternation and to apply to the output of reduplication.

This prediction might seem to be contravened by the data in (69), in which /e/-
raising applies to both copies of a reduplicated root.

(69) Root Unreduplicated Reduplicated

/ena:pi-/ ina:pi-wa ina-h-ina:piwa ‘he looks [there]’ 216
/enowe:-/ ino-h-inowe:-wa ‘he says [thus]’ 216
/ešawi-/ iša-h-išawi-wa ‘he does [thus]’ 216

This might be interpreted as an instance of overapplication of a Word-level

alternation in service of base-reduplicant identity, but the data in (70) argue
against an identity-driven overapplication analysis. In these similar forms, the
reduplicated root is prefixed, rendering the first copy non-initial in the word.
As a result, /e/-raising does not apply to the first copy. But it still applies to the
second copy:

(70) /ena:pi-/ ne-t-ena:pi ne-t-ena-h-ina:pi ‘I look [there]’ 216

/ešawi-/ ne-t-eša-h-išawi ‘I do [thus]’ 216
/enowe:-/ ne-t-eno-h-inowe ‘I say [thus]’ 216

Here, the apparently opaque application of /e/-raising in the second copy

serves to differentiate the two copies of the root, not to make them more sim-
ilar. Base-reduplicant identity is clearly irrelevant to the application of /e/-
raising. We instead adopt Dahlstrom’s view (p. 216) that /e/-raising applies
within phonological words (Pwords), and that the daughters of reduplication
are separate Pwords. Prefixes incorporate into the following Pword, explaining
why root-initial /e/-raising is blocked by prefixation.22
The question then arises: why doesn’t junctural /h/ syllabify with the root,
in the above examples, make it consonant-initial? Our proposal is that both /h/-
insertion and /e/-raising apply at Pword boundaries in the Word cophonology,
and interact in a counterbleeding fashion. (On approaches to counterbleeding in
Optimality Theory, see, for example, McCarthy 1996; Sprouse 1997; McCarthy
1999.)

5.5.3 Summary of Fox

We have seen three types of junctural phonological alternations in Fox, sum-
marized below:

Root-sfx junctural alternations: reduplicated

Prefix Stem junctural alternations: not reduplicated

Word Level, at phonological word junctures: not reduplicated
 Conclusion: morphology underlies opacity 173

MDT correctly predicts the reduplicability of these alternations. Those applying

internal to stems will be doubled when the stems are doubled; those triggered in
cophonologies external to the morphological constituent created by reduplica-
tion will apply transparently (relative to reduplication). The morphology of the
Fox word, plus the assumption that cophonologies are associated with morpho-
logical constituent types, correctly predicts the scope of a given phonological
alternation. It is only when morphology is ignored that overapplication of the
kind we have seen in Fox becomes a mystery and requires extra theoretical
mechanisms to describe.

5.6 Conclusion: morphology underlies opacity

We have seen in all of the case studies in this chapter that morphology
often explains what phonology alone cannot. Once the role of morphology–
phonology layering is recognized in reduplication, the need for special mecha-
nisms like BR-Faith vanishes. Not only is BR-Faith unnecessary; it has in
fact become an impediment to understanding the nature of reduplicative opacity,
in that it predicts unattested types of opaque reduplicative phonology.
It is sometimes claimed, though without supporting evidence, that opacity
(over- and underapplication) is much more common in reduplication than in
other kinds of constructions. The MDT approach is in many ways similar to
the rule-ordering approach to reduplicative opacity discussed in, for example,
Wilbur 1973, Dudas 1976, and other sources of that era. In those approaches,
a given phonological rule could be ordered after reduplication, in which case
it would apply transparently, or before reduplication, potentially giving rise
to overapplication or underapplication effects. Kenstowicz (1981), comparing
such approaches unfavorably to Wilbur’s Identity Principle in generating over-
and underapplication, writes that ordering theories

[do] not provide an explanation for why reduplication should have this ten-
dency to late ordering while other morphological processes such as suffixation,
infixation, etc. typically apply before most if not all phonological rules. The
reason appears to be the “iconic” nature of reduplication. (p. 441)

Since the writing of this passage, of course, many cases have been docu-
mented of phonology applying to the inputs to affixation and other morpho-
logical processes; indeed the entire literature on cyclicity is of this kind. It is
difficult, however, to evaluate Kenstowicz’s claim that opacity is more com-
mon in reduplication than elsewhere. Even if this is true, it is still not obvious
that a grammatical BR-identity requirement is responsible. The prevalence of
174 Morphologically driven opacity in reduplication

opacity in reduplication could simply be due to the historical development of

truncation in partial reduplication, from which, it is our impression, most cases
of synchronic opacity result. Blevins (2004) lays out a detailed historical expla-
nation of the development of C duplication from CV reduplication, but the
diachronic pathway to more severe truncation is not well enough understood to
make speculation on this matter fruitful here (see, however, Niepokuj 1997).

5.7 The question of backcopying

As seen in this chapter, MDT makes more restrictive, and accurate, predictions
regarding the opacity of junctural phonological effects in reduplication than
does BR correspondence theory. In this section we examine one last prediction
that distinguishes BRCT from MDT: the question of mutual influence between
base and reduplicant.
Possibly the most striking claim of BRCT is that the output form of the
reduplicant can influence the output form of the base via BR correspondence, a
predicted phenomenon referred to as backcopying (McCarthy & Prince 1995a).
We examined, and rejected, one potential case of backcopying underapplication
in Klamath in Chapter 4. A schematic example of backcopying overapplication
is presented in (71). This hypothetical language laxes vowels in closed syllables,
as shown in (a). CVC reduplication of a CVCV stem, as shown in (b), results
in one copy (the reduplicant) which consists of a closed syllable, and another
(the base) which consists of two open syllables. The transparent application of
vowel laxing would result in a lax vowel in the reduplicant and a tense vowel in
the base. BRCT, however, makes it possible for the reduplicant to influence the
base: if the constraint requiring closed syllable laxing (Lax) and BR-Ident
both outrank IO-Faith, the result is a lax vowel in the base as well, as shown
in (b):

(71) Backcopying of internal alternation

a. Language has closed syllable laxing:
/kem-ta/ → [kεmta] Lax » IO-Faith
b. Laxing overapplies to open syllable in base (backcopying)
/red-kema/ → kεm-kεma BR-Faith, Lax » IO-Faith

Coupling the possibility of backcopying with the possibility, predicted in

BRCT, that internal or external junctural phonology may be rendered opaque
by BR correspondence, derives an even more dramatic set of predicted opac-
ity effects. In the example in (72), also from a hypothetical language with
CVC reduplication, nasal place assimilation (NPA) applies across prefix-root
 The question of backcopying 175

junctures (a). NPA also applies at the reduplicant-base juncture, as in (b). Here,
we see that by virtue of the ranking BR-Faith, NPA » IO-Faith, assimilated
place features of the reduplicant-final nasal are backcopied to the corresponding
nasal in the base.

(72) Backcopying of internal junctural alternation

a. Language has NPA:
/in- kana/ → [iŋkana]
b. Internal BR juncture, backcopying overapplication
/red-kana/ → [kaŋ-kaŋa] BR-Faith, NPA » IO-Faith

In a similar vein, BRCT can derive the backcopying of junctural phonologi-

cal effects triggered at the juncture between the reduplicant and another affix.
To illustrate this predicted phenomenon we modify our hypothetical language
slightly so that it has VC suffixing reduplication, as shown in (73a). In (b)
we see the word in (a) followed by the suffix -ta, to which the reduplicant-final
nasal assimilates in place. Backcopying, resulting from the ranking BR-Faith,
NPA » IO-Faith, causes the base-final nasal to take on the assimilated features
of the reduplicant-final nasal:

(73) External juncture backcopying

a. /pekam-red/ → [pekam-am]
b. /pekam-red-ta/ → [pekan-an-ta] NPA, BR-Faith » IO-Faith

MDT strongly predicts these effects not to occur in morphological reduplica-

tion.

5.7.1 Lack of evidence for backcopying in morphological reduplication

Backcopying in morphological reduplication appears rare, if not nonexistent. To
our knowledge there have been no claimed cases of stem-internal backcopying
(71). Two examples of apparent internal junctural backcopying (72) have been
described in the literature, namely Johore Malay (Onn 1976) (note 18, this
chapter) and Chaha (Kenstowicz & Petros Banksira 1999), discussed below
in §5.7.2. Alleged cases of external junctural backcopying (73) are the most
numerous in the literature, and we will focus on these here.
Thus far we have in fact already seen two cases that superficially appear
to meet the description of External Juncture backcopying. One was the pur-
ported instance of underapplication in Klamath (McCarthy & Prince 1995a),
which we reanalyzed (following Zoll 2002) in Chapter 4. Rather than instan-
tiating backcopying underapplication, Klamath is a case of non-application;
176 Morphologically driven opacity in reduplication

the mother node cophonology associated with reduplication fails to impose a

syncope alternation which some other prefixes in the language trigger.
The other case is Eastern Kadazan, analyzed earlier in §5.2.2, which also
could be described in terms of backcopying overapplication. Recall forms like
the following, from (34), above (Hurlbut 1988:107):

(74) UR of unreduplicated stem Reduplicated stem

/m-pi-siN-alud/ misingangalud

The fact that the prefix-final consonant, “ng,” is doubled along with the
following root-initial vowel in this example could be described as a case of
backcopying overapplication. In fact, McCarthy and Prince (1995a) give pre-
cisely such an analysis to parallel phenomena in Chumash and Tagalog. Using
Eastern Kadazan data in place of Chumash or Tagalog, their analysis would sit-
uate the reduplicant immediately preceding the root and require the reduplicant
to be a light syllable, as below:

(75) /m-pi-siN-Red-alud/ Red = ␴ ␮

Since its base is alud, the reduplicant might be expected to be a. However,

because of universal syllabification constraints, the preceding prefix-final con-
sonant (‘ng’) must syllabify with the vowel of the reduplicant. McCarthy and
Prince (1995a:310) propose for a different case, Chumash, that resyllabifica-
tion results in morphological reaffiliation; the prefix-final consonant acquires a
morphological affiliation with the reduplicant morpheme, which in the Eastern
Kadazan case in question would consequently consist of the well-formed syl-
lable nga. Backcopying from reduplicant to base results in the “ng” segment
being duplicated at the left edge of the base, yielding m-i-si-nga-ngalud.
As seen in §5.2.2, however, there is independent evidence to support a com-
pletely different account of Eastern Kadazan in which reduplication is a late
infixing construction, targeting an internal prosodic constituent (Proot) within
potentially complex words. The prefix-root juncture in (75) is internal to the
stem that is the subject of reduplication, and its internal syllable structure is
respected by the infixing process that produces a CV copy of the Proot.
Chapter 6 argues that this same analysis is in fact the correct one for
Tagalog, and Chumash as well. Involving both infixation and truncation, East-
ern Kadazan, Tagalog, and Chumash perfectly illustrate the factors which we
have argued in this chapter lead to reduplicative opacity. Their analysis does
not require backcopying.
 The question of backcopying 177

5.7.2 Backcopying as phonological assimilation

Although we contend that there are few or no viable cases in which BR identity
results in backcopying of the types discussed above, it is undeniable that in
some cases the constituent called “base” can assimilate phonologically to the
one labelled “reduplicant,” giving the superficial appearance of backcopying.
However, as has been documented by Zuraw (2002), the types of assimilation
involved are not specific to reduplication but are found in nonreduplicative con-
texts as well. There is no need to appeal to BR identity to force their existence.
This point is illustrated, below, by an example in which base assimilates to redu-
plicant in morphological reduplication, followed by an example in which source
assimilates to copy in phonological duplication. Both examples yield to a sim-
ilar analysis in which BR correspondence per se plays no role. What is instead
involved is phonologically mandated string-internal correspondence between
similar segments. Walker (2000a), Hansson (2001), and Rose and Walker (2001)
have variously proposed, in their analyses of long-distance consonant interac-
tions, that segments in the same word can be required to correspond if they
meet specified proximity and similarity thresholds (the closer together and more
similar two segments are, the likelier they are to correspond); corresponding
segments can then be compelled to be identical (or non-identical) in certain
respects, yielding harmony (or dissimilation). String-internal correspondence
is purely phonological, unrelated to BR correspondence.

5.7.2.1 Chaha biliteral roots

As described by Kenstowicz and Petros Banksira (1999), a near-allophonic
alternation between [k] and [x] in Chaha exhibits apparent backcopying effects
in reduplication. The general rule in Chaha is that the velar obstruent surfaces as
[x] unless a fricative follows, in which case it surfaces as [k]. (Some exceptions
to this generalization in certain cells of verbal paradigms derive from histori-
cal or abstract gemination of [x], yielding opaque [k].) We follow Kenstowicz
and Petros Banksira in assuming that the velar is underlyingly [x], though a
representation unspecified for [continuant] is also plausible. The relevance for
backcopying is the following: when a biliteral root such as /xt/ ‘cut’ is redupli-
cated to flesh out a CVCCVC template, the velar fricative in the second copy
feeds hardening, to [k], of the velar in the first copy, and the velar in the second
copy consequently hardens sympathetically to [k], producing, for the root /xt/,
the output [kətk t]. In such forms Kenstowicz and Petros Banksira analyze the
first copy as the reduplicant and the second as the base, making this, in their
view, an instance of backcopying of an internal junctural effect in reduplication.
The relevance of this particular example to the question of BR identity hinges,
178 Morphologically driven opacity in reduplication

however, on whether or not the attested hardening and assimilation processes

can be independently motivated outside of a reduplicative context. Kenstow-
icz and Petros Banksira cite data showing that an initial velar obstruent in a
(nonreduplicated) triliteral root must surface as [k] if either the second or third
root consonant is a fricative; for example, ‘hash,’ a root they analyze underly-
ingly as /xtf/, surfaces with [k] in the jussive: yə-kt f (p. 575). Thus for an input
/xt-xt/ we independently expect hardening of the first /x/ to [k]. The question
is then how to analyze the hardening of the second /x/ to [k] even though no
fricative follows. Our perusal of Kenstowicz and Petros Banksira 1999 shows
no words – reduplicated or not – in which a [k] is followed in a word by [x], or
indeed any words in which the two different allophones co-occur. This distri-
butional generalization easily follows if we assume a surface correspondence
relation between velars (in this case, velars in adjacent onsets), subject to a
typical phonological agreement constraint, entirely independent of reduplica-
tion, which requires velar obstruents to agree in [continuant]. Constraints such
as this are needed independently to handle everyday, nonreduplicative conso-
nantal co-occurrence restrictions of the sort discussed in Hansson 2001 and the
many references cited therein; see especially the discussion of stricture harmony
(p. 137).
In summary, requiring [continuant] agreement of velars in Chaha obviates
the need for BR correspondence in the analysis of Chaha [k] ∼ [x] alternations
and relates the facts to the well-known typology of consonant harmony.

5.7.2.2 Hausa participle formation

Hausa forms adjectival participles by adding to a verbal base a suffix consisting
of the vowels a and e , separated by a geminate consonant which is featurally
identical to the root-final consonant (Newman 2000:19):

(76) Verb Participle

dafà dàf-affe ‘cook/cooked’

cikà cı̀k-akke ‘fill/full’
gasà gàs-as sshe ‘roast/roasted’

An autosegmental analysis would spread a root node (or all of the features it
dominates) to the suffixal consonant position. In Optimality Theory, a variety of
approaches exist to accomplish the same representational end. One possibility
is to rank Dep,the ban on feature insertion, higher than the ban on multiple
indexation of output features, which would favor feature sharing (copying) over
 The question of backcopying 179

the insertion of a completely new set of consonant features. The other possibility
is to hand the task of copying over to the constraints compelling string-internal
correspondence (Walker 2000a; Hansson 2001; Rose & Walker 2001) between
segments meeting specified thresholds of similarity and proximity.23 Either
approach will result in a representation like that in (77), in which the suffix
consonant features are co-indexed with those of the root-final consonant:

(77) di afj -affj e

In the representation in (77), root-final f and suffixal ff correspond to each

other. This is precisely the relation that, in reduplication, BRCT predicts to give
rise to backcopying (although here, of course, the correspondence is driven
by the phonology, not by a special BR-identity imperative). There is indeed
evidence in Hausa that phonological alternations affecting the copy consonant
can be replicated on the root-final consonant. Hausa palatalizes coronals before
front vowels (e.g. /sa t-e / ‘steal-pronoun object’ → [sà tʃe ], /o fı̀s-n/ ‘office-
def’ → [o fı̀ʃin], etc.; see Newman 2000:307, 414ff., 627ff.). The participial
suffix has a front vowel, setting up the palatalization environment in case the
root it combines with copies a coronal consonant into the suffix. As noted
in McCarthy 1986; Newman 2000:417, the palatalization which, as expected,
targets a suffixal coronal is occasionally extended to the stem-final consonant,
resulting in the overapplication of palatalization:

(78) fasà → [fàs-aʃʃe ] ∼ [fàʃ-aʃʃe ] ‘roast/roasted’

In summary, the phonological interaction between similar segments in the

same string is not restricted to cases of morphological reduplication. String-
internal correspondence is, according to Walker, Rose, and Hansson, implicated
in all phonological harmony (or dissimilation) effects. There is no reason to
expect reduplication to be any different from reduplicative morphology in its
susceptibility to harmony.
As the Hausa example shows, phonological duplication is itself an extreme
effect of long-distance phonological correspondence, the situation in which
the quality of an essentially epenthetic consonant is determined by surface
correspondence.
Consonant harmony, speech errors (to which Hansson explicity links conso-
nant harmony), and phonological duplication all involve the same mechanism.
BR correspondence is not required; rather, only sufficient similarity (featural
and positional) and proximity are needed for the segments to affect each other.
180 Morphologically driven opacity in reduplication

5.7.3 Implications
Most of the examples of backcopying in morphological reduplication cited in
the BRCT literature turn out instead to be reanalyzable in one of two directions.
Either the phenomenon in question is actually close-range phonological assim-
ilation of a type attested outside of reduplication (and therefore not reducible
to BR identity), or the phenomenon in question is illusory, the result of mor-
phological misanalysis. Chapter 6 is devoted to demonstrating the latter for
two famous cases of apparent backcopying, and to arguing that for these, as
well as for the cases discussed in this chapter, morphological doubling, sans
BR-identity constraints, is the right analysis of morphological reduplication.
6 Case studies

This chapter presents case studies of partial reduplication in Tagalog and

Chumash, two languages which have been cited in the BRCT literature in
support of backcopying overapplication. As we saw in Chapter 5, backcopying
is not a possible phenomenon in MDT, and it is therefore incumbent on MDT
to provide superior alternative analyses of both cases.
Our investigations have succeeded in doing so. Closer inspection of the mor-
phological properties of Tagalog and Chumash reduplication reveals an infixing
pattern exactly comparable to that of Eastern Kadazan, discussed in Chapter
5. All three languages exhibit apparent overapplication of phonological effects
occurring at the juncture between a prefix and a preposed reduplicant. The rea-
son is that the alternation applies internal to the morphologically complex stem
which is reduplicated. Part of the opacity in Tagalog and Chumash results from
the fact that, as in Eastern Kadazan and Javanese, the reduplicant is truncated to
(some portion of) the Proot, which closely corresponds to but does not exactly
match the morphological root. This near match has led to the understandable,
but mistaken, assumption in past BRCT analyses of Javanese, Tagalog, and
Chumash that reduplication targets the morphological root.
The apparent overapplication effects in Tagalog and Chumash result from a
conspiracy of three factors: a mismatch between the morphological root and
the Proot; truncation; and infixation. We have seen all three factors at work in
reduplicative opacity in Chapter 5. They are sufficient to account for the opacity
in Tagalog and Chumash as well. The increased power of backcopying in BRCT
is not justified by these cases.

6.1 Tagalog
Like many other Austronesian languages, Tagalog has partial reduplication
in the vicinity of the root, as well as a phonological process of nasal fusion
operating at prefix junctures. The interaction of these two phenomena creates
opacity in reduplication.

181
182 Case studies

McCarthy and Prince (1995a) cite the forms in (1) in support of BRCT’s
prediction of backcopying overapplication. Nasal fusion, conditioned by some
ŋ-final prefixes, converts a sequence of /ŋ/ + /p,k,t,s/ to a single nasal consonant
at the relevant place of articulation (/m,ng,n,n/); nasal fusion also optionally
collapses /ŋ-b/ to /m/ and /ŋ-ʔ/ to /ŋ/).1

(1) a. pa-mu-mutul ‘a cutting in quantity’ MP95:255

cf. pa-mutul < /paŋ-putul/ ‘that used for cutting’
cf. putul ‘cut (n.)’
b. na-ŋi-ŋisda ‘is/are going fishing’2 MP95:308, SO:364
cf. maŋ-isda ‘go fishing’ SO:364

McCarthy and Prince (1995a) take the position, consistent with descriptions
in the primary sources (e.g. Schachter & Otanes 1972; marked “SO” above)
that Tagalog has root reduplication. They assume morphological inputs of the
type in (2):

(2) /paŋ-Red-putul/
/naŋ-Red-isda/

Given these inputs, McCarthy and Prince are naturally concerned with the fact
that the prefix affects the shape not only of the reduplicant, which it immediately
precedes, but also the base, to which it is not adjacent. In (1a), the result of nasal
fusion at the reduplicant-base juncture is reflected in the base as well, and in
(1b), the final consonant of the prefix is actually doubled, serving as onset not
only to the reduplicant vowel but also to the base.
McCarthy and Prince’s (1995a:60) solution to the doubling problem is back-
copying: the reduplicant interacts transparently with the prefix and, as a result
of BR identity, the phonological results are mirrored in the base: /paŋ-Red-
putul/ → [pa-mu-mutul].

6.1.1 Alternatives to backcopying

Given the morphological analysis in (2), backcopying would seem to be the
only analytical option. However, a number of researchers – Carrier-Duncan
1984; Aronoff 1988; Lathroum 1991; Booij & Lieber 1993; Cole 1994, as well
as French 1988 – have proposed an alternative account: essentially, reduplica-
tion, which is internal, applies to the output of prefixation. The morphological
analysis on this account is very different from that assumed by McCarthy and
Prince, as shown below for the same two forms considered above:
 Tagalog 183

(3)
Input to reduplication: [pamutul] (< /paŋ-putul/) [naŋisda] (< /naŋ-isda/)
↓ ↓
Output of internal reduplication: [pa-mu-mutul] [na-ŋi-ŋisda]

Because reduplication operates on prefixed stems, the output of nasal fusion is

thus present in the input to reduplication. The fact that its effects are doubled
is not a mystery.3

6.1.2 Prefixation vs. infixation

Three arguments, all derived from close study of reduplicated stems bearing the
prefix maŋ- (on which there is abundant data), support the infixation + normal
application analysis in (3) over the prefixation + backcopying analysis in (2).
First, reduplication appears to be dependent on prefixation, in that in most
cases an independent maŋ-stem (with a transparently related meaning) exists,
while an independent reduplicated stem, with a meaning such that it could serve
as the input to maŋ-prefixation in these words, does not. Illustrative examples
from English’s (1986) Tagalog–English dictionary are provided in (4); forms
or glosses also cited by Schachter and Otanes (1972:103) are marked “SO”.4

(4) Stem Unreduplicated maŋ-Stem Reduplicated maŋ-Stem

a. bayan ma-mayan ma-ma-mayan

‘town,’ ‘country’ ‘to live or reside ‘resident of a city or town’
(SO) in a town’ ‘citizen’ (SO)
b. ibig ma-ŋibig ma-ŋi-ŋibig
‘love, fondness’ ‘to court, to be a suitor’ ‘beau, suitor,’ ‘lover’ (SO)
c. tahi ma-nahi ma-na-nahi
‘sew’ ‘to engage in sewing’ ‘one who sews; tailor,
seamstress . . .,’
‘dressmaker’ (SO)
d. sayaw ma-nayaw ma-na-nayaw
‘dance (n.)’ ‘to take up dancing as a ‘a professional
profession’ dancer’
e. damboŋ man-damboŋ man-da-ramboŋ
‘loot, plunder (n.)’ ‘to loot, plunder’ ‘looter, plunderer,
‘armed robbery’ pillager,’ ‘bandit’ (SO)
(SO)
f. kalakal ma-ŋalakal ma-ŋa-ŋalakal
‘merchandise, ‘to engage in trading ‘trader, merchant,
commodity,’ or commerce’ businessman’
‘business’ (SO)
184 Case studies

These data suggest that the maŋ-stem is a subconstituent of the word con-
taining both maŋ- and reduplication, whereas the bare reduplicated stem is not.
The second, semantic, argument supports this conclusion. Although the
semantic evidence is not always cut-and-dried, in examples like those in (4)
the meaning of the word with both reduplication and maŋ- is based on the
meaning of the maŋ- stem in a way that supports treating the maŋ-stem as a
subconstituent of the word. The prefixed stem ma-mayan means ‘to reside in a
town’; ma-ma-mayan is the nominalization of that verb. The same relationship
holds between maŋibig (< /maŋ-ibig/) ‘to court, be a suitor’ and its nominalized
counterpart ma-ŋi-ŋibig ‘beau, suitor.’
The third argument in favor of infixation is phonological. Some stems
containing disyllabic prefixes, when subject to reduplication, show variation
between stem reduplication and second syllable reduplication (which happens
to target the prefix). The data below are taken from Schachter and Otanes
1972:370:5

(5)
Basic form Contemplated aspect
UR, no reduplication Stem reduplication Second syllable reduplication

a. /ika-takbo/ ika-ta-takbo ∼ i-ka-katakbo

‘cause to run’ ‘will cause to run’
b. /ipag-linis/ ipag-li-linis ∼ i-pa-paglinis
‘clean for’ ‘will clean for’
c. /maka-halata/ maka-ha-halata ∼ ma-ka-kahalata
‘notice’ ‘will notice’

Second syllable reduplication is clearly infixing, showing that infixation has

to play some role in Tagalog reduplication; this lends plausibility to an infixing
analysis of reduplication in other cases as well.
If we accept infixation as the right analysis for Tagalog prefixation generally,
the question then becomes one of mechanics: to what is the infixing reduplica-
tion being infixed in the case of stems with monosyllabic prefixes (e.g. maŋ-)?
Aronoff (1988) proposes that reduplication looks into the word for the morpho-
logical head. Booij and Lieber (1993), Cole (1994), and Downing (1998d) all
propose, with minor variations, a prosodic solution for the mismatch between
the morphological root and what actually reduplicates: essentially, reduplica-
tion targets not the root but a prosodic constituent which corresponds closely but
in some cases imperfectly to the morphological root. We build on this proposal
in developing an analysis similar to what we proposed in Chapter 5 for Javanese
and Eastern Kadazan: reduplication truncates the word down to the first CV of
 Chumash 185

its Proot. In Tagalog, the Proot consists of the morphological root plus a pre-
ceding prefix-final consonant to serve as a syllable onset, where needed. An
analysis of an apparent overcopying example, maŋiŋibig ‘suitor,’ is provided
below. This structure is composed of three independent constructions: prefixa-
tion, which produces maŋ-stems, truncation (which produces the truncated stem
type called for in the first daughter), and reduplication:

(6) ma{ŋi}{ŋibig}
← Infixation
{ŋi} ma{ŋibig}
Truncation → | |

ma{ŋibig} ma{ŋibig}
| | ← Prefixation
/maŋ -ibigRt/ /maŋ -ibigRt/

Once it is recognized that prefixes are present in the input to reduplication and
that the reduplicant is a truncated version of the Proot, which can include prefixal
material, the doubling of onsets and of nasal fusion in Tagalog reduplication
is no longer a mystery. Rather, it is normal application of phonology, rendered
opaque by truncation. No backcopying is required.

6.2 Chumash
Chumash, like Tagalog, has been described as exhibiting root reduplication
with opaque overcopying of prefix material to the left of the reduplicative
prefix (McCarthy & Prince 1995a). Our investigations have revealed a very
different picture: instead, as we found for Tagalog and, in Chapter 5, for Eastern
Kadazan, Chumash exhibits infixing reduplication which targets the Proot of a
morphologically complex word. Our conclusions echo those of Cole (1994) and
Downing (1998d), both of whom also recognize the role of the prosodic stem in
Chumash reduplication. Downing, in particular, observes that an infixing Pstem
analysis makes McCarthy and Prince’s backcopying analysis unnecessary.
Our sources for Chumash data are Applegate 1972; 1976, on the Ineseño
dialect, and Wash 1995, on the Barbareño dialect. We will refer to these data
sources in examples as A72, A76, and W95, respectively.6 Most discussions in
the literature are based on the Ineseño dialect; any examples from Barbareño
will be explicitly identified.
The data in (7) illustrate the apparently prefixing stem reduplication in
nouns and verbs. Reduplication pluralizes nouns and renders verbs repetitive,
186 Case studies

distributive, intensive, or continuative (Applegate 1972:383). To assist in pars-

ing these complex words, roots are in bold in unreduplicated forms, and redu-
plicants are in bold in reduplicated forms:7

(7)
čh umaš ‘islander’ čh um+čh umaš ‘islanders’ A76:273
k-ni-č’eq 1subj-trans-tear kni+č’eq+č’eq ‘I’m tearing it up’ A76:282
k-wi-č’eq 1subj-by hitting-tear kwi+č’eq+č’eq ‘I pound it to pieces’ A72:387

Chumash stem reduplication has played a prominent role in the BRCT liter-
ature because of data like those in (8). McCarthy and Prince (1995a:308) cite
these forms, taken (apparently via Mester 1986) from Applegate (1976:279),
as manifesting backcopying overapplication in reduplication, a signature pre-
diction of BRCT:8

(8) Plain Reduplicated

a. s-ikuk sik+sikuk ‘he is chopping, hacking’ (cf. *s-ik-ikuk)

b. s-iš-expeč ši+šex+šexpeč ‘they two are singing’ (cf. *š-iš-ex-expeč)
c. k-ʔanš k’an+k’anš ‘my paternal uncles’ (cf.*k-ʔan-ʔanš
→ k’anʔanš)

In the forms in (8), an alternation conditioned at the prefix-reduplicant junc-

ture is reflected in the base, where it is not conditioned. In (8a) and (8b), the
alternation consists of syllabifying the prefix-final consonant into the onset
of the syllable containing the VC reduplicant. It is the doubling of this onset
consonant after the reduplicant, where it surfaces as onset to the base-initial
syllable that McCarthy and Prince (1995a) analyze in terms of backcopying.
In (8c), the relevant alternation consists of fusing the prefixal consonant with
the reduplicant-initial glottal. This alternation appears to overapply in that the
fused consonant appears following the reduplicant, at the beginning of the base,
where it is not expected.
Similar phenomena occur in Barbareño Chumash (Wash 1995). As in
Ineseño, a sequence of identical stops or fricatives across a morpheme boundary
is converted to a single, aspirated consonant.9

(9) /s-šuphuč/ → šh uphuč ‘it is full of earth/dirt’ W95:98

/k-kutiy-šaš/ → kh utiyšaš ‘I saw myself [in the mirror]’ W95:98

Aspiration appears to overapply in the Barbareño forms in (10), where a

sequence of identical stops across the prefix-reduplicant boundary reduces to
a single, aspirated stop. In such cases, aspiration is reflected in the base of
 Chumash 187

reduplication as well. Surface forms and glosses are taken from Wash 1995
(pp. 31–32, 167):
(10)
Plain stem Reduplicated

/s-/ prefix (3sg subject) a. /s-suʔnan/ → sh uʔnan sh un+sh uʔnan W95:32

‘he would continue’
b. /s-šutowič/ → šh utowič šh ut+šh utowič W95:167
‘he was very quick’
/k-/ prefix (1sg subject) c. /k-kalaš/ → kh alaš kh al+kh alaš W95:31
‘I am breathing’
/p-/ prefix (2sg subject) d. /p-paš-/ → ph aš ph aš+ph aʔš W95:32
‘You are vomiting’

6.2.1 Alternatives to backcopying

McCarthy and Prince (1995a), who discuss the Ineseño forms in (8), develop a
backcopying overapplication analysis, driven by a requirement that reduplicant
and base begin with identical segments. Their analysis crucially relies on the
assumption that CVC reduplication targets the morphological stem, schema-
tized again below:
(11) Prefixation analysis: /Pfx-Red-Stem/

In situations in which the prefix-final consonant interacts with the reduplicant,

base-reduplicant faithfulness constraints require the effects of this alternation to
be duplicated on the corresponding segments of the base. Below, three possible
outputs for the input /s-iš-Red-expeč/ (8b) are evaluated for well-formedness
and BR identity:
(12) * si.še.x-ex.peč Onset satisfied, but BR identity violated (š = e)
* siš.ex.-ex.peč Onset violated, though BR identity maintained (e=e)
) si.š ex.-šex.peč Onsetsatisfied and BR identity maintained (š = š)10
Insofar as backcopying of effects from reduplicant to base can only be described
if reduplicant and base segments are in correspondence, effects such as this
are clear support for Correspondence Theory and its family of BR-Faith
constraints.
But linear or hierarchical order is never straightforward to determine for
morphological constructions consisting of prosodic operations, and the logically
possible alternative to (11) is that Chumash CVC reduplication is infixing. As
in Tagalog, reduplication could be targeting the stem – or, more accurately, the
Pstem – inside of a morphologically complex word:
188 Case studies

(13)
Input to reduplication: [si.šex.peč] (> /s-iš-expeč/) [k’anš] (> /k-ʔanš/)
↓ ↓
Output of internal [si-šex-šexpeč] [k’an-k’anš]
reduplication:

On the infixing account, by contrast, all the relevant phonological alternations

are transparently conditioned, if we assume that, as in Tagalog, what Chumash
reduplication targets is a prosodic constituent which matches the morphological
stem but can be mismatched in case a preceding prefix ends in a consonant and
the stem begins with a vowel. Indeed, this very Pstem proposal has already been
made by Downing (1998d), who observes that if the PStem is the domain of
reduplication, backcopying becomes unnecessary in the analysis of overcopying
effects. In the MDT account, reduplication operates after, or on the output of,
prefixation; the prefix-final consonant is adjacent to the stem-initial segment
in the input to reduplication. Thus any doubling of effects at the prefix-stem
juncture are the result of normal application of Chumash syllabification rules.
No overcopying, and crucially no backcopying, takes place.
The infixation account also corresponds closely to the account favored by
Applegate (1976), who writes (p. 278) that “CVC reduplication is a very low-
level phonological process which applies to the output of the main block of
phonological rules.”
The choice between the prefixing and infixing alternatives is highly signif-
icant to Morphological Doubling Theory, which lacks the descriptive power
of backcopying. If Morphological Doubling Theory is to be able to account
for Chumash, it must adopt the infixing hypothesis in (13). The Morphological
Doubling Theory account would look something like (14), in which the input to
reduplication – the input to each daughter in the construction – contains all the
morphemes involved in the phonological alternations in question. Each stem
thus contains (in its input) the context for the resyllabification/fusion alterna-
tion, even though truncation of the reduplication renders this context opaque:

(14) Infixation analysis, in Morphological Doubling Theory

si{
ex}{
expe}
⇐ Infixation

{
ex} si{
expe}
Truncation ⇒ | |
si{
expe} {
expe}
Stem-formation ⇒ | | ⇐ Stem-formation
/s-i
-expe/i /s-i
-expe/i
 Chumash 189

Two different morphological hypotheses about Chumash reduplication are

now in competition, one (Prefixing) which motivates backcopying and there-
fore BRCT, and one (Infixing) which is compatible with the more restrictive
MDT approach. Is Chumash stem reduplication strictly prefixing, in which
case backcopying is entailed, or is it infixing, in which case the phonological
doubling effects are simply ordinary reduplication? Our answer, that infixation
is the correct analysis, requires some background in Chumash morphology to
explain.

6.2.2 Chumash verb morphology

A detailed description of Ineseño Chumash morphology is found in Applegate’s
(1972) PhD dissertation, based primarily on the field notes of J. P. Harrington,
in which he discusses six types of reduplication in Ineseño Chumash. Of these
only the most well-known, CVC prefixing reduplication, is discussed here. Wash
(1995) provides a thorough survey of productive CVC prefixing reduplication
in Barbareño Chumash, allowing for comparison between the two dialects.
Although Applegate (1972) is the best source of information about the
morphological structure of complex words, most theoretical discussions of
(Ineseño) Chumash rely on Applegate’s short (1976) paper on reduplication, in
which morpheme breaks are not given. Such discussions are thus not informed
by the morphological evidence to be presented here.
Chumash verbal morphology is heavily prefixing (there are also a small num-
ber of suffixes, both derivational and inflectional). Applegate (1972) partitions
verb prefixes into three main classes: Inner, Outer, and Personal, with inner
prefixes occurring closest to the root and outer prefixes, farthest away:

(15) Outer prefixes – Personal prefixes – Inner prefixes – [root – suffixes]Stem

Outer prefixes mark things like negative, tense, nominalization/relativization,

clause subordination, and sentential adverbs. Personal prefixes mark person
and number of subject. Inner prefixes are quite various, including aspectual,
instrumental, action classifiers, spatial orientation, verbal force, and other things
(see Applegate 1972:301ff.).
This segregation accords closely with the level ordering system for affixes
proposed by Wash for Barbareño. According to Wash, Level 1 includes deriva-
tional suffixes; Level 2 includes derivational prefixes (including classifier pre-
fixes, causative, and diminutive); Level 3 includes inflectional suffixes and
prefixes (including tense, aspect, reflexive, stativizer/nominalizer, pronominal
suffixes, venitive, desiderative, and habitual); and Level 4 prefixes, which Wash
190 Case studies

analyzes as clitics, include tense, indefinite, subordinate, plural, emphatic, arti-

cle, deictic, pronominal, and connectives.
The difference between Inner prefixes and all other prefixes is relevant to
reduplication. Where possible, we will use a hyphen “-” to mark Inner prefixa-
tion boundaries and an equals sign “=” to mark Outer and Personal prefixation
boundaries. Reduplication boundaries are marked with a plus sign “+”.

6.2.3 Which prefixes can contribute an overcopying final C?

In Ineseño, it appears that any preceding prefix will contribute its final C to a
V-initial reduplicating syllable or fuse its final consonant, as appropriate, with
the initial consonant of the reduplicating syllable.11

(16)
Gloss Unreduplicated Reduplicated Source

a. ‘he is naked’ s=ʔamn’ → s’amn’ s’am+s’amn’ A72:389,

A76:279
b. ‘they two are singing’ s=iš-ex-peč ši+šex+šexpeč A76:279
c. ‘they are doing it’/ s=iy-eqwel si+yeq+yeqwel A72:386/
‘they are making . . .’ A76:279
d. ‘basket-maker’ ʔal-aqnp’ ʔa+laq+laqnp’ A72:210

In Barbareño, a Morpheme Integrity Constraint (as in Kinande or Javanese)

appears to prevent the comparable phenomenon, although consonantal prefixes
are amenable to reduplicating along with a following root. See Wash 1995
(especially Chapter 4) for discussion.

6.2.4 Inner (Level 2) prefixes in the reduplication domain

As seen earlier, Applegate’s (1976) characterization of Chumash CVC redu-
plication as stem reduplication plays a central role in McCarthy and Prince’s
(1995a) prefixing analysis. In fact, however, as Applegate (1972) is especially
careful to document, reduplication often targets Inner prefixes as well: “With
the exception of a few prefixes which regularly shift reduplication to the follow-
ing morpheme . . . CVC reduplication falls on the first CVC sequence following
person-number markers” (Applegate 1972:386). This is a highly significant fact,
since it dispels any notion that the reduplicant simply targets the morphological
root. Reduplicating prefixes are part of the Pstem, whose initial CVC portion
is what the reduplicant is truncated down to.
Examples from Ineseño and Barbareño in which an Inner prefix is included
in the reduplicant are given in (17). In (17a–b), the Causative su-, an Inner
prefix, is part of the string of which the initial CVC is reduplicated. In (17c),
 Chumash 191

the agentive prefix ʔal- is doubled; in (17d), the reduplicant is instantiated by

material from the prefix wi- ‘apart.’ Curly brackets in the unreduplicated forms
demarcate the Pstem; Pstem-internal Inner prefixes are shown in boldface.

(17)
No reduplication Reduplication

Ineseño a. k={su-pšeʔ} k+šup+šupšeʔ A76:282

1subj-caus-to be extinguished ‘I’m putting out (a fire)’
b. k={su-towič} k+šut+šutowič A76:282
1subj-caus-? ‘I’m doing it fast’
c. {ʔal-aqša+Vn+ʔ} ʔal+ʔalaqšan’ A72:210
‘the dead’
d. s-{pil-kowon} s+pil+pilkowon A76:281
‘it is spilling’
Barbareño d. {wi-phatatan} wih+wiphatataʔniš W95:128
of hitting-fall apart = ‘pieces’
to destroy, take apart’

A difference between Ineseño and Barbareño, attributed above to Morpheme

Integrity, is that in Barbareño, the reduplicant apparently cannot partially trun-
cate two different morphemes (Wash 1995:128). In (17b), an /h/ is inserted to
supply the reduplicant with a final C, rather than taking the initial stem C for a
coda, as Ineseño would do. (See Downing 1998d for further discussion of the
significance of the dialectal difference for the BRCT approach to reduplication.)
Forms in which reduplication targets a Pstem containing a prefix, which
appear to be common in both dialects, show clearly that the Prefixation picture
in (11) cannot be right. Reduplication is not targeting the morphological root,
and cannot therefore be treated as a morphological prefix to the root; instead,
reduplication is targeting a potentially morphologically complex constituent,
the Pstem.

6.2.5 A split among Inner prefixes in Ineseño

Although some Inner prefixes are Pstem-internal, as seen above, others are not.
As part of his comprehensive inventory of verbal affixes, Applegate (1972)
classifies each Inner prefix according to whether or not it can reduplicate.12
Building on the analysis developed so far, this split corresponds to whether
an Inner prefix is Pstem-internal or Pstem-external. According to Applegate
(1972), “the great majority” of Inner prefixes are Pstem-internal (p. 387). Below
are a number of examples containing Pstem-external Inner prefixes (in boldface)
that do not reduplicate.13
192 Case studies

(18)
Without reduplication With reduplication Gloss Source

a. k=wi-{č’eq} kwi+č’eq+č’eq ‘I pound it to pieces’ A72:387

1subj-by hitting-tear
b. s=uti-{lp’n} suti+lp+lp’n ‘the ground is uneven’ A72:390
c. s=wati-{k’ot} swati+k’ot+k’ot ‘is broken to pieces’ A72:384
/s=wati-{lok’in-š}/ šwati+lok+lok’ič
‘it falls apart’ A72:388
d. s=pili-{paq’} spili+paq+paq’ ‘a travelling company A72:389
splits up’
e. s=ni-{wy} šni+wy+wy ‘he is cutting notches on it’ A72:384
k=ni-{č’eq} kni+č’eq+č’eq ‘I’m tearing it up’ A76:282
ni-{ph at} ni+ph at+ph at ‘to break to pieces’ A72:390
f. s-am=ti-{lok’in} samti+lok+lok’in ‘they cut it off’ A72:387
g. s=akti-{kuti} sakti+kut+kuti ‘he comes to wash’ A72:387

The same word can contain both Pstem-external and Pstem-internal Inner
prefixes. As expected, reduplication targets the Pstem, duplicating the Pstem-
internal, but not the Pstem-external, Inner prefixes. Inner prefixes are, as before,
in boldface in the unreduplicated forms:

(19)
Without reduplication With reduplication Gloss Source

a. k=sili-{pil-wayan} ksili+piw+piwayan ‘I want to swing’ A72:387

b. s-iy=ak{t-aqu-smon} siyak+taq+taqusmon ‘they come to A72:388
< /s-iy-akti-{aqu-smon}/ gather it’

The form in (19b) shows that the final consonant of a Pstem-external prefix will
incorporate into the Pstem, as expected, in case the Pstem would otherwise be
V-initial.
When more than one Inner prefix is present in a word, the Pstem contains the
leftmost Inner prefix and all following material. In (20), the root is preceded by
a string of up to three Inner prefixes, each independently known to be Pstem-
internal. The reduplicant (in bold) is the initial CVC portion of this Pstem.
The form in (20c) shows that Pstems which include Inner prefixes behave in
the same way as those based on roots. If the Pstem would be vowel-initial
but is preceded by a consonant-final prefix (subject to the dialectal conditions
discussed above), the consonant joins the Pstem. In the unreduplicated form,
Inner prefixes are shown in boldface and the Pstem is delimited with curly
brackets:
 Chumash 193

(20)
Without reduplication With reduplication Gloss Source

s={max-keken} s+mex+mexkeken ‘he is spreading it open’ A72:387

{s=ali-max-keken} sal+salimexkeken ‘he is stretching it out’ A72:387
{s=uti-ali-max-keken} sut+sutalimexkeken ‘he suddenly gives it A72:387
a stretch’

6.2.6 Implications of the split among Inner prefixes

The split between reduplicating and nonreduplicating prefixes is easy to model
on the Infixation account, as seen above: reduplication, a late morphological
process, calls for two instances of the entire word, truncating the first copy
down to the first CVC portion of its Pstem. Reduplicating Inner prefixes are
within the Pstem; nonreduplicating Inner prefixes are not.
The split in the behavior of Inner prefixes is much more difficult to capture
on the Prefixing account crucially underlying McCarthy and Prince’s (1995a)
backcopying analysis. On the Prefixing account, the Red morpheme immedi-
ately precedes the string whose first (C)VC it copies. To distinguish redupli-
cating from nonreduplicating Inner prefixes on this account, it would have to
be the case that reduplicating Inner prefixes are linearly ordered closer to the
Stem than nonreduplicating Inner prefixes, and that reduplication is ordered in
between the two prefix classes:
(21) Nonreduplicating Inner prefixes – Red – Reduplicating Inner prefixes – Stem

The Prefixation hypothesis, with ordering as its only possible account of the
split behavior of Inner prefixes, makes a striking morphological prediction: all
reduplicating Inner prefixes should be ordered after (inside, closer to the stem
than) all nonreduplicating Inner prefixes. The Infixation hypothesis makes no
such prediction, as it draws no structural morphological distinction between
Pstem-internal and Pstem-external Inner prefixes.
The ordering prediction of the Prefixation account appears to be inconsistent
with the information provided by Applegate on the relative linear order of the
Inner prefixes. Applegate partitions Inner prefixes into position classes which
correlate roughly with syntactic function and semantic transparency. Although
there are not as many examples containing both types of Inner prefixes as
one would like, the forms that do exist give the clear impression that there
is no fixed linear ordering between reduplicating and nonreduplicating Inner
prefixes.
Example (19) illustrates words in which nonreduplicating Inner prefixes pre-
cede reduplicating Inner prefixes. These data are consistent with the Ordering
194 Case studies

hypothesis. But Inner prefixes also occur in the opposite order, contrary to
the Ordering hypothesis. The examples in (22) illustrate the ubiquitous, highly
productive Causative prefix su- seen earlier in (17), (20). A Pstem-internal,
reduplicating prefix (see, for example, (17)), su- can precede other prefixes
which Applegate classifies as nonreduplicating or, in the terms used here, Pstem-
external. In (22), reduplicating (Pstem-internal) prefixes are subscripted as “R”;
nonreduplicating (Pstem-external) prefixes (on which see (18)) are subscripted
as “N”. One or two of the Inner prefixes in (22) are unclassified, due to a lack
of sufficient data.

(22)
a. /maqR -suR -niN -apayRt / maqsunapay ‘to raise a line, string a bow’ A72:379
/wi-suR -niN -apayRt / wisunapay ‘[sea] to cast ashore’ A72:383
b. /k=suR -watiN -lok’inRt / ksuwatilok’in ‘I cut it in passing A72:368
(I’ll be back for it later)’
c. /suR -tiN -wyRt / sutiwy ‘to sing a charm, spell’ A72:324
/suR -wašR -tiN -aq-peyRt / šuwešteqpey ‘to mend, fix’ A72:383

When verbs like these are subject to reduplication, the leftmost Pstem-internal
prefixes define the left edge of the Pstem, which includes all following material,
including Inner prefixes that would otherwise be Pstem-external. This can be
seen in example (23), in which the Pstem-internal prefix xul- precedes the
Pstem-external transitive prefix ni- (see (18e)). The reduplicant is xun, which
consists of the prefix xul plus the initial consonant of ni-, with which the final
l of the prefix fuses.14

(23) Plain Reduplicated Source

/k=xulR -niN -yw/ → k{xuniyw} k+xun+xuniyw A72:384

‘1subj=?-trans-?’ ‘I am looking all over for it’

The inescapable conclusion from examples like (22) and (23) is that the split
in reduplicability within Inner prefixes is not reflected in their ordering relations;
both reduplicating > nonreduplicating and nonreduplicating > reduplicating
orders are well-attested in Ineseño Chumash. Furthermore, a normally nonredu-
plicating prefix can be “promoted” to a reduplicating prefix if it is sandwiched
between a reduplicating prefix and the stem. This behavior follows naturally
from the Pstem analysis; it is contrary to the predictions of the Prefixation
account, and therefore supports the Infixation account of reduplication.15
Although evidence from semantics is not strong, it tends to support the claim
of the Infixation account that reduplication is a late morphological process tar-
geting whole words, rather than having scope over only stems (as the Prefixation
 Chumash 195

account would predict). Applegate (1972) describes CVC-verb reduplication

as endowing the verb with “a repetitive, distributive, intensive, or continuative
force” (pp. 383–84). Examples like the following are consistent with the pre-
diction of the Infixation account that reduplication should have semantic scope
over the whole verb:16

(24)
Plain Reduplicated Source

a. /s={tal-memen}/ štelmemen š+tel+telmemen A72:384

3subj=of grasping-? ‘he touches it’ ‘he [is] groping
around’
b. /s=wati-{k’ot}/ swatik’ot swati+k’ot+k’ot A72:384
3subj=of disintegration-? ‘is broken’ ‘is broken to pieces’
c. /ni-{ph at}/ niph at ni+ph at+ph at A72:390
trans-? ‘to take apart’ ‘to break to pieces’

Consider the nonreduplicating Inner prefix wati- in (24b). Although it is not

reduplicated phonologically, its meaning appears to be affected by reduplicative
semantics. According to Applegate wati- means ‘apart, of disintegration,’ which
is transparent in the unreduplicated word swatik’ot ‘is broken.’ Judging by the
meaning of the reduplicated verb with wati-, i.e. swatik’otk’ot, reduplication
is acting semantically on the prefixed verb. The meaning ‘is broken to pieces’
is the expected result of applying a distributive suffix to a form meaning ‘is
broken.’ This is exactly as the Infixation account would predict. By contrast,
the Prefixation account would predict wati- to attach after reduplication, and
therefore to be outside its semantic scope. (It would of course help to know
what k’ot means.)

6.2.7 Summary
In Chumash, as argued above, the whole word is subject to reduplication; the
first copy is truncated to the initial CVC portion of its Pstem, which is then
infixed before the Pstem of the second copy. The Pstem contains the stem and
any Pstem-internal prefixes (as well as other prefixes sandwiched in between).
Because of consonant fusion and onset requirements, Pstems can also contain
the final consonant of a preceding prefix which is otherwise Pstem-external.
The potential mismatch between Pstem and morphological constituent is what
gives rise to the appearance of onset overcopying, or overapplication of conso-
nant fusion. However, both effects are normal application within the MDT
analysis, for which we have provided independent support from Chumash
morphology.
196 Case studies

This analysis is, interestingly, very close to Applegate’s own characterization

of reduplication as “a very low-level phonological process which applies to
the output of the main block of phonological rules” (Applegate 1976:278);
“Reduplication occurs only after all other applicable rules have determined the
final shape of the terminal CVC sequence to be reduplicated” (Applegate 1976:
279).
Chumash is a clear illustration of the theme sounded throughout this book.
Understanding the phonology of reduplication requires a solid understanding of
the morphology of reduplication. In Chumash, as in Tagalog, Eastern Kadazan,
Fox, Javanese, Klamath, and many of the other case studies in this book, mor-
phology explains phonological phenomena which would be puzzling from a
purely phonological perspective, and which in some cases have prompted the
introduction of reduplication-specific technologies which are unnecessary in a
theory which approaches reduplicative phonology from the broader perspective
of morphologically conditioned phonology.
7 Final issues

A fundamental claim of Morphological Doubling Theory is that redupli-

cation involves semantic, rather than phonological, identity. If this is true,
then wholesale, potentially long-distance phonological string copying, devel-
oped expressly to account for reduplication (see, for example, Marantz 1982;
Clements 1985; Mester 1986; Steriade 1988, inter alia), is not needed in phono-
logical theory. As argued in Chapter 1, however, there is still motivation for
small-scale, proximal, phonologically motivated phonological copying, both
within and outside of reduplication. This chapter returns to the distinction
between morphological duplication and phonological copying, firming up the
defining criteria and exploring the potential for interaction between the two
types of duplication.

7.1 Criteria distinguishing phonological copying from

morphological reduplication
The four criteria distinguishing between phonological copying and morpholog-
ical reduplication, originally presented in Chapter 1, are given below:
(1) Phonological copying Morphological reduplication
1. Serves a phonological purpose Serves a morphological purpose
2. Is phonologically proximal Is not necessarily phonologically
proximal
3. Involves single phonological Involves morphological constituents
segments
4. Is driven by phonological identity Is not driven by phonological identity
imperative imperative

7.2 The purpose and nature of phonological copying

Accounting for phonological assimilation is central to any theory of phonol-
ogy. In autosegmental phonology, assimilation is viewed as spreading; well-
formedness constraints on autosegmental association ensure that assimilation
is proximal, meaning that a target segment will assimilate to the closest eligible

197
198 Final issues

trigger, not (for example) the farthest. Recent work in Optimality Theory has
cast assimilation in terms of correspondence; for Walker 2000a; b; Hansson
2001; Walker to appear, assimilation occurs when a grammar imposes identity
requirements on corresponding segments; a key component of the theory is that
segment-to-segment correspondence is established in the first place only among
segments that are sufficiently similar and sufficiently proximal in the input. In
this approach, as in the autosegmental approach, assimilation can occur at a dis-
tance only if it also occurs locally; assimilation is therefore always proximal. In
Turkish vowel harmony, for example, a harmonic suffix agrees with the closest
root vowel, not the farthest, as can be demonstrated with disharmonic roots:
thus anne ‘mother’ is anne-ler and elma is elma-lar. There is no language with
vowel harmony in which these plurals could be anne-lar and elma-ler (unless
dissimilation is involved, but then the identity of the initial vowel would be
irrelevant).
The view of phonological segment copying adopted in this book is that phono-
logical copying is an extreme form of phonological assimilation. It conforms to
the same generalizations about proximity, and it is handled with the same the-
oretical mechanisms. What distinguishes phonological copying from ordinary
epenthesis of a default segment is the phonological correspondence established
between the introduced segment and its nearest comparable neighbor. Zuraw
(2002) has observed that phonological correspondence itself can be a desidera-
tum; where faithfulness to input features is not an issue, an epenthetic segment
which corresponds to another is, from that perspective, more desirable than one
which does not.
To illustrate the ways in which phonological segment copying patterns with
phonological assimilation, rather than with morphological reduplication, we
may fruitfully contrast Hausa noun pluralization (2a) and Ponapean prefixation
(2b) with Madurese noun pluralization (2c):

(2) a. fensir-o
ri
‘pencils’

mo
t-o
ci
‘cars’
te
bur-o
ri
‘tables’
b. ak-pwung → akupwung ‘petty’
ak-dei → akedei ‘to engage in a throwing contest’
ak-tantat → akatantat ‘to abhor’
c. w̃ã-mōw̃ã ‘faces’
ỹāt-nẽỹāt ‘intentions’
w␥-kh uw␥ ‘caves’

The Hausa requirement that syllables begin with consonantal onsets leads to
consonant epenthesis in the plural suffix.1 As discussed in Chapter 1, phonologi-
cal copying derives from the need for this consonant to be featurally specified; as
 The morphological purpose of reduplication 199

expected in phonological assimilation, it draws all of its features from the nearest
consonant. In a similarly cross-linguistically common pattern, Ponapean inserts
an epenthetic copy vowel when “an impermissible consonant cluster is cre-
ated through the process of affixation” (in these examples, involving the prefix
ak-, meaning ‘to make a demonstration of’) and the following syllable is heavy
(Rehg 1981:70, 92). The epenthetic vowel takes on the quality of the vowel in
the immediately following syllable. Proximity of this kind is typical of phono-
logical copying generally. It is not, however, a requirement of morphological
reduplication, as illustrated by the “opposite-edge” morphological reduplica-
tion occurring in Madurese (Stevens 1985, via McCarthy & Prince 1995a:278),
where, as mentioned in Chapter 5, the first copy in plural noun reduplication
is truncated to its last – not its initial – syllable. Although opposite-side effects
are a minority pattern in reduplication, the important point is that they exist at
all; opposite-edge effects never occur in phonological assimilation.2
In summary, phonological copying serves a phonological purpose; it involves
single segments, and when copying is involved, the copying is proximal. In all
of these ways, phonological copying is distinct from morphological copying,
which involves morphological constituents – often long strings, not single seg-
ments – is potentially nonproximal, and serves a morphological purpose.
Despite being different – or perhaps because they are different – phonolog-
ical copying and morphological reduplication are not mutually exclusive. For
example, Hausa pluractional formation (discussed in Chapters 1 and 4) involves
both phenomena. Pluractional verbs are formed in Hausa by doubling the verb
stem and truncating the first copy to CVC. This is morphological reduplication.

(3) Plain stem Reduplicated stem

‘open mouth widely’ wa
gè
waw-wà
ge

‘step on’ ta
kà
tat-tà
ka

‘oppress’ dannè
dad-dànne

‘go out’ fı̀tá
fı̀f-fı̀tá

‘sell’ sayar sas-sayar

The final consonant of the truncated first copy, however, is subject to a form of
phonological copying: it assimilates totally to the following consonant, under
appropriate conditions discussed in Chapter 4. This is phonological copying.

7.3 The morphological purpose of reduplication

In saying that morphological reduplication serves a morphological purpose it is
important to understand that morphological reduplication need not always be
associated with a specific meaning, as it is in Hausa pluractional reduplication
200 Final issues

in (3). Morphological reduplication can also serve a morphological purpose by

creating a stem type that is called for by other morphological constructions. For
example, the Bella Coola diminutive (4a) selects for reduplicated stems (see
Raimy 2000:61), as do many Ilokano affixes, including the adjective intensify-
ing prefix illustrated in (4b) (Rubino 2001).

(4) a. Bella Coola diminutive suffix /i∼yi/ (Nater 1984:109)

qwtulh qw-tulh-tlh-i ‘cradle basket’ 108
kwpalh kw-palh-plh-i ‘liver’ 108
sum su-sum-ii ‘trousers’ 109
qlhm q-lhm-lhm-ii ‘black cod’ 109
b. Ilokano adjective intensification (Rubino 2001)
naángot smelly naka-ang-ángot ‘stinking very much’
nasakı́t ‘sore’ naka-sak-sakı́t ‘very sore’
katáwa ‘laughter’ naka-kat-katáwa ‘funny’

The Roviana construction deriving instrumental or locational nouns from

verbs, discussed in Chapter 4, is another such case; the construction is instan-
tiated simultaneously by reduplication and the suffix -ana (Corston-Oliver
2002:472–80). Thus, for example, ham bo ‘sit’ nominalizes as ham bo-ham botu-
ana ‘chair’ (p. 469). Reduplication without suffixation has a completely dif-
ferent meaning (verb affect intensification); the reduplication seen with -ana
serves no semantic function which can be distinguished from that of -ana, and
is best analyzed as a requirement of the suffix.
In MDT, morphologically required reduplication of this sort is handled by
positing a semantically null reduplication construction whose only morpholog-
ical function is to create stems of a type that certain constructions call for.

(5) Stem-forming reduplication: [ ]F, ‘stem type x’

[ ]F [ ]F

Just like semantically null affixation, semantically null reduplication is con-

strained by structural economy constraints (e.g. *struc) from occurring within
words unless required; in the cases seen just above, the requirement is an instruc-
tion on the part of affixation constructions that their stem be of type “x,” where
“x” is the arbitrary name of the stem type issuing from the semantically null
reduplication construction.
Occasionally morphological reduplication is recruited to bring a stem up
to the minimal size required for it to participate in another morphological
 CV reduplication 201

construction, as in Chukchee.3 The absolutive singular is marked in Chukchee

by root reduplication, which truncates the second copy of the root to its ini-
tial CVC portion. Interestingly, this contentful reduplication construction is
recruited, minus its absolutive singular connotation, in absolutive plurals, which
are normally marked with the suffix -t. Exactly when the root is CVC in shape,
the absolutive plural shows the same reduplication found throughout the abso-
lutive singular (Krause 1980):
(6) Root shape Absolutive singular Absolutive plural

CVCV nute-nut nute-t ‘earth’

tala-tal tala-t ‘pounded meat’
CVC čot-čot čot-čot-te ‘pillow’
tam-tam tam-tam-ət ‘growth’

Krause (1980) and Kiparsky (1986) attribute the phonologically conditioned

plural reduplication to a prosodic minimality condition on noun roots; to be
inflected (e.g. with plural -t), a noun stem should be larger than CVC. Some
languages might epenthesize a vowel in this situation; Chukchee appeals instead
to semantically null reduplication.
The fact that reduplication is obligatory in the absolutive singular makes
a morphological analysis of the plural reduplication particularly plausible. If,
as is commonly assumed, absolutive is the unmarked case and singular is the
unmarked number, there are already grounds to propose that the morphologi-
cally required reduplication construction generating absolutive singular stems
is semantically vacuous. It is therefore available to be recruited to assist CVC
roots in satisfying the disyllabicity requirements of the plural suffix. Morpho-
logical reduplication in Chukchee functions, in the absolutive plural, like the
semantically empty affixes in Ndebele, Chichêwa, and other languages dis-
cussed in Chapter 2. It serves the ultimately morphological purpose of building
a stem of the type for which some other morphological construction selects.

7.4 CV reduplication
Although for the majority of examples it is clear whether phonology or mor-
phology is driving the observed duplication, it is inevitable that there will be
cases where the two analyses are similarly compelling, without a good way to
choose between them. The area of greatest potential ambiguity involves dupli-
cation of strings small enough to be manageble phonologically but large enough
to be plausible as truncated morphological constituents – in particular, CV redu-
plication. CV reduplication has generally been analyzed as morphological in
202 Final issues

this work (see, for example, Paamese, in Chapter 2, Roviana and Tawala, in
Chapter 3; Tohono O’odham and Chamorro, in Chapter 4; Eastern Kadazan,
in Chapter 5, Tagalog, in Chapter 6, and others). In some of these cases, e.g.
Paamese, there is evidence from allomorphy that the process is clearly morpho-
logical reduplication; in other cases, we have simply made the assumption that
morphological reduplication is taking place, in the absence of evidence to the
contrary.
It is theoretically possible that, in the unusual event that vowel and con-
sonant epenthesis are conditioned simultaneously and copy epenthesis is pre-
ferred to default segment epenthesis, CV duplication could be derived phono-
logically. In Miya, for example, pluractional stems are formed in a variety
of ways which Bissell (2002) unifies under the analysis of mora prefixation.
As described by Schuh (1998), CaC roots form the pluractional via vowel
lengthening, e.g. tlakə → tláakə ‘scrape’ (p. 176), while for all other CVC
roots the pluractional is formed through Ca- prefixation, where the C is a
phonological copy of the root-initial consonant, e.g. pa-pə́rà ‘cut’ (p. 175),
kwa-kwı́yà ‘catch’ (p. 176). The lengthening evident in tláakə could be seen as
copy vowel epenthesis, while the prefix consonant in pa-pə́rà could be seen as
copy consonant epenthesis. If both happened to co-occur in the same syllable,
purely phonological processes could add up to what appears to be CV redu-
plication. A similar phenomenon occurs in the English construction Yu (2003)
terms “Homeric infixation,” in which the infix -ma- follows a metrical foot:
saxophone → saxa-ma-phone, secretary → secre-ma-tary. The infix imposes
another condition, however, which is that at least one syllable must follow. Thus
in cases such as music or tuba, which consist of only one foot, consonant dupli-
cation and schwa (spelled “a”) insertion take place to satisfy the demands of the
infix: mu-sa-ma-sic, tu-ba-ma-ba. This instantiates the phonological insertion
of a CV syllable; it is not exactly CV duplication, since the vowel is fixed, as
in Miya, but, like Miya, it comes close.
The potential for CV duplication to occur for phonological reasons is con-
sistent with the essential bifurcation we have argued for between phonological
duplication and morphological reduplication. The ambiguous nature of CV
duplication is also consistent with evidence from language change. Niepokuj
(1997) has argued for four stages in the diachronic evolution of reduplication.
Stage 1 is total; in Stage 2, one copy is reduced. In Stage 3, reduplication
is affixal (= partial), and by Stage 4, the reduplicant is so reduced as to be
realized as gemination. Although Niepokuj does not have evidence from recon-
structions to support all aspects of this schema, she does cite two cases that
strongly support the diachronic path from Stage 3 CV prefixing reduplication
 The question of rhyme 203

to Stage 4 consonant gemination, possibly by way of CV prefixing reduplica-

tion (with a reduced, nonreduplicated vowel). A larger body of similar cases
is also discussed in Blevins (2004), in the context of the evolution of gemi-
nate consonants. Thus, for example, where Nez Perce (representing the more
conservative CV stage) forms distributives through CV prefixing reduplication
(7a), its sister language Sahaptin geminates the initial consonant (7b). Where
proto-Micronesian exhibited CV prefixing reduplication, modern Trukese gem-
inates initial consonants (7c) (Goodenough 1963, via Niepokuj 1997:63 and
Blevins 2004:169, 172). Bellonese exhibits synchronic CV ∼ V variation, asso-
ciating CV reduplication with a slower speech rate and C gemination with a
faster speech rate (7d) (Elbert 1988, via Blevins 2004:37, 45, 50).

(7) a. Nez Perce (simplex/distributive) tı́lu ti-tı́lu ‘big’ N34

b. Sahaptin (simplex/distributive) k’aiwá k-k’aiwá ‘short one’ N35
c. Proto-Micronesian > Trukese *ca-canu > ccen ‘watery/wet’
d. Bellonese (slow/fast) ␤a-␤aŋe ␤-␤aŋe ‘to play’
sa-saka s-saka ‘to beg’
ha-hatu h-hatu ‘to untie’

It stands to reason that if CV reduplication can change into Cə reduplication

or C gemination, there is likely to be a stage in which a part or all of the CV
reduplicant is analyzed as a phonological copy segment. Our assumption is still
that the null hypothesis, for duplication above the level of the segment, should
be morphological reduplication, but we leave open the possibility that in certain
cases where it is phonologically plausible, a phonological analysis may also be
considered. The analysis of phonological copying is, of course, possible only
when the duplication is proximal and involves single instances of phonological
structure types. Chaha CVC reduplication, discussed in Chapter 5, is potentially
a case of this kind. CVC roots double to flesh out a CVCCVC template. This
could be analyzed either as semantically vacuous double root selection, of the
kind discussed in §7.3, or as copy epenthesis (with the epenthetic onset matching
the existing onset feature for feature, and the same for the epenthetic coda; see
note 2). In this case there is no empirical evidence that could shed light on the
question.

7.5 The question of rhyme

The final criterion in (1) is phonological identity. It has been argued exten-
sively in this book that phonological identity is not a requirement in morpho-
logical reduplication. Phonological identity plays a direct role only in phono-
logical copying, although it may exert an auxiliary effect on elements in a
204 Final issues

morphological reduplicant that sit in close enough proximity to be subject to

surface correspondence of the type described in Walker 2000a; b; Hansson
2001; Rose & Walker 2001; Walker to appear.
This stance departs sharply from the position of Yip (1999), who, extending
observations by Kiparsky 1973b and Holtman 1996, has argued that redupli-
cation is formally identical to, and should be analyzed in the same terms as,
rhyme, alliteration, and related devices (such as those summarized in Fabb
2002). As Yip (1999) puts it, “humans have both an aptitude and a taste for cre-
ating repetitive sequences” (p. 4). For Yip, then, phonological surface identity
requirements of the type utilized in language art are also strongly ensconced in
language proper.
Allowing the grammar of a language to require morphological constituents
to alliterate, or rhyme, or exhibit assonance or consonance, goes far beyond the
usual assimilatory principles according to which phonetically similar things
become more similar, formalized as string-internal output correspondence in
Walker 2000a; b; Hansson 2001; Rose & Walker 2001; Walker to appear and, in
a somewhat different guise, by Zuraw 2002. Evidence that phonological agree-
ment can be imposed by fiat on any kind of word or pair of words, regardless of
their original phonetic similarity, would be evidence for the kind of phonolog-
ical correspondence imposed by BRCT and, therefore, evidence against MDT.
While alliteration and rhyme clearly seem to play a role in processing and
are aesthetically pleasing, the evidence suggests that rhyming, alliteration, and
other similar correspondences do not play a significant role in productive mor-
phology. If rhyming constraints were part of grammar, we would expect to
find phenomena like the following: suppletive root allomorphy whose distri-
bution is governed by rhyme with preceding prefixes or following suffixes;
suffixal allomorphs distributed according to alliteration with the stem-initial
consonant; words in which all primary or secondary stressed syllables must
alliterate; and so forth. Setting aside language games, this does not occur. The
cases Yip discusses are ambiguous between rhyme and reduplication. However,
unambiguous cases of rhyme, alliteration, assonance, and so forth are not found
in morphological derivation.
There are several exceptions that prove this rule by virtue of how extraordi-
nary they are and by virtue of the fact that none actually crosses the threshold
of rhyme, alliteration, etc., despite seeming to come close.
Rhyme sometimes plays a role in nickname formation, a morphological
phenomenon which can cross the boundary into language play, where rhyme
is known to figure prominently (see, for example, Bagemihl 1988; Weeda
1992). Watson (1966:79) describes nickname formation in Pacoh as follows: “A
 The question of rhyme 205

Pacoh nickname is formed by choosing two words which are commonly used
together . . . the second of which rhymes with the given name of the person
being nicknamed. The appropriate sex-title, ku ‘male’ or kán ‘female’ is then
chosen to precede the paired words and the given name.”

(8) Ku-tur kânsı́ng Kı́ng ‘male cobra’s back King’

Kán sêl târla Pa ‘female peel potatoes Pa’

This example, while compelling in its illustration of the aesthetic role rhyme
plays for humans, does not pass the substitution test: it would be very sur-
prising to find a pattern which phonologically resembled the Pacoh nickname
construction but instead formed, for example, agentive compounds. Our take on
the Pacoh construction is based on an observation made by Ourn and Haiman
2000, namely that in some cases, instances of a particular grammatical con-
struction that happen to rhyme are preferred stylistically to ones that do not.
Apte (1968), for example, describes a productive construction in Marathi
which couples words whose meanings are the opposite of each other and which
rhyme, e.g. jewha-tewha ‘every now and then, lit. when-then’ (p. 54). Alliter-
ation is common in synonym compounds, as well, e.g. in the Khmer examples
in (9) (from Ourn & Haiman 2000, discussed in Chapter 2); alliteration is not,
however, an absolute requirement of the construction.

(9) treek ‘enjoy’ + trɑʔaal ‘enjoy’ ‘enjoy’ 503

sruəl ‘easy’ + srɑnok ‘easy’ ‘easy’ 503
cuəp ‘meet’ + cum ‘meet’ ‘meet’ 502
proət ‘depart’ + praah ‘depart’ ‘depart’ 502

Rhyme can function as a stylistic filter on stylized constructions, e.g. choos-

ing the best nickname for a given individual. Rhyme should not be expected to
function as a filter on, or precondition for, truly general grammatical construc-
tions like noun-noun compounding, verb serialization, causativization, and so
forth.
The conclusion that rhyme and reduplication are not the same thing does
not, however, require the complete abandonment of Yip’s generalization that
the structural elements involved in versification (onsets, rimes, feet) are similar
to those involved in reduplication. As argued extensively in Kiparsky 1973b;
Hanson & Kiparsky 1996, inter alia, versification in any particular language is
intimately connected to the structures in use in that language’s grammar. The
connection between reduplication and rhyme is that both make use of the same
structural alphabet. It does not have to follow from this that both are regulated
by the same constraints.
206 Final issues

A truly grammatical phenomenon which comes close to the threshold

of rhyme, but significantly does not cross it, is so-called “alliterative con-
cord,” which Dobrin (1998) explicitly relates to reduplication (p. 60). Dobrin
(1998; 1999; see also Aronoff 1992; 1994) analyzes a system of agreement in
Arapeshan dialects originally described by Fortune (1942). According to
Dobrin, Arapeshan requires agreement, on the part of various syntactic ele-
ments, with a noun’s final consonant. Dobrin cites examples like the following:

(10) a. ulubun šaku-šaku-n b. anauwip-ih ñmineh

hardwood palm(6) small-small-6 six-13.pl day(13.pl)
‘small hardwood palm’ ‘six days’

In (10a), the modified noun ‘hardwood palm’ ends in the consonant n; the
agreeing adjective takes an n suffix as well. Were it modifying a noun ending
in a different consonant, it would end in the corresponding consonant. In (10b),
the head noun ‘days’ ends in h; the concord element in the preceding adjective
ends in the consonant h.
The concord elements in (10) are assigned numbers in morpheme glosses.
These numbers represent noun classes. Dobrin makes the important point that
the concord in Arapeshan is mediated, to a very high degree, by noun class,
as analyzed by Aronoff (1992; 1994). Following Fortune (1942), Aronoff and
Dobrin posit thirteen noun classes in Arapeshan. Masculine and feminine human
nouns are assigned to fixed classes (7 and 4, respectively), regardless of the final
consonant they end in. Most other nouns have class membership consistent with
their final consonant.

(11) Class Final phonological string of singular noun

I by ahoryby ‘knee’
II bør wabør ‘village’
III g aijag ‘leg’
IV ku babweku ‘grandmother’
V Vm daudam ‘spider’
VI n lawan ‘tree snake’
VII n ašuken ‘older brother to a man’
VIII ñ abotiñ ‘long yam’
V sumo ‘flying fox’
C pas ‘taro pounder’
IX pu ilupu ‘feast’
X r∼l awhijar ‘bat’
XI t nybat ‘dog’
XII uh∼uh baweuh ‘mountain road’
XIII ah ∼uh atah ‘ear’
 The question of rhyme 207

Aronoff and Dobrin present a variety of convincing reasons for thinking of

the agreement system as noun class agreement, rather than as a phonological
requirement that final consonants of nouns and their agreeing elements be iden-
tical. First, for masculine and feminine human nouns there is no consonance.
Masculines fall into class 7 regardless of final consonant; feminines fall into
class 4. Any phonological agreement pattern operating at the phrase or utter-
ance level would have to be sensitive not only to part of speech but also to
properties of gender and humanness, properties to which postlexical phono-
logical rules or constraints have not independently been shown to have access.
Second, some nouns are exceptional. The noun kwo
ar, though it ends in r and
would be expected to trigger class 10 concord, instead takes default class 8
concord (Dobrin 1998:65). This pattern is inexplicable phonologically but is
easily captured morphologically by prespecifying the noun root as morpholog-
ical class 8, instead of letting the grammar assign it to class 10 on phonological
principles.
Setting aside exceptional nouns and human nouns, there are still serious
barriers to a phonological account of the apparent consonance effect. First, the
size of the phonological string shared by agreeing elements varies considerably.
In some cases it is a single consonant; in other cases it is a syllable, e.g. bør
(class 2) or a syllable rime, e.g. Vm (class 5). No single statement of phonological
identity could cover all of these cases. Second, the phonological position of the
agreeing string does not fall in the same phonological location in the various
types of words with which a noun agrees. Dobrin, citing Fortune, provides a
templatic description of all the concord elements in Arapeshan, in which “C”
stands for the constant phonological string. For example, the possessor noun
suffix is -iC with final “C,” e.g. polisipepimin-it musket ‘the policeman’s rifle,’
eguh-ib tinab ‘cans of fish’ (Dobrin 1998:66). The demonstrative pronoun used
in ‘near me’ contexts, aCudaʔ, has a medial “C,” e.g. aguda dybarig amwi-eg
‘whose garden is this?’ (Dobrin 1998:63). In yet other cases, e.g. the adjective
ending -Cali, the “C” is morpheme-initial, as shown in (12), an example which
illustrates the wide range of phonological positions in which the agreeing “C”
is found (Dobrin 1998:67).

(12) əgdak nebe-g-ali trag

this(3) big-3-lasting quality truck(3)
‘this big truck’

Again, no single phonological identity statement could cover all of these

cases.
208 Final issues

A third phonological barrier to a phonological agreement analysis, and an

argument in support of a noun class analysis, is that vowel-final nouns pattern
with nouns ending in ñ in triggering ñ-based concord, e.g. taia ñ-etemu əmañ
‘tire(8) 8-be heavy-8 = the tire was heavy’ (Dobrin 1998:67). If agreement were
essentially phonological, we would instead expect to find a vowel, or simply
nothing, in place of “C” in the words agreeing with vowel-final nouns. ñ-based
concord makes no sense in terms of phonological identity, but it makes perfect
sense in terms of morphological concord, as Dobrin herself observes. Class 8
is the default class; it happens to be instantiated, on agreeing elements, by ñ.
As Aronoff and Dobrin make clear, a noun class system is independently
needed anyway in order to capture generalizations about the relationship
between the singular and plural form of any given noun. The plural form of any
given noun is generally predictable from the singular, though plurals and singu-
lars do not agree phonologically. Rather, there is simply a set of often arbitrary
correspondences between the endings of singulars and plurals of the sort that
noun class is designed to capture. Class 1 singular nouns, which end in by , end in
bys in the plural: ahoryby ∼ ahorybys ‘knee(s)’; class 2 singular nouns, which
end in bør, end in ryb in the plural: wabør ∼ waryb ‘village(s).’ The correspon-
dences can be quite arbitrary, as the class 2 forms show; moreover, some noun
classes admit a variety of plural forms. Thus class 3, whose singulars all end in
g, have two options for plurals: s and gas, which Aronoff allocates to the sub-
classes 3a and 3b: for each singular noun, which subclass of 3 it belongs to must
be learned, e.g. aijag ∼ aijas ‘leg(s)’ (class 3a); nubarig ∼ nubarigas‘garden(s)’
(Class 3b). For some other classes there is an even greater number of
subclasses.
What gives the system the appearance of grammatically enforced “allitera-
tion” (for which “consonance” would be a more accurate term) is, primarily,
the fact that the set of agreement markers for any given noun class tends to
intersect in the phonological string ending singular nouns of that class. This
intersection presumably has a historical explanation of the kind that has been
offered for similar patterns in Bantu languages, in which the use of a particular
head-modifying morpheme was extended to other syntactic elements, resulting
in agreement. Dobrin provides two arguments that the phonological conso-
nance generalization, while overridable, should nonetheless be a part of the
synchronic grammar of Arapeshan. The first argument is that borrowed nouns
ending in a consonant not found finally in native nouns have spawned a new
agreement pattern. Borrowings from Tok Pisin which end in s cannot (or could
not, originally) simply be slotted into an existing class for s-final nouns, as
none occur natively in Arapeshan. What has happened is that such nouns take
 The question of rhyme 209

s-agreement. The “C” position in words agreeing with s-final nouns is instan-
tiated with s.

(13) ənan-is kes Dobrin 1998:68

pro.7-poss suitcase
‘his suitcase’

But this pattern does not demand a phonological agreement analysis. An

equally viable approach would be to view the pattern as the result of morpho-
logical analogy, or backformation, which occurred in the recent history of the
language, setting up a new noun class and set of associated concord markers
along the lines of those for existing classes.
Dobrin’s second argument that consonance is phonologically enforced comes
from an alternation specific to class 3 nouns and agreeing elements. The class
3 plural ending, s, optionally alternates with ʔ when word-final. According to
Dobrin, “the glottal stop does not appear in place of s on agreeing elements
unless it appears in place of s on the noun; nor [does s appear] on final agreeing
elements when the noun itself is pluralized with glottal stop.” Intriguing as
this is, it still supports a noun class analysis. Class 3a has two subclasses, call
them 3ai and 3aii, in either of which a class 3 noun may be realized. If a noun is
optionally realized as 3ai, it must take the 3ai singular ending and 3ai agreement;
likewise for 3aii. Crucially, as Dobrin points out, only class 3a s behaves in
this manner; the plural s of other classes do not. This is strong evidence that
noun class, not phonology, plays the central role. If Arapeshan really were
using consonance in the way it is used in versification, all nouns ending in s
would trigger s-consonantism in their modifiers; s-consonantism would be in
a fixed location, e.g. initial or final. This is not the case. By coming close to
but not achieving the description of consonance, Arapeshan is an exception
which proves the rule that rhyme, alliteration, consonance, and assonance are
not viable grammatical principles.
As mentioned above, Kiparsky has often made the point that the connec-
tion between rhyme and reduplication is their mutual use of the phonological
structures that the language independently makes available (Kiparsky 1973b;
Hanson & Kiparsky 1996). This is not, however, license to equate the two
phenomena. Reduplication is morphological, and rhyme is phonological; close
inspection shows their manifestations to be sharply different. In a larger context,
this is the same point made in Chapter 5 about morphological reduplication vs.
phonological duplication. Though there is a small set of data for which the two
analyses converge, in general the criteria for morphological vs. phonological
identity are sharply different, and a good theory must treat them separately.
210 Final issues

7.6 The question of anti-identity

Anti-identity effects are just as relevant as identity effects to the question of
whether grammars can enforce phonological identity across morphemes or
words. Any grammar with the power to compare phonological strings across
morphemes and words has the power to enforce either identity or anti-identity.
It is our view that anti-identity is no more integral in morphology than is iden-
tity; anti-identity effects do not support the need for generalized phonological
correspondence between morphological constituents.
One apparent type of anti-identity effect in morphology is classified under
the Repeated Morph Constraint (Menn & MacWhinney 1984), a ban on the
occurrence of consecutive homophonous morphemes. It is true that some such
sequences are banned in particular languages. However, the problem with
attributing such bans to anti-identity is that languages also often ban the occur-
rence of adjacent morphemes which are not homophonous, and it is far from
clear whether the bans on homophonous morphemes have any special status.
Turkish highlights this issue particularly dramatically. In Turkish, the third
possessive suffix (e.g. oda-sı ‘room-3poss’) and the suffix marking productive
compounding (yemek oda-sı ‘dining room-sfx’) are homophonous: following
a vowel, both surface as -sI (with a harmonic high vowel). Compound nouns
can be possessed. However, it is not possible to have two -sI suffixes in a row;
the possessed form of ‘dining room’ is homophonous with the nonpossessed
form: yemek oda-sı. The doubly marked *yemek oda-sı-sı is ungrammatical.
This might seem to be an ironclad case of the RMC. However, looking at the
example in the broader context of Turkish morphology shows that the gen-
eralization is not, in fact, anti-identity-driven. The first piece of evidence is
that both the compound marker and the third person possessive suffix have
two suppletive allomorphs, -sI and -I, which occur after vowels and conso-
nants, respectively (thus oda-sı ‘his/her/its room’, okul-u ‘his/her/its school’).4
A compound ending in -I cannot take a third person possessive -sI, and vice
versa; thus, even when identity is not at issue, forms containing both suffixes
in sequence are ungrammatical (kılıç balıǧ-ı ‘sword fish’ = ‘his/her/its sword
fish,’ *kılıç balıǧ-ı-sı ‘his/her/its sword fish’). Even more significantly, the ban
on the third possessive suffix following compound-marking -sI∼-I is only a
small part of a larger picture: Turkish also prevents the compound-marking
suffix from co-occurring with any other possessive suffixes; thus ‘my dining
room’ is yemek oda-m, not *yemek-oda-sı-m. This ban cannot be attributed
to the RMC. Further afield, even certain affixes which are altogether outside
the possessive paradigm are unable to co-occur with compound markers. One
 The question of anti-identity 211

example is the ‘with’ suffix /-lI/: yemek-oda-lı ‘endowed with a dining room,’
not *yemek oda-sı-lı. What these facts show is that the constraint barring *yemek
oda-sı-sı cannot be a prohibition on adjacent identical morphs; it is something
more arbitrary and unrelated to phonological form. Given that languages can
ban arbitrary sequences of morphemes (as position class languages attest in the
extreme), careful statistical analysis would have to be undertaken to show that
the RMC is not an illusion, the result of linguists paying attention to morpheme
co-occurrence constraints when the morphemes are identical, and not paying
as much attention otherwise. Another window onto the psychological reality
of the RMC could be reanalysis; if languages that used to allow sequences of
identical morphs start banning them, it could be that the RMC has reared its
head. According to Yoquelet (2004), however, synchronic instances of what
appears to be the RMC tend strongly, if not universally, to derive historically
from morphemes that have acquired multiple functions but retain the distribu-
tional properties of their shared ancestor. If Yoquelet is right, then the RMC is
not a real grammatical constraint even diachronically.
Anti-identity effects have been claimed to hold elsewhere in morphology
as well. Kurisu (2001) argues that much nonconcatenative morphology can be
reduced to an anti-identity requirement holding between words, where one word
is the morphological base of the other. If no overt affixation marks the derived
word, anti-identity requires some other modification to take place. Kurisu pro-
poses that the ranking of anti-identity requirements in the markedness hierarchy
determines what modification will be performed: truncation, ablaut, metathesis,
stress shift, etc. Even if we set aside the cases in which homophony does occur
in paradigms, however, Kurisu’s approach cannot account for cases in which
truncation, ablaut and other effects co-occur with overt affixation which itself is
sufficient to ensure non-homophony between related words. Further, Kurisu’s
approach incorrectly predicts that words in which anti-identity forces modifica-
tion should differ minimally, in being less marked along some single dimension,
from their nonderived counterparts. In any event, all of Kurisu’s data can be
handled by input–output dissimilation; it is universally recognized in generative
phonology (including Optimality Theory) that there is phonological correspon-
dence between inputs and outputs.
It is beyond the scope of this chapter to review all of the claims in the
literature to the effect that the synchronic grammar acts to avoid identity across
related words, a topic that has recently come into focus (see, for example,
Crosswhite 1997; 1999; Suzuki 1999). We can only suggest that such claims
be strongly scrutinized.5 Opening the door to phonological correspondence,
whether used to enforce or to prohibit identity, endows the theory with a great
212 Final issues

deal more descriptive power, much of which is clearly unnecessary, e.g. the
ability to describe overapplication of junctural effects (see Chapter 5, §5.5.2),
backcopying (see Chapter 5, §5.8), and other unattested phenomena.

7.7 Beyond reduplication

The natural next step for the MDT approach to reduplication, a construction
whose daughters are required to be semantically identical, would be to extend
the approach to morphological and syntactic constructions whose daughters are
in a different, but equally obligatory, semantic relationship. Chapter 4 opened
the door to this possibility by observing that reduplication forms a natural class
with antonym constructions.
One clear area for future research, which our framework predicts possible,
is that a construction could require sisters to be in a semantic or morphological
subset relationship, where one sister is more specific than the other in some
way. For example, the formation of reciprocal verbs in Malay calls for two
verb stems, one consisting of the bare verb root and the other prefixed with
meng-, a prefix which marks agent focus on transitive verbs and patient focus
on intransitives (Mintz 1994:130–32). Data are from Mintz 1994:275:
(14) tulis-menulis ‘to write to one another’
tolong-menolong ‘to help one another’
pandang-memandang ‘to observe one another’

A neutral description of this construction would be that the second daughter

is specified for thematic role focus, while the first is unspecified. This kind of
relationship, where one copy of the relevant constituent is unspecified relative
to the other, is probably very common; indeed, a number of the examples of
syntactic doubling discussed in Chapter 1 were of this kind. Noun incorpora-
tion is a likely area in which to find more. So-called “classifier incorporation”
(Mithun 1984; Rosen 1989; Mithun 2000) can involve two copies of a head
noun, one, which is uninflected for case and number, in the verb, and the other,
fully inflected, heading the corresponding noun phrase in the syntax.
We have argued in this book that reduplicative phonology is best under-
stood in the larger context of morphologically conditioned phonology. The
same principle applies to reduplication as a morphological construction. Further
investigation into parallelism in grammar is sure to shed further light on the
small corner of morphology we have investigated here.
 Notes

1 Introduction
1. Weber (1989) characterizes the meaning of adverbial clause reduplication as “rep-
etition of the event referred to by the reduplicated element” (p. 323).
2. The Hausa data are taken from Newman 2000 (p. 424); the Amele data are from
Roberts 1991 (p. 135), and the Warlpiri data are based on Nash 1986 (p. 130).
3. On the quality of the prefixal vowel, see Akinlabi 1993; Pulleyblank 1998; Alderete
et al. 1999.
4. Other languages cited by Landau as having similar verb copying effects include
Haitian, Vata, Yoruba, Brazilian Portuguese, Yiddish, and Russian.
5. See Raimy 2000; Frampton 2003 for a phonological approach to reduplication that
does not invoke a Red morpheme.
6. One exception is the literature on a-templatic reduplication, which offers criteria
for distinguishing templatic from a-templatic morphological reduplication. See, for
example, Gafos 1998b; Urbanczyk 1998; Hendricks 1999, as well as Zuraw 2002.
We return to this issue in Chapter 7.
7. Following McCarthy and Prince 1995a, this conviction has been bolstered by
renewed attention to apparent identity effects described by Wilbur 1973; Moravcsik
1978; Marantz 1982; Clements 1985; Kiparsky 1986; Mester 1986; Steriade 1988;
McCarthy and Prince 1993; 1995a; 1999b, and elsewhere. These will be discussed
at length in subsequent chapters.
8. Chechen data are presented in the practical orthography developed by Johanna
Nichols; for more information, see http://socrates.berkeley.edu/∼chechen.
9. See Chapter 2 for a fuller discussion of Sye and other cases of divergent allomorphy.
The same general phenomenon (in Sanskrit and other languages) is also discussed
in Saperstein 1997.
10. Blust (2001) argues, with examples from Thao (Formosan), that triplication often
does have an iconic function of intensifying whatever meaning is associated with
ordinary reduplication.
11. For recent overviews of Optimality Theory, see Kager 1999; McCarthy 2002b. It is
important to note that the use of Optimality Theory in this work differs from standard
implementations in two ways: (a) some practitioners of Optimality Theory reject the
use of cophonologies, though all admit morphologically conditioned phonology in
some guise, and (b) the Base-Reduplicant Correspondence Theory (BRCT) devel-
oped for Optimality Theory by McCarthy and Prince 1995a is not used here; MDT

213
214 Notes to pages 21–42

is intended as an alternative to BRCT. Cophonologies are highlighted in Chapters

3 and 4; BRCT is addressed most directly in Chapters 4, 5, and 6.
12. These roots come from forms with underlying schwa, which has undergone a process
of unstressed vowel deletion. See Black 1996 for discussion.

2 Evidence for morphological doubling

1. Piñon demonstrates that reduplicated preverbs behave differently from conjoined
preverbs in respect to the word orders allowed in sentences containing them and
suggests that preverb reduplication may occur in the syntax. If so, then Hungar-
ian provides further evidence for the kind of syntactic reduplication discussed in
Chapter 1.
2. Roberts (1987) and Roberts (1991) use slightly different transcription systems for
Amele; the forms cited here reflect these differences. In the forms taken from Roberts
1987, “c” = [ʔ] and “q” = gb .
3. Some roots, instead of reduplicating, lengthen a following vowel instead. These are
not shown here; see Roberts 1987:254–55, Roberts 1991 for discussion.
4. Double reduplication of affix and root, with the same overall meaning as reduplica-
tion of just one of those constituents, would appear to violate general principles of
morphological economy; see, for example, §2.2.1.
5. Compare to the distinction drawn by Bloomfield (1933) (and also, more recently,
Saperstein 1997), who associated partial reduplication with affixation and total
reduplication with compounding.
6. GTT was developed not so much to capture generalizations about root and affix size
as to avoid the necessity to directly stipulate the prosodic shape of any given redu-
plicant. The reason for avoiding such statements is the so-called “Kager–Hamilton”
problem, discussed in more detail in Chapter 3.
7. We are grateful to Galen Sibanda for discussion of Ndebele; his dissertation paints
a much fuller picture of all of the phenomena discussed here.
8. For further discussion of this suffix, see Chapter 7. We are grateful to Orhan Orgun
for help with our understanding of Turkish phonology and morphology.
9. Rice and McDonough analyze the inserted material as phonologically epenthetic,
rather than as constituting a lexically listed empty morph, the approach we have
taken to such data in this chapter. It is not always easy to decide whether a given
segment-zero alternation is morphologically conditioned epenthesis or due to the
presence vs. absence of a semantically empty, lexically listed morph. The fact that
Ndebele has two distinct phonological strings alternating with zero supports an
empty morph analysis for at least one of them.
10. Our approach also has important points of contact with the Enriched Input theory of
Sprouse 1997, originally designed to handle phonological opacity. Enriched Input
theory provides competing phonological inputs based on a single underlying form.
11. Srichampa 2002, citing various data sources, describes a number of other rhyme-
replacing reduplicative constructions in Vietnamese; this particular one is not men-
tioned.
12. Transcriptions are converted, using Bosson’s romanization (p. 11), from the Cyrillic.
 Notes to pages 44–75 215

13. See Saperstein 1997 for an earlier proposal which allows the two daughters in
reduplication to differ in stem type.
14. Steriade (1988:76), citing Wiesemann 1972, describes a similar process in Kaingang,
where pluralization can be marked by g-insertion, raising a penultimate vowel,
and/or reduplication.
15. -eʔ and -oʔ are infinitive suffixes; -doʔ is a combination of object marker (-do-) and
infinitive suffix. See Roberts 1987:272, 281ff. for discussion of the generalizations
over which vowels will be replaced by which other vowels.
16. Tier replacement is the analysis given in Inkelas 1998 of “consonant extraction”
in Modern Hebrew (see, for example, Bat-El 1994 and references therein) and is
also arguably involved in the well-known Javanese habitual-repetitive construction,
about which Yip has written extensively (see, for example, Yip 1997; 1998), and
in which a particular vowel in one of the copies is replaced; thus eliŋ → elaŋ-eliŋ
‘remember,’ tuku → tuka-tuku ‘buy,’ udan → udan-uden ‘rain’ (Yip 1997).
17. For an analysis in this same spirit of Sanskrit reduplication in terms of different
stem allomorphs, see Saperstein 1997.
18. Crowley notes (p. 199) that verb roots beginning with t and k exhibit allomorphy
only in the active, not in the stative.
19. We are grateful to Terry Crowley for discussing the Sye data with us.
20. Crowley (1982) notes (p. 155) that for verbs which participate in both CV- and
CVCV- reduplication, it is typical for CV- reduplication to be associated with
habitual/random or habitual/uninterrupted semantics, and for the CVCV- variant to
be associated with detransitivizing or habitual/interrupted meanings. However, this
association does not hold for verbs which reduplicate in only one way.
21. Singh calls these “partial reduplication”; however, as no truncation appears to be
associated with the process, we use the more standard term of “echo reduplica-
tion.” This particular pattern of echo reduplication, in which the onset of the sec-
ond member is replaced by a labial, occurs throughout Asia; see Abbi 1991 for
a typological overview, and §2.2.4 for discussion of echo formation as Melodic
Overwriting.
22. See also Malkiel’s (1959) discussion of the semantics of irreversible binomials.

3 Reduplicative phonology: daughters

1. Donca Steriade, in unpublished work, has recently argued that there are outside-in
effects in which the phonology of derived words can influence, synchronically, the
phonology of words serving as their bases; similarly, Bochner (1992) has argued for
treating morphological backformation on a par with other morphological derivation.
We agree that sporadic outside-in and backformation phenomena exist but, following
Kiparsky 1982c, view them as diachronic events, rather than as patterns regulated
by synchronic grammar.
2. Some verbal or nominal suffixes impose different patterns, such that only a sub-
set of verb stem-forming affixes are in fact associated with the cophonology in
(9a); similarly, only a subset of noun stem-forming suffixes are associated with the
cophonology in (9b). The fact that we are simplifying the picture here should not
216 Notes to pages 76–89

detract from our basic observation that different morphological constructions can
be associated with different cophonologies.
3. Some reduplication constructions combine bare morphemes, not derived stems, and
thus do not have cophonologies associated with the daughters; see especially the
discussion of affix reduplication in Chapter 2. The construction in (11) should be
understood as the maximally complex reduplication construction.
4. Only full-grade forms are shown here. Zero-grade allomorphs of the root may
exhibit other differences between the two copies, which Steriade (1988) argues to
be a consequence of differences in the syllabification of full vs. zero grades. See
also Saperstein 1997 for an analysis more in line with MDT’s double stem selection
approach.
5. Transcriptions are based on Weeda 1992: 162–63. “HBC” stands for the Hooper
Bay/Chevak dialect; other forms are from the General Central Alaskan Yupik com-
plex of dialects.
6. Alderete (1999; 2001) invokes input–output anti-identity constraints to drive accen-
tual dominance effects, but the phenomenon he is analyzing is one of neutralization
(accent deletion), rather than dissimilation proper.
7. Hausa data are presented in Hausa orthography, with three exceptions: vowel length
is marked with a colon rather than a macron; following Newman 2000, tapped /r̃/
is marked with a tilde to distinguish it from retroflex /r/; following Hausa linguistic
tradition and Newman 2000, Low tone is marked with a grave accent (High is
unmarked).
8. The coronals /z/ and /d/ are related via coronal palatalization, which applies before
front vowels.
9. Truncation and vowel raising do not occur in polysyllabic bases. Niepokuj
(1997:24ff.) takes this as evidence that monosyllabic and polysyllabic reduplication
are different morphological constructions (prefixing and suffixing, respectively).
10. A possible alternative analysis, after Downing 2004, might be to analyze the Tarok
pattern as the result of the base melody being mapped, via the regular left-to-right
association algorithm of the language, to the prosodic word which contains both
base and reduplicant. However, for monosyllabic Low-toned bases the tone pattern
in reduplication is Low-Mid; according to Sibomana 1980:201, Low-Mid is not a
legal result of regular tone distribution.
11. In response to similar ranking paradoxes in reduplication in other languages (e.g.
Kwakwala and Lushootseed), Struijke (2000a; 2000b) has proposed Existential
Faithfulness (∃-Faith), which differs from standard input–output faithfulness in
being satisfied as long as input structures have correspondents anywhere in the
output; they could be in the reduplicant or in the base. While ∃-Faith would permit
BRCT to describe the Tarok facts, it is argued on other grounds in Chapter 4 that
∃-Faith is not a generally viable approach to reduplicative phonology.
12. The adoption of reduplicant-specific markedness constraints, e.g. *V]red , might
seem to be another alternative, but it is also essentially identical to introducing
cophonologies and just as incompatible with the foundational assumptions of BRCT.
13. In any case, nothing in BRCT, even with GTT, could rule out the reduplication of,
for example, a truncated nickname (e.g. Jonathan → Jon → Jon-Jon), which is
 Notes to pages 91–101 217

surface identical to directly truncating reduplicant and base (Jonathan → Jon-Jon).

MDT predicts both scenarios to be possible.
14. Our source for the Stanford truncations is our personal knowledge as well as the
website http://www.stanford.edu/dept/undergrad/uac/freshmen/glossary.htm. Our
source for the English truncations is the website http://ccat.sas.upenn.edu/plc/echo.
15. Healey, Isoroembo, and Chittleborough (1969:35) describe the semantics of redu-
plication as follows: “The stem of a verb may be modified to indicate a repetitive
action or (if transitive) a plural object by the use of a reduplicative prefix or a suffix
or both.”
16. On infixing reduplication in MDT see Chapters 5 and 6.
17. One possible example of base-dependence occurs in Ponapean, where the prosodic
quantity of reduplicants and bases appears to be (inversely) correlated; see McCarthy
and Prince 1999b for an overview. The facts are more complicated than this, but
in any case any sort of prosodic relationship between reduplicant and base could
also be analyzed as a relationship between the input and output forms of each
daughter.
18. A minority pattern is to reduplicate both vowels: daona ‘long (verb)’ → dao-daona
‘long (adj.),’ gao ‘gap’ → gao-gao ‘week.’ Note the semantic irregularity of these
forms; this set also includes apparently lexically reduplicated roots like woe-woe
‘to paddle’ (Ezard 1997:42).

4 Reduplicative phonology: mothers

1. See §4.4 below for more discussion of Lushootseed reduplication. The transcription
system used here is that of Urbanczyk 1996; “X” represents a uvular fricative.
2. The presentation here is highly simplified for purposes of illustration. For a full
discussion of the morphophonemics of Japanese accent, see, for example, McCawley
1968; Poser 1984.
3. For similar reasons, Kim (2003) proposes construction-specific constraints to
explain suffix-specific requirements on the shape of their base in Nuu-chah-nulth.
4. On cophonologies, see, for example, Itô and Mester 1995b; 1996b; Mester and
Itô 1995; Orgun 1996; Inkelas, Orgun, and Zoll 1997; Orgun 1997; Inkelas 1998;
Inkelas and Orgun 1998; Orgun 1999; Yu 2000; on morphologically specific output–
output correspondence constraints, see Downing 1998a, 1998b; also Downing 1997;
1999a,b; 2000; 2001; on indexed constraints, see, for example, Benua 1997; Smith
1997; Alderete 1999; Itô and Mester 1999; Alderete 2001; on affix-specific corre-
spondence relations, see Alderete 1999; 2001. A detailed comparision of constraint
indexation and cophonologies can be found in Inkelas and Zoll 2003; the issue is
also discussed in Chapter 3.
5. Dakota allows a variety of onset clusters. See Patterson 1990:67 for details.
6. The absence of the final vowel is not a special property of the reduplication con-
struction. The final vowel fails to appear whenever a consonant-final stem is the first
member of a lexical compound or whenever a consonant-final noun is incorporated
into a verb (Shaw 1980:119). Notice that this vowel is also not subject to stress
(Shaw 1980:32).
218 Notes to pages 101–13

7. Since dissimilation occurs only at the juncture, a full account of this effect in any
framework must include a mechanism to restrict the alternation to derived environ-
ments. On derived environment effects, see Mascaró 1976; 1982b; 1993 and, more
recently, Burzio 1997; Lubowicz 1999; Inkelas 2000; Anttila in press; Cho in press.
8. A potential alternative to this ranking paradox, ∃-Faith, is discussed and rejected
in §4.4.
9. Cohn and McCarthy’s (1998) more detailed account of Indonesian stress utilizes
the following constraint on the relationship between two PrWds (after Chomsky &
Halle 1968) to force stress subordination in compounds:
Nuc-Str
In PrWd1 PrWd2 , PrWd2 is more prominent
A general Culminativity constraint is sufficient for the argument here, but see
their paper for justification of Nuc-Str.
10. Kenstowicz (1995) and Cohn and McCarthy (1998) use a comprehensive BR-Max
constraint.
11. One might wish to account for the seeming typological oddity of the opposite pattern,
whereby stress demotion would occur in reduplication but not in compounds. At
present, however, its existence is predicted possible by any theory, including BRCT,
which could derive it as a standard TETU effect by changing the relative ranking of
BR-Faith and IO-Faith with respect to markedness. Ruling it out would require
as yet undiscovered meta-constraints on constraint indexation/reranking.
12. Another potential contributing factor to underapplication is opaque interactions
between daughter and mother cophonologies; we discuss a case from Javanese in
Chapter 5.
13. “TETU” stands for “The Emergence of the Unmarked,” the phenomenon whereby
a low-ranking markedness constraint M can have an effect when a higher-ranking
constraint, obedience to which normally prevents satisfaction of M, is itself nec-
essarily violated in some context, permitting satisfaction of M. See McCarthy and
Prince 1994a; Alderete et al. 1999.
14. The candidate bukú-bukú-ña, where stress shifts in both Red and Base, is ruled out
by a constraint favoring initial stress. See Kenstowicz 1995 and Cohn and McCarthy
1998 for details.
15. Not all cases previously described as underapplication will yield to the same analysis
as Klamath. See Chapter 5 for cases of underapplication in Javanese which have a
different origin.
16. This process, described in detail by Barker 1963; Barker 1964, has been discussed
extensively in the linguistics literature; see, for example, Kisseberth 1972; Kean
1973; White 1973; Thomas 1975; Feinstein and Vago 1981; Clements and Keyser
1983, inter alia.
17. Transcriptions of Klamath throughout this discussion follow Barker, except for
phonetically reduced symbols in brackets; uppercase letters represent voiceless
sonorants.
18. The “Intensive” has a variety of meanings. Barker (1964: 120) states that it can
signify repeated action, such as “twinkling, flickering, trembling,” or “emotional
and bodily conditions which persist: e.g., being nauseated, feeling bad, burning (in
taste), etc.”
 Notes to pages 118–37 219

19. See Raimy 2000 for discussion of Southern Paiute, cited in McCarthy and
Prince (1995a) as exhibiting identity-induced underapplication. As Raimy and,
independently, Gurevich (2000) demonstrate, consideration of additional data from
Sapir 1930 shows the effect to be nonexistent.
20. “Reduce” stands for the constraint(s) driving reduction; see Urbanczyk 1996;
Struijke 2000b for a detailed Optimality Theory account of the specifics of vowel
reduction.
21. Hausa data are presented in Hausa orthography, with three exceptions: vowel length
is marked with a colon rather than a macron; following Newman 2000, tapped /r̃/
is marked with a tilde to distinguish it from retroflex /r/; following Hausa linguistic
tradition and Newman 2000, Low tone is marked with a grave accent (High is
unmarked).
22. See the detailed discussion of Lushootseed in Park 2000 as well.
23. On derived environment effects, see Mascaró 1976; 1982b; 1993 and, more recently,
Burzio 1997; Lubowicz 1999; Inkelas 2000; Anttila in press; Cho in press.
24. Fitzgerald (2002), a response to Yu 2000, focuses on theoretical objections to
cophonologies; for discussion of the issues Fitzgerald raises, see Chapter 3 and
Inkelas and Zoll 2003.
25. CVCV reduplication is attested in nouns, adjectives, and verbs; in the case of verbs,
it is intensifying (Corston-Oliver 2002:470, 483).
26. Prefixes also attract stress. Corston-Oliver states that roots of three or more
syllables are stressed on the first and second syllables, an observation which
suggest that “stress” in Roviana may either consist of, or be accompanied by,
tone.
27. The suffix -ana derives instrumental or locational nouns from verbs; it is always
accompanied by verb root reduplication (Corston-Oliver 2002:472). Otherwise, verb
root reduplication “indicates intensive affect” (p. 480). See Chapter 7 for further
discussion.
28. For discussion of compound reduction in another Tibeto-Burman language, see
Nagaraja 1979 on Khasi.
29. In pluractional CVG verbal reduplication (G = gemination of following conso-
nant), the reduplicant agrees in tone with the following syllable; however, in CVG-
reduplication deriving adjectives of sensory quality, the reduplicant is uniformly L
toned, e.g. zurfi ‘heat’ → zùz-zurfa ‘very deep’ (Newman 2000:511). Both patterns
are non-tone-integrating.

5 Morphologically driven opacity in reduplication

1. Javanese has a ten-vowel surface inventory (i, i, e, ε, ə, a, ɔ, o, υ, u; see, for example,
Dudas 1976:5). Transcriptions of Javanese consonants are orthographic. ‘tj’ = [ts];
‘dj’ = [dz]; ‘q’ = [ʔ], ‘nj’ = [ ], ‘ng’ = [ŋ]; ‘j’ = [j]. Conversions are as follows;
‘t’ and ‘d’ are dental; ‘t.’ and ‘d.’ are alveolar.
2. The issue of whether the /a/∼/ɔ/ alternation is one of /a/-raising or /ɔ/-lowering
is extensively discussed by Dudas (1976). Although we follow Dudas in adopt-
ing a raising analysis, we note that the opacity arguments would be similar if the
alternation were reanalyzed as /ɔ/ lowering.
220 Notes to pages 140–51

3. A parallel situation occurs in Warlpiri, as documented by Nash (1986). Verb redu-

plication, in which the first copy is truncated to the first two syllables of the verb,
lends “speed, vigor or distributivity” to the meaning of the verb (p. 136). Nash shows
that inflectional suffixes surface in the first copy only when they are needed to sat-
isfy disyllabicity, thus yirra-ka ‘put-imp’ → yirra-yirra-ka (p. 136) vs. pu-ngka
‘hit-imp’ → pungka-pungka (p. 137). Warlpiri has a process of anticipatory vowel
harmony by which root /i/ vowels assimilate to suffixal /u/. This process affects both
syllables of reduplicated disyllabic verb roots even though the triggering inflectional
suffix surfaces only after the second copy, thus, for example, pangi-pangi-rni ‘dig
(reduplicated)-npast’ vs. pangu-pangu-rnu ‘dig (reduplicated)-past’ (p. 142).
The opaque harmony in the first copy follows naturally from an account in which
reduplication targets suffixed stems; the suffix is truncated in most cases by the
general process reducing the first copy to two syllables.
4. A few roots retain their final vowel before -an, e.g. gawe-an ‘work, activity’
(H123).
5. Horne (1961:127) documents double -(a)n suffixation in some cases of reduplica-
tion, e.g. aŋgɔ-n-aŋgɔ-n-an (nominalization of aŋgɔ ‘use, wear’).
6. Sumukti (1971:98) cites two intriguing examples in which the Nominalizer shows
up (in its [-n] form) on both vowel-final members of compounds: kakɔ-n + atε-n
‘hot tempered,’ cf. kaku ‘stiff,’ ati ‘heart’; and gugɔ-n + tuhɔ-n ‘superstitious,’ cf.
gugu ‘to obey,’ tuhu ‘earnest.’ Although it is not clear to us whether semantically
nominalization would be expected to occur before or after compounding, these data
do suggest that the Nominalizer can be present in the input to compounding, just as
it can in the input to reduplication. Phonological identity is an unlikely explanation
for double nominalization within compounds.
7. Dudas and Horne differ in their characterization of liquid-initial stems; according
to Horne, the active prefix is nge- or me- in such cases, while according to Dudas,
it is nge- or ng-. Horne (1961:103) reports that monosyllabic roots take a syllabic
allomorph of the prefix (nga- or nge-).
8. Cole’s analysis is more like the one sketched here, in that for Cole, the Pstem is what
reduplicates. For Downing, such conscription represents a misalignment between
reduplicant and prosodic structure.
9. This could be seen as an anti-opacity effect; since causative allomorphy is sensitive
to whether the root ends in a consonant or a vowel, /h/-deletion would render
allomorph selection opaque.
10. Horne (1961) provides the same form, with glosses, but transcribes the vowels
differently (p. 71): d.ajoh ‘guest,’ d.ajoh-d.ajoh ‘guests,’ d.aju-d.aju-e ‘the guests.’
11. We have no explanation for the discrepancy in the medial consonants in the two
copies in this form; it may be an aberration in the data, but in any case does not
affect the issue at hand.
12. Stressed CV reduplication serves a variety of functions, including nominalization
and the marking of continuative aspect. Final CV reduplication intensifies adjec-
tives, negatives, and directional adverbs (Topping 1973:183). Topping marks only
irregular stress; in addition, regular penultimate stress is marked on all relevant
forms here.
 Notes to pages 152–79 221

13. Klein (1997) proposes a BRCT analysis of final CV reduplication in which the
opaque antepenultimate stress pattern is attributed to BR identity; penultimate stress,
on Klein’s analysis, would make base and reduplicant metrically unequal. As dis-
cussed in Chapter 4, however, Klein’s analysis founders once the facts of stressed
CV reduplication are considered.
14. Lexical items are taken from sentences in Hurlbut 1988, who provides interlinear
glosses for morphemes but not for words; therefore word glosses are not attempted
here.
15. Hurlbut 1988 does not gloss the prefix to-; however, she mentions the existence
of a root ruol (p. 60) and cites, in the glossary, the forms o-ruol ‘be painful’ and
poko-ruol ‘curse, lit. cause pain’ (p. 133).
16. The counterexamples to this generalization all involve single-segment reduplication,
which, as argued in Chapter 7, falls under the category of phonological duplication,
rather than morphological reduplication.
17. See Kiparsky 2000:9 for critical discussion of Benua’s analysis of this example;
Kiparsky’s objections, however, do not affect the point made here.
18. There are two reported examples, to our knowledge, of overapplication of a BR-
juncture effect. One, in Chaha, is reanalyzed in §5.7 as phonological agreement,
not necessarily reduplication-specific. The other is a case from Malay, discussed in
Kenstowicz 1981; McCarthy and Prince 1995a. In Malay, nasality spreads rightward
from a nasal consonant to any following vowel or glide; it is blocked by nonglides.
In forms like hamã → hãmã-hãma, nasal harmony applies across the BR juncture
(hamã-hamã → hamã-hãmã) and is then apparently reflected to the first vowel
of the first copy (→ hãmã-hãmã), even though it is not transparently conditioned
there. This dramatic example of overapplication has attracted much attention in the
literature. To our knowledge Onn 1976 is the only source of this data, which has
not been replicated in any other publication.
19. For a parallel example, see the discussion of Warlpiri in note 3.
20. As Richard Rhodes has pointed out to us, the first form in (55c) exhibits double
reduplication; the initial ne of nenehkememewa is an instance of monosyllabic
reduplication, a less productive pattern than the disyllabic reduplication discussed
here. On double reduplication in Fox, see Dahlstrom 1997.
21. The /e∼i/ alternation in the roots in (59a) and the suffixes following the null roots in
(59b) is due to “Initial Change” discussed in §5.5.2.1. The /e/ vowel is underlying;
/i/ is derived.
22. An alternative view is offered by Burkhardt 2002, who proposes output-output
faithfulness as a means of handling the overapplication of /e/-raising.
23. Zuraw (2002) develops what is in many ways a similar theory, though gives a
base-reduplicant analysis to the corresponding strings. Her view is that learners
reanalyze words containing similar segments or sequences as reduplicative, which
then leads them to make the strings in question even more similar (hence the name
for her theory, namely “Aggressive Reduplication”). However, since the effect of
the reanalysis (increased similarity) is comparable to what is yielded by the purely
phonological correspondence analyses of Hansson 2001 and Rose and Walker 2001,
it seems unnecessary to posit a reduplicative morpheme as part of the process.
222 Notes to pages 182–94

6 Case studies
1. These data are taken from McCarthy and Prince 1995a. The rest of the data cited in
our discussion come from Bloomfield 1933; Schachter and Otanes 1972; English
1986; French 1988. Data are orthographic, except for ‘ng,’ which is converted to
[ŋ]. Stress and vowel length, not represented in orthography, are not represented
here either, despite being marked in some of the sources.
2. This form, whose gloss comes from Schachter and Otanes 1972:364, is the imper-
fective of maŋ-isda ‘go fishing.’ According to Schachter and Otanes, imperfectives
of verbs beginning with m-initial prefixes are formed by a two-part operation: redu-
plication and replacement of m with n.
3. Another alternative is offered by Saperstein (1997), who suggests a possible process
of reanalysis by which stems such as putul have developed nasal-initial allomorphs
(e.g. mutul), for which reduplication selects.
4. On the predictable d ∼ r alternation illustrated in (4e), see Schachter and Otanes
1972.
5. In some cases, e.g. magpa-kain → mag-pa-pakain, second syllable reduplication is
the only option.
6. Both Applegate and Wash use as their data source the field notes of J. P. Harrington;
Wash, in addition, draws from tape recordings made by Madison Beeler.
7. Applegate does not typically provide morpheme glosses. Applegate 1972 con-
tains extensive discussion of Chumash affixational morphology, and with work
one can usually identify the affixes in a given verb. However, there is enough
homophony among the affixes to make this tricky in some cases. We have supplied
morpheme glosses here and there, but have not yet attempted to do this for every
form.
8. The glosses in (8) are taken directly from Applegate 1976; some minor typographical
data errors in Mester 1986:204, carried over in McCarthy and Prince (1995a:308,
313), are also corrected here.
9. Wilbur (1973:17) says that geminate aspiration applies transparently in Chumash,
citing the form /k-kut-kuti/ → kh utkuti ‘to look.’ We have not been able to confirm
this form in Applegate’s work.
10. Note that this analysis requires the prefix-final consonant to switch morphemic
allegiance from prefix to reduplicant, a transformational process which could not
occur in MDT.
11. A CV or CVʔ stem will have a CVh reduplicant.
12. For only a minority of cases is supporting evidence given in the dissertation. How-
ever, there are examples to back up all of the crucial arguments made here.
13. The prefix in (18a) does reduplicate in Barbareño; see (17d). In (18f) we correct an
apparent typographical error by Applegate, whose surface form omits the vowel in
the first syllable.
14. As elsewhere, Applegate does not gloss these individual morphemes. There is a
prefix qili-/qulu- ‘of seeing, vision’ (p. 362) that could be the source of xul-. The
consonants /q/ and /x/ alternate freely in Chumash, and the final /i/ of qili-/qulu-
could be epenthetic (see p. 327), rather than underlying.
 Notes to pages 194–201 223

15. In an affix-ordering chart, Applegate (1972:380) indicates that ni- precedes su-, a
claim consistent with such examples as
/ni-su-wal-tun/ [nisuwatun] ‘to put something over something else’ A72:332
/ni-su-tap/ ‘to put into; to stuff (e.g. a doll)’ A72:332

The prefixes ni- and su- thus occur in both orders. Several explanations are possible –
free variation, different orderings with different meanings, two different su- prefixes,
or two different ni- prefixes – but it is impossible to decide among these on the basis
of the available data. A similar discrepancy involves su- and wati-, occurring below
in the opposite order from what is exhibited in (22):
/wati-su-axsil-š/ [watišaxšilš] ‘to go fishing every now and then’ A72:383

16. The surface form of ‘my arm is broken’ is constructed from Applegate’s underlying
representation.

7 Final issues
1. On Hausa, see Newman 2000. Another recently discovered example of onset-driven
phonological copying occurs in American English “Homeric” infixation examples,
discussed by Yu 2003.
2. Proximal assimilation includes the possibility of onset-to-onset assimilation (pos-
sibly across an intervening coda) or coda-to-coda assimilation (possibly across an
intervening onset); Temiar, recently the subject of several insightful studies by Gafos
(1998a; 1998b), appears to be a case of the latter type.
3. Two other such cases occur in Kinande (Mutaka & Hyman 1990) and Nancowry
(Radhakrishnan 1981). The Kinande facts are amply discussed by Mutaka and
Hyman, and we will not repeat them here. Nancowry has been described by
Hendricks 1999 as a system in which reduplication of monosyllabic roots occurs
“to create a stem that is usable for other processes, such as inflection” (p. 251),
though in fact only one affix (the causative, which is derivational) selects for
reduplicated roots. There is no general requirement that the base of affixa-
tion must be disyllabic. In his grammar of Nancowry, Radhakrishnan 1981 dis-
cusses five different affixes, three of which have two suppletive allomorphs,
making a total of eight lexically listed affixes. Five of these eight affixes are
able to combine directly with monosyllabic roots; in fact the possessive -u
(e.g. kan-u ‘possessing [married to] a woman’; p. 65), the objective -a (e.g. wiʔ-a
‘a thing made’; p. 66), the causative ha- (e.g. ha-káh ‘to cause to know’; p. 54),
and the instrumental -an- (e.g. s-an-ák ‘spear’; p. 61) combine only with mono-
syllabic roots. Two of the affixes, causative -um (e.g. p-um-ʔ´y ‘to cause to have
bad smell,’ cf. paʔ´y ‘smell’; p. 54) and instrumental -in- (e.g. t-in-kuác ‘tracer,’
cf. takuác ‘to have a trace’; p. 62), combine only with lexically disyllabic roots;
reduplication of a monosyllabic root is not an option for these. But it is for
one prefix, the /ma-/ allomorph of the agentive. This prefix combines only with
reduplicated monosyllabic roots (p. 58), thus /ʔáp/ ‘to be closed’ → m-up-ʔáp
224 Notes to pages 210–11

‘one that is closed,’ where up-ʔáp is the reduplicated root. (On the phonology of
Nancowry reduplication, see Steriade 1988; Alderete et al. 1999; Hendricks 1999.)
4. Following standard practice in the literature on Turkish, we use uppercase letters (e.g.
“I”) for vowels whose specifications for [back] and [round] are predictable from the
context.
5. For example, Alderete (1999; 2001) invokes anti-identity constraints in analyzing
accentual dominance effects; anti-identity is used only to accomplish tone deletion,
not actually to toggle the values for tones or to insert tones into toneless bases.
Simple tone deletion (as proposed for the same data in Inkelas 1998) might be a more
straightforward analysis. Crosswhite (1997; 1999) invokes anti-identity constraints
to prohibit homophony within inflectional paradigms in Bulgarian and Russian. For
example, the suffix vowel in the Russian verb /stávj -at/ ‘place-3pl’ reduces to /ə/
(stáv j ət), rather than the expected /i/, ostensibly in order to avoid homophony with
the third singular form stáv j it. As Alan Timberlake (personal communication) points
out, however, other suffix /a/ vowels also reduce to /ə/ in the same phonological
context even when homophony is not at issue; the target of reduction appears to
be morphologically conditioned, making anti-homophony unnecessary as a causal
factor.
 References

Abbi, Anvita. 1991. Reduplication in South Asian languages: an areal, typological and
historical study. New Delhi: Allied Publishers Limited.
Akinlabi, Akinbiyi. 1985. Tonal underspecification and Yoruba tone. PhD thesis,
University of Ibadan.
Akinlabi, Akinbiyi. 1993. Underspecification and the phonology of Yoruba /r/. Linguistic
Inquiry 24:139–60.
Alderete, John. 1999. Morphologically governed accent in Optimality Theory. PhD
thesis, University of Massachussets, Amherst.
Alderete, John. 2001. Dominance effects as transderivational anti-faithfulness. Phonol-
ogy 18:201–53.
Alderete, John, Jill Beckman, Laura Benua, Amalia Gnanadesikan, John McCarthy,
and Suzanne Urbanczyk. 1999. Reduplication with fixed segmentism. Linguistic
Inquiry 30:327–64.
Anderson, Stephen R. 1974. The organization of phonology. San Diego: Academic Press.
Anderson, Stephen R. 1992. A-morphous morphology. Cambridge: Cambridge Univer-
sity Press.
Anttila, Arto. 1997. Deriving variation from grammar. In Frans Hinskens, Roeland
Van Hout, and W. Leo Wetzels (eds.), Variation, change and phonological theory.
Amsterdam: John Benjamins, 35–68.
Anttila, Arto. 2002. Morphologically conditioned phonological alternations. Natural
Language and Linguistic Theory 20:1–42.
Anttila, Arto. In press. Derived environment effects in Colloquial Helsinki Finnish. In
Kristin Hanson and Sharon Inkelas (eds.), The nature of the word: essays in honor
of Paul Kiparsky. Cambridge: MIT Press.
Anttila, Arto and Young-mee Yu Cho 1998. Variation and change in Optimality Theory.
Lingua 104:31–36.
Applegate, Richard. 1972. Ineseño Chumash grammar. PhD thesis, University of
California, Berkeley.
Applegate, Richard. 1976. Reduplication in Chumash. In Margaret Langdon and Shirley
Silver (eds.), Hokan Studies. The Hague: Mouton, 271–83.
Apte, Mahadeo. 1968. Reduplication, echo formation and onomatopoeia in Marathi.
Poona: Deccan College Post-Graduate and Research Institute.
Arnott, David W. 1970. The nominal and verbal systems of Fula. Oxford: Oxford
University Press.

225
226 References

Aronoff, Mark. 1988. Head operations and strata in reduplication: a linear treatment.
Yearbook of Morphology 1:1–15.
Aronoff, Mark. 1992. Noun classes in Arapesh. Yearbook of Morphology 1991:21–32.
Aronoff, Mark. 1994. Morphology by itself: stems and inflectional classes. Cambridge:
MIT Press.
Austin, Peter. 1981. A grammar of Diyari. Cambridge: Cambridge University Press.
Bagemihl, Bruce. 1988. Alternate phonologies and morphologies. PhD thesis, University
of British Columbia.
Barker, Muhammad Abd-al-Rahman. 1963. Klamath Dictionary, University of Califor-
nia Publications in Linguistics, vol. 31. Berkeley and Los Angeles: University of
California Press.
Barker, Muhammad Abd-al-Rahman. 1964. Klamath grammar. University of Califor-
nia Publications in Linguistics, vol. 32. Berkeley and Los Angeles: University of
California Press.
Barnes, Jonathan. 2002. Positional neutralization: a phonologization approach to typo-
logical patterns. PhD thesis, University of California, Berkeley.
Bat-El, Outi. 1994. Stem modification and cluster transfer in Modern Hebrew. Natural
Language and Linguistic Theory 12:571–96.
Bates, Dawn. 1986. An analysis of Lushootseed diminutive reduplication. In N. Niki-
foridou, M. VanClay, M. Niepokuj, and D. Feder (eds.), Proceedings of the Berkeley
Linguistics Society 12. Berkeley, CA: Berkeley Linguistics Society, 1–13.
Bates, Dawn and Barry F. Carlson. 1998. Spokane (Npoqi’nis&cn) syllable structure
and reduplication. In Ewa Czaykowska-Higgins and Dale Kinkade (eds.), Sal-
ish languages and linguistics: theoretical and descriptive perspectives. Trends in
Linguistics: Studies and Monographs 107. Berlin: Mouton de Gruyter, 99–123.
Bates, Dawn, Thom Hess, and Vi Hilbert. 1994. Lushootseed dictionary. Seattle:
University of Washington Press.
Benua, Laura. 1997. Transderivational identity: phonological relations between words.
PhD thesis, University of Massachusetts.
Bhaskararao, Peri. 1977. Reduplication and onomatopoeia in Telugu. Poona: Deccan
College.
Bissell, Teal. 2002. “Avoidance of the marked” in Miya pluractional allomorphy. In
Aniko Csirmaz, Zhiqiang Li, Andrew Nevins, Olga Vaysman, and Michael Wagner
(eds.), Phonological answers. Cambridge, MA: MITWPL, 1–42.
Black, Deirdre. 1996. The morphological and phonological structure of Spokane
lexemes. PhD thesis, University of Montana.
Blevins, James. 2003. Stems and paradigms. Language 79:737–67.
Blevins, Juliette. 1996. Mokilese reduplication. Linguistic Inquiry 27:523–30.
Blevins, Juliette. 1999. Untangling Leti infixation. Oceanic Linguistics 38:383–403.
Blevins, Juliette. 2004. Evolutionary phonology: the emergence of sound patterns.
Cambridge: Cambridge University Press.
Bloomfield, Leonard. 1933. Language. New York: Henry Holt.
Blust, Robert. 2001. Thao Triplication. Oceanic Linguistics 40:324–36.
Boas, Franz and Ella Deloria. 1941. Dakota Grammar, vol. 23, part 2: Memoirs of the
National Academy of Sciences.
 References 227

Bochner, Harry. 1992. Simplicity in generative morphology. Publications in Linguistic

Sciences 37. Berlin: Walter de Gruyter.
Booij, Geert. 1984. Coordination reduction in complex words: a case for prosodic
phonology. In H. van der Hulst and N. Smith (eds.), Advances in non-linear
phonology. Dordrecht: Foris, 143–60.
Booij, Geert. 2000. Morphology and phonology. In Geert Booij, Christian Lehmann, and
Joachim Mugdan (eds.), Morphologie – morphology. Berlin: Walter de Gruyter,
335–43.
Booij, Geert and Rochelle Lieber. 1993. On the simultaneity of morphological and
prosodic structure. In Sharon Hargus and Ellen Kaisse (eds.), Phonetics and phonol-
ogy 4: studies in lexical phonology. San Diego: Academic Press, 23–44.
Bosson, James E. 1964. Modern Mongolian: a primer and reader. The Hague:
Mouton & Co.
Bowen, Donald J. 1969. Beginning Tagalog: a course for speakers of English. Berkeley:
University of California Press.
Breen, Gavan and Rob Pensalfini. 1999. Arrernte: a language with no syllable onsets.
Linguistic Inquiry 30:1–25.
Broselow, Ellen. 1983. Salish double reduplications: subjacency in morphology. Natural
Language and Linguistic Theory 1:317–46.
Bruening, Benjamin. 1997. Abkhaz mabkhaz: m-reduplication in Abkhaz, weightless
syllables, and base-reduplicant correspondence. In Benjamin Bruening, Yoonjung
Kang, and Martha McGinnis (eds.), PF: papers at the interface. Cambridge, MA:
MIT Working Papers in Linguistics, 291–330.
Burkhardt, Petra. 2002. Multiple faithfulness relations in Fox (Central Algonquian)
reduplication. In Marjo van Koppen, Joanna Sio, and Mark de Vos (eds.), Console
X. University of Leiden, 33–48.
Burzio, Luigi. 1997. Cycles, non-derived environment blocking, and correspondence. In
Joost Dekkers, Frank van der Leeuw, and Jeroen van de Weijer (eds.), The pointing
finger: conceptual studies in Optimality Theory. Oxford: Oxford University Press.
Bybee, Joan. 1985. Morphology: a study of the relation between meaning and form.
Amsterdam: John Benjamins.
Carlson, Katy. 1997. Sonority and reduplication in Nakanai and Nuxalk (Bella Coola).
Available online from the Rutgers Optimality Archive, ROA# ROA 230-1197.
Carrier-Duncan, Jill. 1984. Some problems with prosodic accounts of reduplication. In
Mark Aronoff and Richard Oehrle (eds.), Language sound structure. Cambridge,
MA: MIT Press, 260–86.
Cho, Young-mee Yu. In press. Derived environment effects in Korean. In Kristin Hanson
and Sharon Inkelas (eds.), The nature of the word: essays in honor of Paul Kiparsky.
Cambridge, MA: MIT Press.
Chomsky, Noam. 1995. The minimalist program. Cambridge, MA: MIT Press.
Chomsky, Noam. 2000. Minimalist inquiries: the framework. In Roger Martin, David
Michaels, and Juan Uriagereka (eds.), Essays on minimalist syntax in honor of
Howard Lasnik. Cambridge, MA: MIT Press, 89–155.
Chomsky, Noam and Morris Halle. 1968. The sound pattern of English. New York:
Harper and Row.
228 References

Clements, G. N. 1985. The problem of transfer in nonlinear morphology. Cornell

Working Papers in Linguistics 7:38–73.
Clements, G. N. and Samuel J. Keyser, 1983. CV Phonology: a generative theory of the
syllable. Cambridge, MA: MIT Press.
Cohn, Abigail. 1989. Stress in Indonesian and bracketing paradoxes. Natural Language
and Linguistic Theory 7:167–216.
Cohn, Abigail. 1990. Phonetic and phonological rules of nasalization. PhD thesis,
UCLA.
Cohn, Abigail and John McCarthy. 1998. Alignment and parallelism in Indonesian
phonology. Working papers of the Cornell Phonetics Laboratory 12:53–137.
Cole, Jennifer. 1994. A prosodic theory of reduplication. PhD thesis, Stanford University.
Collins, Chris. 1994. The factive construction in Kwa. Travaux de recherche sur le créole
haı̈tien 23:31–65.
Comrie, Bernard. 2000. Morphophonological alternations: typology and diachrony. In S.
Bendjaballah, W. U. Dressler, O. E. Pfeiffer, and M. D. Voeikova (eds.), Morphology
2000: selected papers from the ninth morphology meeting, Vienna, 24–28 February
2000. Amsterdam: John Benjamins, 73–89.
Conathan, Lisa and Jeffrey Good. 2000. Morphosyntactic reduplication in Chechen and
Ingush. In Arike Okrent and John P. Boyle (eds.), Proceedings of the Chicago
Linguistic Society 36–2: the panels. Chicago: Chicago Linguistic Society, 49–61.
Conathan, Lisa and Esther Wood. 2003. Repetitive reduplication in Yurok and Karuk:
semantic effects of contact. In H. C. Wolfart (ed.), Papers of the thirty-fourth
Algonquian Conference. Winnipeg: University of Manitoba, 19–33.
Conteh, Patrick, Elizabeth Cowper, and Keren Rice. 1985. The environment for conso-
nant mutation in Mende. In G. J. Dimmendaal (ed.), Current Approaches to African
Linguistics, vol. III. Dordrecht: Foris Publications, 107–16.
Corston-Oliver, Simon. 2002. Roviana. In John Lynch, Malcolm Ross, and Terry Crow-
ley (eds.), The Oceanic languages. Richmond: Curzon Press, 467–97.
Crosswhite, Katherine. 1997. Avoidance of homophony in Trigrad Bulgarian vowel
reduction. Paper presented at Southwest Workshop on Optimality Theory, Univer-
sity of California, Los Angeles.
Crosswhite, Katherine. 1999. Intra-paradigmatic homophony avoidance in two dialects
of Slavic. In Matthew K. Gordon (ed.), UCLA working papers in linguistics,
vol. I. Papers in phonology 2. Los Angeles: UCLA Department of Linguistics,
48–67.
Crowley, Terry. 1982. The Paamese language of Vanuatu. Pacific Linguistics Series
B-87. Canberra: Australian National University.
Crowley, Terry. 1991. Parallel development and shared innovation: some developments
in Central Vanuatu inflectional morphology. Oceanic Linguistics 30:179–222.
Crowley, Terry. 1998. An Erromangan (Sye) grammar. Oceanic linguistics special pub-
lication, no. 27. Honolulu: University of Hawaii Press.
Crowley, Terry. 2002a. Raga. In John Lynch, Malcolm Ross, and Terry Crowley (eds.),
The Oceanic languages. Richmond: Curzon Press, 626–37.
Crowley, Terry. 2002b. Serial verbs in Oceanic: a descriptive typology. Oxford: Oxford
University Press.
 References 229

Crowley, Terry. 2002c. Southeast Ambrym. In John Lynch, Malcolm Ross, and Terry
Crowley (eds.), The Oceanic languages. Richmond: Curzon Press, 660–70.
Crowley, Terry. 2002d. Sye. In John Lynch, Malcolm Ross, and Terry Crowley (eds.),
The Oceanic languages. Richmond: Curzon Press, 694–722.
Dahlstrom, Amy. 1997. Fox (Mesquakie) reduplication. International Journal of Amer-
ican Linguistics 63:205–26.
DeLacy, Paul. 2002. The formal expression of markedness. PhD thesis, University of
Massachusetts, Amherst.
Delancey, Scott. 1991. Chronological strata of suffix classes in the Klamath verb. Inter-
national Journal of American Linguistics 57:426–45.
Delancey, Scott. 1999. Lexical prefixes and the bipartite stem construction in Klamath.
International Journal of American Linguistics 65:56–83.
Dixon, Robert. 1972. The Dyirbal language of North Queensland. Cambridge:
Cambridge University Press.
Dixon, Robert. 1988. A grammar of Boumaa Fijian. Chicago: University of Chicago
Press.
Dobrin, Lise M. 1998. The morphosyntactic reality of phonological form. Yearbook of
Morphology 1997:59–81.
Dobrin, Lise M. 1999. Phonological form, morphological class, and syntactic gender:
the noun class systems of Papua New Guinea Arapeshan. PhD thesis, University
of Chicago.
Dolbey, Andrew. 1996. Output optimization and cyclic allomorph selection. In Brian
Agbayani and Sze-Wing Tang (eds.), Proceedings of the fifteenth West Coast
Conference on Formal Linguistics. Stanford: CSLI Publications, 97–112.
Downing, Laura. 1997. Correspondence effects in Siswati reduplication. Studies in the
Linguistic Sciences 25:81–95.
Downing, Laura. 1998a. Morphological correspondence constraints on Kikerewe redu-
plication. In Emily Curtis, James Lyle, and Gabriel Webster (eds.), Proceedings
of the sixteenth West Coast Conference on Formal Linguistics. Stanford: CSLI
Publications, 161–75.
Downing, Laura. 1998b. Onset motivated overcopy in reduplication. In E. van Gelderen
and V. Samiian (eds.), Proceedings of WECOL. Fresno, CA: California State
University Press, 81–96.
Downing, Laura. 1998c. On the prosodic misalignment of onsetless syllables. Natural
Language and Linguistic Theory 16:1–52.
Downing, Laura. 1998d. Prosodic misalignment and reduplication. Yearbook of
Morphology 1997:83–120.
Downing, Laura. 1998e. Prosodic stem
= prosodic word in Bantu. In Tracy Hall and
Ursula Kleinhenz (eds.), Studies on the phonological word. Current Studies in
Linguistic Theory. Amsterdam: John Benjamins, 73–98.
Downing, Laura. 1999a. Morphological constraints on Bantu reduplication. Linguistic
Analysis 29:6–46.
Downing, Laura. 1999b. Verbal reduplication in three Bantu languages. In Harry van der
Hulst, René Kager, and Wim Zonneveld (eds.), The prosody-morphology interface.
Cambridge: Cambridge University Press, 62–89.
230 References

Downing, Laura. 2000a. Morphological and prosodic constraints on Kinande verbal

reduplication. Phonology 17:1–38.
Downing, Laura. 2000b. Morphological correspondence in Kinande reduplication. Pro-
ceedings of the Berkeley Linguistics Society 23. Berkeley, CA: Berkeley Linguistics
Society.
Downing, Laura. 2001. Ungeneralizable minimality in Ndebele. Studies in African
Linguistics 30:33–58.
Downing, Laura. 2004. The emergence of the marked: tone in some African reduplicative
systems. In Bernhard Hurch (ed.), Studies in reduplication. Berlin: Mouton.
Dressler, Wolfgang. 1985. Morphonology, the dynamics of derivation, ed. Kenneth
C. Hill. Ann Arbor: Karoma Publishers.
Dudas, Karen Marie. 1976. The phonology and morphology of modern Javanese. PhD
thesis, University of Illinois.
Durie, Mark. 1985. A grammar of Acehnese: on the basis of a dialect of North Aceh.
Dordrecht: Foris.
Early, Robert. 2002. Niuafo’ou. In John Lynch, Malcolm Ross, and Terry Crowley (eds.),
The Oceanic languages. Richmond: Curzon Press, 848–64.
Egbokhare, Francis O. 2001. Phonosemantic correspondences in Emai attributive ideo-
phones. In F. K. Erhard Voeltz and Kilian-Hats (eds.), Ideophones. Amsterdam:
John Benjamins, 87–96.
Elbert, Samuel H. 1988. Echo of a culture: a grammar of Rennell and Bellona.
Oceanic Linguistics Special Publication, No. 22. Honolulu: University of Hawaii
Press.
Emeneau, Murray. 1955. Kolami: a Dravidian language. Berkeley: University of
California Press.
English, Leo J. 1986. Tagalog–English dictionary. Manila: Congregation of the Most
Holy Redeemer.
Ezard, Bryan. 1980. Reduplication in Tawala. Kivung 12:145–60.
Ezard, Bryan. 1997. A grammar of Tawala. Pacific Linguistics Series C-137. Canberra:
Research School for Pacific and Asian Studies, Australian National University.
Fabb, Nigel. 2002. Language and literary structure: the linguistic analysis of form in
verse and narrative. Cambridge: Cambridge University Press.
Feinstein, Mark H. and Robert M. Vago. 1981. Non-evidence for the segmental cycle
in Klamath. In Didier L. Goyvaerts (ed.), Phonology in the 1980s. Ghent: Story-
Scientia, 119–45.
Fillmore, Charles, Paul Kay, and Mary Catherine O’Connor. 1988. Regularity and
idiomaticity in grammatical constructions. Language 64:501–38.
Fitzgerald, Colleen M. 1999. Loanwords and stress in Tohono O’odham. Anthropolog-
ical Linguistics 41:193–208.
Fitzgerald, Colleen M. 2000. Vowel hiatus and faithfulness in Tohono O’odham redu-
plication. Linguistic Inquiry 31:713–22.
Fitzgerald, Colleen M. 2001. The Morpheme-to-Stress principle in Tohono O’odham.
Linguistics 39:941–72.
Fitzgerald, Colleen M. 2002. Tohono O’odham stress in a single ranking. Phonology
19:253–71.
Ford, Alan and Rajendra Singh. 1983. On the status of morphophonology. In John
F. Richardson, Mitchell Marks, and Amy Chukerman (eds.), Papers from the
 References 231

parasession on the interplay of phonology, morphology and syntax. Chicago:

Chicago Linguistic Society, 63–78.
Fortune, R. F. 1942. Arapesh. Publications of the American Ethnological Society,
no. 19. New York: J. Augustin.
Frampton, John. 2003. Root vowel syncope and reduplication in Sanskrit. Paper pre-
sented at the thirty-ninth meeting of the Chicago Linguistics Society, University of
Chicago.
French, Koleen. 1988. Insights into Tagalog: reduplication, infixation and stress from
nonlinear phonology. Dallas: SIL and University of Texas at Arlington.
Gafos, Diamandis. 1998a. Eliminating long-distance consonantal spreading. Natural
Language and Linguistic Theory 16:223–78.
Gafos, Diamandis. 1998b. A-templatic reduplication. Linguistic Inquiry: 515–25.
Geraghty, Paul. 2002. Nadrogā. In John Lynch, Malcolm Ross, and Terry Crowley (eds.),
The Oceanic languages. Richmond: Curzon Press, 833–47.
Gerdts, Donna. 1998. Incorporation. In Andrew Spencer and Arnold Zwicky (eds.), The
handbook of morphology. Oxford: Blackwell, 84–100.
Goldberg, Adele. 1995. Constructions. Chicago: Chicago University Press.
Goodenough, Ward H. 1963. The long or double consonants of Trukese. Proceedings of
the ninth Pacific Science Congress, 1957 3:77–86.
Gorgoniev, Yu. A. 1976. The relationship between reduplication and some other gram-
matical means in Khmer. In Stanley Starosta (ed.), Austroasiatic studies, part 1.
Honolulu: University Press of Hawaii, 309–21.
Green, Antony Dubach. 2002. Word, foot and syllable structure in Burmese. Available
online from the Rutgers Optimality Archive, ROA# 551-1002.
Gurevich, Naomi. 2000. Reduplication in Southern Paiute and Correspondence
Theory. In Roger Billerey and Brook Lillehaugen (eds.), Proceedings of the
nineteenth West Coast Conference on Formal Linguistics. Somerville: Cascadilla
Press, 167–77.
Haiman, John. 1980. Hua: a Papuan language of the Eastern Highlands of New Guinea.
Studies in Language Companion Series, vol. 5. Amsterdam: John Benjamins.
Haiman, John. 1998. Repetition and identity. Lingua 100:57–70.
Hanson, Kristin and Paul Kiparsky. 1996. A parametric theory of poetic meter. Language
72:287–335.
Hansson, Gunnar. 2001. Theoretical and typological issues in consonant harmony. PhD
thesis, University of California, Berkeley.
Haraguchi, Shosuke. 1977. The tone pattern of Japanese. Tokyo: Kaitakusha.
Hardman, M. J. 2000. Jaqaru. Languages of the world/materials 183. Munich: Lincom
Europa.
Hargus, Sharon. 1988. The lexical phonology of Sekani. Outstanding dissertations in
linguistics series. New York: Garland.
Hargus, Sharon. 1995. The first person plural subject prefix in Babine/Witsuwit’en.
Available online from the Rutgers Optimality Archive, ROA# 108-0000.
Harris, B. P. and G. N. O’Grady. 1976. An analysis of the progressive morpheme in
Umpila verbs. In P. Sutton (ed.), Languages of Cape York. Canberra: Australian
Institute of Aboriginal Studies, 165–212.
Harrison, Sheldon P. 1973. Reduplication in Micronesian languages. Oceanic Linguistics
12:407–54.
232 References

Harrison, Sheldon P. 1976. Mokilese Reference grammar. Honolulu: University Press

of Hawaii.
Hayes, Bruce. 1981. A metrical theory of stress rules. Bloomington: Indiana University
Linguistics Club.
Healey, Alan, Ambrose Isoroembo, and Martin Chittleborough. 1969. Preliminary notes
on Orokaiva grammar. Papers in New Guinea Linguistics 9:33–64.
Hendricks, Sean Q. 1999. Reduplication without template constraints: a study in bare-
consonant reduplication. PhD thesis, University of Arizona.
Hess, T. M. and V. Hilbert. 1978. Lushootseed. Seattle: Daybreak Star Press.
Hess, Thom. 1977. Lushootseed dialects. Anthropological Linguistics 19:403–19.
Hess, Thom. 1993. A schema for the presentation of Lushootseed verb stem. In A.
Mattina and T. Montler (eds.), American Indian linguistics and ethnography in
honor of Laurence C. Thompson. University of Montana Occasional Papers in
Linguistics No. 10, 113–25.
Hess, Thom. 1995. Lushootseed reader with introductory grammar: volume 1: Four
Stories by Edward Sam. University of Montana Occasional Papers in Linguistics
II.
Hill, Jane H. and Ofelia Zepeda. 1992. Derived words in Tohono O’odham. International
Journal of American Linguistics 58:355–404.
Hill, Jane H. and Ofelia Zepeda. 1998. Tohono O’odham (Papago) plurals. Anthropo-
logical Linguistics 40:1–42.
Hockett, Charles. 1954. Two models of grammatical description. Word 10:210–31.
Holtman, Astrid. 1996. A generative theory of rhyme: an Optimality approach. OTS
Dissertation Series Utrecht: LEd.
Horne, Elinor. 1961. Beginning Javanese. New Haven: Yale University Press.
Hualde, José. 1988. A lexical phonology of Basque. PhD thesis, University of Southern
California.
Hurlbut, Hope M. 1988. Verb morphology in Eastern Kadazan. Pacific Linguistics Series
B-97. Canberra: Department of Linguistics, Research School of Pacific Studies,
Australian National University.
Hyman, Larry M., Sharon Inkelas, and Galen Sibanda. To appear. Morphosyntactic cor-
respondence in Bantu reduplication. In Kristin Hanson and Sharon Inkelas (eds.),
The Nature of the Word: Essays in Honor of Paul Kiparsky. Cambridge, MA: MIT
Press.
Hyman, Larry M. and Al Mtenje. 1999. Prosodic morphology and tone: the case of
Chichewa. In Rene Kager, Harry van der Hulst, and Wim Zonneveld (eds.), The
prosody-morphology interface. Cambridge: Cambridge University Press, 90–133.
Inkelas, Sharon. 1990. Prosodic constituency in the lexicon. New York: Garland.
Inkelas, Sharon. 1995. The consequences of optimization for underspecification. In Jill
Beckman (ed.), Proceedings of the Northeastern Linguistics Society 25: Amherst,
MA: GLSA, 287–302.
Inkelas, Sharon. 1998. The theoretical status of morphologically conditioned phonology:
a case study from dominance. Yearbook of Morphology 1997:121–55.
Inkelas, Sharon. 2000. Phonotactic blocking through structural immunity. In Barbara
Stiebels and Dieter Wunderlich (eds.), Lexicon in focus. Studia Grammatica 45.
Berlin: Akademie Verlag, 7–40.
 References 233

Inkelas, Sharon and Cemil Orhan Orgun. 1995. Level ordering and economy in the
lexical phonology of Turkish. Language 71:763–93.
Inkelas, Sharon and Cemil Orhan Orgun. 1998. Level (non)ordering in recursive mor-
phology: evidence from Turkish. In Steven Lapointe, Diane Brentari, and Patrick
Farrell (eds.), Morphology and its relation to phonology and syntax. Stanford, CA:
CSLI Publications, 360–92.
Inkelas, Sharon, Cemil Orhan Orgun, and Cheryl Zoll. 1997. Implications of lexical
exceptions for the nature of grammar. In Iggy Roca (ed.), Constraints and deriva-
tions in phonology. Oxford: Clarendon Press, 393–418.
Inkelas, Sharon and Cheryl Zoll. 2003. Is grammar dependence real? Unpublished ms.,
UC Berkeley and MIT. Available online from the Rutgers Optimality Archive,
ROA# 587-0303.
Innes, G. 1971. A practical introduction to Mende. London: School of Oriental and
African Studies.
Itô, Junko. 1990. Prosodic minimality in Japanese. In Michael Ziolkowski, Manuela
Noske, and Karen Deaton (eds.), Papers from the twenty-sixth regional meeting of
the Chicago Linguistics Society, vol. II: The parasession on the syllable in phonetics
and phonology. Chicago: Chicago Linguistics Society, 213–39.
Itô, Junko and Armin Mester. 1986. The phonology of voicing in Japanese: theoretical
consequences for morphological accessibility. Linguistic Inquiry 17:49–73.
Itô, Junko and Armin Mester. 1995a. The core-periphery structure of the lexicon
and constraints on reranking. University of Massachussetts Occasional Papers in
Linguistics 18. Amherst, MA: GLSA.
Itô, Junko and Armin Mester. 1995b. Japanese phonology: constraint domains and
structure preservation. In John Goldsmith (ed.), The handbook of phonological
theory. Blackwell Handbooks in Linguistics Series Cambridge, MA: Blackwell,
817–38.
Itô, Junko and Armin Mester. 1996a. Rendaku 1: constraint conjunction and the OCP.
Available online from the Rutgers Optimality Archive, ROA# 144-0996.
Itô, Junko and Armin Mester. 1996b. Structural economy and OCP interactions in local
domains. Paper presented at the Western Conference on Linguistics (WECOL).
University of California, Santa Cruz.
Itô, Junko and Armin Mester. 1997. Correspondence and compositionality: the Ga-gyo
variation in Japanese phonology. In Iggy Roca (ed.), Derivations and constraints
in phonology. Oxford: Clarendon Press, 419–62.
Itô, Junko and Armin Mester. 1999. The phonological lexicon. In N. Tsujimura (ed.),
The handbook of Japanese linguistics. Malden, MA: Blackwell, 62–100.
Itô, Junko, Armin Mester, and Jaye Padgett. 1995. Licensing and underspecification in
Optimality Theory. Linguistic Inquiry 26:571–613.
Jeanne, LaVerne Masayesua. 1982. Some phonological rules of Hopi. IJAL 48:
245–70.
Johanson, Lars and Eva Csato, eds. 1998. The Turkic languages. New York: Routledge.
Jones, William. 1919. Ojibwa Texts, Part 2. New York: G. E. Stechert.
Kager, Rene. 1996. On affix allomorphy and syllable counting. In Ursula Kleinhenz
(ed.), Interfaces in phonology. Studia grammatica 41. Berlin: Akademie Verlag,
155–71.
234 References

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

Kaisse, Ellen and Patricia Shaw. 1985. On the theory of Lexical Phonology. Phonology
Yearbook 2:1–30.
Kanerva, Jonni. 1989. Focus and phrasing in Chichewa phonology. PhD thesis, Stanford
University.
Kawu, Ahmadu Ndanusa. 1998. The emergence of the unmarked in Niger-Congo verb
reduplication. Unpublished ms., Rutgers University.
Kay, Paul. 2002. English subjectless tagged sentences. Language 78:453–81.
Kean, Mary-Louise. 1973. Non-global rules in Klamath Phonology. Quarterly Progress
Report, MIT Research Laboratory in Electronics 108:288–310.
Kelepir, Meltem. 2000. To be or not to be faithful. In Aslı Göksel and Celia Kerslake
(eds.), Proceedings of the ninth International Conference of Turkish Linguistics.
Wiesbaden: Otto Harrassowitz, 11–18.
Kenstowicz, Michael. 1981. Functional explanations in generative phonology. In
Didier L. Goyvaerts (ed.), Phonology in the 1980s. Ghent: E. Story-Scientia, 431–
44.
Kenstowicz, Michael. 1995. Cyclic vs. non-cyclic constraint evaluation. Phonology
12:397–436.
Kenstowicz, Michael. 1996. Base identity and uniform exponence: alternatives to
cyclicity. In Jacques Durand and Bernard Laks (eds.), Current trends in phonology:
models and methods. Salford: ESRI, 363–94.
Kenstowicz, Michael. 1997. Uniform exponence: extension and exemplification. In Viola
Miglio and Bruce Moren (eds.), University of Maryland working papers in lin-
guistics: selected phonology papers from the Johns Hopkins Optimality Theory
Workshop/Maryland Mayfest. Baltimore: Department of Linguistics, University of
Maryland, 139–55. Available electronically from the Rutgers Optimality Archive,
ROA# 218-0997.
Kenstowicz, Michael and Charles Kisseberth. 1979. Generative phonology. New York:
Academic Press.
Kenstowicz, Michael and Degif Petros Banksira 1999. Reduplicative identity in Chaha.
Linguistic Inquiry 30:573–86.
Kim, Eun-Sook. 2003. Patterns of reduplication in Nuu-chah-nulth. In M. Kadowaki
and Shigeto Kawahara (eds.), Proceedings of the Northeastern Linguistics Society
33. Amherst, MA: GLSA, 127–46.
Kiparsky, Paul. 1973a. Phonological representations: abstractness, opacity and global
rules. In Osamu Fujimura (ed.), Three dimensions in linguistic theory. Tokyo: TEC,
56–86.
Kiparsky, Paul. 1973b. The role of linguistics in a theory of poetry. Proceedings of the
American Academy of Arts and Sciences 102:231–44.
Kiparsky, Paul. 1982a. From cyclic phonology to lexical phonology. In Harry van
der Hulst and Norval Smith (eds.), The structure of phonological representations,
part I. Dordrecht: Foris, 131–75.
Kiparsky, Paul. 1982b. Lexical morphology and phonology. In I.-S. Yang (ed.),
Linguistics in the morning calm. Linguistics Society of Korea. Seoul: Hanshin,
3–91.
 References 235

Kiparsky, Paul. 1982c. Word-formation and the lexicon. In Frances Ingemann (ed.), 1982
Mid-America linguistics conference papers. Lawrence, KS: University of Kansas,
3–29.
Kiparsky, Paul. 1986. The phonology of reduplication. Unpublished ms., Stanford
University.
Kiparsky, Paul. 1993. Blocking in non-derived environments. In Sharon Hargus and
Ellen Kaisse (eds.), Phonetics and phonology 4: studies in lexical phonology. San
Diego: Academic Press, 277–313.
Kiparsky, Paul. 1994. Remarks on markedness. Paper presented at TREND, University
of California, Santa Cruz.
Kiparsky, Paul. 1997. Templatic morphology: Base-reduplication Identity constraints vs.
cyclicity. Class handout from Scandinavian Summer School in Generative Phonol-
ogy, Hvalfjardarströnd, Iceland, June 1997.
Kiparsky, Paul. 2000. Opacity and cyclicity. The Linguistic Review 17:351–67.
Kisseberth, Charles W. 1972, Cyclical rules in Klamath Phonology. Linguistic Inquiry
3:3–33.
Kiyomi, Setsuko. 1993. A typological study of reduplication as a morpho-semantic
process: evidence from five language families (Bantu, Australian, Papuan,
Austroasiatic and Malayo-Polynesian). PhD thesis, Indiana University.
Klein, Thomas B. 1997. Output constraints and prosodic correspondence in Chamorro
reduplication. Linguistic Inquiry 28:707–16.
Koenig, Jean-Pierre. 1992. Shared structure vs. constructional autonomy in construction
grammar. Proceedings of the fifteenth International Congress of Linguistics, 356–
59.
Koenig, Jean-Pierre and Daniel Jurafsky. 1995. Type underspecification and on-line type
construction in the lexicon. In Raul Aranovich, William Byrne, Susanne Preuss,
and Martha Senturia (eds.), Proceedings of the thirteenth West Coast Conference
on Formal Linguistics. Stanford: CSLI Publications, 270–85.
Kornfilt, Jaklin. 1997. Turkish. London: Routledge.
Krause, Scott. 1980. Topics in Chukchee Phonology and Morphology. PhD thesis,
University of Illinois Urbana-Champaign.
Kroeger, Paul. 1990. Discontinuous reduplication in vernacular Malay. Proceedings of
the Berkeley Linguistics Society 15. Berkeley: Berkeley Linguistics Society, 193–
202.
Kurisu, Kazutaka. 2001. The phonology of morpheme realization. PhD thesis, UC Santa
Cruz.
Landau, Idan. 2003. Modular recoverability: chain resolution in Hebrew VP-fronting.
Unpublished ms., Ben Gurion University of the Negev and Tel Aviv University.
Lathroum, Amanda. 1991. Resolving the disparity between linear and structural
order in words: a study of reduplication in Austronesian. PhD thesis, Harvard
University.
Lefebvre, Claire and A. M. Brousseau. 2002. A grammar of Fongbe. Berlin: Mouton de
Gruyter.
Levin, Juliette. 1985a. A metrical theory of syllabicity. PhD thesis, Massachusetts
Institute of Technology.
236 References

Levin, Juliette. 1985b. Reduplication in Umpila. In Diana Archangeli, Andrew Barrss,

and Richard Sproat (eds.), MIT working papers in linguistics, vol. VI: papers
in theoretical and applied linguistics. Cambridge: Massachusetts Institute of
Technology, 133–59.
Lewis, Geoffrey. 1967. Turkish grammar. Oxford: Oxford University Press.
Li, Charles and Sandra Thompson. 1981. Mandarin Chinese: a functional reference
grammar. Berkeley: University of California Press.
Lieber, Rochelle. 1980. On the organization of the lexicon. PhD thesis, Masachusetts
Institute of Technology.
Lombardi, Linda and John McCarthy. 1991. Prosodic circumscription in Choctaw mor-
phology. Phonology 8:37–71.
Longtau, Selbut R. 1993. A formal Tarok phonology. Afrika und Übersee 76:15–40.
Lubowicz, Anna. 1999. Derived environment effects in OT. In Kimary Shahin, Susan
Blake, and Eun-Sook Kim (eds.), Proceedings of the seventeenth West Coast Con-
ference on Formal Linguistics. Stanford: CSLI Publications, 451–65. Available
electronically from the Rutgers Optimality Archive, ROA# 239-0198.
Lynch, John. 2002a. Niuafo’ou. In John Lynch, Malcolm Ross, and Terry Crowley (eds.),
The Oceanic languages. Richmond: Curzon Press, 865–76.
Lynch, John. 2002b. Ulithian. In John Lynch, Malcolm Ross, and Terry Crowley (eds.),
The Oceanic languages. Richmond: Curzon Press, 792–803.
Lynch, John and Rex Horoi. 2002. Arosi. In John Lynch, Malcolm Ross, and Terry
Crowley (eds.), The Oceanic languages. Richmond: Curzon Press, 562–72.
Lynch, John and Malcolm Ross. 2002. Banoni. In John Lynch, Malcolm Ross, and Terry
Crowley (eds.), The Oceanic languages. Richmond: Curzon Press, 440–55.
Lynch, John, Malcolm Ross, and Terry Crowley. 2002. The Oceanic languages.
Richmond: Curzon Press.
Malkiel, Yakov. 1959. Studies in irreversible binomials. Lingua 8.
Marantz, Alec. 1982. Re reduplication. Linguistic Inquiry 13:483–545.
Marantz, Alec and Caroline Wiltshire. 2000. Reduplication. In Geert Booij, Christian
Lehmann, and Joachim Mugdan (eds.), Morphologie – morphology. Berlin: Walter
de Gruyter, 557–66.
Mascaró, Joan. 1976. Catalan phonology and the phonological cycle. Bloomington:
Indiana University Linguistics Club.
Matteson, Esther. 1965. The Piro (Arawakan) language. University of California publi-
cations in linguistics, vol. 42. Berkeley, CA: University of California Press.
McCarthy, John. 1986. OCP effects: gemination and antigemination. Linguistic Inquiry
17:207–63.
McCarthy, John. 1993. The parallel advantage. Paper presented at Rutgers Optimality
Workshop 1, Rutgers University.
McCarthy, John. 1996. Remarks on phonological opacity in Optimality Theory. In J.
Lecarme, J. Lowenstamm, and U. Shlonsky (eds.), Studies in Afroasiatic grammar:
papers from the second Conference on Afroasiatic Linguistics, Sophia Antipolis,
1994. The Hague: Holland Academic Graphics, 215–43.
McCarthy, John. 1999. Sympathy and phonological opacity. Phonology 16:331–99.
McCarthy, John. 2002a. Against gradience. Available online from the Rutgers Optimality
Archive, ROA# 510-0302.
 References 237

McCarthy, John. 2002b. A thematic guide to Optimality Theory. Cambridge: Cambridge

University Press.
McCarthy, John and Alan Prince. 1986. Prosodic morphology. Unpublished ms.,
University of Massachusetts, Amherst and Brandeis University.
McCarthy, John and Alan Prince. 1990. Foot and word in Prosodic Morphology: the
Arabic broken plural. Natural Language and Linguistic Theory 8:209–83.
McCarthy, John and Alan Prince. 1993. Prosodic Morphology I: constraint interac-
tion and satisfaction. Unpublished ms., University of Massachusetts, Amherst and
Rutgers University.
McCarthy, John and Alan Prince. 1994a. The emergence of the unmarked. In Mercè
Gonzàlez (ed.), Proceedings of the Northeastern Linguistic Society 24. Amherst,
MA: GLSA, 333–79.
McCarthy, John and Alan Prince. 1994b. Generalized alignment. In Geert Booij and
Jaap van Marle (eds.), Yearbook of Morphology 1993. Dordrecht: Kluwer, 79–
153.
McCarthy, John and Alan Prince. 1994c. An overview of Prosodic Morphology part
I: templatic form in reduplication. Paper presented at Workshop on Prosodic
Morphology, Utrecht University.
McCarthy, John and Alan Prince. 1994d. An overview of Prosodic Morphology part
II: Template satisfaction. Paper presented at Workshop on Prosodic Morphology,
Utrecht University.
McCarthy, John and Alan Prince. 1995a. Faithfulness and reduplicative identity. In
Jill Beckman, Laura Dickey, and Suzanne Urbanczyk (eds.), University of Mas-
sachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory.
Amherst, MA: GLSA, 249–384.
McCarthy, John and Alan Prince. 1995b. Prosodic Morphology. In John Goldsmith (ed.),
Handbook of phonological theory. Cambridge, MA: Blackwell, 318–66.
McCarthy, John and Alan Prince. 1999a. Faithfulness and identity in Prosodic Mor-
phology. In Rene Kager, Harry van der Hulst, and Wim Zonneveld (eds.),
The prosody-morphology interface. Cambridge: Cambridge University Press,
218–309.
McCarthy, John and Alan Prince. 1999b. Prosodic Morphology (1986). In John Gold-
smith (ed.), Phonological theory: the essential readings. Malden, MA: Blackwell,
238–88.
McCawley, J. D. 1968. The phonological component of a grammar of Japanese. The
Hague: Mouton.
McDonough, Joyce. 1990. Topics in the phonology and morphology of Navajo verbs.
Amherst: GLSA.
McGuckin, Catherine. 2002. Gapapaiwa. In John Lynch, Malcolm Ross, and Terry
Crowley (eds.), The Oceanic languages. Richmond: Curzon Press, 297–321.
McLaughlin, Fiona. 2000. Consonant mutation and reduplication in Seereer-Siin.
Phonology 17:333–63.
Menn, Lise and Brian MacWhinney. 1984. The repeated morph constraint: toward an
explanation. Language 60:519–41.
Mester, Armin. 1986. Studies in tier structure. PhD thesis, University of Massachusetts,
Amherst.
238 References

Mester, Armin and Junko Itô. 1995. The core-periphery structure of the lexicon and con-
straints on reranking: University of Massachussets Occasional Papers in Linguistics
18. Amherst, MA: GLSA.
Mintz, Malcolm W. 1994. A student’s grammar of Malay and Indonesian. Singapore:
EPB Publishers Pte Ltd.
Mithun, Marianne. 1984. The evolution of noun incorporation. Language 60:847–94.
Mithun, Marianne. 2000. Incorporation. In Geert Booij, Christian Lehmann, and Joachim
Mugdan (eds.), Morphologie – morphology. Berlin: Walter de Gruyter, 916–28.
Mohanan, K. P. 1986. Lexical phonology. Dordrecht: Kluwer.
Mohanan, K. P. 1995. The organization of the grammar. In John Goldsmith (ed.), The
handbook of phonological theory. Cambridge, MA: Blackwell, 24–69.
Moravcsik, Edith. 1978. Reduplicative constructions. In Joseph Greenberg (ed.),
Universals of human language. Stanford: Stanford University Press, 297–334.
Mtenje, Al. 1988. On tone and transfer in Chichewa reduplication. Linguistics 26:125–
55.
Mutaka, Ngessimo and Larry M. Hyman. 1990. Syllable and morpheme integrity in
Kinande reduplication. Phonology 7:73–120.
Nagaraja, K. S. 1979. Contraction of Khasi nouns in compounds. Indian Linguistics
40:18–23.
Nash, David. 1986. Topics in Warlpiri grammar. New York: Garland.
Nater, Henry. 1984. The Bella Coola language: Canadian Ethnology Service Paper 92.
Ottawa: Museum of Man.
Nater, Henry. 1990. A concise Nuxalk–English dictionary: Canadian Ethnology Paper
115. Ottawa: Museum of Civilization.
Nespor, Marina and Irene Vogel. 1986. Prosodic phonology. Dordrecht: Foris.
Newman, Paul. 1986. Tone and affixation in Hausa. Studies in African Linguistics
17:249–67.
Newman, Paul. 1989a. The historical change from suffixal to prefixal reduplication in
Hausa pluractional verbs. Journal of African Languages and Linguistics 11:37–44.
Newman, Paul. 1989b. Reduplication and tone in Hausa ideophones. In Kira Hall,
Michael Meacham, and Richard Shapiro (eds.), Proceedings of the fifteenth annual
meeting of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society,
248–55.
Newman, Paul. 2000. The Hausa language: an encyclopedic reference grammar. New
Haven: Yale University Press.
Newman, Stanley. 1971. Bella Coola reduplication. International Journal of American
Linguistics 37:34–38.
Nguyen, Dı̀nh-hoà. 1997. Vietnamese: Tieng Viet Khong Son Phan. Amsterdam: John
Benjamins.
Niepokuj, Mary Katherine. 1997. The development of verbal reduplication in Indo-
European. Washington: Institute for the Study of Man.
Noonan, Michael. 1992. A grammar of Lango. Berlin: Mouton de Gruyter.
Nunberg, Geoffrey, Ivan A. Sag, and Thomas Wasow. 1994. Idioms. Language 70:491–
538.
Onn, Farid M. 1976. Aspects of Malay phonology and morphology: a generative
approach. PhD thesis, University of Illinois.
 References 239

Orgun, Cemil Orhan. 1994. Monotonic cyclicity and Optimality Theory. In Mercè
Gonzàlez (ed.), Proceedings of the Northeastern Linguistic Society 24. Amherst,
MA: GLSA, 461–74.
Orgun, Cemil Orhan. 1996. Sign-based morphology and phonology: with special atten-
tion to Optimality Theory. PhD thesis, University of California, Berkeley.
Orgun, Cemil Orhan. 1997. Cyclic and noncyclic effects in a declarative grammar.
Yearbook of morphology 1997:179–218.
Orgun, Cemil Orhan. 1999. Sign-based morphology: a declarative theory of phonology-
morphology interleaving. In Ben Hermans and Marc van Oostendorp (eds.),
The derivational residue in phonological Optimality Theory. Amsterdam: John
Benjamins, 247–67.
Orgun, Cemil Orhan. 2002. Phonology-morphology interaction in a constraint-based
framework. Unpublished ms., University of California, Davis.
Orgun, Cemil Orhan and Sharon Inkelas. 2002. Reconsidering bracket erasure. Yearbook
of Morphology 2001:115–46.
Ourn, Noeurng and John Haiman. 2000. Symmetrical compounds in Khmer. Studies in
Language 24:483–514.
Park, Miae. 2000. Surface opacity and phonological issues in Klamath and Lushootseed.
PhD thesis, University of Wisconsin-Madison.
Parker, G. J. 1968. Southeast Ambrym verb inflection and morphophonemics. Papers
in Linguistics of Melanesia. Pacific Linguistics Series A-15:27–40.
Patterson, Trudi Alice. 1990. Theoretical aspects of Dakota morphology and phonology.
PhD thesis, University of Illinois.
Payne, David. 1981. The phonology and morphology of Axininca Campa. Arlington,
TX: Summer Institute of Linguistics.
Pesetsky, David. 1979. Russian morphology and lexical theory. Unpublished ms.,
Massachusetts Institute of Technology.
Peterson, David. 2001. The Ingush clitic ʔa: the elusive type 5 clitic? Language 77:144–
55.
Piñon, Chris. 1991. Falling in paradise: verbs, preverbs and reduplication in Hungarian.
Handout from talk delivered at Syntax Workshop, 5/21/91, Stanford University.
Poser, William J. 1984. The phonetics and phonology of tone and intonation in Japanese.
PhD thesis, Massachusetts Institute of Technology.
Poser, William J. 1989. The metrical foot in Diyari. Phonology 6:117–48.
Prince, Alan and Paul Smolensky. 1993. Optimality theory: constraint interaction in
generative grammar. Cognitive Science Center Technical report TR-2, Rutgers
University.
Pulleyblank, Douglas. 1986. Tone in lexical phonology. Dordrecht: D. Reidel.
Pulleyblank, Douglas. 1988. Vocalic underspecification in Yoruba. Linguistic Inquiry
19:233–70.
Pulleyblank, Douglas. 1998. Yoruba vowel patterns: deriving asymmetries by the ten-
sion between opposing constraints. Available online from the Rutgers Optimality
Archive, ROA# 270-0798.
Pulleyblank, Douglas. To appear. Patterns of reduplication in Yoruba. In Kristin Hanson
and Sharon Inkelas (eds.), The nature of the word: essays in honor of Paul Kiparsky.
Cambridge, MA: MIT Press.
240 References

Pykllänen, Liina. 1999. On base-reduplicant identity. In P. Norquest (ed.), Proceed-

ings of the eighteenth West Coast Conference on Formal Linguistics. Somerville:
Cascadilla Press, 448–61.
Radhakrishnan, R. 1981. The Nancowry word: phonology, affixal morphology and roots
of a Nicobarese language. Edmonton: Linguistic Research, Inc.
Raimy, Eric. 2000. The phonology and morphology of reduplication. Berlin: Mouton.
Raimy, Eric and William Idsardi. 1997. A minimalist approach to reduplication in
Optimality Theory. In Kiyomi Kusumoto (ed.), Proceedings of the Northeastern
Linguistic Society 27. Amherst, MA: GLSA, 369–82.
Ramos, Teresita V. and Resty M. Caena. 1990. Modern Tagalog: Grammatical expla-
nations and exercises for non-native speakers. Honolulu: University of Hawaii
Press.
Rehg, Kenneth L. and Damian G. Sohl. 1981. Ponapean Reference Grammar. Honolulu:
University of Hawaii Press.
Rice, Keren. 1989. A grammar of Slave. Berlin: Mouton de Gruyter.
Riehemann, Susanne. 2001. A constructional approach to idioms and word formation.
PhD thesis, Stanford University.
Roberts, John. 1987. Amele. London: Croom Helm.
Roberts, John. 1991. Reduplication in Amele. Papers in Papuan Linguistics 1:115–46.
Robins, R. H. 1959. Nominal and verbal derivation in Sundanese. Lingua 8:337–69.
Robinson, J. O. 1976. His and hers morphology: the strange case of Tarok possessives. In
Larry Hyman, L. Jacobson, and Russell Schuh (eds.), Studies in African Linguistics,
Supplement 6. Los Angeles: Department of Linguistics, UCLA, 201–09.
Rose, Sharon. 1999. The (non)interaction of lenition and reduplication. Papers from the
Regional Meetings, Chicago Linguistic Society 35:277–92.
Rose, Sharon and Rachel Walker. 2001. A typology of consonant agreement as corre-
spondence. Unpublished ms., University of California, San Diego; and University
of Southern California. Available online from the Rutgers Optimality Archive,
ROA# 405–0800.
Rosen, Sara. 1989. Two types of noun incorporation. Language 65:294–317.
Ross, Malcolm. 2002. Kaulong. In John Lynch, Malcolm Ross, and Terry Crowley (eds.),
The Oceanic languages. Richmond: Curzon Press, 387–409.
Rubino, Carl. 2001. Iconic morphology and word formation in Ilocano. In F. K. Erhard
Voeltz and Christa Kilian-Hatz (eds.), Ideophones. Amsterdam: John Benjamins,
303–20.
Sag, Ivan and Thomas Wasow. 1999. Syntactic theory: a formal introduction. Stanford:
CSLI Publications.
Saperstein, Andrew. 1997. A word-and-paradigm approach to reduplication. PhD thesis,
Ohio State University.
Sapir, Edward. 1930. Southern Paiute, A Shoshonean language. Proceedings of the
American Academy of Arts and Sciences 65, no. 1, Cambridge, MA.
Schachter, Paul and Fe Otanes. 1972. Tagalog reference grammar. Berkeley: University
of California Press.
Schmidt, Hans. 2002. Rotuman. In John Lynch, Malcolm Ross, and Terry Crowley
(eds.), The Oceanic languages. Richmond: Curzon Press, 815–32.
Schuh, Russell. 1998. A grammar of Miya. University of California Publications in
Linguistics 130. Berkeley: University of California Press.
 References 241

Selkirk, Elisabeth. 1982. The syntax of words. Cambridge, MA: MIT Press.
Sezer, Engin. 1981. The k/Ø alternation in Turkish. In George N. Clements (ed.),
Harvard Studies in phonology. Bloomington: Indiana University Linguistics Club,
354–82.
Shaw, Patricia A. 1980. Theoretical issues in Dakota phonology and morphology: Out-
standing dissertations in linguistics. New York: Garland.
Shaw, Patricia A. 1985. Modularisation and substantive constraints in Dakota Lexical
Phonology. Phonology 2:173–202.
Sherrard, Nicholas. 2001. Blending and reduplication. PhD thesis, University of Essex.
Sibanda, Galen. 2004. Issues in the phonology and morphology of Ndebele. PhD thesis,
University of California, Berkeley.
Sibomana, Leo. 1980. Grundzüge der Phonologie des Tarok (Yergam). Afrika und
Übsersee 63:199–206.
Sibomana, Leo. 1981. Tarok II: Das Nominalklassensystem. Afrika und Übersee 64:25–
34.
Sietsema, Brian M. 1988. Reduplications in Dakota. In Papers from the regional meeting,
Chicago Linguistic Society. Chicago: Chicago Linguistic Society, 337–52.
Singh, Rajendra. 1982. On some “redundant compounds” in Modern Hindi. Lingua
56:345–51.
Smith, Jennifer. 1997. Noun faithfulness: on the privileged status of nouns in phonology.
Available online from the Rutgers Optimality Archive, ROA# 242-1098.
Smith, Jennifer. 1998. Noun faithfulness and word stress in Tuyuca. In Jennifer Austin
and Aaron Lawson (eds.), Proceedings of ESCOL 97. Ithaca: CLC Publications,
180–91.
Smolensky, Paul. 1993. Harmony, markedness and phonological activity. Rutgers
University: Rutgers Optimality Workshop 1.
Soltész, Katalin J. 1959. As ösi magyar igekötök [The ancient Hungarian preverbs].
Budapest: Akadémia Kiadó.
Sommer, Bruce. 1981, The shape of Kenjen syllables. In Didier L. Goyvaerts (ed.),
Phonology in the 1980s. Ghent: E. Story-Scientia, 231–44.
Spaelti, Philip. 1997. Dimensions of variation in multi-pattern reduplication. PhD thesis,
University of California, Santa Cruz.
Spencer, Andrew. 1998. Morphophonological operations. In Andrew Spencer and
Arnold Zwicky (eds.), Handbook of morphology. Oxford: Blackwell, 123–
43.
Spring, Cari. 1990. Implications of Axininca Campa for Prosodic Morphology and
reduplication. PhD thesis, University of Arizona.
Sproat, Richard. 1986. Malayalam compounding: a non-stratum ordered account. In
Mary Dalrymple, Jeffrey Goldberg, Kristin Hanson et al. (eds.), Proceedings of the
fifth West Coast Conference on Formal Linguistics. Stanford: Stanford Linguistics
Association, 268–88.
Sprouse, Ronald. 1997. A case for enriched inputs. Available online from the Rutgers
Optimality Archive, ROA# 193–0597.
Sprouse, Ronald. In preparation. Allomorphy in the lexicon: blocking, opacity, and
ungrammaticality. PhD thesis, University of California, Berkeley.
Srichampa, Sophana. 2002. Vietnamese verbal reduplication. In Robert S. Bauer (ed.),
Collected papers on Southeast Asian and Pacific languages. Canberra: Pacific
242 References

Linguistics, Research School of Pacific and Asian Studies, Australian National

University, 37–47.
Steriade, Donca. 1988. Reduplication and syllable transfer in Sanskrit. Phonology 5:73–
155.
Steriade, Donca. 1997. Lexical conservatism and its analysis. Selected papers from
SICOL 1997. Seoul: Hanshin, 157–79.
Steriade, Donca. 1999. Lexical conservatism in French adjectival liaison. In Marc
Authier, Barbara Bullock, and Lisa Reed (eds.), Formal perspectives on Romance
linguistics. Amsterdam: John Benjamins, 243–70.
Stevens, Alan M. 1985. Reduplication in Madurese. In Proceedings of the Second East-
ern States Conference on Linguistics. Columbus: Linguistics Department, the Ohio
State University, 232–42.
Steyaert, Marcia. 1976. Verb reduplication in Dakota. Minnesota working papers in
linguistics and philosophy of language. Minneapolis, MI: University of Minnesota
Linguistics Department, 127–43.
Struijke, Caro. 1998. Reduplicant and output TETU in Kwakwala. In H. Fukazawa,
F. Morelli, C. Struijke, and Y. Su (eds.), University of Maryland working papers,
vol. 7: Papers in phonology. College Park: University of Maryland, 150–78.
Struijke, Caro. 2000a. Existential Faithfulness: a study of reduplicative TETU, feature
movement, and dissimilation. PhD thesis, University of Maryland.
Struijke, Caro. 2000b. Why constraint conflict can disappear in reduplication. In Mako
Hirotani (ed.), Proceedings of the Northeastern Linguistics Society 30. Amherst,
MA: CLSA, 613–26.
Stump, Gregory. 1998. Comments on the paper by Inkelas and Orgun. In Steven Lapointe,
Diane Brentari, and Patrick Farrell (eds.), Morphology and its interaction with
syntax and phonology. Stanford, CA: CSLI Publications, 393–405.
Sumukti, Rukmantoro Hadi. 1971. Javanese morphology and phonology. PhD thesis,
Cornell University.
Suzuki, Keiichiro. 1999. Identity avoidance vs. identity preference: the case of Sun-
danese. Paper presented at annual meeting of the Linguistic Society of America,
Los Angeles.
Tauli, V. 1966. Structural tendencies in Uralic languages. London: Mouton.
Thomas, Linda May Kopp. 1975. Klamath vowel alternations and the segmental cycle.
PhD thesis, University of Massachusetts, Amherst: GLSA.
Topping, Donald. 1973. Chamorro reference grammar. Honolulu: University of Hawaii
Press.
Uhlenbeck, E. M. 1953. Word duplication in Javanese. Bijdragen tot de Taal-, Land-
en Volkenkunde 109:52–61. Reprinted in Uhlenbeck, 1978, Studies in Javanese
morphology. The Hague: Martinus Nijhoff, pp. 89–97.
Uhlenbeck, E. M. 1954. Duplication in the morphology of the Javanese verb. Bijdra-
gen tot de Taal-, Land- en Volkenkunde 110:369–87. Reprinted in E. M. Uhlen-
beck (ed.), 1978, Studies in Javanese morphology. The Hague: Martinus Nijhoff,
pp. 98–115.
Urbanczyk, Suzanne. 1996. Patterns of reduplication in Lushootseed. Amherst:
GLSA.
 References 243

Urbanczyk, Suzanne. 1998. A-templatic reduplication in Halq’eméylem. In Kimary

Shahin, Susan Blake, and Eun-Sook Kim, (eds.), Proceedings of the Seventeenth
West Coast Conference on Formal Linguistics. Stanford: CSLI Publications, 655–
69.
Vanbik, Kenneth. 2003. Junctural and parasitic voicing in Burmese. Paper presented at
Berkeley Linguistics Society 29.
Vaux, Bert. 1996. Abkhaz Mabkhaz: m-reduplication in Abkhaz and the Problem of
Melodic Invariance. Unpublished MS, Harvard University.
Vaux, Bert. 1998. The Phonology of Armenian. Oxford: Clarendon Press.
Vaux, Bert. Forthcoming. The Cwyzhy dialect of Abkhaz. Stanford: CSLI Publications.
Walker, Rachel. 2000a. Long-distance consonantal identity effects. In Roger Billerey
and Brook Lillehaugen (eds.), Proceedings of the nineteenth West Coast Conference
on Formal Linguistics. Somerville: Cascadilla Press, 532–45.
Walker, Rachel. 2000b. Yaka nasal harmony: spreading or segmental correspondence?
In Lisa Conathan, Jeff Good, Darya Kavitskaya, Alyssa Wulf, and Alan C.-L. Yu
(eds.), Proceedings of the Berkeley Linguistics Society 26. Berkeley, CA: Berkeley
Linguistics Society, 321–32.
Walker, Rachel. To appear. Consonantal correspondence. Proceedings of the workshop
on the lexicon in phonetics and phonology. Edmonton: University of Alberta.
Walsh, D. S. 1982. Variation of verb-initial consonants in some Eastern Oceanic lan-
guages. Pacific Linguistics Series C-74:231–42.
Wash, Barbara. 1995. Productive reduplication in Barbareño Chumash. MA thesis, Uni-
versity of California, Santa Barbara.
Watson, Richard L. 1966. Reduplication in Pacoh. MA thesis, Hartford Seminary Foun-
dation.
Weber, David John. 1989. A grammar of Huallaga (Huánuco) Quechua. Berkeley:
University of California Press.
Wee, Lionel. 1994. Identifying the reduplicant. Paper presented at Trilateral Phonology
Weekend (TREND) II, Stanford University.
Weeda, Donald Stanton. 1992. Word truncation in prosodic morphology. PhD thesis,
University of Texas, Austin.
Wells, Margaret. 1979. Siroi grammar. Pacific Linguistics Series B 51.
White, Robin Barbara Davis 1973. Klamath phonology. Studies in Linguistics and
Language Learning, vol. 12. Seattle: University of Washington.
Wierzbicka, Anna. 1986. Italian reduplication: cross-cultural pragmatics and illocution-
ary semantics. Linguistics 24:287–315.
Wiesemann, U. 1972. Die phonologische und grammatische Struktur der Kaingang
Sprache. The Hague: Mouton.
Wilbur, Ronnie. 1973. The phonology of reduplication. Bloomington: Indiana University
Linguistics Club.
Wise, Mary Ruth. 1986. Grammatical characteristics of PreAndine Arawakan languages
of Peru. In Desmond C. Derbyshire and Geoffrey K. Pullum (eds.), Handbook of
Amazonian languages, vol. I. Berlin: Mouton de Gruyter, 567–642.
Yip, Moira. 1982. Reduplication and CV skeleta in Chinese secret languages. Linguistic
Inquiry 13:637–61.
244 References

Yip, Moira. 1997. Repetition and its avoidance: the case of Javanese. In Keiichiro Suzuki
and D. Elzinga (eds.), Arizona phonology conference Proceedings, vol. V: Pro-
ceedings of the South Western Workshop on Optimality Theory (SWOT). Tucson:
University of Arizona, Department of Linguistics coyote Papers, 238–62. Available
electronically from the Rutgers Optimality Archive, ROA# 82-00002.
Yip, Moira. 1996. Lexicon optimization in languages without alternations. In Jacques
Durand and Bernard Laks (eds.), Current trends in phonology: models and methods.
Salford: University of Salford Publications, 757–88.
Yip, Moira. 1998. Identity avoidance in phonology and morphology. In Steven LaPointe,
Diane Brentari, and Patrick Farrell (eds.), Morphology and its relations to phonol-
ogy and syntax. Stanford: CSLI Publications, 216–46.
Yip, Moira. 1999. Reduplication as alliteration and rhyme. GLOT International 4:1–7.
Yoquelet, Corey. 2004. Grammaticalization and the epiphenomenon of the so-called
repeated morpheme constraint. Unpublished ms., University of California, Berke-
ley.
Yu, Alan C.-L. 2000. Stress assignment in Tohono O’odham. Phonology 17:117–35.
Yu, Alan C.-L. 2003. The phonology and morphology of infixation. PhD thesis, Univer-
sity of California, Berkeley.
Zepeda, Ofelia. 1983. A Papago grammar. Tucson: University of Arizona Press.
Zepeda, Ofelia. 1984. Topics in Papago morphology. PhD thesis, University of Arizona.
Zigmond, Maurice L., Curtis G. Booth, and Pamela Munro. 1990. Kawaiisu: a grammar
and Dictionary with Texts. Berkeley: University of California Press.
Zimmer, Karl and Barbara Abbott. 1978. The k/Ø alternation in Turkish; some experi-
mental evidence for its productivity. Journal of Psycholinguistic Research 7:35–46.
Zoll, Cheryl. 2002. Vowel reduction and reduplication in Klamath. Linguistic Inquiry
33:520–27.
Zuraw, Kie Ross. 2000. Exceptions and regularities in phonology. PhD thesis, University
of California, Los Angeles.
Zuraw, Kie Ross. 2002. Aggressive reduplication. Phonology 19:395–439.
 Index of languages

Abkhaz 42, 79–80 Emai 14

Acehnese 63–64, 67 English 1, 32, 42, 91, 105, 156–57, 202
Amele 1, 29, 30, 31, 45–46
Amuesha 32 Fongbe 5
Apma 49, 50 Fox 124, 165–73
Arapeshan 206–09 Fula 51
Armenian 42, 80
Arosi 14 Gapapaiwa 29, 30
Arrernte 36–37
Axininca Campa 32 Haitian 213
Hausa 1, 2, 19–21, 46, 83–84, 86–87, 90, 91,
Babine 78 122–23, 127, 132–33, 198–99
Banoni 14–15 Hindi 59–60, 61
Bantu family 6, 32, 33, 38–39, 65, 208 Hopi 105
Barbaren.o Chumash (see also Hua 43–44, 82–83
Chumash) 186–87, 188, 189, 190–91 Hungarian 28
Bella Coola 127, 200
Bellonese 203 Ilokano 200
Bierebo 49 Indonesian 98–99, 103–04
Boumaa Fijian 29–30 Ineseño Chumash (see Chumash)
Brazilian Portuguese 213 Ingush 8
Bulgarian 224
Burmese 78, 131–32 Japanese 67, 74–75, 91, 99–100
Jaqaru 37
Celtic family 51 Javanese 136, 159–60, 215
Chaha 175, 177–78, 203 Johore Malay 145, 175, 221
Chamorro 106–08, 151, 155, 161
Chechen 8–9 Kaingang 215
Chichewa 33, 34–38 Kashaya 105
Chinese 43, 60 Kaulong 67
Chukchee 156, 201 Kawaiisu 48
Chumash 4, 156, 185–96 Khasi 36, 219
Khmer 8, 62–63, 205
Dakota 100–03, 119 Kikerewe 38
Diyari 79 Kinande 38, 39, 223
Dyirbal 27, 28, 30, 31 Klamath 113–18, 119, 175–76
Kolami 42, 59
Eastern Kadazan 152–55, 156, 176 Kwakwala 216

245
246 Index of languages

Lango 78 Russian 213, 224

Latin 32
Leti 97 Sahaptin 203
Lushootseed 98, 119–21, 125–27, 216 Salish family 127
Sanskrit 63, 76–77, 78, 213, 215
Madurese 22, 157, 198–99 Sekani 105
Malay (see also Johore Malay, Ulu Muar Serbo-Croatian 32–33
Malay) 212 Sereer 51
Malayalam 67, 105 Siroi 15, 44–45, 46
Mangarayi 96 Siswati 38
Marathi 205 Slave 33
Marquesan 14, 15 Southeast Ambrym 54–55, 57
Mende 51 Southern Paiute 219
Miya 202 Spokane 21
Modern Hebrew 3, 215 Sundanese 157–58
Mokilese 96 Sye 9–10, 14, 47, 52–54, 55, 57
Mongolian 42
Tagalog 78, 80–81, 156, 181–85
Nadrogā 15 Tamil 36
Nakanamanga 49–50 Tarok 14, 84–86, 119
Namakir 49 Tawala 67, 94–95, 96, 164–65
Nancowry 223–24 Telugu 42
Navajo 33 Temiar 223
Nāti 49, 50–51 Thao 213
Ndebele 10–11, 31, 32, 33, 38–41, 43, 65 Tibeto-Burman 131, 219
Nez Perce 203 Tohono O’odham 129–30
Nilotic 80 Tok Pisin 208–09
Niuafo’ou 15 Trukese 203
Nuu-chah-nulth 217 Turkic 42, 80
Turkish 33, 42, 59, 61, 67, 72, 105, 156, 198,
Oceanic languages 94 210–11
Orokaiva 92, 93–94
Oykangand 96 Ulithian 15
Ulu Muar Malay 96
Paamese 54, 55–59 Umpila 22, 156
Pacoh 37–38, 204–05 Urdu 60
Pali 63
Pingding (Mandarin) 97 Vanuatu, languages of (Central) 49, 52, 53, 54
Piro 32, 105–06 Vata 213
Ponapean 198–99, 217 Vietnamese 42, 43, 60, 62

Quechua (Huallago) 1, 36 Warlpiri 1, 4, 5–6, 8, 181, 220

Raga 49, 50 Yiddish 213

Rotuman 14 Yoruba 2–3, 21, 213
Roviana 94, 96, 130–31, 200 Yupik 79
 Index of names

Abbi, A. 36, 60, 61, 79, 80, 215 Burzio, L. 218

Abbott, B. 72 Bybee, J. 165
Akinlabi, A. 2–3
Alderete, J. 3, 42, 74, 75, 78, 81, 83, 99, 100, Caena, R. 78
216, 224 Carlson, K. 21, 127
Anderson, S. 12, 18, 31, 158 Carrier-Duncan, J. 182
Anttila, A. 70, 71, 105, 118, 218 Chittleborough, M. 93
Applegate, R. 4, 185–86, 188, 189, 190, 191, Cho, Y. 70, 218
195, 196 Chomsky, N. 3
Apte, M. 79, 205 Clements, G. 4, 135
Arnott, D. 51 Cohn, A. 98–103, 104, 108, 110–11, 112,
Aronoff, M. 32, 45, 90, 182, 184, 206–09 157, 158, 218
Austin, P. 79 Cole, J. 140, 145, 182, 184, 185
Collins, C. 5
Bagemihl, B. 204 Comrie, B. 67
Barker, M. 113, 114, 116, 118, 218 Conathan, L. 8–9, 166
Barnes, J. 78 Conteh, P. 51
Bat-El, O. 215 Corston-Oliver, S. 94, 130, 200
Bates, D. 21, 98, 119 Cowper, E. 51
Benua, L. 65, 74, 75, 157, 158 Crosswhite, K. 211, 224
Bhaskararo, P. 42 Crowley, T. 9, 10, 49, 50–51, 52–53, 54,
Bissell, T. 202 55–57, 58, 215
Black, D. 21 Csato, E. 42, 80
Blevins, James 45, 90
Blevins, Juliette (= Levin, J.) 96, 97, 156, 203 Dahlstrom, A. 124, 165–67, 168–70, 171, 172
Bloomfield, L. 214, 222 DeLacy, P. 124
Blust, R. 213 Delancey, S. 114
Boas, F. 100 Deloria, E. 100
Bochner, H. 12, 215 Dixon, R. 27, 28, 29, 30
Booij, G. 67, 140, 182, 184 Dobrin, L. 206–09
Booth, C. 48 Dolbey, A. 34
Bosson, J. 42 Downing, L. 6, 10, 11, 31, 33, 38, 39, 41, 65,
Bowen, D. 78 83, 89, 132, 140, 145, 184, 185, 188, 191,
Breen, G. 37 216
Broselow, E. 119, 127 Dressler, W. 67
Brousseau, A. 5 Dudas, K. 137, 138, 139, 144–45, 147–48,
Bruening, B. 42, 80 150, 173
Burkhardt, P. 124, 221 Durie, M. 63–64, 67

247
248 Index of names

Early, R. 15 Inkelas, S. 6, 10, 11, 12, 16, 31, 33, 34, 38, 39,
Egbokhare, F. 14 40–41, 43, 48, 65, 67, 70, 71, 72, 100, 105,
Elbert, S. 203 118, 132–33, 140, 156, 215, 218, 224
Emeneau, M. 42, 79 Innes, G. 51
English, L. 183 Isdardi, W. 118, 127
Ezard, B. 67, 94–95, 164 Isoroembo, A. 93
Itô, J. 34, 67, 70, 74, 91
Feinstein, M. 218
Fillmore, C. 16 Jeanne, L. 105
Fitzgerald, C. 119, 129, 130, 219 Johanson, L. 42, 80
Ford, A. 67 Jones, W. 124
Fortune, R. 206–07 Jurafsky, D. 45
Frampton, J. 159, 213
French, K. 182 Kager, R. 4, 34, 89, 112
Kaisse, E. 70
Gafos, D. 21, 213, 223 Kanerva, J. 33
Geraghty, P. 15 Kawu, A. 2–3
Gerdts, D. 61 Kay, P. 16
Goldberg, A. 16 Kean, M. 218
Good, J. 8–9 Kelepir, M. 79
Goodenough, W. 203 Kenstowicz, M. 65, 72, 74, 104, 105, 108,
Gorgoniev, Y. 62 109, 112, 173, 177–78
Green, A. 131 Keyser, S. 218
Gurevich, N. 219 Kim, E. 217
Kiparsky, P. 4, 17, 70, 71, 105, 124, 135, 136,
Haiman, J. 8, 43–44, 47, 59, 60, 61, 62–63, 150, 156, 165, 201, 204, 205, 209, 213, 215,
82, 205 221
Halle, M. 218 Kisseberth, C. 105, 218
Hamilton, P. 89 Kiyomi, S. 14
Hanson, K. 205 Klein, T. 107, 108, 221
Hansson, G. 21, 177, 178, 179, Koenig, J.-P. 16, 45
198 Kornfilt, J. 33
Haraguchi, S. 75 Krause, S. 22, 156, 201
Hardman, M. 37 Kroeger, P. 96
Hargus, S. 71, 78 Kurisu, K. 80, 211
Harris, B. 22, 156
Harrison, S. 96 Landau, I. 3, 5, 6
Hayes, B. 71 Lathroum, A. 182
Healey, A. 93 Lefebvre, C. 5
Hendricks, S. 21, 213, 214, 223 Levin, J. (= Blevins, Juliette) 22, 96, 174, 203
Hess, T. 98, 119 Lewis, G. 33, 42
Hilbert, V. 98, 119 Li, C. 60
Hill, J. 129 Lieber, R. 12, 13, 17, 140, 182, 184
Hockett, C. 32 Lombardi, L. 155
Holtman, A. 204 Longtau, S. 84
Horne, E. 137, 138, 139, 144–45 Lubowicz, A. 219
Horoi, R. 14 Lynch, J. 14, 15, 49
Hualde, J. 71
Hurlbut, H. 152, 154–55, 176 MacWhinney, B. 210–11
Hyman, L. 5, 6, 10, 11, 31, 33, 38, 39, 40–41, Malkiel, Y. 215
65, 143, 223 Marantz, A. 4, 27, 68, 100
 Index of names 249

Mascaro, J. 218 Petros Banksira, D. 177–78

Matteson, E. 105 Piñon, C. 28
McCarthy, J. 4, 5, 18, 20, 22, 24, 26, 27, 30, Poser, W. 67, 75, 79, 99, 100, 217
32, 42, 68, 77, 78, 79, 85, 86, 89, 92, 93, Prince, A. 4, 5, 18, 20, 24, 26, 27, 30, 32,
96–97, 99, 104, 108, 109–10, 111, 112, 34, 42, 68, 77, 78, 79, 85, 86, 89, 92, 93,
113–14, 115, 116, 118, 125, 135, 136, 138, 96–97, 99, 111, 113, 114, 115, 116, 118,
144, 155, 156, 161, 163, 172, 174, 176, 179, 124, 129, 135, 136, 137, 138, 144, 155, 156,
182, 185, 186, 187, 193, 218 161, 163, 174, 176, 182, 185, 186, 187,
McCawley, J. 67, 75, 217 193
McDonough, J. 33, 214 Pulleyblank, D. 2, 6, 70, 213
McGuckin, C. 29, 30 Pyllkänen, L. 140
McLaughlin, F. 51
Menn, L. 210 Radhakrishnan, R. 223
Mester, A. 4, 34, 67, 70, 74, 135, 158 Raimy, E. 118, 127, 159, 200, 213, 219
Mintz, M. 212 Ramos, T. 78
Mithun, M. 61, 212 Rehg, K. 199
Mohanan, K. P. 67, 70, 71 Rhodes, R. 221
Moravcsik, E. 7, 14, 28, 37 Rice, K. 33, 51, 214
Mtenje, A. 33, 38 Riehemann, S. 16
Munro, P. 48 Roberts, J. 1, 29, 30, 45, 213
Mutaka, N. 38, 143, 223 Robins, R. 157
Robinson, J. 14, 84
Nagaraja, K. 219 Rose, S. 21, 119, 177, 179
Nash, D. 4, 220 Rosen, S. 212
Nater, H. 127 Ross, M. 15, 49, 67
Nespor, M. 140 Rubino, C. 200
Newman, P. 2, 3, 19, 46, 83, 86, 90, 122, 127,
132, 178, 179 Sag, I. 16
Newman, S. 127 Saperstein, A. 6, 8, 215, 216, 222
Nguyen, D. 8, 59, 60 Schachter, P. 78, 80–81, 182, 183, 184
Nichols, J. 213 Schmidt, H. 14
Niepokuj, M. 14, 83, 84, 174, 202, 216 Schuh, R. 202
Noonan, M. 78 Selkirk, E. 12
Nunberg, G. 16 Sezer, E. 72
Shaw, P. 70, 100–01, 135
O’Connor, C. 16 Sherrard, N. 6
O’Grady, G. 22, 156 Sibanda, G. 6, 10, 11, 31, 32, 33, 38, 39,
Onn, F. 137, 145 40–41, 65
Orgun, O. 6, 12, 16, 45, 61, 67, 70, 71, 72, 74, Sibomana, L. 84
105, 118 Sietsema, B. 101
Otanes, F. 78, 80–81, 182, 183, 184 Singh, R. 47, 59, 60, 67, 215
Ourn, N. 8, 47, 59, 60, 61, 62–63, 205 Smith, J. 75, 105
Smolensky, P. 20, 34, 111, 124, 129
Padgett, J. 34 Soltész, K. 28
Park, M. 114, 219 Sommer, B. 96
Parker, G. 55 Spaelti, P. 118, 119
Patterson, T. 100–01 Spencer, A. 67
Payne, D. 32 Srichampa, S. 214
Pensalfini, R. 37 Spring, C. 32
Pesetsky, D. 70 Sproat, R. 140
Peterson, D. 8 Sprouse, R. 34, 156, 172, 213, 214
250 Index of names

Steriade, D. 4, 6, 19, 34, 65, 68, 69, 76, 77, Wasow, T. 16

78, 81, 197, 213, 215, 216, 224 Watson, R. 37, 204
Stevens, A. 156, 199 Weber, D. 1, 36
Steyaert, M. 129 Wee, L. 96
Struijke, C. 77, 99, 118, 119, 120, 125, 216 Weeda, D. 5, 79, 157, 204
Stump, G. 71, 118 Wells, M. 15, 45, 46
Sumukti, R. 137, 138, 139, 220 White, R. 114, 118
Suzuki, K. 211 Wierzbicka, A. 47
Wiesemann, U. 215
Tauli, V. 28 Wilbur, R. 18, 68, 77, 135, 136, 158, 161, 173,
Thomas, L. 114, 218 213, 222
Thompson, S. 60 Wiltshire, C. 4
Timberlake, A. 224 Wise, M. 32
Topping, D. 106–07, 151 Wood, E. 166
Woodbury, A. 79
Uhlenbeck, E. 137, 145
Urbanczyk, S. 3, 30, 86, 89, 98, 119, 125–26, Yip, M. 6, 34, 43, 79, 80, 204–09, 215
213 Yoquelet, C. 211
Yu, A. 16, 70, 97, 129–30, 154, 155, 156, 202,
Vago, R. 218 217, 219, 223
Vanbik, K. 131
Vaux, B. 42, 80 Zepeda, O. 129
Vogel, I. 140 Zigmond, M. 48
Zimmer, K. 72
Walker, R. 21, 177, 179, 198 Zoll, C. 16, 70, 71, 114, 117–18, 175, 217,
Walsh, D. 49, 50 219
Wash, B. 185, 186–87, 189–90, 191 Zuraw, K. 21, 53, 70, 177, 198, 204, 213, 221
 Index of subjects

ablaut 17 meta-constructions 12, 118

affix reduplication 27–31 phonology of: see cophonologies
‘Aggressive Reduplication’ 221 semantics of 13–16, 28
alliteration (see also rhyme, consonance) 204 cophonologies 16–18, 19–20, 24, 69, 70–88,
‘alliterative concord’ 205–09 89, 97, 99, 100, 102, 103–04, 106, 115,
allomorphy in Optimality Theory 34 117–18, 120, 123, 124, 127, 129–30, 133
A-morphous Morphology 31 cyclic/layered nature of 103, 136, 148–49,
anti-identity effects 210–11 151–52, 154–56
in reduplication (see also inside-out nature of (see also cyclicity) 74,
dissimilation) 79–80 75
antonym constructions (see also synonym Master Ranking of 71
constructions, near-synonym of daughter nodes in reduplication (see also
constructions) 47, 61–62, 63, 64 the Independent Daughter Prediction) 18,
a-templatic reduplication 213 19, 25, 69, 75–97, 140, 161
of mother node in reduplication (see also
backcopying 24, 113–14, 174–75, 182, the Mother Node Prediction) 19, 25, 69,
186–87 70, 71, 75–77, 95, 96, 98, 155, 161
in morphological reduplication 114–18, scope of 74, 75, 100, 133, 162–73
175–76, 182–85, 193 vs. constraint indexation 74–75, 100, 106
in phonological assimilation 177–79 Copy and Association theory of reduplication
‘base-dependence’ in reduplication 92–97 (see also templates in reduplication) 3–4,
‘base-priority’ in Optimality Theory 74 68
Base-Reduplicant Correspondence Theory Correspondence Theory: see Optimality
(BRCT) (see also backcopying, Coerced Theory
Identity theories, Kager–Hamilton cyclicity (see also layering) 16, 20, 51, 70, 75,
problem) 4, 8, 23, 24, 26, 41, 65, 68, 77, 99, 110, 133, 136
80, 82, 85–86, 88–89, 99, 102–03 as source for opacity 135–36, 138–43,
opacity in 104, 106, 107–08, 109–12, 146–47, 148–50, 153–55, 156–58, 161,
113–14, 115–16, 135, 138, 149, 158–60, 169, 172, 173
161, 165
derived environment effects 105, 129–30
Coerced Identity theories 18, 23, 24, 68, 77, dissimilation, input–output (see also Melodic
81, 104, 107–08, 149 Overwriting) 78, 80–81, 211
consonance (vs. reduplication) 204 divergent allomorphy in reduplication 8–11,
constructions (morphological) 7, 11–20, 25, 40
45, 61, 201 semantically empty morphs 10–11, 27, 31,
in reduplication 7–11, 15–16, 19–20, 34, 36
37, 44, 46, 57–59, 60–62, 68, 185, 188 (see also Melodic Overwriting)

251
252 Index of subjects

divergent allomorphy in reduplication (cont.) Lexicon Optimization: see Optimality

suppletive allomorphy 9–10, 11, 47–51, Theory
53–54, 55, 56–59, 64 linker morphs 36–37, 43
divergent modification: see phonology of
reduplication ‘Master Ranking’: see cophonologies
Divergent Modification Ranking Paradox 102 Melodic Overwriting 27, 42, 46, 79, 83
dominant-recessive affixes 99–100, 133 dissimilation in 79–80
double reduplication 29–30 double Melodic Overwriting 43–45, 83,
122
echo reduplication: see Melodic Overwriting tier replacement 45–46
empty morphs: see semantically empty meta-constructions: see constructions
morphology Morpheme Integrity Principle 143, 190, 191
Enriched Input Theory 214 morphological constructions: see
Existential Faithfulness (∃-Faith) 24, 99, constructions
118–33 morphologically conditioned phonology (see
problems for 121–33 also cophonologies) 12, 16, 24, 45, 67,
exocentricity 13, 15, 63 70–97, 98–100, 105–06, 114–15, 118,
128–33
feature percolation 13 morphosyntactic (MS) feature duplication 2,
Full Copy theory of reduplication 26, 68 3, 4, 5–6, 8, 9, 20, 22, 104
vs. phonological copying: see semantic
Generalized Phonology Prediction 69, 70, 78, identity vs. phonological identity
79, 81–82, 86, 105 Mother Node Prediction 70
Generalized Template Theory 30, 86, 89 mutation 51, 54

iconicity in reduplication: see semantic Native Identity theories 18, 68, 77–78, 81,
functions of reduplication 104, 110
identity: see phonological identity, semantic ‘near-synonym constructions’ (see also
identity synonym constructions) 47, 61–65
Identity Principle (Wilbur) 68, 135, 158–59 nonderived environment blocking: see derived
ideophones 1 environment effects
idiomatic meaning 12, 13, 15
Independent Daughter Prediction 69, 82 opacity effects in phonology (see also opacity
independent inputs in reduplication 26 effects in reduplication) 82, 135,
infixation 93–94, 155–56, 193 156–58
as source of phonological opacity in opacity effects in reduplication (see also
reduplication 153–54, 157–58, 183–85, cyclicity/layering, Base-Reduplicant
187–89, 193–95 Correspondence Theory) 20, 24, 135,
internal reduplication (see also 136–38, 141–42, 144–46, 147–48, 151,
infixation) 156, 182 152, 168, 172, 181–82, 186–87
item-based morphology 6, 12, 18, 46, 73, overapplication 24, 123, 135, 146, 150,
91 159–60, 161, 162, 164, 182, 186
underapplication 24, 99, 104, 105–06, 135,
‘Kager–Hamilton problem’ 89, 214 137, 148, 150, 159–60, 161, 162
opposite-edge reduplication 22, 156, 199
layering: see also cyclicity 102 Optimality Theory 6, 20, 21, 79, 92, 99
level ordering theory 17, 71, 105, 189–90 allomorphy in 34, 58
Lexical Conservatism 6 epenthesis in 124, 125
Lexical Morphology and Phonology (see also Lexicon Optimization 34
level ordering theory) 70–71 markedness in 124
 Index of subjects 253

opacity in 172 realizational morphology 6, 12, 18, 31, 73, 91

output–output correspondence in 65, 221 Reduplicative Blending Theory templates in
reduplication in: see Base-Reduplicant reduplication (see also copy and
Correspondence Theory association theory) 6, 21
output–output correspondence: see Optimality ‘redundant compounds’ (see also synonym
Theory constructions) 59
overapplication: see opacity effects in Repeated Morph Constraint 210–11
reduplication repetition 1
rhyme, compared with reduplication 204–05,
percolation: see feature percolation 209
phonological copying 2–4, 5, 6, 9, 20–21, 22,
26, 27, 51 Scopal Opacity Prediction 162
as distinct from morphological semantic function of reduplication 13–16
reduplication 3–4, 48, 197–203 derivational 14–15, 65
as side effect of semantic identity 18, 22, iconic semantics 13–14, 27
23, 66 non-iconic semantics 14
criterion for phonological copying 22–23, semantically vacuous 22, 200
197 semantic identity in reduplication 6, 7–10, 11,
identity as criterion for 23, 197 18, 25, 26, 27, 31, 38, 40, 44, 47, 62, 65,
not enforced anti-identity effects 23, 25, 36, 66, 91, 92, 97, 197
64, 92, 150, 210 vs. phonological identity 10–11, 27, 31, 36,
parallels with phonological 43, 44, 54, 65, 197, 209
assimilation 198–99 semantically empty morphology (see also
phonological purpose of phonological linker morphs, Melodic Overwriting,
identity effects in reduplication 21, 22, semantically vacuous reduplication) 27,
197–99 31–47, 90
vs. semantic identity, in reduplication 22, ‘semantically symmetrical compounds’
26, 37, 41, 47, 48 (see also synonym compounds) 59
phonological subcategorization 155 Sign-Based Morphology 6–7, 12–13, 72
phonology of reduplication 4, 18–20, 25, single-segment reduplication 1, 21
67–70, 75–97, 99, 106–18, 119, 122–23, stem type 44, 45, 53, 57–59, 83, 87–88, 90
124–34 stratal ordering: see level ordering theory
divergent modification 82–86 Stratum Domain Hypothesis 71
junctural effects 98, 100–04, 124–25, Strong Domain Hypothesis 71
161–65 suppletive allomorphy in reduplication: see
not unique to reduplication 24, 67–69, divergent allomorphy in reduplication
77–82, 97, 99, 115, 127–33 ‘synonym compounds’ (see also synonym
opacity effects in (see opacity effects in constructions) 59
reduplication) synonym constructions (see also antonym
parallel modification 86–91 constructions) 7–8, 27, 47, 59–61,
tonal effects 46, 83–85, 90, 133 62–63, 64
phrasal reduplication 1 syntactic doubling/reduplication 5, 8
process morphology: see realizational
morphology The Emergence of the Unmarked
prosodic circumscription 155 (TETU) 109, 111
prosodic minimality 33, 140, 200–01 Thesis of Independent Inputs 25, 59
prosodic root, stem, word 140–41, 145, 147, Thesis of Morphological Targets 22, 25, 27,
153–55, 156, 172, 184–85, 187–88, 190, 30, 66
191–93 Thesis of Semantic Identity 25, 26, 31, 47,
pseudoreduplication 2, 3 66
254 Index of subjects

tone in reduplication: see phonology of underapplication (see also opacity effects in

reduplication reduplication)
triplication 14 as nonapplication 106–08, 110, 112,
truncation 4, 13, 17, 71, 78 117–18, 175–76
as source of opacity in reduplication 138,
153, 188, 190 word and paradigm morphology 6, 12
as source of partial reduplication 5, 79, 138,
140, 147, 156–57, 191 zero-derivation 17, 71