Fonologiajapanese

Title Pages
University Press Scholarship Online
Oxford Scholarship Online
The Phonology of Japanese

Laurence Labrune
Print publication date: 2012

Print ISBN-13: 9780199545834
Published to Oxford Scholarship Online: May 2012
DOI: 10.1093/acprof:oso/9780199545834.001.0001
Title Pages
The Phonology of Japanese The Phonology of the World’s Languages The
Phonology of Japanese
General Editor: Jacques Durand
Published
The Phonology of Icelandic and Faroese
Kristján Árnason
The Phonology of Danish
Hans Basbøll
The Phonology of Dutch
Geert Booij
The Phonology of Standard Chinese, second edition
Page 1 of 5
Title Pages
San Duanmu
The Phonology of Polish
Edmund Gussmann
The Phonology of English
Michael Hammond
The Phonology of Italian
Martin Krämer
The Phonology of Norwegian
Gjert Kristoffersen
Laurence Labrune
The Phonology of Portuguese
Maria Helena Mateus and Ernesto d’Andrade
The Phonology and Morphology of Kimatuumbi
David Odden
The Lexical Phonology of Slovak
Jerzy Rubach
The Phonology of Hungarian
Péter Siptár and Miklós Törkenczy
The Phonology of Mongolian
Jan-Olof Svantesson, Anna Tsendina, Anastasia Karlsson, and Vivan Franzén
The Phonology of Armenian
Bert Vaux
The Phonology and Morphology of Arabic
Janet Watson
Page 2 of 5
Title Pages
The Phonology of Catalan
Max Wheeler
The Phonology of German
Richard Wiese
In preparation
The Phonology of Tamil
Prathima Christdas
The Phonology of Welsh
S. J. Hannahs
The Phonology of Turkish
Bariş Kabak
The Phonology of Latin
Giovanna Marotta
The Phonology of Spanish
Iggy Roca
The Phonology of Greek
Anthi Revithiadou
The Phonology of Swedish
Tomas Riad
The Phonology of Washo
Alan C. L. Yu
(p.iv)
Page 3 of 5
Title Pages
Great Clarendon Street, Oxford OX2 6DP
Oxford University Press is a department of the University of Oxford.

It furthers the University’s objective of excellence in research,
scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi

Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press

in the UK and in certain other countries
Published in the United States

by Oxford University Press Inc., New York
© Laurence Labrune 2012
The moral rights of the author have been asserted

Database right Oxford University Press (maker)
First published 2012
All rights reserved. No part of this publication may be reproduced,

stored in a retrieval system, or transmitted, in any form or by any
means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the
appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights
Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose the same condition on any acquirer
British Library Cataloguing in Publication Data
Page 4 of 5
Title Pages
Data available
Library of Congress Cataloging in Publication Data
Data available
Typeset by SPI Publisher Services, Pondicherry, India

Printed in Great Britain
on acid‐free paper by
MPG Books Group, Bodmin and King’s Lynn
ISBN 978–0–19–954583–4
1 3 5 7 9 10 8 6 4 2
Acknowledgements

Laurence Labrune

Print ISBN-13: 9780199545834
DOI: 10.1093/acprof:oso/9780199545834.001.0001
(p.v) Acknowledgements
This work is a substantially revised and updated version of my book in French entitled La
phonologie du japonais, jointly published by the Paris Linguistic Society and Peeters
editions in 2006.
For his constant support and enthusiasm, I would like to express my gratitude to Jacques
Durand, who supervised this work from its very beginnings and later gave me the
opportunity to publish it at Oxford University Press.
Many thanks are also due to the following friends and colleagues for their comments and
help on earlier versions in French or in English of this book or on parts of it: the late Nick
Clements, Marc Plénat, Takayama Tomoaki, Catherine Garnier, François Dell, Tanaka
Shin’ichi, Irène Tamba, Elsa Gomez-Imbert, Martin Kramer and several anonymous
readers.
I am especially indebted to Kamiyama Takeki who read the entire final manuscript with
great care, making many valuable comments and suggestions which helped me correct a
number of mistakes.
Particular mention must also be made of Abe Junko, Hiraide Naoya, Wakasa Anju,
Page 1 of 3
Acknowledgements
Furihata Atsuko, Nakamura Yayoi, Kawaguchi Yuji, and many other friends and
colleagues who kindly provided information on the Japanese examples, of Joan Busquets
for his help in editing the figures, and of Michel Vieillard-Baron for assistance with the
poetic materials.
I am also most grateful to all the Japanese scholars who have provided me with their
teaching, advice, support, and help throughout the last twenty years or so during my
research stays in Japan, in particular Komatsu Hideo, Kitahara Yasuo, Hayashi Chikafumi,
Haraguchi Shôsuke, Jôo Hakutarô, Kondô Takako, Aoki Saburô, and I am especially
grateful to Takayama Tomoaki who was always willing to share his vast knowledge of the
phonology of Modern and Ancient Japanese with me. I owe a special and old debt to
Akinaga Kazue thanks to whom I discovered the joy of Japanese phonology at Waseda
University during the years 1987–1989.
I acknowledge with gratitude several scholarships from the Japanese Ministry of

Education and The Japan Foundation, which allowed me to conduct research in Japan at
Waseda University and Tsukuba University on several occasions. These institutions gave
me the precious opportunity to carry out most of the preliminary investigation for this
work. My research has also benefited from the constant scientific and financial support of
my CNRS research team in Bordeaux and in Toulouse, CLLE ERSS (UMR 5263) and the
University of Bordeaux 3 which I also want to thank.
(p.vi) I also thank Teddy Auly, a cartographer at the Centre National de la Recherche
Scientifique and the University of Bordeaux 3, who edited the two maps included in this
book.
Finally, my sincere thanks go to John Davey and his staff at Oxford University Press for
their editorial support and everlasting patience.
None of these persons, of course, necessarily agrees with the analyses I propose. All
errors and omissions are mine.
Tables, Figures, and Maps

Laurence Labrune

Print ISBN-13: 9780199545834
DOI: 10.1093/acprof:oso/9780199545834.001.0001
(p.xi) Tables, Figures, and Maps
Table 1.1. Hiragana (basic symbols) 8

Table 1.2. Katakana (basic symbols) 9
Table 3.1. Consonantal phonemes of Japanese 59
Table 4.1. Summary of blocking patterns among Yamato noun–noun compounds
123
Table 6.1. The 103 distinctive moras of Modern Standard Japanese in phonological
transcription 144
Table 7.1. Location of accent in nominal Yamato and Sino-Japanese words
(according to Sibata, 1994), in relation to length of lexemes 187
Table 7.2. Accent of simplex Yamato nouns 194
Table 7.3. Accentual effect of particles 195
Table 7.4. Accent of verbs 198
Table 7.5. Accent of -i adjectives 199
Table 7.6. Accent of compounds made up of a numeral + Sino-Japanese specifier
246
Table 7.7. Cross-dialectal accent correspondences for bimoraic nouns for the five
Kindaichi word classes 256
Figure 2.1. Spectrogram and oscillogram of a ki kara (with devoiced i) 35
Page 1 of 2
Tables, Figures, and Maps
Figure 2.2. Spectrogram and oscillogram of a ki demo (no devoicing of i) 36

Figure 2.3. Final vowel shortening in Western clippings 48
Figure 2.4. Token frequency of vowels in Archaic Japanese 57
Figure 2.5. Token frequency of vowels in Modern Japanese 57
Figure 3.1. Textual frequency (in %) of Archaic Japanese consonants 99
Figure 3.2. Lexical frequency (in %) of Archaic Yamato Japanese consonants for
the initial of words 100
Figure 3.3. Lexical frequency in absolute value of consonants according to their
position in bimoraic Yamato nouns in the modern language 100
Figure 3.4. Textual frequency of modern Japanese consonants 101
Figure 7.1. Accent curve (F0) of hana-ga ‘flower’ 182
Figure 7.2. Accent curve (F0) of hana-ga° ‘nose’ 182
Map 1. Administrative Japan xiv
Map 2. Geographical distribution of accent types 252
Notes on Transcription, Abbreviations, and Other Matters

Laurence Labrune

Print ISBN-13: 9780199545834
DOI: 10.1093/acprof:oso/9780199545834.001.0001
(p.xii) Notes on Transcription, Abbreviations, and Other Matters

The system of romanization adopted throughout the book is the Hepburn system (ヘボン ,
hebon-shiki), except for the notation of the bilabial fricative [F] which is written as h
before u, and of vowel length. Long vowels are transcribed as ou, aa, ii, ei, or ee,uu
(rather than ô,â,î, ê, û), except in proper names, linguistic terms, and in the bibliography.
This transcription, which has been calqued on the kana writing, has the advantage of
allowing for a more adequate notation of accent by dissociating the two parts of a long
vowel. It has one drawback, which is that it does not allow for a distinction between to u
‘tower’ (actually pronounced as [ˡtoː]) and to u う ‘to ask’ ([ˡ¹toɯ]) which are both spelled
as とう in hiragana. IPA transcription will be provided for disambiguation of ou sequences
in the text when necessary.
When needed, the phonological transcription (see Table 6.1, section 6.1) is used, as well
as phonetic transcriptions in the International Phonetic Alphabet (IPA). For instance, the
word しゅうり( in kanji) meaning ‘repair’ will be transcribed as shu uri in adapted
Hepburn, /syuRri/ in phonological transcription, and [ˡɕɯːɾi] in IPA.
In Hepburn romanizations, the accented mora appears in bold. In IPA transcriptions, the
sign ˡ is placed before the accented mora following the usual practice in the IPA: kokoro
[koˡkoɾo] ‘heart’, kyo uto [ˡkjoːto] ‘Kyôto’. Atonic words are followed by the symbol °:
Page 1 of 4
sakura° ‘cherry tree’. Many Japanese words display several possible accent patterns.
Generally, only the most frequent pattern is given for a word, except when accent
variation may be relevant to the discussion. Accent will not be provided for ancient,
dialectal, or invented forms (except when relevant for the discussion), for non-
independent morphemes and in cases where the form has to be considered
independently of its accentuation. In the citation of examples taken from other scholars
who do not provide accent information, I have automatically added the accent patterns if
necessary.
The source word of Western loans is given between braces: konpyuutaa ‘computer’
{computer}.
The components in transparent compound words are separated by a hyphen: kodomo-

beya ‘children's room’ when relevant to the discussion.
The following abbreviations are used:
(p.xiii) intr.= intransitive verb
tr.= transitive
µ= any mora
m= deficient (weak) mora
M= regular mora
σ= syllable
π= foot
V= vowel
C= consonant
C1= initial constituent, C2 = final constituent (in compounds)
#= word boundary
*= unattested form (or, reconstructed forms in passages dealing with historical matters)
jp= Japanese
ch= Chinese
rk= Ryûkyûan
H= high (tone), or heavy syllable
Page 2 of 4
L= low (tone), or light syllable
AJ= Archaic Japanese
OJ= Old Japanese.
In Chapter 3, which is devoted to the consonantal system, the notation of classical (linear)
generative phonology is used. For instance, the formula x → y / z reads as x becomes y
when occurring before z.
Old Chinese reconstructions come from Tôdô (1996) except when otherwise specified.
Japanese personal names are given in the following order: family name, personal name.
They are cited under the romanized form which appears in the original publication.
Authors’ names of books and papers published in Japanese have been transcribed
following the Hepburn system, except for those people who have chosen some other
transcription (when this other transcription is known to me).
The spectrograms and the oscillograms were made using the Praat software developed
by Paul Boersma and David Weenink, Amsterdam.
Finally, note that contrary to a majority of recent Western works on Japanese phonology,
I do not recognize the existence of the syllable in this language, although I will
occasionally provide syllabic information or representations for comparative purposes
when needed. The view retained in this book is that of the native Japanese tradition in
phonology, which holds that only the mora is relevant. (p.xiv)
Map 1. Administrative Japan
Page 3 of 4
Introduction

Laurence Labrune

Print ISBN-13: 9780199545834
DOI: 10.1093/acprof:oso/9780199545834.001.0001
Introduction
Laurence Labrune
DOI:10.1093/acprof:oso/9780199545834.003.0001
Abstract and Keywords
This first chapter provides a general introduction to the book, presenting its aims,
methods, theoretical background, the status, origins, and periodisation of the Japanese
language, as well as the previous scholarship written on the subject of Japanese
phonology. It also presents the writing system of modern Japanese, made of a mixture of
Chinese characters, two kana syllabaries and Latin alphabet, and describes and discusses
the issue of the lexicon partition into Yamato, Sino-Japanese and Western words.
Keywords: Japanese language, periodisation, Japanese writing system, Chinese characters, two kana
syllabaries, Latin, Yamato words, Sino-Japanese words, Western words
The Phonology of Japanese offers a comprehensive overview of the phonological

structure of modern Japanese from its segmental to its prosodic and accentual structure.
The purpose of the book is twofold.
Page 1 of 23
Introduction
First, it will present the actual ‘state of the art’ of Japanese phonology, based on a
compilation of recent and older Western and Japanese materials, reflecting current
debates in Japanese phonology. The aim is to provide a synthesis of two major research
streams: that of Japanese traditional linguistics and philology, kokugogaku , which is
characterized by its data-oriented approach, a strong philological background, and
careful attention to the empirical realities of the language, but which, unfortunately,
seems to be largely ignored outside Japan in spite of its excellence and remarkable
achievements (see the seminal works by Kindaichi Haruhiko, Hashimoto Shinkichi, Hattori
Shirô, Hamada Atsushi, Kamei Takashi, and many others); that of Western scholarship, for
which Japanese has often served as a test ground for newly developing theories. One
should recall that many aspects of Japanese phonology have contributed to the advance
of modern phonological theory in a significant manner. Without aiming at exhaustivity, let
us mention the works of James McCawley in the 1960s (classical generative phonology),
Haraguchi Shôsuke in the 1970s (non-linear phonology), Itô Junko and Armin Mester in
the 1980s and 1990s (underspecification theory, Optimality Theory), and Kubozono
Haruo in the 1990s and 2000s (Optimality Theory).
In sum, the main ambition of this book is to survey the achievements by scholars
belonging to different linguistic schools and traditions, to assess them critically, and to
integrate them into a uniform approach in order to make the results available to a larger
scientific community. It is hard to simply grasp the quantity and quality of native research
when one has no access to it, and it is even harder to evaluate it, be it in the field of
phonology or of any other area of linguistics. It should also be acknowledged that some
recent Western works often fail to give credit to the richness and excellence of this
tradition.1 This is (p.2) why it has appeared essential to devote so much attention to
Japanese contributions through an approach that attempts to blend and reconcile, in a
unifying perspective, two ways of doing linguistics that usually ignore each other. This
stand by no means precludes our casting a critical eye over one or other approach.
Further, this book aims to offer new analyses and data concerning some of the central
issues of Japanese phonology in a theoretically oriented approach. Issues for which new
analyses are proposed in this volume are those of the mora and syllable, the notion of
‘special mora’, compound noun accentuation, default accentuation (through a case study
of Western borrowings), the underlying accent of some Sino-Japanese morphemes, the
status of diphthongs, the consonant /r/, and the interaction of moras and feet.
The aim is thus to provide both a critical synthesis of the state of the art in Japanese
phonology and to provide theoretically oriented description and analyses in its main
areas. However, the purpose is not to promote a given theoretical or formal framework
set in advance and to which the data of Japanese would be forcefully moulded. Rather,
what I have tried to do is to provide general—albeit precise—information on the
phonological structure of the Japanese language in all its complexity and, whenever it
appears relevant, to point to the analytical and theoretical extensions of the issue likely to
be considered. Therefore it is why priority is always given to the presentation of the
linguistic data. I have nevertheless chosen to give a tighter theoretical and formal
Page 2 of 23
Introduction
treatment to a small number of specific issues that have appeared to deserve more
thorough treatment due to their importance in the field.
There are unfortunately a number of issues that I could not address as I would have
liked to. Notably, there is no in-depth treatment of intonation. The morpho-phonology of
verbal flexion would also probably have merited a whole chapter. However, this aspect of
Japanese morpho-phonology being generally introduced in Japanese grammars and even
textbooks, it is relatively easy to find good descriptions of it outside specialized phonology
or morphology works.
This book is intended for a general audience of students and linguists with no specialized
knowledge of the Japanese language, and to non-linguist Japanologists who want to obtain
up-to-date information in the field of Japanese phonology.
For the needs of the latter audience, Japanese terminology has been provided both in
roman transcription and in the original writing (kana or kanji), and priority has been given
to first-hand sources and references in the Japanese language.
(p.3) 1.1 Theoretical Background

The general framework of our reflection and analyses will be that of generative phonology
in the broad sense as it has been developed from the end of the 1960s onwards,
although some parts of the book also owe a great deal to structural phonology, a current
that was widely followed in Japan in the 1940s, 1950s, and even later, in the works of
outstanding Japanese linguists like Hattori Shirô and Kindaichi Haruhiko, whose analyses
will be often referred to in the following pages. But whatever framework lies behind our
discussions, a distinction is always made between an underlying form (or input) and a
surface form (output). In order to account for the formal relation which exists between
these two levels, we adopt a non-derivational approach which is that of Optimality Theory
(see Prince and Smolensky, 1993, Kager, 1999, for introductions), in which the
relationship between the input and the output is viewed as the result of the interaction of
constraints rather than sequential rule application as in the traditional generative model
(Chomsky and Halle, 1968). Such an approach proves to be particularly effective for the
treatment of phenomena relating to prosodic morpho-phonology, and it will be used, in
particular, for the formal analysis of default accentuation and compound nouns
accentuation which will be offered in Chapter 7.
As regards contents representations, be it the internal structure of segments in terms of

distinctive features or the architecture of the prosodic components, the references are
clearly those of traditional non-linear and autosegmental phonology. Optimality Theory has
actually very little to say about the contents and nature of representations, and is
compatible with various representational conceptions.
Phonology being a relatively technical and formal discipline, it was not possible within the
limits of this work to provide definitions and explanations of all the concepts used here. It
is assumed that the basic notions of articulatory phonetics and of phonological analysis are
known. Readers who want to acquaint themselves with the discipline are invited to consult
Page 3 of 23
Introduction
for example the reference works of Kenstowicz (1994a), Goldsmith (1990, 1995), and
Hayes (2009), which provide good introductions to various aspects of phonological
theory.
In the pages devoted to the presentation of segmental phonology (Chapters 2, 3, and 4),
the theoretical background of the description and analyses will be cast in a classical (and
rather neutral) framework in terms of features and statements. A broadly generativist
phonological framework will be adopted, such as the one introduced in Kenstowicz
(1994a). For the mora and syllable analysis (Chapter 6), the autosegmental, non-linear
framework will be used. I will refer especially to the conceptions developed by Larry
Hyman (2003 [1985]) regarding (p.4) the status of the TBU (tone-bearing units), i.e. the
moras, for the analysis, but other standard models will also be reviewed for the sake of
comparison.
The accentual analyses of compound nouns and of Western borrowings in Chapter 7

(sections 7.2.5 and 7.3.2) are cast within the framework of Optimality Theory. Some other
current phonological frameworks will occasionally be referred to when necessary, for
instance when previous scholarship and analyses concerning some of the problems of
Japanese phonology provide alternative and arguably more insightful views of the
phenomena under consideration.
1.2 The Japanese Language

Japanese is spoken by about 130 million speakers, nearly all living in the Japanese
archipelago. Its genetic affiliation is dubious. It has often been classified as a Ural-Altaic
language, but the reality of its origins is more complex. Prehistoric Japanese is probably
the result of hybridization between an Austronesian and an Altaic language, with some
possible other continental influences.
The language closest to Japanese is Ryûkyûan, spoken in the Ryûkyû Islands, southwest
of Kyûshû. Together, they form the Japonic family.
The language closest to Japanese outside Ryûkyûan is Korean. There exist strong
typological resemblances between the two languages, which suggest a genetic
relationship, although well-established regular phonetic correspondences are hard to
establish (see Martin, 1966 for an attempt).
Two main Japanese dialect groups are recognized: Eastern dialects (Tôkyô type) and
Western dialects (Kyôto-Ôsaka type), and Japan is still a country with great dialectal
diversity.
This book is primarily concerned with Modern and Contemporary Standard Japanese.
Japanese linguists generally refer to that variety as hyôjungo ( ) ‘standard language’,
kyôtsûgo ( ) ‘common language’, or Tôkyôgo ( ) ‘the Tôkyô dialect’. It corresponds
roughly to the language spoken in the districts of the area known as Yamanote in Tôkyô
and in the national media, in particular the NHK (Japanese Broadcasting Corporation).
Page 4 of 23
Introduction
We will also refer to dialectal varieties of the language and to historical developments
when necessary for an understanding of the synchronic facts.
For the periodization of Japanese, the following labels are adopted. These divisions also
correspond to standard major political divisions in Japanese political history:
– Archaic Japanese (jôdaigo ): before 794 (until the end of the Nara
period)
– Old Japanese (chûkogo ): 794–1350 (Heian and Kamakura)
(p.5) – Middle Japanese (chûseigo ): 1350–1603 (Muromachi, Azuchi-
Momoyama)
– Pre-modern Japanese (kinseigo ): 1603–1868 (Edo)
– Modern Japanese (kindaigo ): 1868–1945 (from Meiji to World War II)
Archaic Japanese was the period when Chinese characters were first massively imported
into Japan. The materials of those times are written exclusively using Chinese characters,
read in a Chinese or a Japanese manner (Man’yôgana, ‘Chinese characters used
only for their phonetic value’, see below).
Old Japanese saw the development of the kana syllabaries, and the flourishing of a
national literature written exclusively in kana with very few words of Chinese origin. The
language reflected in the materials is primarily that of the Kyôto aristocracy and has
served as the basis for the prestige written language for centuries.
The middle of the fourteenth century can be seen as the major turning point between
Ancient and Modern Japanese, to the extent that it is sometimes sufficient to oppose Old
Japanese (the language before the fifteenth century) to Modern Japanese (the language
after the fifteenth century). Middle Japanese underwent significant changes due to the
spread of the Sino-Japanese vocabulary and the generalized use of the kanji to write it.
The so-called kanji-kana majiri bun ( じり ‘kanji and kana mix style’), based on a
mixture of kanji and kana as in Modern Japanese (see section 1.5) became the most
common style of writing. Middle Japanese is also a period of major modification in the
verbal and adjectival flexional system as well as in the phonological system with the
establishment of the special segments (see Chapter 5), as a result of the sound changes
known as onbin , whose first occurrences can be traced back to Old Japanese.
Pre-modern Japanese, in the Edo period, is known to us through a huge number of

different types of materials reflecting the colloquial and dialectal diversity of the time,
including a number of foreign descriptions of the Japanese language, principally European
ones, with the publication of dictionaries and grammar books, but also accounts made by
Chinese and Korean scholars (such foreign descriptions of Japanese actually started in
the fifteenth century). Modern Japanese, starting with the Meiji Restauration in 1868, has
been influenced by Western languages. It also corresponds to the spread of Tôkyô
Japanese as the standard language, and the development of a new form of written
language closer to the spoken one.
Page 5 of 23
Introduction
‘Contemporary Japanese’ (gendaigo ) can be used more specifically to refer to the

variety of language which developed after World War II.
(p.6) 1.3 Particular Status of Japanese for Linguistic Science

A word should be said here about the status of Japanese in the field of linguistics.
Japanese is no doubt one of the best-documented non-Indo-European languages in the
world, if not the best-documented. In addition, it has a rare characteristic: most
specialists of Japanese linguistics are native speakers of the language, who, moreover,
have been working within their own rich linguistic tradition in a cumulative manner,
without ignoring the achievements of general linguistics outside their country. This
tradition, it should be emphasized, did not develop in an intellectual environment
completely sealed off from the rest of the world. It has been nourished by Chinese,
Indian, European, and American contributions throughout its long history. Descriptive
and cumulative work has thus been conducted in an optimal manner, although one might
have the feeling that, in very recent years, even the major works by outstanding scholars
such as Arisaka Hideyo, Hashimoto Shinkichi, Kindaichi Haruhiko, and Hattori Shirô, for
example, are no longer part of the compulsory reading of younger Japanese linguists
trained in the West.
Last but not least, Japanese linguistic research has enjoyed quite a favourable economic
environment. For decades, the various academic institutions of the country such as
research centres and universities have devoted an impressive number of material means
to research on the national language and its dialects, with the result that one can benefit,
in the case of Japanese, from an exceptional quantity of quality data and documentation
(even more, it seems, than for English or French, which have also been extensively
studied). The accumulation of descriptive and analytical materials is completely
bewildering, and contemporary phonology would be much worse off if it did not take
account of the contributions of the Japanese academic tradition.
1.4 Previous Western Literature On the Phonology of Japanese

There exist few general references in European languages relating to the phonology of
Japanese, in comparison to the huge number of studies carried out in Japan. I will only
mention here studies of a general and broad character, but naturally there are a fair
number of articles and some monographs relating to specific aspects of the phonology of
Japanese (mainly in English).
The excellent book by Timothy J. Vance, An Introduction to Japanese Phonology,

published in 1987, constitutes the best descriptive reference of the (p.7) discipline in
the English language. Unfortunately, it has been out of print for a number of years, and
therefore hard to get. Timothy Vance is also the author of The Sounds of Japanese,
published in 2008, which is a handbook designed for English-speaking students. It is of
course impossible not to mention James McCawley’s thesis, The Phonological Component
of a Grammar of Japanese, published in 1968, which was one of the first studies seeking
to apply to a language other than English the generativist framework of the Sound Pattern
of English (Chomsky and Halle, 1968). This reference remains invaluable, and much of the
data and analyses of McCawley have not lost their interest, but a lot of water has gone
Page 6 of 23
Introduction
under the bridge of phonological theory since 1968, so the framework is a bit outdated.
One should also mention the monograph by Samuel E. Martin, Morphophonemics of
Standard Colloquial Japanese, published in 1952, that of Günther Wenck, The Phonemics
of Japanese—Questions and Attempts (1966) as well as Japanese Phonetics (1997) and
Japanese Phonology (2000) by Akamatsu Tsutomu. Wenck is also the author of a
monumental Japanische Phonetik in four volumes, written in German (1954–1959). In
French, one should mention Haruhiko Kinda-ichi (= Kindaichi) and Hubert Maës’s,
Phonologie du japonais standard, published in 1978, which consists in fact of a translation
and adaptation by the second author of an original Japanese text by the first author
(Kindaichi), one of the most eminent Japanese phonologists. In addition to the fact that it is
out of print, this work, which is rather short (59 pages), is theoretically outdated. I am the
author of La phonologie du japonais, published in 2006 by the Société de Linguistique de
Paris (Peeters, Leuven). The present book is a substantially updated and modified
version of this 2006 French edition. All these books, except for Labrune (2006) and
Vance (2008), have sadly been out of print for a number of years.
1.5 Overview of the Writing System

Throughout this book, we will occasionally refer to the orthographical status which some
of the phonological units of the language have received in the native writing system of
Japanese. This is because the written dimension provides an interesting background to
the phonological reality of these units. The graphemic system often reflects the phonemic
one, and, vice versa, since phonology in turn can be influenced by the writing system, or,
to put it in Suzuki’s words (Suzuki, 1977), writing can become a formative agent of the
language. This is especially true for Japanese. Kess and Miyamoto (1999:32) observe that
the nature of the multi-faceted Japanese orthography must be viewed as a formative
agent that exerts some influence, if not power, over the spoken language itself. However,
it goes without saying that the orthographical criteria should not be held up as definite
proof of the phonological status of a given element.
(p.8) This being said, a general presentation of the writing system of Japanese will be
given in the following pages, but readers with no specific interest in the issue may skip
this section and proceed directly to the following one.
The Japanese writing system is composed of four different scripts. First, it has two
original ‘syllabaries’2 of 48 signs each (of which 46 only are presently in common use), the
hiragana and the katakana , which are referred to under the generic term of
kana . Katakana and hiragana were created by the Japanese. They both took as their
basis Chinese characters used only for their phonetic value (the man’yôgana ; see
Seeley, 1991 for a general presentation of the history and development of the Japanese
writing system in English). Hiragana and katakana are based on the mora3 and take as
their basis the same units, so that a given mora of Japanese can be denoted by the
corresponding letter of either set. The elaboration of these two sets of kana symbols was
more or less achieved around the tenth century.
In addition, several thousands of ideographic characters originally borrowed from

Chinese, the kanji , are used. The Latin alphabet, rôma-ji ローマ , and Arabic numerals are
Page 7 of 23
Introduction
also part of the modern writing system. The writing of a Japanese text is done today by
using in a joint and complementary way the first
Table 1.1. Hiragana (basic symbols)

a あ i い u う e え o お
ka か ki き ku く ke け ko こ
sa さ shi し su す se せ so そ
ta た chi ち tsu つ te て to と
na な ni に nu ぬ ne ね no の
ha は hi ひ hu ふ he へ ho ほ
ma ま mi み mu む me め mo も
ya や yu ゆ yo よ
ra ら ri り ru る re れ ro ろ
wa わ wi ゐ we ゑ wo を
N ん
(p.9)
Table 1.2. Katakana (basic symbols)

a ア i イ u ウ e エ o オ
ka カ ki キ ku ク ke ケ ko コ
sa サ shi シ su ス se セ so ソ
ta タ chi チ tsu ツ te テ to ト
na ナ ni ニ nu ヌ ne ネ no ノ
ha ハ hi ヒ hu フ he ヘ ho ホ
ma マ mi ミ mu ム me メ mo モ
ya ヤ yu ユ yo ヨ
ra ラ ri リ ru ル re レ ro ロ
wa ワ wi ヰ we ヱ wo ヲ
N ン
three systems (hiragana,katakana, and kanji), and, in an accessory manner, the latter
two.
The writing of Japanese is unanimously recognized as one of the most complex, or even
the most complex, of all known systems. As Kess and Miyamoto (1999:13) put it, ‘it is no
stretch of the imagination to declare Japanese one of the most intricate, most elegant and
yet most difficult writing systems in the modern world’. Complexity lies first of all in the
fact that the structure and the orthographical principles of these various scripts are
fundamentally different. Kana and the Latin alphabet have in common the fact that they
Page 8 of 23
Introduction
are phonographic. However, the kana adopt as a basic unit the mora, while the alphabet
is based on the phoneme. Chinese characters, on the other hand, are primarily
logographic (ideographic) symbols, like the Arabic numerals. Moreover, the way Chinese
characters are used in Japanese writing is the source of another complexity, since, as we
shall see below, most characters can be read in at least two fashions, depending mostly
on the context in which they occur.
Tables 1.1 and 1.2 present the hiragana and katakana according to the traditional order
of the gojûonzu ( ‘table of the fifty sounds’4). As we shall see in more detail in
Chapters 2, 3, and 4, the core phonemic system of Japanese consists of five vowels (a, i,
u, e, o) and fourteen consonants (p, b, t, d, k, g, s, z, h, m, n, r, y, w). It will be noted that
the combinations starting with p,b,d,z,g, which are derived from the corresponding
unvoiced kana letter by (p.10) addition of a diacritic symbol, as well as those comprising
a palatalization, do not appear here (we will reconsider this point a little further; a table of
all the Japanese moras is provided in Chapter 6, Table 6.1).
Hiragana mainly denote grammatical elements or elements with no stable referents such
as enclitic particles, verbal and adjectival inflexions, functional names, interjections,
connectors, and a number of adverbs. They are sometimes used to write lexical
morphemes that the scripter does not want (or does not know how) to write in
characters. Texts for children are thus transcribed exclusively in hiragana, which are the
first writing symbols taught to Japanese children.
Katakana are generally reserved for the transcription of recent foreign loans. They are
also sometimes used to write mimetic words, the names of plants or animals, dialectal or
slang forms, and sometimes also erudite words. They may also be employed to highlight
an element in a sentence, somewhat like the italics in the Latin alphabet script, to mark
irony, or even to give a more colloquial, oral flavour to a text.
The modern versions of hiragana and katakana (Tables 1.1 and 1.2) comprise 46 or 48
signs if one takes into account the two kana denoting the moras wi and we that are in
principle encountered only in texts written prior to 1946. The characteristic of the
hiragana and the katakana is initially, as the term ‘syllabary’ reflects, that they transcribe
‘syllables’ with the traditional Japanese direction of the term, that is, moras. A second
characteristic is that they use diacritics rather than distinct letters to denote the
difference between voiceless and voiced obstruents. Another 58 additional moras can
thus be written by the addition of a diacritic symbol, or combination of two existing kana.
Voiced obstruents are marked by two small strokes, the dakuten or nigoriten り ,
placed at the upper right corner of the matrix of a given kana (1a, see also sections 3.7.1
on the correspondence between h and b, and Chapter 4 on voicing in general). A
comparable device is used to represent the moras pa,pi, pu, pe, and po: a small circle,
the handakuten (literally ‘semi-voicing dot’, 1b) is added at the top of the kana
transcribing the h series.
(1)
Page 9 of 23
Introduction
a. Notation of obstruent voicing
t : d ta chi tsu te to たちつてと : da ji zu de do だぢづでど

s : z sa shi su se so さしすせそ : za ji zu ze zo ざじずぜぞ
k : g ka ki ku ke ko かきくけこ : ga gi gu ge go がぎぐげご
h : b ha hi hu he ho はひふへほ : ba bi bu be bo ばびぶべぼ
b. Notation of moras starting with /p/
h : p ha hi hu he ho はひふへほ : pa pi pu pe po ぱぴぷぺぽ
The palatalized combinations (kya,kyu, and so on) are transcribed by adding onto the
right side of a kana containing the -i vowel the kana ya,yu, and yo in (p.11) reduced
size, as shown below (in the following examples, capitals are used to reflect full-size kana,
while small letters transcribe reduced-size kana):
(2)
KI : KIya (kya) SHI : SHIyu (shu) CHI : CHIyo (cho)

き : きゃし : しゅち : ちょ
Hiragana and katakana are indeed different scripts but they are almost identical as far as
the principles that underlie their internal structure, organization, and phonemic
referential units are concerned. The only difference between the two is a tiny one. It lies
in the fact that vocalic length is not treated identically. In katakana, it is uniformly
represented by an horizontal line (vertical in cases where the text is written from top to
bottom), while in hiragana, it is transcribed differently according to the quality of the long
vowel: the kana letter for u is added after the moras containing -u and -o when the length
results from the fall of a consonant followed by u (the most frequent case), the letter for a
is added after a, that for i after -e and -i. For instance (here the hyphens mark mora
boundaries):
(3)
toukyou° ‘Tôkyô’ pronounced [toːkjoː]

written TO-U-KIyo-U とうきょう ( )
guuzen° ‘fortuity’ pronounced [ɡɯːzeN]
written GU-U-ZE-N ぐうぜん ( )
reisei° ‘calm’ pronounced [ɾeːseː]
written RE-I-SE-I れいせい ( )
There exist some particular uses, which one encounters, for instance, in cases of native
Page 10 of 23
Introduction
Japanese words where the lengthening of the long vowel [oː] corresponds to the loss of a
consonant originally followed by o: the vocalic length is noted in that case by means of the
letter o, as in the word ookii (〈 ohokii) ‘large’. One can also mention words like oneesan
‘older sister’, whose long vowel is written ee and not ei.
In katakana, vowel length is indicated by means of a horizontal or vertical bar following

the vowel (according to the direction of the writing), whatever the quality of the vowel
concerned, for example ‘super’ su upaa スーパー.
The reduced-size kana tsu (っ in hiragana and ッ in katakana) is employed to write the
first part of a non-nasal geminate consonant, whatever it may be. Thus a tta あった ‘had’ is
noted A-tsu-TA in hiragana, while ka tto カット ‘cut’ {cut} is noted KA-tsu-TO in katakana.
Apart from rare exceptions, the kana spelling of Japanese words is simple and
straightforward.
The Chinese characters, or kanji ( ), are generally used to write the non-variable part
of lexical morphemes of Chinese or Japanese origin. The stem of a flexional word of
Japanese origin such as a verb or an adjective is written by (p.12) means of one or
several Chinese characters, while the variable part is transcribed in kana. Thus the kanji
represents the idea of ‘reading’ but it can be read in different ways. As shown in the
examples in (4), is employed to represent the stable part of the various inflected forms
of the verb ‘to read’ in Japanese and it is read yo-, according to its native Japanese
reading (kun’yomi み ‘meaning-reading’). The endings which undergo variation will be
noted in hiragana. The same character is also used in compound nouns such as do
kusha ‘reader’, tokuhon° ‘reading book’, koudoku° ‘subscription’, and many
others, with the Sino-Japanese reading (on’yomi み ‘sound-reading’) doku or toku.
(Accents are ignored in the following examples.)
(4)
yomu む .MU ‘to read’

yonda んだ .N.DA ‘read (past tense)’
yomanai まない .MA.NA.I ‘do(es) not read’
dokusha ‘reader’
tokuhon ‘reading book’
koudoku ‘subscription’
It is generally considered that the knowledge of approximately 2000 characters is

sufficient for the reading of current Japanese texts. However, this figure is extremely
relative, and represents in fact a minimum threshold.5
The difficulty raised by the Japanese sinograms comes from the fact that, on the one
hand, a character almost always has several readings (see below, section 1.6.2), and that
Page 11 of 23
Introduction
on the other hand, a lexeme can almost always be written using different characters. For
instance, is read yo-,doku, or toku. But the verb yo mu ‘to read’ can be written む or
む.
A written Japanese sentence is thus composed of an arrangement of kanji and kana, and
it is not, moreover, uncommon that a text contains some sequences in the Latin alphabet
or Arabic numerals. Texts are written from top to bottom vertically, starting from the
rightmost side of the page, or horizontally, from left to right. One occasionally encounters
horizontal inscriptions, generally made up of a couple of Chinese characters, written from
right to left.
Each symbol (kana,kanji,rôma-ji, or figure) is separated by a blank. There does not exist
any special demarcating device to separate words or syntagms. It is thus only the
alternation between Chinese characters, hiragana, (p.13) katakana, rôma-ji, and Arabic
numerals, as well as the use of punctuation, that helps the segmentation of the various
elements of the sentence.
1.6 The Stratification of the Lexicon

The lexicon of Japanese is stratified into morphemes belonging to different classes
corresponding to distinct morpho-phonological, semantic, and pragmatic systems. This
organization is fundamental for the description and comprehension of Japanese as a
whole. Lexicon stratification plays a central role in the grammar because it entails major
structural as well as pragmatic (register) differences. Words belonging to different
classes may undergo different rules or constraints. For instance, one of the best-known,
and most often cited, examples is rendaku (‘sequential voicing’), which applies differently
according to the stratum (see Chapter 4).
Japanese linguistics traditionally distinguishes a minimum of three lexical classes:
– Wago , or Yamato lexemes, the class of native words within which one
might possibly put the subclasses of the mimetic words and other expressive
words such as childish, familiar, or slang vocabulary.
– Kango , Sino-Japanese lexemes. They are loans from Chinese introduced
massively in Japan starting from the fourth century at least. This class
comprises many words of the erudite and abstract vocabulary, as well as
concepts and objects borrowed from Chinese culture, but it also contains
other more common, unmarked items.
– Gairaigo , which are lexemes that have been recently borrowed from
foreign languages, primarily Western languages from the sixteenth century.
They contain mainly technical, scientific terms or refer to modern objects and
concepts with a Western connotation.
It is sometimes useful to distinguish a fourth stratum, that of mimetic words 6:

onomatopoeias (giseigo ) and ideophones (gitaigo ).
The overwhelming majority of mimetic words are etymologically of native origin.7 For this
Page 12 of 23
Introduction
reason, they belong to the Yamato class in the strict sense, even if (p.14) they display a
number of properties which may lead one to categorize them in a specific subclass. In this
book, when necessary, we will make a distinction, within the Yamato class, between non-
mimetic and non-expressive words (the Yamato class stricto sensu), and mimetic and
expressive words (a distinct class for some authors).
To the Yamato, Sino-Japanese, and Western strata, the class of non-integrated foreign
words (gaikokugo ) is sometimes added. These are words whose degree of
adaptation into the Japanese language is not as advanced as that of the gairaigo. They
consist of direct quotes from a Western language in the Latin alphabet. Some scholars
also distinguish between formal Sino-Japanese and vulgarized Sino-Japanese (Takayama,
2005).
It is also necessary not to forget the existence of a mix or hybrid class (konshugo ),
which comprises compounds made up of words or morphemes belonging to different
classes, for instance wago + kango as in ni motsu ‘luggage’, kango + wago as in juu-
bako° ‘superposable meal box’, gairaigo + wago as in demoru ‘to demonstrate (in the
streets)’ (from de mo,demonsutoreeshon {demonstration} + -ru, ‘verbal suffix’). Finally,
note that a certain number of lexemes of Sanskrit, Ainu, or other origins do not fall into
any of these categories and have unclear status as to the lexical class they belong to.
This partition is largely determined by etymology, but it would be imprudent to adopt too
narrow a vision and a simply historical approach to the problem. Often, the supposed
etymology is more determining than the real one, and the actual ‘phonological profile’ of
the word plays a more important role than its origin (not to mention its semantic or
pragmatic profiles). Actually, the differences between the lexical classes are extremely
delicate to handle. First of all, it is difficult to establish with precision the origin of certain
lexemes. Second, we know almost nothing about the history of the Japanese language
before the fifth century, and in particular we are ignorant of the true nature of the
contacts between the spoken language in Japan and the spoken language(s) in the Korean
peninsula or elsewhere. One should also take into account the fact that loans from foreign
languages (especially from Chinese) have had a deep influence and have considerably
modified the morpho-phonology of the Yamato lexemes. Moreover, it is not unusual that
the linguistic intuitions of non-linguist speakers regarding which lexical class a given
lexeme belongs to are in clear contradiction with the true etymology. For example shio
‘salt’ or mu gi ‘wheat’ are actually very old loans from Chinese, but they are handled and
behave like Yamato words. The same applies to kappa° ‘raincoat’ or kasutera° ‘pound
cake’, which are words of Portuguese origin but treated as Yamato lexemes. Which is
more important, the etymological data or speakers’ intuitions? As Takayama (2005)
observes, in order to determine the lexical stratum to which a given word belongs, one
has to consider both word forms (phonotactic patterns) and connotation, that is, (p.15)
whether the word is culturally associated with a foreign background. I would add that the
writing may constitute another strong clue to determining which stratum a word belongs
to.
Lastly, it will be necessary to question the manner this partition is acquired by Japanese
Page 13 of 23
Introduction
children. It is not clear whether native speakers acquire this intuition through education,
especially the knowledge of Chinese characters and of the difference between Sino-
Japanese and Japanese readings of the characters, and acquisition of katakana and
hiragana (remember that katakana are used primarily for the notation of Western
loanwords, while kanji and kana are used to transcribe Yamato and Sino-Japanese
words), or if it is of a deeper, truly linguistic nature. Probably, both dimensions are
involved, and education only serves to reinforce and stabilize a robust difference. How
does the child manage to internalize the difference between Yamato words, Sino-
Japanese words, and Western words? Is this knowledge of a metalinguistic nature, that is,
acquired through education and literacy, and particularly thanks to the mastering of the
writing system? Ota (2004) provides a good discussion of the issue of the learnability of
lexicon partition in Japanese, and points out the unrealistic scenario of phonological
learning that is implied by OT constraint-based models (Itô and Mester, 1995a, b; Itô and
Mester, 1999; Itô, Mester, and Padgett, 1999, for instance).
In spite of these problems, the partition of the lexicon plays a key role in the grammar of
the language. It is conveyed in the writing, and constitutes an important component of the
metalinguistic knowledge of any Japanese speaker. In principle, Yamato words are written
in hiragana or kanji,kango in kanji, and gairaigo in katakana. However, a well-integrated
gairaigo can be written in hiragana or even in kanji, and a ‘yamatoized’ kango can end up
being written only in hiragana.
According to the statistics provided by the Shinsen Kokugo Jiten dictionary (8th edition,
2002), wago represent 33.8% of the entries of the dictionary, kango 49.1%, gairaigo
8.8%, and hybrid words 8.4%. The words of Chinese origin are thus the most numerous
in the lexicon. In textual frequency (corpus of the written language drawn from the press)
the proportions are roughly similar with respect to type frequency (KKK, 1964). On the
other hand, wago are most frequent in speech: 46.9% compared with 40% for the kango
(Hayashi O., 1982). The proportion of wago goes up to 71.8% in token frequency. This is
evidently explained by the fact that words of the basic lexicon, and those that fulfil a
grammatical function (auxiliaries, particles, etc.), which are frequently repeated, almost all
belong to the Yamato class. Moreover, some studies have shown that the proportions
between the strata could vary according to the sex of the speakers. The survey by
Tsuchiya (1965) reveals indeed that kango are employed more (p.16) frequently by
male than by female speakers, at least at the time of the investigation.
It is frequently the case that the same referent can be referred to by a wago, a kango, or
a gairaigo, for instance:
(5)
Wago Kango Gairaigo

tegami° / shokan° / re taa レター ‘letter, missive’
meshi / go han / ra isu ライス ‘rice’
Page 14 of 23
Introduction
odori° り / buyou° / da nsu° ダンス ‘dance’
However, the three lexemes in each set have different connotations, and sometimes also
semantic specializations. The kango are generally felt to be more formal, more precise,
and belonging to a higher register than the wago or the gairaigo. The gairaigo generally
refer to Western realities. For example, dansu can only refer to a Western type of dance,
contrary to odori° and buyou°, and raisu designates some rice presented or cooked in a
Western way. But there are exceptions. For instance, the word kappu nuudoru {cup
noodle} (originally a trade mark) indicates an ‘instantaneous noodle dish cooked in an
Asian manner’. Here, the connotation brought in by the use of gairaigo is modernity. The
gairaigo also tend to refer to concrete, material entities, whereas kango are preferred
for the abstract (Loveday, 1996). In addition, gairaigo frequently appear as compound
formatives. For example re taa is more often used in expressions such as rabu retaa
{love letter} or retaa peepaa {letter paper} than in isolation.
1.6.1 Wago
In its diachronic sense, the term Yamato refers to the ‘original’, native Japanese language
with no elements of Chinese or from any other foreign origin. The most operational
definition of what a Yamato word is seems to be as follows: a Yamato morpheme is a
morpheme which does not result from a loan posterior to the fifth century of our era.8
In the old language, the following properties were characteristic of Yamato words:
– structure of the basic prosodic unit = V or CV;

– prohibition of hiatus (onsetless vowels were allowed only word initially);
– absence of words starting with a voiced obstruent (/b/, /d/, /g/, /z/) or with
/r/;
– impossibility of having two voiced obstruents, or two /r/, within the same
root;
(p.17) – scarcity of the /e/ vowel, in particular at the beginning of words
longer than two moras;
– existence of vowel harmony;
– simplex lexemes from two to three mora long.
Most of these characteristics remain today only as a residue. In modern Japanese, words
of Yamato origin are characterized by the absence of /p/, the absence of /h/ in word
internal position, the impossibility of finding geminated voiced obstruents and geminated
/r/, and by the constrained distribution of voiceless consonants after the mora nasal /N/.
One will also note the scarcity of palatalized consonants.
Whenever the same referent can be referred to either by a lexeme of Yamato or Sino-
Japanese origin, the connotations brought in by the Yamato word are generally
associated with the register of intimacy, the expression of sensations and emotions. They
are also considered more poetic than Sino-Japanese or Western words, and constitute
Page 15 of 23
Introduction
the core lexicon of Japanese traditional poetry (haiku,tanka).
1.6.2 Kango
Sino-Japanese words, or kango ( ), are words which are written using one or more
Chinese characters pronounced in a Sino-Japanese manner. Kango are ‘words’ (go ),
but they are above all meaningful written units associated with one or more Sino-
Japanese readings. If the character corresponding to a word of Chinese origin is no
longer used to write the word, it becomes difficult to regard the word in question as a
kango. One can mention the case of the lexeme se i ‘fault, reason’, which, in spite of its
Chinese origin, is never written in characters ( ) in contemporary Japanese, or the word
se sse to, ‘assiduously’ ( (Nakada and Hayashi, 1982). Consequently, most Japanese
speakers are surprised to learn that these words are actually kango.
Many kango are jukugo ( ), that is, Sino-Japanese compound words made up of two to
four kanji. A majority of one-character (one kanji) Sino-Japanese lexemes only occur as
bound morphs, that is, as components of a jukugo, and never occur in an autonomous
way, like sho ‘write’ in to sho ‘book’ or shokan° ‘letter’. However, a few one-
character kango (ichiji kango ) also function as autonomous words, like ho n
‘book’, niku ‘meat’, or ki° ‘spirit’.
In the modern language, kango lexemes are often characterized by the following
phonological properties:
– presence of many palatalized consonants;

– presence of mora nasals;
– presence of geminations (only in compound kango);
– presence of long vowels;
(p.18) – absence of non-geminated /p/;
– absence of geminated voiced obstruents;
– morpheme (stem) length from one to two moras.
Here are typical kango lexemes: gakkou° ‘school’, nippon ‘Japan’, gyuuniku°
‘beef’, shu ukyou ‘religion’.
Sino-Japanese morphemes are organized around a vowel, possibly preceded by a

consonant, palatalized or not. This group may be followed by a mora nasal (noted /N/), by
a vocalic length (noted /R/), by a front high vowel /i/, or by an extra mora containing /t/ or
/k/ followed by the vowels /i/ or /u/. This structure can be synthesized with the following
formula where the symbols between the braces indicate non obligatory elements:
(6)
Examples:
/hoN/ hon
/koR/ kou
Page 16 of 23
Introduction
/ai/ ai
/botu/ botsu, /kiti/ kichi, /kyaku/ kyaku, /teki/ teki9
Thus one has: i ‘stomach’, ya ‘house’, ki ‘spirit’, ryo ‘travel’, un ‘fate’, man ‘ten
thousand’, jun /zyuN/ ‘pure’, sou /soR/ ‘grass’, kyou /kyoR/ ‘to teach’, nai ‘inside’,
botsu /botu/ ‘rejection’, kyaku ‘guest’, kichi /kiti/ ‘good fortune’, reki ‘passing of
time’, and so on.
Loans from Chinese were made through three successive waves from the end of
Antiquity, over a vast period covering nearly one thousand years, and starting at least
from the fourth century onwards. This is the reason why it is not uncommon for a given
character to have two or even three different Sino-Japanese readings (on’yomi み).
According to Vance (1987:169), about 13% of the 1850 currently used kanji have more
than one Sino-Japanese pronunciation. In addition, the majority of kanji also have at least
one native Japanese reading (kun’yomi み). The different types of Sino-Japanese
readings are:
– the go readings (or Wu readings, goon ), which are linked to the

introduction of Buddhism in Japan and correspond to loans made during the
fifth and sixth centuries, probably via Korea. The exact geographic source is
not always clear, but it seems to have been somewhere in Southern China,
near the mouth of the Yangtze river.
– the kan readings (or Han readings, kan’on ), which are by far the most
important. They correspond to loans dating back to the seventh and eighth
(p.19) centuries. Kan’on are derived from the pronunciation of the Tang
capital Chang’an (presently Xi’an).
– the tô readings (or Tang readings, tôin or tôon , sometimes referred to
as sôon or tôsôon ). They concern later borrowings, from different
Chinese provinces, which explains why they are less homogeneous from the
point of view of their pronunciation.
To this list, one should add the so-called ‘usage readings’ (kan’yôon ), which
correspond to alterations of kan or tô readings, and which represent irregular Sino-
Japanese evolutions from the original Chinese pronunciations (see Vance, 1987:167ff. for
a presentation in English, and Nakada and Hayashi, 2000 in Japanese). One should also
add a couple of loans dating back to a period earlier than the fifth century such as uma
‘horse’, e (〈 we) ‘picture’, or ki nu ‘silk’, which have been perfectly adapted to the
Yamato phonology, so that nothing in their phonological structure hints at the fact that
they are indeed words of Chinese origin. Hence, they are often regarded as Yamato
words.
So a given character is likely to possess several different Sino-Japanese readings (on’yomi
Page 17 of 23
Introduction
み). Some characters only have one Sino-Japanese reading, generally the kan reading,
some have two, a kan and a go reading, and a small number of characters even have
three, four, or even more readings, since there may be several kan,go,tô, and usage
readings attached to a single character. Most speakers are not capable of saying whether
a given pronunciation is go,kan,tô, or usage. It seems that they simply memorize the
different Sino-Japanese readings of a given character, and the contexts in which each
reading is employed. Here are some examples, which illustrate some of the different
cases one is likely to encounter, and which provide an idea of the complexity of the issue
(Old Chinese reconstructions are from Tôdô, 1996):
(7)
to go ten to
mirror north clearness thousand various supply
Go – hoku myou 〈 gyou 〈 mon – nou 〈
reading myau gyau nahu
Kan kan 〈 hoku mei kou 〈 ban zou dou 〈
reading kamu kau dahu
Tô – – – an – zou 〈 na
reading zahu
Usage – – – – man zatsu, tou 〈
reading zou tahu, nan
Old Ch. *klǎm *puək *mıǎŋ *hǎŋ *mıuǎn *dzəp *nəp
(p.20) There are other alternations typical of Sino-Japanese lexemes, which have a more
synchronic and morpho-phonemic status, such as the CV/Q alternation, the i/u alternation,
the h/p alternation, and so forth. Those will be addressed in the relevant chapters of this
book.
The Sino-Japanese lexicon is rich in possibilities of lexical creation even if, at the present
time, those are not as exploited as they used to be. This is because, nowadays, gairaigo
constitute another privileged source of word-coining. Note also that a fair number of the
kango currently in use are actually Japanese lexical creations (rather like neo-classical
compounds with Greek or Latin roots are in European languages). These are called wasei
kango ‘Chinese words coined in Japan’. This practice of lexical creation has been
attested since the Heian period. These new lexemes are generally pure neologisms (for
example denwa° ‘telephone’), but they may also represent the semantic calques of
existing Yamato words. For example ka ji ‘fire’ is simply the Sino-Japanese reading of
the Yamato expression hi no koto の . During the Meiji era, literal translations of words
belonging to Western languages by way of Chinese characters were extremely common,
like byouin° ‘hospital’, a calquing from Dutch zieken-huis (‘illness’ + ‘public house’,
cited by Loveday, 1996:71). These neologisms containing Chinese characters have
frequently made their way back to Modern Chinese or Korean, where they have
received the corresponding Sino-Korean or Modern Chinese readings.
Page 18 of 23
Introduction
In comparison to Yamato and Western lexemes, kango are considered to belong to a

more formal and learned register. They correlate with intellect, distance, authority,
formality, and are used in juridical, academic, or scientific texts, in preference to Yamato
words when the alternative exists.
1.6.3 Gairaigo
What is a gairaigo? It is generally a word that has been borrowed in Japanese after the
sixteenth century, and mainly during the twentieth century, from a language which does
not use Chinese characters. For example tabako° ‘tobacco, cigarette’ (Portuguese
{tabaco}), mi sa ‘mass’ (Latin {missa} through Portuguese), zubon / zubon ‘trousers’
(French {jupon}), meetoru° ‘metre’ (French {mètre}), bi iru ‘beer’ (Dutch {bier}),
arubaito ‘(student) job’ (German {Arbeit}), interi° ‘intellectual’ (Russian {intelligentsija}),
ba taa ‘butter’ (English {butter}). It may also be—although more rarely—a word
borrowed recently from a modern Asian language using Chinese characters like Chinese
or Korean, but in Japanese the loan has kept a pronunciation which is close to that of the
source language. That means that even if the word can be written in sinograms, those will
not be read according to the conventional Sino-Japanese reading but with a pronunciation
which attempts to be faithful to that of the (p.21) modern source language, for instance
maajan° / maajan ‘mahjong’ (from a Chinese dialect, this word was borrowed at the
beginning of the twentieth century), cho ngaa ‘old boy’ (Korean {chhonggak}), kochujan
‘pepper paste’ (Korean {kochhujang}). Lastly, it is important to mention that some
gairaigo are nothing more than lexical forms coined by the Japanese, and involving
Western roots. They are made up either by combining morphemes existing in one or
more foreign languages or by truncating a borrowed form. Japanese lexicographers call
these wasei eigo (‘English word created in Japan’) or wasei yôgo (‘Western
word created in Japan’). They are usually written in katakana. Such words are extremely
numerous, for example sarariiman {salaryman} ‘a company worker’, gouruden-wiiku
{golden week} ‘a succession of several holidays around the end of April or beginning of
May’, wo ukuman {walkman}, na itaa {nighter} ‘night game (baseball)’, paso-kon°
(abbreviation of {personal computer}) ‘personal computer’, depaato (abbreviation of
{department store}) ‘department store’.
It is also necessary to mention acronyms and Latin-alphabet-based creations, which also

constitute a source of derivation and lexical coinage. The basis for a number of acronyms
are Yamato or Sino-Japanese morphemes, whose initial letter in the Latin alphabet is used
for acronymization: Nippon Housou Kyoukai -〉 NHK / enu-ecchi-kee ‘NHK
(Japan Broadcasting Corporation)’, hentai° -〉 H / e tchi ‘pervert’, ofisu redii -〉 OL / ou
eru° ‘female employee’.
The gairaigo often display phonotactic combinations that are not found in wago or kango
(see section 3.14). They can also be very long, and contain many long vowels and
geminate consonants.
Gairaigo are used primarily to refer to new objects, or to concepts borrowed from
foreign cultures. They are particularly frequent in the fields of fashion and cosmetics,
Page 19 of 23
Introduction
sport, non traditional arts, gastronomy, technology, and sciences. There often exists a
native Japanese or Sino-Japanese equivalent of a gairaigo. The use of a gairaigo can also
be dictated by pragmatic or stylistics factors. The connotations associated with this
vocabulary are: the West, modernity, innovation, and also in certain cases refinement and
sophistication.
It might be necessary to distinguish gairaigo, which are Japanized foreign loans, written
in katakana, pronounced in a Japanese way, and likely to be integrated in a Japanese
sentence like any Yamato or Sino-Japanese lexeme, from gaikokugo , which are
borrowed ‘wholesale’ from a foreign language. Gaikokugo are written in the alphabet and
their pronunciation is not yet Japanized. Their lexical categorization also remains fuzzy.
Such non-integrated loans are generally used in advertisements, without being
integrated into a sentence. They are still at the margins of the language.
(p.22) It follows from this definition that in Japanese, a gairaigo, literally a ‘word coming
from the outside’, is not necessarily a foreign word—many gairaigo are Japanese lexical
creations, or ‘Japenglish’—and that the major group of lexemes of foreign origin, namely
the kango, do not belong to the category of gairaigo.
1.6.4 Other Types

Japanese also contains a number of words of Sanskrit origin, most of which pertain to the
Buddhist vocabulary. These words were often borrowed via Chinese: for example
kawara° {kapāla} ‘tile’, daruma° {(bodhi-) dharma}, or danna° {dāna} ‘master,
husband’.
Words of Ainu origin, for example sa ke or sha ke {sakipe 〈 sakuipe} ‘salmon (literally
‘summer food’)’, ka ba, kanba {kaniha} ‘birch’, rakko°, rakko {rakko} ‘sea otter’, have
dubious status. Certain lexicologists classify them among Yamato words, others among
the gairaigo. The difference generally comes from the epoch at which the loan was made:
the older it is, the more likely the word will be regarded as a wago. It is necessary finally
to mention a number of prehistoric loans from Korean, such as kushi {kusil} ‘a comb, a
spit’ or tera {tSɔl} ‘temple’. These words are generally considered to be Yamato words.
From the morpho-phonological point of view, they do not display any particular
characteristic which distinguishes them from Yamato words.
1.6.5 The Limits of Stratum Categorization

The formal boundaries between wago,kango, and gairaigo tend to attenuate as time goes
by, through a process of lexicon homogenization. Thus the characteristics that were
originally specific to kango, such as palatalization, the presence of the mora nasal /N/,
gemination, the presence of the /r/ consonant or of a voiced obstruent word-initially,
ended up extending to words of the Yamato stratum. The presence of these elements
thus no longer constitutes, in itself, a proof that a word is of Chinese origin, even if it
remains generally possible to determine the origin of a lexeme just by its phonological
structure. On the other hand, most gairaigo, especially most recent ones, generally have
a phonological structure which makes them immediately identifiable as such. But the
Page 20 of 23
Introduction
oldest gairaigo or the ones which are in very frequent or daily use are more Japanized
than those of more recent introduction or those less frequently employed.
The classes are thus not discontinuous. They are organized rather like a continuum:
certain words belonging etymologically to one of the classes can ‘move’ to another one,
or borrow in a more or less occasional way some of its morpho-phonological features.
Certain words do not have the phonological, orthographical, or semantic profile of their
true etymology, and they belong de facto to some other class than to the one that history
should have confined them. (p.23) For instance, kappa° ‘raincoat’ is an old loan from
Portuguese, which is treated as a wago since it undergoes rendaku (sequential voicing) in
the compound words ama-gappa ‘raincoat’ or biniiru-gappa ‘rainwear made of plastic’
(Takayama T., 2005). Niku ‘meat’ is originally a Sino-Japanese word, but it takes a Yamato
polite prefix o-, instead of go- normally used as the polite prefix of Sino-Japanese words.
The reading yo / yo n for the numeral ‘four’ is etymologically a Yamato word, but it is
chosen instead of the Sino-Japanese reading shi ‘four’ in many numeral compounds using
Sino-Japanese components (for instance yo-nin rather than *shi-nin ‘four persons’), so
that it behaves like a Sino-Japanese word. Finally, let us mention the word kouhii コーヒー
‘coffee’, a Dutch loan {koffie}, which is sometimes written in kanji ( ) instead of
katakana. The effective placing of a lexeme in such or such a class can also vary according
to the speaker. Thus there do not exist absolute criteria to determine which class a
lexeme belongs to. In many cases, whether an item belongs to a given stratum or not
cannot be determined on the basis of surface distribution patterns (Ota, 2004). For
instance, there is nothing in the surface phonology of the Yamato word tonbo ‘dragonfly’
which suggests that it does not belong to the same stratum as te nba ‘flying horse’, which
is a Sino-Japanese word, or ko nbo ‘combo’, a Western borrowing. The script is often one
of the crucial elements for native speakers, alongside the phonological characteristics of
the word.
Itô and Mester (1995a, 1999) have proposed a concentric model, or core–periphery
organization of the lexicon, whose internal structuring is governed by the interaction of
constraints. The lexicon is viewed as an abstract space with a core and a periphery. At the
periphery stand the lexical items which are least assimilated (gairaigo and gaikokugo), in
the centre, the native lexemes (Yamato). Sino-Japanese lexemes appear in intermediate
position. In this model, the maximum set of lexical constraints holds in the core lexical
domain, occupied by lexical items traditionally labelled as Yamato. Itô and Mester (1995b)
propose a slightly different implementation of this view. They postulate that all lexical items
obey the same markedness constraints but that there exist different versions of stratum-
specific faithfulness constraints. The constraints which demands faithfulness for Yamato
words are ranked lower than those demanding faithfulness for kango, which in turn are
lower than those demanding faithfulness to gairaigo. So as the peripheral zone of the
lexicon is approached, many of the constraints cease to hold (are ‘turned off’) or are
weakened in various ways. However, such a concentric conception of the lexicon
organization does not solve all the problems, whatever the type of implementation which
is adopted (cophonology approach or the indexed faithfulness approach).
Page 21 of 23
Introduction
As Itô and Mester observe, it is not possible to impose a total ordering on vocabulary
strata because the domain of application of constraints is not continuous, since a
constraint might hold for two strata which are separated by another (p.24) stratum.
There exist certain constraints which apply to more peripheral members and not to those
located at the centre, for instance the constraint governing the length of Sino-Japanese
words: they are one or two mora long, whereas Yamato words, on the one hand, and
Western loans, on the other hand, can be longer than two moras. One of the issue
currently debated in recent papers dealing with lexical classes in OT works is whether
one of them should be considered as more marked than the others, or unmarked (in the
OT sense, which is also, mutatis mutandis, the structuralist sense). For instance, Itô and
Mester (1995a, b) argue that wago are the most unmarked items, following the
traditionally implicit assumption of Japanese linguistics, while Kawahara et al. (2003), on the
contrary, claim that it is the Sino-Japanese stratum which is the most unmarked. Yet, it
seems that none of the strata should be regarded as more marked or unmarked with
respect to others. Each stratum should be seen as a coherent class, independently and
absolutely defined, and associated with a set of associated formal and pragmatic (usage)
properties such as the ones mentioned in this chapter. The real question is whether a
given lexical item is marked with regard to the features that prototypically characterize
the stratum it belongs to etymologically. For instance, as a kango,haka (in ha kase
‘doctor’) is marked because it is bimoraic and does not end with i or u, ta npopo
‘dandelion’ is marked as a Yamato lexeme because it contains an unvoiced stop after the
mora nasal, and so on.
Notes:
(1 ) For instance, it is somewhat surprising that a 520-page book entitled The Handbook of
Japanese Linguistics, recently published by Blackwell, includes only about 10% of
Japanese titles in its abundant list of references.
(2) As we will see in Chapter 6, ‘syllabary’ is not a proper term, since kana denote moras
rather than syllables. However, for the sake of convenience, I will use the term syllabary
to refer to katakana and hiragana thoughout this book.
(3) The basic and most important prosodic unit of Japanese is the mora. The mora
corresponds to one phonological beat. For instance, sequences such as okinawa°
[okinawa] ‘Okinawa’, toukyou° [toːkjoː] ‘Tôkyô’, konpon° [kompoN] ‘basis’, and gakkari
[ɡakˡkaɾi] ‘disappointed’ all count as four beats, i.e. four moras, according to native
speakers’ intuitions. See Chapter 6 for an in-depth presentation of the prosodic
components of Japanese, and a discussion on the structure and nature of the mora.
(4) The terms ‘fifty sounds’ refers to the theoretical number of letters given the number
of rows and lines (5 x 10). It was coined before the invention of the kana ん /N/, which was
added later on. The table actually never contained more than 47 or 48 items.
(5) An average adult Japanese speaker is exposed to about 4000 characters in his or her
daily life. It has to be noted however that he or she does not necessarily know all the
Page 22 of 23
Introduction
meanings or readings of a given character. Word-processing software contains between

6000 and 7000 characters, in conformity with the set of lists approved by the computer
industry, which has agreed on a standard encoding for Japanese, the JIS (Japanese
Industrial Standard). The most complete character dictionary in Japan, the Daikanwa Jiten
(published by Taishûkan) contains about 50,000 characters.
(6) A number of recent Japanese researchers sometimes use the term onomatope オノマトペ
{onomatopée} (from French) as a cover term for the mimetic class of words. However,
since onomatopée originally refers only to words which are supposed to imitate a sound
of the extra-linguistic world, the terms mimetic or ideophone seem more appropriate in
the case of Japanese, because a large number of Japanese mimetic words are not sound
imitation but rather express feelings, sensations, attitudes, visual states, and so on in an
iconic manner.
(7) However, a number of mimetic words based on Chinese or Western lexemes exist,
for instance go u-gou 々 ‘with a rumbling sound’ (ch.), chi ku-taku {tic tac}, or rabu-rabu°
{love love} ‘in love’.
(8) Of course, there exist Yamato words which have been created after the fifth century,
and even very recently. They are nearly always the result of compounding, for instance
odakagata° ‘oxytonic’, or kakitome° ‘registered mail’.
(9) The status of final i and u in Sino-Japanese bimoraic morphemes will be addressed in
section 2.4.
Vowels

Laurence Labrune

Print ISBN-13: 9780199545834
DOI: 10.1093/acprof:oso/9780199545834.001.0001
Vowels
Laurence Labrune
DOI:10.1093/acprof:oso/9780199545834.003.0002
Chapter 2 presents the five vowels /a, i, u, e, o/ and their corresponding long vowels in
Tôkyô Japanese, with a look at other modern dialectal systems and the ancient Kyôto
language. The special status of the high vowels /i/ and /u/, which undergo frequent elisions
or insertions, and are the locus of devoicing, a phenomenon of major relevance in the
phonology of Japanese, is discussed in detail. This chapter also provides a detailed
description of prosodic lengthening and shortening, and proposes a discussion and
analysis on the status of the so-called diphthongs.
Keywords: prosodic shortening, prosodic lengthening, devoicing, high vowels, elision, insertion, Japanese
vowels
2.1 The Vowels of Standard Japanese: Outline of the System And General
Characteristics
Standard Japanese has the following five vowels:
Page 1 of 32
Vowels
i [i] front high unrounded vowel

u [ɯ] back high unrounded vowel
e [e] front mid unrounded vowel
o [o] back mid rounded vowel
a [a] low central vowel
The distinction between short and long vowels is relevant: [i] vs. [iː], [e] vs. [eː], [ɯ] vs.
[ɯː], [o] vs. [oː], [a] vs. [aː] (see section 2.7). There is no significant quality difference
between the short and long members of a pair.
Generally, the articulation of Japanese vowels is rather clear, but not very tense.
Japanese /i/, /e/, and /o/ are slightly less tense than the corresponding cardinal vowels. /o/
is the most rounded and most posterior of all Japanese vowels. /u/ is generally rather
unrounded, especially in Tôkyô Japanese. Its phonetic quality varies between [ɯ], [ɯ̈], [ᶶ],
and [ᶤ]. However, roundedness is at least phonologically present, since the fricative /h/ is
always bilabial ([ϕ]) before /u/, while it is not before /a/, /o/, /i/, and /e/. This would be
unexplainable if /u/ did not contain a certain degree of phonological labiality, that is, of
roundedness. One encounters a centralized allophone [ɯ̈] after /s/, /t/, /z/, and after the
palatalized consonants (Cy), for example in the word gyuunyuu° ‘milk’ [gjɯ̈ːnjɯ̈ː].
The low vowel /a/ is generally central, sometimes even slightly posterior ([a]) after velar
consonants but lip-rounding is not very marked.1
The five vowels of this system can be classified, from the most open to closest, in the
following order: a 〉 o〉 e 〉 u 〉 i (Mabuchi, 1971) or, from longest to shortest: a 〉 e 〉 o 〉 i 〉
u (Shimizu Han, 1962). According to Shimizu Han (1962), /a/ is almost 1.5 times longer
than /u/. We will see further in section 2.4 that /i/ (p.26) and /u/ display a number of
remarkable properties due to their closeness and brevity.
From the point of view of general linguistics, Japanese thus displays a rather classical
triangular vowel system with five elements, the commonest system among the languages
of the world (Maddieson, 1984), except for the presence of the unrounded back high
vowel [ɯ].
(1) Vowel system of Japanese
The three distinctive features [high], [low], and [front] are sufficient to characterize this
system:
Page 2 of 32
Vowels
i u e o a
high + + − − −
low − − − − +
front + − + − −
2.2 Old And Dialectal Vowel Systems

According to Hashimoto Sh. (1928), Archaic Japanese—until the beginning of the ninth
century approximately—had an eight-vowel system whose elements can be denoted, for
want of anything better, as i1 , i2, u, e1 , e2, o1 , o2, a. The transcription i, ï, u, e, ë, o, ö, a is
also commonly used in Japanese historical linguistics. The vowels ï, ë, and ö are called
otsu, or type A, vowels (otsu boin ) in opposition to the kô, type B, vowels (kô boin
). The exact number as well as the phonetic quality of these vowels constitute one of the
most debated issues of Japanese historical phonology (in addition to works by Hashimoto
Shinkichi, see also Arisaka, 1955, Ôno, 1980, Morishige, 1975, Matsumoto, 1975, Hattori,
1976, among others, and Yasuda, 1982, for a review of the various assumptions).
At the turn of the seventeenth century, the Japanese language as described by the
Iberian missionaries was characterized by the existence of a distinction between a close-
mid o and an open-mid o, both long, /oː/ and /ᵓː/, transcribed as ô and ǒ in the roman
alphabet texts of the time. These two vowels are termed respectively gôon ( ) and kaion
( ). The former was pronounced more or less like present-day long o [oː] and
corresponded to the pronunciation of (p.27) sequences transcribed as (C)ou, (C)eu (ou
オウ、kou コウ、sou ソウ、eu エウ、keu ケウ、seu セウ, and so on), while the latter [ᵓː] corresponded to the
realization of sequences originally comprising the written group (C)au (au アウ、kau カウ, etc.),
all very frequent in kango. A number of modern dialects have maintained this opposition,
which has disappeared in Standard Japanese where both vowels are now realized with a
long o ( [oː]).
Three-vowel systems (for instance /i, u, a/ in the Ryûkyû dialect of Yonaguni, Okinawa
prefecture), four-vowel systems (/i, u, o, a/, Miyakejima dialect, Tôkyô prefecture, or /i, ï,
u, a/, Amami dialect, Kagoshima prefecture), six-vowel systems (/i, u, e, o, ▱, a, Sendai),
seven-vowel systems (/i, u, e, o, ▱, æ, a/, North of Honshû and Niigata area), and up to
eight-vowel systems (/i, y, u, e, ∅, o, æ, a/, in which /æ/, /y/, and /∅/ are phonetically long,
Nôbi dialect, Aichi prefecture) have been reported. These systems are generally
regarded as representative of the relatively recent innovations starting from the five-
vowel triangular system /i, u, e, o, a/ presented above, rather than as direct descendants
of the archaic system with eight vowels. Additional phonemes in the dialectal systems
having more than five vowels generally result from the coalescence of two simple vowels:
for example, in Aichi Japanese, /y/ ([yː]) results historically from /ui/, /∅/ [∅ː] from /oi/, and
/æ/ ({ː]) from /ai/.
2.3 Distributional Characteristics of /e/

The vowel /e/ stands out from the other Japanese vowels by its special distribution in
Yamato and mimetic words. In morphemes longer than one mora, it seldom occurs in the
Page 3 of 32
Vowels
initial, and it never appears twice in a root. Thus forms such as *meni or *kaseme are ill-
formed as mono-morphemic Yamato words. This suggests that /e/ was probably absent in
the proto-system. Besides, a number of modern words ending in /e/ display an
alternating, non-independent form in /a/, for instance a me ‘rain’ occurs as ama- in
ama-do (‘rain’ + ‘door’) ‘shutter’, sake° ‘alcohol’, as saka- in saka-gura° (‘alcohol, sake’ +
‘cellar’) ‘sake cellar’. The same type of alternation occurs, but to a lesser extent, between
/i/ and /o/: ki ‘tree’ but ko-kage° ‘shade of the trees’. Such alternations are no longer
productive in the modern language.
/e/ is also the least frequent of all the five Japanese vowels (see section 2.9).
In mimetic words, /e/ generally brings a negative connotation, as in hera-hera ‘in a

meaningless manner’ or be ta-beta ‘sticky’.
Another characteristic of /e/ is that it cannot be palatalized, i.e. preceded by the semi-
consonant y (see section 3.11) in the Yamato and Sino-Japanese strata. The moras
kye,mye,hye, etc. do not exist and they are also rare in gairaigo.
(p.28) The long version of /e/ also deserves a number of remarks, which will be made
further in section 2.7.1. We will also see in section 2.6. that /e/ is, of all the Japanese
vowels, the least likely to undergo devoicing.
2.4 Phonological Status of /i/ And /u/

The high vowels /i/ and /u/ are characterized by a number of remarkable properties.
First, they have devoiced allophones in certain contexts, for example hiku° [Ci̥kɯ] ‘to
draw’, de su [desɯ̥], or even sometimes [des] or [desμ] (Copula). This is a phenomenon
of major importance for the phonetics and phonology of Japanese, which also concerns the
other vowels but to a lesser extent. It will be discussed in more detail in section 2.6. In
addition, the presence of the vowels /i/ and /u/ modifies in a significant way the phonetic
realization of the consonants /t/, /d/, /h/, /s/ and /z/ which occur before them. This leads to
the neutralization of a number of phonological oppositions which will be considered in
Chapter 3 devoted to the description of the consonant system.
It is necessary also to mention the very frequent cases of neutralization between /i/ and
/u/ after /s/ and /z/ (sh and j) in Tôkyô Japanese, even though they are seldom reflected in
the spelling: for example shu jutsu /syuzyutu/ ‘surgical operation’ is often realized as shi
jitsu /sizitu/, shinjuku° as shinjiku° ‘Shinjuku (a district of Tôkyô)’, geshuku° as geshiku°
‘pension, hotel’. This phenomenon, called chokuonka (? ? ), seems to occur particularly
frequently before the moras ku,tsu, and ju. It also appears in Western loans in a larger
variety of contexts: rejume° /rezyume/ {résumé} may become rejime°. This confusion
dates back at least one century, since it is already mentioned by Aston (1904).
Lastly, /u/ and /i/ function as the epenthetic vowels par excellence. They are also the most
likely to undergo deletion under certain conditions. These two aspects, which also
concern, though to a lesser degree, the other vocalic segments, will be the topic of the
following section.
Page 4 of 32
Vowels
2.5 Vowel Insertions And Deletions

Insertion and deletion of vowels are extremely frequent in Japanese, but the data are
various and complex. We shall consider here the issue of vowel epenthesis occurring in
Western loans, for example re sutoran {restaurant}, as well as the alternations between
CV and /Q/ triggered by vowel deletion in some Sino-Japanese morphemes containing an
etymologically epenthetic vowel, as in (p.29) gakusei° ‘student’ / gakki° ‘school
term’, two Sino-Japanese words which both start with the morpheme gak(u) ‘study’.
The lexemes belonging to the Western stratum contain many epenthetic vowels. The goal
of such epentheses is obviously to get surface forms which conform to the prosodic
structure of Japanese, and in particular to avoid word final consonants and illicit
consonant clusters. In the first loans from European languages, the quality of an
epenthetic vowel was often obtained by copying that of a surrounding vowel: for example
garasu° 〈 {glas} (Dutch) ‘glass’, kirishitan 〈 {cristão} (Portuguese) ‘Christian’,
gorofukuren 〈 {grof grein} (Dutch) ‘camlet’, paatere (accent unknown) 〈 {pater} (Latin)
‘priest’. In Modern Japanese, it is mainly /u/ which is used, but the vowel /i/ is also
encountered after sh and ch and occasionally after k, as well as /o/ after t and d: huransu°
‘France’ {France}, su kasshu ‘squash’ {squash}, suto raiki ‘strike’ {strike}, suto raiku
‘strike (in base ball)’ {strike}, shichuu° ‘stew’ {stew}, or te kisuto ‘textbook’ {text}.
At the level of phonetic realization, some of these vowels can undergo drastic reduction,
and even become almost inaudible, including, quite unexpectedly, in contexts which are
not normally devoicing contexts. This is the case, for instance, of /u/ following /r/ in amusu
terudamu {Amsterdam}.
One of the problems raised by these vowels to phonological theory is the following:
although phonetically present, they are sometimes—but not always (hence the problem)
—invisible at the phonological level (Kubozono, 1996, 2001b, 2006b). For example,
whereas four-mora long Western loans ending with a -CVCV sequence are normally
atonic (e.g. arizona° {Arizona}), those ending with an epenthetic /u/ behave differently,
displaying the tonic pattern, with penultimate or initial accent. Thus words having a
(C)VCVCVCV pattern like su toresu {stress} or a doresu {address}, with final epenthetic
u, are tonic, as are words with a (C)VCVCVC pattern ending with the mora nasal, such as
gu ratan {gratin} and ri mujin {limousine}. It looks like the final vowel of su toresu and a
doresu was absent at the phonological level, and that these words had a (C)VCVCVC
structure, like gu ratan and ri mujin. In Chapter 6, we shall see that the moras containing
such epenthetic vowels must be considered as belonging to the set of what I call deficient
moras.
It also happens that an accent which is expected to fall on a vowel of epenthetic origin is
shifted one mora leftward: we thus have amusu terudamu {Amsterdam} instead of
*amusuterudamu, which would be expected by application of the antepenultimate accent
rule (see section 7.2.4), or andesu-kai ‘club of the Andes’ instead of *andesu-kai, which is
the form in conformity with the general rule of compound accentuation. The exact
characterization of these vowels thus poses very serious problems to current
phonological theories, which still await a proper solution.
Page 5 of 32
Vowels
(p.30) In the Sino-Japanese stratum, non-etymological /i/ and /u/ occur after /t/ and /k/,
at the end of bimoraic morphemes, as stated before in section 1.6.2, and as the examples
below illustrate.2
(2) Non-etymological /i/ and /u/ in Sino-Japanese morphemes
Old Chinese Japanese
*kiet /kiti/, /kitu/ kichi, kitsu ‘good fortune’

*buət /butu/ butsu ‘Buddha’
*tıok /tiku/ chiku ‘bamboo’
*ɦıuək /iki/, /yoku/ iki, yoku ‘territory’
In almost all cases, the choice of the vowel depends on the context (Itô and Mester,
1996; Tateishi, 1990). After /t/, /u/ is normally used (for example /butu/ butsu ‘Buddha’,
/setu/ setsu ‘to touch’, /katu/ katsu ‘energy’). It is the same after /k/ ( /tiku/ chiku
‘bamboo’, gaku ‘study’, huku ‘good fortune’), except when the Sino-Japanese
morpheme in question contains the front vowel /e/ in its first mora, in which case /i/ is
selected after k: teki ‘enemy’, reki ‘passing of time’. The only context in which non-
predictable variation between /i/ and /u/ occurs is that of Sino-Japanese morphemes which
may contain an /i/ in the first mora: kichi or kitsu (/kiti/ or /kitu/) for ‘good fortune’, iki or
yoku for ‘territory’. These differences between the two forms of a pair are generally
explained by the fact that the two phonetic forms were borrowed at different historical
periods.
In the modern language, such final i and u do not appear before a voiceless consonant
when the morpheme is used in compounding. The preceding consonant k or t remains as
the first part of a geminate (/Q/). In the case of /k/, the phenomenon occurs only before
another /k/ (3a) whereas it may occur before all voiceless consonants in the case of /t/
(3b).
(3) CV / Q alternation in Sino-Japanese
a.
gaku + kai 〉 gakkai° ‘congress’

gaku + sei 〉 gakusei° / *gassei ‘student’
b.
ichi /iti/ + satsu /satu/ 〉 issatsu /iQsatu/ ‘one book’

ichi /iti/ + kai /kai/ 〉 ikkai /iQkai/ ‘once’
ichi /iti/ + tou /toR/ 〉 ittou° /iQtoR/ ‘first class’
ichi /iti/ + hon /hoN/ 〉 i ppon3 /iQpoN/ ‘one long object’
Page 6 of 32
Vowels
(p.31)
The first issue to consider is whether these non-etymological vowels must, in synchrony,
be considered as epenthetic or not. The traditional position of Japanese linguists and
philologists is that they should not: /i/ and /u/ are regarded as belonging to the underlying
lexical form of the Sino-Japanese morpheme. However, Itô and Mester (1996) defend an
opposite and radical point of view. They argue that the final V which appears after t and k
in bimoraic Sino-Japanese stems should be regarded as epenthetic even in contemporary
Japanese. Their principal argument rests on the observation that the quality of the second
vowel of the Sino-Japanese morphemes is almost always predictable.
However, not only does this line of analysis fail to explain alternations of the kichi / kitsu
type, which is totally non-predictable, but it also neglects the fact that a great number of
these morphemes carry a final accent at the lexical level, for example jitsu ‘truth’ (see
also section 7.2.3 about this type of word), a fact which is difficult to account for
theoretically, since one hardly sees how a segment which is absent at the underlying level
could receive an accent, especially in consideration of the fact that epenthetic vowels are
known to resist accent attribution, in Japanese as well as in other languages.
Kubozono (2001b) discusses another argument which could adduce evidence for the
absence of these vowels at the phonological level. He considers that the morphemes at
stake sometimes behave in an exceptional way, as if they had a CVC structure (their final
vowel would thus be, according to Kubozono, invisible) in compound words, because
they involve accent shift to the antepenultimate mora, that is, on the last mora of the first
component, instead of following the rule which stipulates, according to Kubozono’s
theoretical interpretation (see section 7.3.1), that a non-final lexical accent is preserved in
the second component of a compound. For example the compound yoyaku° + seki
(‘reservation’ + ‘seat’) is accented as yoyaku-seki ‘reserved seat’, with accent shift,
rather than *yoyaku-seki in which the accent would be kept on the initial mora of the
second member. However, it seems simpler and more straightforward to treat the
vowel-final allomorphs as basic, since, as pointed out by Vance (1987), it is not
automatically recoverable from the Q-final allomorph. For instance, the form /riQ/ can
correspond to /ritu/ ‘to stand’, /riki/ ‘strength’, or /riku/ ‘land’. In addition, as
pointed out by Tanaka and Yamane (2000), apart from the specific case mentioned by
Kubozono (2001b) and cited above, vowels of epenthetic origin in Sino-Japanese
morphemes behave in all other cases like underlying vowels rather than like epenthetic
ones. This issue nevertheless requires further consideration, but it appears preferable
for the moment to consider that, in Modern Japanese, final /i/ and /u/ in words such as
kichi / kitsu or gaku belong to the underlying representation of bimoric Sino-
Japanese morphemes of the shape CVCV, and that they may be deleted under certain
conditions (see also McCawley 1968:110ff., Vance 1987:158ff., Kurisu 2000 and (p.32)
Nasu1996, among others for discussions and theoretical treatments of this phenomenon
in addition to the references cited above). This by no means prevents us from
recognizing the special status of the segments /i/ and /u/ themselves, or of the prosodic
units (moras) that contain them, as I shall propose in Chapter 6. Besides, the vowels /i/
Page 7 of 32
Vowels
and /u/, be they epenthetic or not, display several other peculiarities in Japanese.
Another problematic issue is that of the non-systematic application of the CV → Q process.

Otaka (2009) argues that Sino-Japanese morphemes of the shape CVCV should be
categorized in three classes with respect to the CV / Q alternation: (a) the morphemes
which always generate a geminate consonant regardless of the onset consonant of the
second morpheme, unless it is voiced, for instance ichi /iti/ ‘one’, jitsu /zitu/ ‘real’, juu
/zyuu/ ‘ten’; (b) the morphemes that generate a geminate consonant when the onset
consonant of the second mora is identical to the onset consonant of the first mora of the
second morpheme and they are both voiceless, for instance soku /soku/ ‘instant’, seki
/seki/ ‘red’; (c) the morphemes that do not generate a geminate consonant, regardless
of the phonological environment, such as shuku /syuku/ ‘inn’, shichi /siti/ ‘seven’.
Vance (1987:159) observes that the CV / Q alternation does not display the same degree
of regularity. While the tsu -〉 Q and ku -〉 Q processes are regular or almost automatic,
chi -〉 Q is inconsistent and ki -〉 Q clearly irregular. Otaka (2009) also invokes frequency
effects to account for the ku / Q alternation as observed for instance with kaku ‘each’.
Kaku + koku ‘country’, which is extremely frequent, is always geminated (ka kkoku
or kakkoku° ‘each country’) while kaku + ko ‘door’, which is seldom used, is less
often so (ka kko or ka kuko ‘each home’ both coexist). There also exists at least one
minimal pair where the presence or absence of gemination may be distinctive: roku ‘six’
+ hou ‘law’ with a geminate (ro ppou / roppou° ) means ‘the Compendium of Law’
while rokuhou , without gemination means ‘six kinds of laws’. Note that ro ppou and
rokuhou are both made up of the same morphemes.
In addition to /u/ and /i/ deletions just discussed, which are listed in dictionaries and
completely lexicalized, there exist vocalic deletions rarely mentioned but which are just
as frequent and interesting. A word such as gaku + sei 〉 gakusei° ‘student’, referred
to above, is frequently realized with an elision of /u/, i.e. [gakseː] rather than [gakɯseː].
The vowel deletion in gakusei° [gakseː] is—quite rightly—regarded as the result of vowel
devoicing, whereas that of gakkai° (gaku + kai) ‘congress’ is not. Admittedly, the fall of
/u/ seems more systematic in gakkai° than it is in gakusei°—even if that remains to be
demonstrated in spontaneous speech for a dialect like Tôkyô Japanese in which vowel
devoicing is particularly frequent. The point is, however, how gaksei° should be analysed
with respect to gakkai°. Are we dealing, in both cases, with the same (p.33)
phenomenon, the only difference being that the vocalic deletion in gakkai° is lexicalized,
recognized by dictionaries, and reflected in the kana spelling, whereas that of gaksei° is
not? Or does one have to consider that two distinct processes are operating? The answer
to this question is likely to have important consequences for the conception and definition
of the basic rhythmic unit of Japanese. Indeed, a form such as gaksei° contains a
phonetically closed syllable, gak, with a coda which is NOT the first part of a geminate, a
structure that would be quite novel in Japanese.
Moreover, note that in gaksei°, the deletion of the vowel cannot be transcribed in
Japanese writing. Indeed, in their actual state, the kana offer no possibility of writing a
consonant that is not followed by a vowel, with the exception of the mora nasal and the
Page 8 of 32
Vowels
first part of a geminate.
According to M. Beckman (1996), the deletion of /i/ in shokikan ‘secretary’ (/syoki/ +

/kan/) does not yield a homophone for shokkan° ‘tactile organ’ (/syoku/ + /kan/),
because the first /k/ of shokikan is still released. However, it remains to be proved that
such a difference exists, and that, if it does, it is indeed perceptible at the auditory level.
But in any case, these two words being accentually different, they could not be
completely homophonous.
The problem is particularly acute because the deletion of a high vowel is not so
systematic and the factors which condition it are not completely identified. Doublets like
tekikaku° / tekkaku° ‘exact’, kakukai° / kakkai° ‘the sumô world’, sankakukei /
sankakkei ‘triangle’, sakkyokuka° / sakkyokka° ‘music composer’, or
shougakukin° / shougakkin° ‘scholarship, grant’, seem to be quite frequent.
Moreover, intra and inter-speaker variation is extremely widespread. The degree of
lexicalization of the compound and its morphological cohesion are also certainly among the
determining factors, but they are probably not the only ones. In many respects, this
phenomenon is reminiscent of mute e (schwa) in French phonology.
Particularly interesting also is the fact that, as noted by many scholars, morphological
structure is relevant to account for the surface form when the Sino-Japanese compound
contains more than two characters (or stems). As Itô and Mester (1996), building on
Vance (1987), McCawley (1968), Martin (1952), and Kubozono (1993b) put it,
contraction (i.e. CV reduction to /Q/) seems to occur at the end of a stem, provided that it
is not the end of a word, or, as Vance (1987:161) states, the Q-final allomorph does not
appear before the major constituent break in Sino-Japanese words of three or more
morphemes. For instance, the character betsu° ‘special’, undergoes gemination in
besseki° ‘assigned seat’ (betsu° + se ki ) but not in tokubetsu-seki / tokubetsu-seki
(tokubetsu° ‘special’ + se ki ) because the structure of the latter is (XX)(X). One
can further observe that the form *tokubesseki , if it was to (p.34) be created,
could be interpreted as ‘a specially assigned seat’, with a (X)(XX) structure.
Another point to be taken into consideration is the fact that the vowel deletions which
generate geminations also occur in the Yamato stratum when the speech rate is fast, and
that they also concern vowels other than /u/ and /i/. For example kaki-komu ‘to swallow
one’s meal’, or do ko ka ‘somewhere’ are frequently realized kakkomu and do kka. In
fact, vowels which occur between two /k/ seem more particularly concerned by such
‘wild’ deletions, whereas those located in a /tVC/ environment undergo deletion only in
highly lexicalized Sino-Japanese morphemes like /iti/ + /kai/ 〉 /iQkai/ ikkai ‘once’ or /iti/
+ /sai/ 〉 /iQsai/ i ssai ‘one year old’. Actually, some of these cases can also be
accounted for by vowel devoicing, a phenomenon which will be presented in the following
section. A general study of Japanese gemination based on oral, spontaneous, and
authentic data, which would also take into consideration the problem of vowel devoicing,
remains to be carried out.
2.6 Vowel Devoicing
Page 9 of 32
Vowels
Vowel devoicing, unvoicing, or devocalization (boin no museika の ) more

particularly characterizes the dialects of Kantô and Kyûshû. The IPA diacritic which
transcribes devoicing is ̥, for example [i̥], [ɯ̥].
Devoicing is not a recent phenomenon in Japanese. Collado (1632) already remarks that
certain Japanese /i/ and /u/ are sometimes inaudible.
Devoicing affects mainly the high vowels /i/ and /u/ in the two following contexts:
1. – when the vowel (be it accented or not) is placed between two voiceless
consonants (this also includes before the first part of a geminate) (4a);
2. – when the vowel is unaccented and placed after a voiceless consonant and
before a pause (4b).
Devoicing is almost compulsory in Tôkyô Japanese, except when several devoiceable

vowels occur in consecutive moras (see below).
/a/ and /o/ also undergo devoicing but in a more occasional manner, and under more
restrictive conditions. They must, in theory, be unaccented, occur between two voiceless
consonants, and, in addition, the same vowel must occur again in the following mora (4c).
The vowel /e/ seems to be the least affected by devoicing (Maekawa, 1993; NHK, 1985;
Akamatsu, 1997). However, Amanuma et al. (1989) mention the existence of realizations
containing a devoiced /e/ (4d).
(p.35) (4)
a.
hiku° [hi̥kɯ], [çi̥kɯ] ‘to pull’

gakusha° [gakɯ̥ɕa] ‘scholar’
tsuki [tsɯ̥ˡki] ‘moon’
pikkoro° [pi̥kkoɾo] ‘piccolo’
b.
ka rasu [ˡkaɾasɯ̥] ‘crow, raven’

a ki [ˡaki̥] ‘autumn’
ke chi [ˡketɕi̥] ‘stinginess’
c.
kokoro [ko̥ ˡkoɾo] ‘heart’

hokori [ho̥ koˡɾi] ‘pride’
haka [hḁˡka] ‘tomb’
d.
Page 10 of 32
Vowels
sekkaku° [se̥ kkakɯ] ‘on purpose’

keshou [ke̥ ˡɕoː] ‘make up’
Figures 2.1 and 2.2 present the oscillograms and spectrograms for the word a ki
‘autumn’, first in a devoicing context (a ki kara ‘since the autumn’, Figure 2.1) with a
devoiced /i/, then in a non-devoicing context (a ki demo ‘even
Figure 2.1. Spectrogram and oscillogram of a ki kara (with devoiced

i) (0.430s)
(p.36)
Figure 2.2. Spectrogram and oscillogram of a ki demo (no devoicing

of i) (0.410s)
in autumn’, Figure 2.2). The instrumental analysis shows that the devoiced vowel is
characterized by the absence of the first formant and of the so-called ‘voice bar’ that
corresponds to the vocal folds vibration in a ki kara, contrary to a ki demo.
Devoicing can lead to total disappearance (deletion) of the vocalic element on the surface
(Vance, 1987). This is particularly obvious when the high vowels /i/ and /u/ occur after a
fricative, especially in word-final position. For instance shita° /sita/ [ɕta] ‘under’, desu
Page 11 of 32
Vowels
[des] ‘Copula’, -masu [mas] (Politeness Auxiliary). However, even in such drastic
examples, the mora containing the orphan consonant preserves its prosodic weight and
still counts as one rhythmic unit. This is one of the reasons why the vowel cannot be
considered to be deleted at the phonological level. Another reason is that the quality of
the reduced vowel can be recovered from the articulation of the consonant which
precedes it (see below). As Faber and Vance (2000) observe, Japanese voiceless vowels
maintain this supralaryngeal integrity regardless of their surface duration, both in
influencing the articulatory and acoustic characteristics of adjacent phonological units and
in mediating longer-distance effects of one segment on another.
(p.37) When an accented vowel is devoiced, the accent frequently shifts to an adjacent
mora, especially in the conservative Tôkyô speech. However, the factors determining the
choice of the new accent location—when a possibility of choice exists—remain unclear.
Note that such accent shifts are becoming less frequent because it is now common to
maintain an accent on a devoiced vowel. This new trend seems to go back to the second
half of the twentieth century, since it is already reported by Akinaga (1967).
(5) Accent shift due to devoicing
ki te 〉 ki̥ te ‘to come and…’

mushiken 〉 mushi̥ ke n ‘exam-free’
sankakukei 〉 sankaku̥ kei ‘triangle’
Tanaka (2001) assumes that, in cases where several adjacent vowels are likely to receive
the accent after it has been displaced, the principle is that the high pitch remains within
the same foot, as the following examples illustrate (brackets indicate feet boundaries).
(6)
(bi)(jutsu)(kan) 〉 (bi)(jutsu̥)(kan) / *(bi)(jutsu̥)(kan) ‘museum’

(shita)(kuchi)(biru) 〉 (shi̥ta)(ku̥chi )(biru) /*(shi̥ta )(ku̥chi)(biru) ‘lower lip’
This is an interesting assumption, and it would be desirable to gather additional data in

order to confirm its validity, since counterexamples, in which the accent moves outside
the foot, or does not move at all, are rather easy to find. For instance, Yokotani (1997)
takes up the following examples: (tei)(kuu)-(hi)(kou)(ki) ‘low-altitude plane’ in which
the accent may move to the right (tei)(kuu)-(hi)(kou)(ki) (the foot parsing is mine), and
(bou)(shi)-(kake) ‘a hat rack’ in which the accent cannot move. It should be remembered
that feet organization is not always evident. It is the case of the mushiken (or mushiken
with devoicing of i) example, quoted above, where one hesitates between a structure
(mushi)(ken), which builds a foot every two moras, whatever the morphological structure
of the word, and a structure (mu)(shi)(ken), which respects the morpho-lexical structure
of the word made up of three Chinese characters, , and where the unit
Page 12 of 32
Vowels
corresponding to a character corresponds to one foot, whatever its length in moras. In

the first case (mushi)(ken) the word appears as an exception, since the accent moves to
another foot. In the second case (mu)(shi)(ken),shi is a monomoraic foot so there is no
other mora available within the same foot, but the choice of the right mora in preference
to the left one for accent shift remains unexplained.
Although the phonetic aspects of Japanese vowel devoicing have been extensively
studied, the factors which condition it remain difficult to capture. This phenomenon also
has important consequences for phonology and morphology. It causes, among others, the
creation of heavy consonant clusters at the (p.38) surface level, as well as accent
displacements. This is what makes the analysis of vowel devoicing particularly complex and
delicate. Moreover, it is worth noting that, as far as the basic description of the facts is
concerned, few works agree. Many scholars point out that devoicing cannot occur
simultaneously in two adjacent moras, but exceptions exist. For instance bakuchiku°
‘firecracker’, can yield [bakɯʨ͡i ̥kɯ] or [bakɯ̥ʨ͡i ̥kɯ] but never *[bakɯ̥ʨ͡ikɯ̥] (Akamatsu,
1997, the transcription is mine). One can also mention the extreme example
tsukutsuku-boushi [ʦ͡ ɯ̥kɯ̥ʦ͡ ɯ̥kɯ̥¹boːɕi] ‘(a variety of) cicada’, which may contain a
succession of four devoiced vowels. One also frequently reads in the literature that
devoicing cannot affect the initial vowel of a word when there is no consonant (i.e. a vowel
in the # position), but here also exceptions can be found, for example ikiru [i̥ kirս] ‘to
live’ (Imada, 1981:82).
In addition to those already mentioned, factors likely to favour vowel devoicing, and
which might be relevant in cases where several consecutive devoiceable vowels occur,
are: speech tempo; absence of accent on the devoiceable vowel; position in the word
(vowels occurring in the initial mora seem more easily devoiced, except when they are
onsetless); presence of /s/ before the vowel; presence of [k] especially, but also of [t], [s],
and [ɕ], after the vowel; presence of [a] in the following mora (Yoshida N., 2002).
Interestingly, these factors do not necessarily cumulate. A vowel placed after [s] easily
gets devoiced, just as a vowel placed before [s] does, but a vowel placed between two
[s]’s remains generally voiced, even though its loss would create a succession of two s’s
similar to a geminate at the surface level, a sequence which is acceptable in Japanese.
On the other hand, all things being equal, accented vowels occurring in a word-internal
mora, those preceding a morphemic boundary (Kondo, 1997; Vance, 1992, quoted by
Tsuchida, 2001), those followed by /h/ (under its allophonic forms [ç], [ϕ], or [h]), are
more resistant to devoicing. Lastly, as mentioned just above, devoicing of vowels
surrounded by two /s/’s is rare (Yoshida N., 2002). The interaction of these various
parameters is particularly delicate to model. Tsuchida (2001) presents an attempt to
analyse vowel devoicing within the framework of Optimality Theory, but she only takes
into account a small number of the factors just mentioned.
High vowel devoicing is a rather common phenomenon across languages (see for example
the case of Canadian French). It is basically due to the fact that /i/ and /u/ are less
sonorous and generally shorter than the other vowels, and also, as Kamiyama Takeki
(p.c.) points out, to aerodynamic and articulatory factors (narrower constriction causing a
Page 13 of 32
Vowels
higher intra-oral pressure, thus a lower trans-glottal pressure, a lower trans-glottal

airflow resulting in vocal fold vibration harder to realize). Moreover, the devoicing of /i/
and /u/ is no doubt favoured in Japanese by the fact that a number of consonants possess
specific allophones before high vowels. These allophones are maintained even when the
vowel is (p.39) totally deleted, so that it is generally possible to recover the quality of
the vowel on the basis of the consonantal allophone which precedes it even if the vowel is
deleted (for a study of how allophones are processed and recognized, see Ogasawara
and Warner, 2009). Thus in the reduced form [ʦ͡ ¹ki] tsuki ‘moon’, the presence of the
affricate allophone [ts‹] of the /t/ consonant is revealing of the underlying presence of /u/,
the only vowel which triggers the affricate realization of the phoneme /t/. The study by
Beckman and Shoji (1984) shows that the deleted vowels /i/ and /u/ can colour the
spectrum of the fricative [ɕ] in the moras shi /si/ and shu /syu/, thus allowing the vowels to
be recovered. The same effect probably occurs after other consonants, and especially
after the affricate ch [tɕ‹]. Because of this, it seems reasonable to assume that the vowel
is not phonologically deleted.
Finally, recall that vowel devoicing, because it may involve complete disappearance of the
vocalic segment at the surface level, entertains close and complex relations with the
phenomenon of consonant gemination, as seen in section 2.5. More generally, in colloquial
Japanese, devoicing also creates phonotactic sequences that are deemed to be illicit
because they contain a sequence of two consonants C1C2 (where C1 is neither nasal nor
homorganic with C2), including word-initially. Let us mention again examples such as desu
[des] CVC and shita [ɕta] CCV. Kondo (2000) rightly observes that, because of the
generalization of vowel devoicing, Japanese syllable structure may have become more
flexible and is now in the process of changing. Kondo (2003) also argues that consecutive
devoicing is prevented because it would create sequences of more than two consonants,
which would upset speech rhythm.
The status and phonological representation of devoiced vowels will be addressed again in
Chapter 6 which deals with prosodic constituents. It will be argued that prosodic units
which comprise a devoiced vowel can be regarded as structurally defective.
2.7 Vowel Length

Vowel length is distinctive in Japanese. Compare: ku ‘district’ / ku u ‘void’, obasan° ‘aunt’
/ obaasan ‘grandmother’, hi ru ‘leech’ / hi iru ‘heel’, tokai° ‘city’ / toukai° ‘destruction’,
kegen° ‘dubious’ / keigen° 4 (realized as [keːgeᶰ]) ‘reduction’.
(p.40) A long vowel is supposed to last twice as long as a short vowel. However,
instrumental analyses show that the ratio between a long and a short vowel is closer to
1:2.5 or even 1:3 (Shimizu Han 1962: 65). As Akamatsu (1997) observes, the point is that
native speakers intend (or are convinced) to pronounce a succession of two isochronous
identical vowels when they articulate a long vowel, and that native listeners perceive them
as such.
On the phonological level, a long vowel equals two rhythmic units, i.e. two moras, whereas
a short vowel is worth only one. The prosodic weight of the sequence kou [koː] is thus
Page 14 of 32
Vowels
phonemically equivalent to that of the sequences kon or koto, that is, twice that of ko.
Except for the particular case of the long /e/ in some of its possible dialectal realizations
(as [ei]), there does not exist any significant quality difference between a short and a long
vowel. Unlike other languages where length is correlated with quality differences, only
duration is distinctive in Japanese.
2.7.1 The case of long e

The case of long e calls for some observations. The long e which is heard in Sino-Japanese
words like sensei [sen¹seː] ‘professor’ or reigi [ɾeː¹gi] ‘courtesy’ goes back to a
sequence e + i realized as [eː] in normal speech. In a more formal, conservative register,
for example in the speech of certain actors or singers, the pronunciation [ei] is frequent.
It is also usual in certain dialects, in particular in the Kyûshû area. One should note that
long e is written as ei in hiragana (えい). This is probably why, under the hyper-corrective
influence of the writing, the pronunciation of long e as [ei] is now spreading among some
speakers of the standard language.
It seems reasonable to consider the phonetic realization [eː] as the output of an

intrinsically long vowel (see section 2.7.3, for the precise definition of what is an
intrinsically long vowel), and [ei] as a sequence of two distinct vowels, /ei/ (see section 2.8
below).
In some words of the native Yamato lexicon, for example oneesan ‘older sister’, as well as
in recent loans like me ekaa ‘maker’ {maker}, long e always remains [eː]. In such words,
the vowel is a true long vowel (intrinsic long vowel) in the sense defined below. The same
applies in principle to long e’s occurring in Yamato words, which result historically from
the loss of an intervocalic consonant, even though such long e’s are transcribed as ei in
kana, for example e i えい from ehi ‘ray (fish)’, ka rei かれい（）from karehi ‘plaice’, me i めい
( ) from mehi ‘niece’. However, one will also frequently encounter the pronunciations
[ˡei], [ˡkaɾei], and [ˡmei]. Orthoepic recommendations concerning this type of words are
vague and even sometimes contradictory.
(p.41) In cases where there exists a morphological boundary between e and i, the
sequence is pronounced as [ei] in theory, not [eː]. Me iro [ˡmeːɾo] ‘labyrinth’ is
segmentally distinct from meiro° [meiɾo] (me+iro) ‘eye colour’. However, Amanuma
et al. (1989) mention a number of exceptions, like keito° ‘wool thread’ (ke+ito) ,
realized as [keito] or [keːto], keiro° ‘hair colour’, (ke+iro) realized as [keːɾo] or
[keiɾo]. I have also heard [ɯkeːˡɾerɯ] ‘to admit’ for uke-ireru (uke+ireru).
Lastly, note that a long e, realized as [eː] (or [▱ː]) can appear in colloquial male language in
place of ai, ae, and oi. One will thus hear ne e for na i ‘there is not’, omee° for omae°
‘you’, or sugee for sugoi ‘terrible’.
2.7.2 Origin And Distribution of Long Vowels Depending On the Lexical Strata
The distribution and frequency of the long vowels vary considerably depending on the
lexical strata. Whereas each of the five long vowels occurs abundantly in Western loans,
Page 15 of 32
Vowels
for example pa atii {party} ‘party’, ko uto {coat} ‘coat’, be esu {base} ‘base’,
konpyuutaa {computer} ‘computer’, only uu,ou, and ee (ei) exist in Sino-Japanese, with
the exception of a couple of rare cases of a long i resulting from an irregular evolution
(examples (7c) below).
The case of long a [aː] in Yamato words requires specific comment in spite of its scarcity.
Long a’s surface as the result of prosodic lengthening (a process that we will examine in
section 2.7.5) or at morpheme boundaries as in the word baai° ‘case’. Baai° is
originally a Yamato compound which is generally realized as [bawai], or even [bajai], with
an epenthetic glide. A long a can also result from the opposite process, that is the
disappearance of the semi-consonant /w/ in an original awa sequence, as in mawaru° 〉
maaru° ‘to turn’ (see section 3.12) but this type of change is not reflected in the
orthography and is considered to belong to a colloquial register.5 According to Takayama
T. (2003), bawai°, which comes from baai°, must be analysed as a hypercorrection, on
the model of maaru° 〈 mawaru°.
Hamada (1986), who provides a very complete picture of the historical development of
long vowels in Japanese, also mentions a number of cases in which aa or aCa sequences
have evolved into a long o (for example mo u 〈 *(i)maa 〈 ima ‘already’ or houki° / houki 〈
hawaki 〈 hahaki ‘broom’), which would confirm the (p.42) existence of a constraint
prohibiting long a’s in Japanese. Actually, there seems to be no example of a long a
historically resulting from consonant loss in Yamato words.
In the Sino-Japanese stratum, lengthening generally represents the evolution of an Old

Chinese final nasal –ŋ (7a) or plosive (7b), which have turned into diphthongal sequences
at the stage of Middle Japanese: -au,-eu,-iu, and -ou. In this stratum, long a’s do not exist
and long i’s are rare. But lengthening in Sino-Japanese morphemes may sometimes
proceed from a specifically Japanese development which targets a single short vowel or a
vowel followed by an element on the nature of which reconstructions are dubious (7c).
Such cases always resort to readings known as kan'yôon ‘usage readings’ (see
section 1.6.2).
(7) Long vowels in Sino-Japanese words (reconstructions of the Old Chinese forms
are from Tôdô, 1996 , and Karlgren, 1957, as cited by Kano, 1998)
Modern Jp. (〈 Old Jp.) Old Chinese

(Tôdô) (Karlgren)
a.
kuu (〈 kuu) *k’uŋ *k’ung ‘void’

sou (〈 sau) *siaŋ *si̯ang ‘thought’
mei (〈 mei) *mieŋ *mi̯ěng ‘name’
b.
Page 16 of 32
Vowels
hou (〈 hahu 〈 papu) *p uǎp *pi̯wǎp ‘law’

hyou (〈 hyau 〈 pyaku) *p’ǎk *p’ǎk ‘beat’
shuu (〈 sihu 〈 sipu) *d̥ iəp *dzi̯əp ‘study, learn’
hyou (〈 heu) *p ɔg *pi̯og ‘surface’
c.
suu (〈 suu) *sïug *sli̯u ‘number’

huu (〈 huu) *p uag *pi̯wo ‘man’
shii (〈 shi) in *sied *siəd ‘four’ (in the expression ‘four seasons’)
shi iji
shii (〈 shi) in *thiəg *śiəg ‘poetry’ (in the expression ‘Chinese and
shi ika Japanese poetry’)
rou (〈 ro) in hi *g̊ lag *glâg ‘reception’
rou
In Yamato words long vowels are less common. Archaic Japanese had neither long vowels
nor diphthongs. The long vowels which occur at the phonetic level in Modern Yamato
words result from the following three processes: (i) the loss of an intervocalic consonant
(this phenomenon is known as onbin , see also section 5.4), yielding a succession of two
consecutive vowels which have then turned into a long vowel (8a); (ii) the fortuitous
juxtaposition of two vowels (p.43) across a compound word boundary (8b); (iii)
prosodic lengthening, the mechanism by which an etymologically short vowel is
lengthened in certain cases (8c).
(8) Long vowels in Yamato words
a. Long vowel resulting from consonant loss (with possible coalescence)
shirohito 〉 shirouto ‘amateur’

kinohu 〉 kinou ‘yesterday’
ahugi 〉 augi 〉 ougi ‘fan’
wakahito 〉 wakauto 〉 wakoudo ‘youngster’
sumahi 〉 sumou° ‘sumô’
kiuri 〉 kyuuri ‘cucumber’
karihito 〉 kariudo 〉 karyuudo ‘hunter’
kehu 〉 keu 〉 kyou ‘today’
kamibe 〉 kaube 〉 koube ‘Kôbe’
こう ikamu 〉 ikau 〉 ikou ‘let us go!’
ありがとう arigataku 〉 arigatau 〉 arigatou ‘thank you’
き(い) ohoki 〉 ooki(i) / ouki(i) ‘big’
Page 17 of 32
Vowels
towo 〉 to o / tou ‘ten’

いる mochiwiru 〉 mochiiru ‘to use’
nihigata 〉 niigata° ‘Niigata’
しい yasashiki 〉 yasashii° ‘gentle’
う suhu 〉 suu° ‘to smoke, to inhale’
いて kikite 〉 kiite° ‘to hear and…’
b. Sequence of two identical vowels at compound word internal boundary
mizu+umi 〉 mizuumi ‘lake (water + sea)’

sato+oya 〉 satooya° ‘foster parent (village + parent)’
ha+ari 〉 haari° ‘winged ant (wing + ant)’
c. Long vowels resulting from prosodic lengthening (see also section 2.7.5)
おさん oneesan ‘older sister’

ひい、ふう、みい hi i huu mii ‘one, two, three’
まーちゃん Ma a-chan ‘little Masaki (diminutive)’
ka a-dou (kaa-dou°) ‘Tuesday and Saturday’
Some of the long vowels in the Yamato and Sino-Japanese examples above represent
special developments due to the coalescence (or fusion) of two vowels of different quality
which triggered compensatory lengthening between the twelfth and nineteenth centuries.
Some of these evolutions, for example karihito 〉 kariudo / karyuudo, do not seem
strongly motivated from an articulatory standpoint, while others are more commonplace
from the cross-linguistic (p.44) point of view. The most frequent and systematic
coalescences are summed up in (9), with approximate dates:
(9) Long vowels resulting from compensatory lengthening
eu [eɯ] 〉 you [joː] (c. 1200–1300)

ou [oɯ] 〉 ou [oː] (c. 1200–1300)
iu [iɯ] 〉 yuu [jɯː] (c. 1400–1500)
au [aɯ] 〉 ou [oː] (c. 1700–1800)
ei [ei] 〉 ee [eː] (c. 1900)
Note, in the examples above, that palatalization developed when the first part of the
diphthong was one of the front vowels /i/ or /e/ followed by /u/.
Palatalization has also occurred in a number of older Western loans, for instance
youroppa ‘Europe’, from Portuguese {Europa}, where the sequence eu [eɯ] has
Page 18 of 32
Vowels
undergone the regular transformation you [joː]. The phonetic evolution of this word
shows in addition that the change eu 〉 you, initiated at the stage of Old Japanese,
extended over several centuries, and at least until 1500, when the first Westerners from
Portugal—who introduced the word ‘Europa’—reached Japan.
The modern writing in hiragana has maintained some of these old diphthongs (in
particular ou and ei) in the current kana script, while the others have vanished without
leaving any orthographic trace in the contemporary language. Moreover, as already
mentioned, the evolution ei 〉 ee, the most recent one, is not yet realized by all speakers
and not reflected in the writing.
2.7.3 Phonological Status And Representation of Long Vowels: Intrinsically Long Vowels Vs.
Double Vowels
For the linguist Kindaichi Haruhiko, as for a majority of Japanese phonologists, the second
part of a long vowel in words such as imouto ‘younger sister’ or kuukou° ‘airport’ is
identified as a ‘special segment’ denoted as /~/ or /R/ (sometimes also /H/) in phonemic
transcription, and as [ː] in phonetic transcriptions (see section 5.3), rather than as a
succession of two identical short vowels. Imouto ‘younger sister’ will thus be transcribed
phonemically /imoRto/ and phonetically [imoː to], kuukou° ‘airport’ /kuRkoR/ and [kɯːkoː].
Within the auto-segmental framework, an intrinsically long vowel of this sort is conceived
as one single segmental unit associated with two prosodic positions. As we shall see in
Chapter 6, where a detailed presentation of the prosodic components of Japanese and
arguments for the structural organization of the mora will be provided, the basic
Japanese rhythmic unit is conceived in this book as a maximally binary structure
corresponding to a mora. So a word such as kou° [koː] ‘this manner’ corresponds to two
rhythmic units (two moras), as shown in (10).
(p.45) (10) Intrinsically long vowel: こう kou [koː]
However, some of the phonetically long vowels of Japanese must be analysed

phonemically as a succession of two distinct segments with identical phonetic quality
(double vowels) rather than as a single vocalic segment associated with two prosodic
positions. The relevant representation appears in (11). Such vowels do not structurally
differ from the sequences of two different vowels presented further in section 2.8.
Simply, the two vowels in question are of identical quality.
(11) Succession of two identical vowels (double vowels): こお koo /koo/ realized as
[koː] or [koo]
Page 19 of 32
Vowels
Such representations allow us to better capture the phonological difference existing

between pairs such as the following which are familiar textbook examples:
(12)
a.
satou-ya° (satôya) [satoːja] *[satoʔoja] ‘sugar shop’

sato-oya° [satoːja] or [satoʔoja] ‘foster parent’
b.
su uri (sûri) [¹sɯːɾi] *[¹sɯʔɯɾi] ‘mathematical theory’

りsu uri (suuri) [¹sɯːɾi] or [¹sɯʔɯɾi] ‘vinegar seller’
The difference between the two items of a pair constitutes an issue much debated in the
domain of Japanese phonology (Kindaichi 1950; Hattori, 1955, 1961; Hamada, 1951, etc.).
In (12a), the first word comprises something which can be analysed as an intrinsically long
vowel, the second one a succession of two identical vowels. Although such sequences are
generally realized in an identical way in normal speech, a phonetic difference between the
two members of each pair may appear in slow or formal speech, that is, it is possible to
have a hiatus, materialized in the form of a pause or a light glottal stop [ʔ], between the
first and the second element of a double vowel, but not between the two parts of an
(p.46) intrinsically long vowel.6 The representations in (10) and (11) account for this
difference in a natural manner: in (11), there are two distinct segments, but only one in
(10).
The sequence uu in su uri ‘vinegar seller’, whose phonetically long vowel results from the
succession of two short u’s, separated by a morphological boundary (su ‘vinegar’ + uri°
‘salesman’), belongs to the second type as illustrated in (11), that of two-vowel
sequences, whereas su uri ‘mathematical principle’ (su u ‘mathematics’ + ri ‘principle’),
which has only one (phonetically and phonemically) long vowel, pertains to the type
represented in (10). The phonological representations of these two words will thus be as
follows:
(13) su uri /suRri/ ‘mathematical principle’ (a) vs. suuri /suuri/ ‘vinegar seller’ (b)
The following pairs are also to be structurally distinguished:
(14)
Page 20 of 32
Vowels
koou° [kooː] or [koʔoː] ‘acting in concert’

ko uo [¹koːo] or [¹koːʔo] ‘likes and dislikes’
They correspond to the representations in (15):
(15) koou° /kooR/ ‘acting in concert’ vs. kouo /koRo/ ‘likes and dislikes’
(p.47)
The vowels ou and uu of Sino-Japanese words seen above in (7), the majority of the
vowels in (8a) which result from vowel coalescence (sometimes caused by the loss of an
intervocalic consonant) with subsequent compensatory lengthening and in some cases
modification of the vocalic quality, as well as the vowels which result from prosodic
lengthening (8c), are to be analysed as intrinsically long vowels, except in cases where the
vowel plays a flexional role as the final i in yasashii° (‘gentle’), the final u in suu° (‘to
smoke’), and the second i in kiite° (‘to hear and…’, cited in 8a).
On the other hand, the vowels shown in (8b) correspond to double vowels, i.e. to a
sequence of two identical vowel nuclei, in which a hiatus is possible.
When a long vowel results from the loss of a consonant surrounded by two identical
vocalic segments, like kohori 〉 koori° ‘ice’ or ohoki 〉 ooki ‘big’, things are more
ambiguous, as the kana spelling reflects. In such words, long o is written as oo おお and
not ou おう. The problem is comparable to that of the long e written as ei, already
discussed. It would be rash to assume that all speakers of Japanese have exactly the
same phonological representations for the long vowels which occur in these types of
words. It is probable that for certain speakers, the [oː] of koori° ‘ice’ from kohori (a
Yamato lexeme) and that of ko uri ‘axiom’ (a Sino-Japanese lexeme) have the same
phonological representation today, with an intrinsically long vowel, while they have a
different representation for other speakers, koori° having a double vowel under the
influence of the spelling. It is necessary to keep in mind that in non-Western words, all the
intrinsically long vowels of the modern language derive from double vowels (following the
evolutionary pattern VCV 〉 VV 〉 Vː as in ahugi 〉 augi 〉 ougi ‘fan’, or VC 〉 VV 〉 Vː in
certain Sino-Japanese words, for instance tong 〉 tou ‘East’), except for the case of
prosodic lengthenings. The natural fate of a double vowel is thus to turn into an
intrinsically long vowel, and the moment of this transformation naturally varies according
to words or individual speakers.
The distinction between an intrinsically long vowel and a double vowel is further justified
by the fact that the latter has a property of which the former is deprived: in double
vowels, the second element can carry an accent, for example, mizuumi ‘lake’, tookereba
Page 21 of 32
Vowels
‘if it is far’, ooi ‘many’, whereas intrinsically long vowels can only bear an accent on the
first part of the vowel.
2.7.4 Prosodic Shortening

A long vowel is frequently reduced to a simple vowel, as in the following examples:
koukou° 〉 kouko° ‘Japanese pickles’, hontou° 〉 honto° ‘true’, konpyuutaa 〉 konpyuuta
‘computer’ {computer}, akanbou° 〉 akanbo° ‘baby’, daijoubu 〉 daijobu ‘all right’, sensei
〉 sense ‘teacher, master’. According to Sukegawa et al. (1999), about one third of all long
vowels undergo this type of (p.48) reduction in spontaneous speech. The phenomenon,
known as ‘prosodic shortening’, occurs primarily in Western loans, and to a lesser extent
in Sino-Japanese words. It is especially common in a non-formal register, and concerns
mainly medial or final vowels in moras which follow another long vowel or a mora nasal as
in kouko° ‘salted white radish’ or honto° ‘true’, respectively derived from koukou° and
hontou° (KKK, 2004). This is the reason why, according to Takeuchi (1999:46), words
such as ryokou° ‘voyage’, or burezaa ‘jacket’, do not undergo prosodic shortening,
whereas hontou° 〉 honto° ‘true’ or konpyuutaa 〉 konpyuuta {computer} are among the
words most frequently shortened. The study by Sukegawa et al. (1999) also establishes
that vowel shortening is rare in word-initial position, including words which can be
regarded as monosyllabic, for example mo u (mou°) ‘already’.
Vowel quality also plays a role with regard to prosodic shortening, but its relevance can
mainly be observed in Western loans, since this is the only stratum which allows all five
long vowels of the Japanese system to occur freely and frequently in any word position.
In Western abbreviations, final prosodic shortening more frequently targets the [−high]
final vowels /a/, /e/, and /o/ than the [+high] vowels /i/ and /u/, as shown by Figure 2.3
(Labrune, 2007).
In 80% of the cases at hand (20 words out of 25), final length tends to be maintained after
/i/ and /u/ word-finally, whereas it is deleted in more than 60% of the cases after /e/, /o/,
and /a/ (58 words out of 88). For instance, kanningu peepaa {cunning paper} ‘cheat
sheet’ becomes ka n-pe / kan‐pe°, with final shortening, rather than *kan‐pee, while bataa
piinatsu {butter peanuts} is clipped as bata-pii°, rather than *bata-pi.
Figure 2.3. Final vowel shortening in Western clippings
(p.49) 2.7.5 Prosodic Lengthening
Page 22 of 32
Vowels
A number of micro-paradigms of the Yamato stratum have long vowels which do not
result from the historical evolutions of the type VCV 〉 VV and/or VV 〉 Vː as described
above. Such long vowels are prosodically derived from a short vowel by secondary
lengthening. The paradigms in question generally have emotional, impressionistic, or
emphatic connotations. They can be divided in three subtypes. The first type includes
family terms based on vocatives, hypocoristics, baby talk, and manner deictics (for
instance okaasan ‘mama’, yuu-chan ‘little Yuko’, kou ‘this manner’); the second type
contains interjections and mimetic adverbs or interrogatives (zuutto ‘continuously’, naani
‘what’); the third group consists of bases which are all one-mora long, and generally
belong to nominal or numerical series (nii ‘two’, kaa ‘Tuesday’). They are described
below in detail for the sake of completeness, but the reader who is not a specialist in
Japanese can skip this section and come back to it at a later stage.
Type 1
A. Kinship terms vocatives:
ka a-san ‘mum’, to o-san ‘dad’, ji i-san ‘grandpa’, ba a-san ‘grandma’, ni i-san ‘older
brother’, ne e-san ‘older sister’, respectively built on the roots *ka,*to,*ji,*ba (which are
found in the reduplicated forms kaka ‘mum’, toto ‘dad’, jiji ‘grandfather’, baba
‘grandmother’), as well as on a ni ‘older brother’ and ane° ‘older sister’.
Let us also mention imouto ‘younger sister’, otouto ‘younger brother’, shuuto° /
shuutome° ‘father-/ mother-in-law’ in which the presence of a long vowel seems to result
from regular evolution of the older forms *imohito, *otohito, *shihito (*imohito 〉 *imouto
〉 imo:to, *otohito 〉 *otouto 〉 oto:to, *shihito 〉 *shiuto 〉 shuuto°), and thus pertains, at
first sight, to the evolution V 1 hV 2 〉 V1V2 〉 Vː rather than to prosodic lengthening.
However, as noted by Hamada (1951), if things had really occurred that way, one would
expect the t of hito to be voiced, that is, one would have had *imobito 〉 *imoudo, *otobito
〉 *otoudo, *shibito 〉 *shuudo, as is the case in other derived terms containing the
element hito, for example akindo,shiryuudo (akihito 〉 akindo ‘trader’, shirihito 〉
shiryuudo ‘acquaintance’). However, since voicing has not occurred in the imouto type
set of words, they must have followed some other evolution pattern. These facts suggest
that analogy may have played a role within this paradigm of kinship terms, and that the
long vowel of imouto,otouto and shuuto° represents the same type of element as that of
to usan, ka asan, etc., rather than the direct result of a straightforward phonetic
evolution from *imohito, *otohito, *shihito.
(p.50) The word hii-mago° (or hii-mago) ‘great-grandson /-daughter’, in which the
vowel of the morpheme hi is etymologically short is also noteworthy. Lengthening in hii
would thus be explained by the fact that the morpheme hi refers to a kinship term
(Wenck, 1966: footnote 26).
Finally, note that the forms jijii ‘grandpa’ and babaa ‘grandma’, with final lengthening, are
vulgar and depreciative (Kubozono, 2000).
B. Hypocoristic of personal names (Poser, 1990):
Page 23 of 32
Vowels
in hypocoristic derivatives such as Yu u-chan (based on the first name Yu mi), or Ma a-

chan (for Ma saki), secondary lengthening of the originally short vowel of the base
occurs in a very regular manner.
C. Baby talk:
one encounters in one variety of obsolete baby talk7 (reported by Hamada, 1951) forms
such as be ebee / be ebe / bebe ‘clothing’ (origin unknown), ha ahaa / haaha ‘tooth’ (from
ha ‘tooth’), hi ihii / hiihi ‘fire’ (from hi ‘fire’). They are all characterized by reduplication
and the presence of one or two long vowels derived from the corresponding short vowel.
D. Manner deictics:
this paradigm contains the following four forms: kou° ‘like that’ (1st degree, near to the
speaker), sou° ‘like that’ (2nd degree, near to the listener)’, aa° ‘like that’ (3rd degree,
far from the speaker and listener), do u ‘how, in which manner?’. According to Hamada
(1951) and a majority of dictionaries, the manner deictic kou° has been derived from ka
ku く (ka ‘3rd degree deictic’ + ku ‘adverbial ending’), according to the following
evolution: ka ku 〉 kau 〉 kou°. This form kou° would then have been reinterpreted as
derived from ko, the 1st degree deictic morpheme which one finds in kore° ‘this’, koko°
‘here’. Later, under the pressure of analogy, the forms sou°,aa°, and do u,8 would have
developed, by means of prosodic lengthening out of the bases so,a,do. Note that the
bases ko,so-,a-, and do- are all monomoraic, a characteristic which no doubt also played a
role in the development of the lengthened series, since, as we will see below, one type of
prosodic lengthening precisely affects monomoraic morphemes.
(p.51) The four series of forms which have just been described do not concern
isolated, single lexemes but rather paradigms whose paradigmatic coherence is precisely
marked by the presence of a long vowel derived from a short one. We are thus dealing
with a rather systematic usage of vocalic lengthening, which may be analysed as an
instance of template morphology.
Type 2
The second type pertains to expressive and emphatic lengthening occurring in
interjections and mimetics, and in a number of interrogative words. The corresponding
forms with short vowels almost always exist. We thus find:
A. Interjections:
for example o ui [ oːi] ‘hey!’, eetto° ‘well’, ne e ‘listen!’.
B. Onomatopoeia and ideophones:
za azaa ‘water pouring heavily’, gee° (expression of dislike), huutto° (puffing sound or
action).
C. Adverbs and interrogatives with emphatic connotation:
zuutto° ‘continuously’, yo uku ‘often, well’, na ani ‘what?’. The long vowels which are
Page 24 of 32
Vowels
phonetically realized in these types of words are only seldom denoted in the conventional
orthography.
Type 3
Lastly, there exists a type of prosodic lengthening of one-mora-long bases, most of which
pertain to enumerative series or lists: days of the week, Sino-Japanese or Yamato
numbers, animals of the Chinese zodiac, as well as reduplicate verbal radicals. It is to be
noted that all these forms are accented on the initial mora.
A. Days of the week:
the expression for ‘Tuesday and Saturday’, a compound made of the first two
monomoraic morphemes of ka-youbi ‘Tuesday’ and do-youbi ‘Saturday’ is realized ka a
dou (or kaa-dou°). Only ‘Tuesday’ and ‘Saturday’ are concerned with prosodic
lengthening, since they are the only weekday bases which are monomoraic.
B. Sino-Japanese numbers in enumeration:
Sino-Japanese numbers present the following forms when they are recited (underlined
forms mark secondary lengthenings): ichi ‘1’, ni i ‘2’, san° ‘3’, shi i ‘4’, go u ‘5’, roku ‘6’,
shichi ‘7’, hachi ‘8’, kyu u ‘9’, ju u ‘10’. (Compare with ichi ‘1’, ni ‘2’, san° ‘3’, shi ‘4’, go
‘5’, roku ‘6’, shichi ‘7’, hachi ‘8’, kyu u ‘9’, ju u ‘10’, the basic forms which are found
elsewhere than in enumeration). The series 4–5–6–2–1 will be thus pronounced shii°-go
u-roku-ni i-ichi. The lengthening in 2, 4, and 5 is never denoted in the kana writing. On
the other hand, that of 9 and 10 is etymological, and does not constitute a case of
prosodic lengthening.
(p.52) C. Yamato numbers:
the same type of phenomenon occurs with the following numbers of Yamato origin
(Komatsu, 1981:115): hi i ‘1’, hu u ‘2’, mi i ‘3’, yo u ‘4’, i i ‘5’, mu u ‘6’, na a ‘7’, ya a ‘8’.9
Compare with the full forms of the more widespread series: hitotsu ‘1’, hutatsu ‘2’,
mittsu ‘3’, yottsu ‘4’, itsutsu ‘5’, muttsu ‘6’, nanatsu ‘7’, yattsu ‘8’.
D. Abbreviated zodiac signs:
ne° ‘rat’, u° ‘rabbit’, mi° ‘snake’, and i° ‘pig, wild boar’ are pronounced respectively ne
e, u u, mi i, and ii° in the enumeration of the twelve Chinese zodiac signs (ne e ushi tora
u u tatsu mi i uma hitsuji° saru tori° inu ii°, ‘Rat, Ox, Tiger, Rabbit, Dragon, Snake,
Horse, Ram, Monkey, Rooster, Dog, Pig’).
E. Reduplicated verb bases:
the vowel of a monomoraic verb base lengthens when it undergoes reduplication in

order to form an adverbial expression indicating simultaneity (Kageyama, 1976/1977;
Martin, 1987; Itô, 1990): mi ru ‘to look’ 〉 mii mii ‘while looking’, suru° ‘to make’ 〉 shii
shii° ‘while making’. Such lengthening never occurs with bimoraic verbal bases.
Page 25 of 32
Vowels
There also exists a small number of other isolated cases of prosodic lengthening, such as
the word se e (written せ se in kana)10 ‘height (of a person)’. Lengthening further
occurs sometimes in familiar speech following the dropping of an enclitic particle after a
monomoraic word, for example hi (wo) totte kure ‘bring me fire!’, which is actually
realized as [ˡhiːˡtottekɯɾe] or [ˡçiːˡtottekɯɾe] (Hayata, 1980). Kubozono (2000) also
reports some other marginal cases in slang or in the secret language zûjâ-go, used by
jazz and rock musicians. In Kansai dialects, prosodic lengthening of monomoraic forms is
systematic: thus we have [hiː] (with a low-high accent) ‘fire’ (standard hi), [ʨ͡iː] (with a
high-high accent) ‘blood’ (standard chi°), etc. even when the word is followed by a
particle, in which case lengthening is merely optional.
It is noteworthy that in all the above-mentioned series, any of the five vowels of Japanese
can undergo lengthening, including the vowel a, of which no other occurrence as a true
long vowel can be found in a Yamato or Sino-Japanese morpheme (except the cases
discussed in section 2.7.2).
(p.53) 2.8 Sequences of Two Different Vowels: the Problem of ‘diphthongs’

Sequences of two different vowels are very frequent in Modern Japanese.11 All the
possible combinations between the five vocalic elements are attested, namely iu, ie, io, ia,
ui, ue, uo, ua, ei, eu, eo, ea, oi, ou, oe, oa, ai, au, ae, ao. Each vowel in these sequences
represents one mora, and is distinctly articulated. However, as noted by Takayama T.
(2003), apart from the V 1 V 2 sequences in which a morpheme boundary separates the
two vowels, existing combinations in modern Yamato or Sino-Japanese words are limited
to a/i/u/o + i/e/o, namely ai,oi,ui,ie,ae,oe,ue,io,ao,uo (ei occurs but it generally
corresponds to a long e, see section 2.7.1). The groups Vu and Va are unattested, for
various reasons. The Vu sequences have all evolved into long vowels, palatalized or not:
iu 〉 yu:, eu 〉 yo:, au 〉 o:, ou 〉 o: (see section 2.7.2), except at morpheme boundary. We
thus find ka-u° ‘to buy’, omo-u ‘to think’, and so on. On the other hand, the Va sequences
never existed in Yamato or Sino-Japanese monomorphemes, except for a few cases of
prosodic lengthening. It is interesting to note, following Takayama T. (2003), that even in
Western loanwords or at morpheme boundary, where Va is likely to occur, a transition
glide is frequently inserted: thus shiai° [ɕ ja ] ‘match’, takuan [taˡkɯʷaN] ‘salted radish’,
itaria° [itaɾiʲa] ‘Italy’. This suggests that Va sequences (ia,ea,oa,ua as well as aa) are
problematic (on aa / a: see also section 2.7.2).
Sequences of three or four vowels, or even more, are easy to find: for example in the
place name aioi , the adjective aoi ‘blue’, the noun ie-ie ‘houses’. We could thus build a
sentence made up uniquely of vowels such as aioi e aa iuu aoi uo o ou ( へああいういをう)
‘to follow such a blue fish towards Aioi’.
The term diphthong (nijûboin or jûboin ) is frequently used in the phonetic

descriptions of Japanese. However, the definition of what the various scholars mean by
‘diphthong’ is not always precise. Amanuma et al. (1989:93–98) operates a distinction
between renboin ‘vowels in succession’ and jûboin ‘diphthong’, and considers
that certain V + i sequences can be qualified as diphthongs (jûboin) in the traditional
sense of the word. However, the evidence for such a distinction seems rather
Page 26 of 32
Vowels
morphological in nature since Vi is regarded as a ‘diphthong’ when there is no

morphological boundary between V and i, as in the word ka i ‘shell’. Kawakami (1977)
considers (p.54) the sequences ai,ui,oi,ae,ao, and oe as diphthongs (nijûboin) while
specifying that one cannot always speak of a true diphthong when a clear morphemic
boundary separates the two vowels. According to Saitô (1997:85), the sequences
ai,ui,ae,au easily turn into the ‘diphthongs’ ai̯,ui̯,ae̯, au̯ except when they are separated
by a semantic boundary (ha isha, ha +isha° ‘dentist’), and except in bimoraic Yamato
words: oi° ‘nephew’, ei ‘ray (fish)’, ou° ‘to follow’. However, in an experiment I
undertook in 2001 with the help of Jôo Hakutarô, the forms ha isha (ha-isha) ‘dentist’
and ha isha (hai-sha) ‘loser’ appeared to have the same phonetic realization, although
a morpho-semantic boundary between a and i exists in the first word but not in the
latter.
The question of whether the existence of a morpheme boundary between the two vowels
of a sequence is relevant is also evoked by Hashimoto M. (1977:27). According to him,
the pairs ka i ‘a meeting’ (one morpheme) and ka i ‘lower rank’ (two morphemes
ka+i), or ku i ‘regret’ (one morpheme) and ku i ‘meaning of a phrase’ (two
morphemes ku+i) are perceived as different in certain dialects. However, this does not
prove that there is a truly phonetic or even phonological difference between them. It is
quite possible that speakers’ perceptions are heavily influenced by writing or
morphological structure. Moreover, accent might also be involved, since there exist
dialects which can operate an accentual distinction between CVi pairs, for instance
between the pairs kai ‘shell’ and ka i ‘a paddle’ (Izu dialect, Uwano, 2003). More
experimental studies are thus necessary in order to clarify this point. It is nonetheless a
major issue because it is closely related to the problem of the interaction between mora
and syllable in Japanese. Indeed, if there were any evidence that two phonetically and
phonemically different sorts of V 1 V 2 sequences exist in Standard Japanese (for example a
difference between ka i and ka i , as already mentioned), one could then argue that
the first type is monosyllabic while the second is dissyllabic. This would imply that both the
mora and the syllable are distinct and relevant units in Japanese, an assumption which is
challenged in this book, as we shall see in Chapter 6.
I consider that it is mistaken to regard V 1 V 2 sequences as diphthongs since I know of no

phonetic or phonological evidence which would prove that they are. They are simply a
succession of two distinct vocalic nuclei, each with its own prosodic weight representing
one mora. A number of arguments can be brought forward in order to justify this
analysis. First, the quality of the vowels involved in such sequences is not notably
different from what it is when they are realized in isolation. There is no significant gradual
change of the quality of the first vowel towards the second one, no crescendo nor
decrescendo between V 1 and V 2, contrary to what generally occurs with diphthongs in
other languages. Second, in slow or careful speech, a pause—sometimes even a glottal
stop—is frequently inserted between the two vowels, something which would be
impossible in the (p.55) case of true diphthongs like those of English or Chinese, for
example. The group [ai] in the English word sky cannot be articulated in two syllables, i.e.
ska.i, whereas the Japanese words ue° ‘above’, a i ‘love’, uo° ‘fish’ respectively [ɯ.e],
Page 27 of 32
Vowels
[ˡa.i], and [ɯ.o] are normally uttered in two beats. Moreover, in such groups, the second
vowel can carry an accent, for example kai-nushi ‘owner (of a pet)’, even though it is true
that a tendency to de-accent vowels which are not preceded by an onset consonant can
be observed,12 as the possible realization of kai‐nushi as ka i‐nushi attests. Lastly, in
Japanese, no sequence of two vowels results from the diphthongization of a
monophthong. There is no example of an evolution V(:) 〉 V 1 V 2 like the ones so frequently
encountered in Romance or Germanic phonology, for instance Latin bene [beːne] 〉
French bien [bj▱̃ ]; old English nama [næːmə] 〉 modern English name [n▱I‸m]), and so
forth.
It is clear that the recognition of true diphthongs in Japanese is determined by one’s

conception and definition of the prosodic units of the language. If one assumes a priori
that syllables exist in Japanese and that syllabic weight distinctions are relevant, one is
rather easily (too easily?) led to talk about ‘diphthongs’, even though there does not exist
any real evidence for their existence. But the assumption that diphthongs are distinct
phonological entities should in turn imply a re-examination of all the vocalic Japanese
sequences because these are not limited to Vi sequences, the elements most frequently
regarded as ‘diphthongs’: one finds for example ae in hae° ‘fly’, oe in ko e ‘voice’, au in
mausu ‘(computer) mouse’, ao in kao° ‘face’. It thus becomes necessary to inquire
whether these vowel sequences belong to the same prosodic unit (to the same ‘syllable’,
if one believes in the existence of the syllable in Japanese) or not. However, very few
authors are cautious enough to consider such cases and to set forth a thorough analysis
of them within the syllabic framework. I am not aware of any strong internal or external
evidence for operating a structural distinction between hai° ‘ash’ and hae° ‘fly’, or
between ka i ‘shell’ and kao° ‘face’.
Curiously enough, there actually exists in Japanese one case for which the label
‘diphthong’ could be correctly used. I have in mind the sequences transcribed as ya, yu,
yo (with short or long vowels) occurring after a consonant in words like hyaku
‘hundred’, ryokou° ‘trip’, or nyu usu ニュース {news} ‘information’, known as yôon ( ) in
the traditional terminology. However, these elements are generally not categorized as
‘diphthongs’ by Japanese (p.56) scholars, even by phoneticians, whereas ai,ui,oi etc.
generally are. Yet, ya,yu,yo present all the properties of phonetic diphthongs as generally
defined: the palatal semi-consonant which occurs in first position can colour the quality of
the following vowel (for example /u/ frequently becomes [ɯ̈]); the CyV groups count as
only one mora, not two; diaeresis is impossible; finally, strict phonotactic rules exist,
which limit the possibilities of combinations between y and the following vowel. Only ya,yu,
and yo are possible. The yi group is unattested, and ye is rare (see section 3.11).
From the point of view of their phonetic realization, these ya, yu, and yo groups could
thus belong to the category of diphthongs in the classical sense of the word. Yet, as we
shall see in 3.11, there is benefit in treating them as palatalizations of the preceding
consonant at the phonological level. The difference simply consists in considering that the
segment y is associated with the consonant (onset), not with the vowel (nucleus). The
adequate phonological representation of such Japanese entities thus corresponds to a
Page 28 of 32
Vowels
palatalized consonant as represented in (16), rather than to a rising diphthong as in (17).
(16) Palatalized consonant (does exist in Japanese)
(17) Rising diphthong (does not exist in Japanese)
In conclusion, there is no real need to talk about phonological diphthongs in Japanese. In

Chapter 6, a theory of the Japanese prosodic unit (the mora) will be further developed,
which considers that there exist only two structural positions in a rhythmic unit: an onset
(C) and a nucleus (V). Only the C position can be the locus of palatalization. It results from
this configuration that, in a phonetic sequence VV, each of the two vowels must be
regarded as pertaining to two different rhythmic units.
(p.57) 2.9 Relative Frequency of Vowels

The vowels of Japanese do not all have the same frequency in the language. It is thus
interesting to examine some quantitative information relative to this question. I shall first
present some statistical data for the archaic language, then some for the modern
language. Both concern token frequency.
For Archaic Yamato Japanese (Nara), which possessed eight vowels, the data are
provided in Figure 2.4 (based on Ôno, 1980). It can be seen that /a/ exhibits the highest
frequency before /i/, /u/, /ö/, and /o/. The vowel /e/ has a relatively low
Figure 2.4. Token frequency of vowels in Archaic Japanese (based

on the data by Ôno, 1980)
Page 29 of 32
Vowels
Figure 2.5. Token frequency of vowels in Modern Japanese (based

on Campbell, 1999)
(p.58) frequency, while /ë/ and /ï/ are extremely rare. By adding the values of each
member of the pairs which have merged after the archaic period, we get the following
classification: /a/ is still at the top, but /o/ now comes second, before /i/ and /u/: a 〉 o (ö +
o) 〉 i (ï + i) 〉 u 〉 e (ë + e).
As for modern Japanese, the data provided in Figure 2.5 are adapted from Campbell
(1999). In contrast with Archaic Japanese, /o/ ranks first, almost equally with /a/, before /i/
then /u/; as in the archaic language, /e/ is still the least frequent of all.
Notes:
(1 ) Note that the lips are not much used in the articulation of Japanese, be it for vowels or
consonants. On this issue, see section 3.12.
(2) Some rare cases in which a surrounding vowel has been copied, rather than /i/ or /u/
inserted, exist in a couple of very old Sino-Japanese loans, for example ha kase
‘doctor’ (also ha kushi). The character , read here haka, is normally pronounced haku.
(3) On the h / p alternation after /Q/, see section 3.7.1.
(4) Recall that vowel length is transcribed by doubling the vowel in the cases of a,i, and u
(aa,ii,uu), and by use of the letter u in the case of o (ou), and by i in the case of e (ei),
except in some special cases like to oru ‘to go through’, oneesan ‘older sister’, etc. where
the transcription remains faithful to the kana spelling (とおる and おねえさん).
(5) These examples seem to illustrate the existence of a conflict between two
contradictory tendencies in Japanese: the first tendency militates for the deletion of w
before a, and can be regarded as the result of a diachronic process of /w/ weakening,
which first affected w before i and o, and is nowadays achieved. The w weakening is
presently affecting /w/ before a (see section 3.12); the second tendency banishes long a’s
in Yamato and Sino-Japanese words. When the first tendency is stronger, w disappears,
when it is the second, w is maintained.
(6) In contrast with Kindaichi, Hattori (1955, 1960, 1961) assumes the presence of a zero
consonant /’/ to distinguish long vowels resulting from the succession of two vowels
separated by a morphological boundary (in our terminology, double vowels) from those
which constitute intrinsically long vowels. According to him, the latter represents a
Page 30 of 32
Vowels
succession of two vowels with no consonant inbetween. Hattori adopts the notation /V’V/
for the former, and /VV/ for the latter, for example /suuri/ ‘mathematical theory’ and
/su’uri/ ‘vinegar seller’. The zero consonant of Hattori can be reinterpreted as an empty
onset.
(7) Japanese baby words are characterized by the following prosodic patterns, all with
initial accent on the first mora of the stem:
MmM (HL in syllabic terms), for instance o nbu ‘piggyback’, ne nne ‘sleep’, a n’yo ‘walk’.
MmMm (HH in syllabic terms), as in o ppai ‘breast’, ha ihai ‘crawl’, po npon ‘belly’.
Finally, a few forms are made by reduplication of a monomoraic stem and o-prefixation,
for instance o-tete ‘hand’, o-meme ‘eye’.
(8) According to Yanagida (1991), aa° is first attested in texts of the end of the Edo
period (nineteenth century), after kou° and sou°, while the first attestation of dou is
older, dating back to 1527.
(9) ‘9’ is pronounced ko no in recitation. It is the only numeral which does not have a
form with a long vowel. The long vowel in to o ‘10’ is etymological, and does not result
from prosodic lengthening (towo 〉 too). Moreover, in this series, ‘4’, ‘5’, and ‘7’ also have
the alternant forms yo n,itsu, and na na.
(10) This word also admits the phonetic form [¹seː] or [¹sei], written se i せい in kana, for
instance in the expression sei-kurabe ‘height comparison’.
(11 ) The situation was very different in Nara Japanese, since the rhythmic units (syllables
or moras) were all of the form V or CV. Onsetless vowels were not allowed except
morpheme‐initially.
(12) However, note that this type of deaccentuation concerns first and foremost the high
vowels /i/ and /u/, and that it also appears under certain conditions when these vowels are
preceded by a consonant (Tanaka, 1998).
Consonants

Laurence Labrune

Print ISBN-13: 9780199545834
DOI: 10.1093/acprof:oso/9780199545834.001.0001
Consonants
Laurence Labrune
DOI:10.1093/acprof:oso/9780199545834.003.0003
This chapter examines in turn each consonant of the Japanese system, presenting its main
allophones and phonotactic characteristics. Special attention is dedicated to the phonology
of the consonant /h/ which has developed out of*/p/, and to the issue of the velar nasal,
through an enlightening comparison of two competing approaches of the problem, that of
Kindaichi Haruhiko (1942) and that of Junko Itô and Armin Mester (1997). This chapter
also offers new insights on the phonology of the Japanese /r/ and on newly introduced
consonants.
Keywords: allophones, phonotactic characteristics, velar nasal, Kindaichi Haruhiko, Junko Itô, Armin
Mester, new consonants
3.1 General Characteristics of the Consonant System

Two major features characterize the consonant system of modern Japanese. First, the
number and phonotactic possibilities of the consonants differ considerably according to
Page 1 of 41
Consonants
the lexical strata. Second, many of the consonants display a high number of allophones.
While none of these characteristics is, by itself, specific to Japanese, it seems that in this
language, the extent of the phenomenon is without common measure with what one
currently observes in most other languages.
Under these conditions, it is somewhat difficult to present a synoptic table of the system
of the Japanese consonants. In Table 3.1 that I propose here, brackets indicate
consonants whose phonemic status is debatable, while braces mark consonants which
occur only in the most recent Western loans.
The two moraic contoids /N/ (the mora nasal) and /Q/ (the first part of an obstruent
geminate), which are generally referred to as ‘special phonemes’ or ‘special segments’ in
the Japanese tradition, may be added to this list as two additional consonant phonemes.
Since their phonology is particular because of their moraic status, they will be dealt with
in detail in Chapter 5.
Consonant allophony occurs mainly before the high vowels /i/ and /u/. It involves
particularly the consonants /t/, /d/, /s/, /z/, and /h/. This propensity to allophony is also
characteristic of the special moraic segments /N/ and /Q/, and, to a lesser extent, of /r/
and /k/. Such allophonies are the source of a number of neutralization processes and also
the origin of several phonemic splits, as we will
Table 3.1. Consonantal phonemes of Japanese

Labials Alveolars Palatals Velars Glottal
Plosives p b t d k g
Fricatives {ϕ β} s z (ɕ, ʑ) h
Affricates {ʦ͡ } (ʨ͡)
Nasals m n (ŋ)
Glides y [j] w
Liquid r
(p.60) see in the next pages. The possibility for such rather important variations in the
articulatory realizations of the Japanese consonants is probably due to the relatively
modest size of the inventory.
Assimilations of a consonant to the following vowel, that is, if one thinks in terms of
prosodic constituents, assimilations of an onset to the nucleus (for example the fact that /t/
undergoes affrication before /u/: [ʦ͡ ɯ]), is a manifestation of the very tight link which
stands between a consonant and the following vowel in Japanese. One can subscribe
entirely to the formulation by Daniels (1958:58, quoted by Coleman, 1998:268), according
to whom Japanese ‘consonants are said to be “prefixed” to vowels rather than to
‘precede’ them because…in Japanese it is necessary to put the speech organs into the
position of the vowel…before producing the consonant’. This type of co-articulation is of
course attested in all languages, but it would seem that Japanese is rather extreme in this
Page 2 of 41
Consonants
respect.
In the following pages, we will successively review all the consonants of Japanese, in the
following order: plosives (p, b, t, d, k, g), fricatives and affricates (s, z, ɕ, ʑ, tɕ, ts, h), nasals
(m, n, ŋ), glides (y, w), the liquid (ɾ), and finally consonants which have recently been
introduced into the language due to the influence of loanwords.
3.2 /P/ And /b/

The bilabial plosives /p/ and /b/ are realized as [p] and [b].
The voiced plosive /b/ calls for no special comment, except for those which apply more
globally to voiced obstruents, to which Chapter 4 is devoted. Let us merely observe that
in Modern Japanese, /b/ functions morpho-phonologically as the voiced counterpart of /h/,
for example under the application of rendaku (see section 4.2). H thus becomes b, as in
yama + ha to 〉 yama-bato° ‘wild pigeon’.
The voiceless bilabial plosive /p/, on the other hand, is remarkable in a number of
respects. It is currently the least frequent of all Japanese consonants (see the data in
section 3.15). It also displays a limited distribution in words of the Yamato and Sino-
Japanese strata. In non-mimetic and non-expressive Yamato words, /p/ generally occurs
only as an alternant of /h/ or /b/, and almost solely under the geminate form [pp]. Singleton
p is found only after the mora nasal /N/ in a limited number of examples. The geminate pp
occurs almost exclusively at morphemic boundaries in Yamato lexemes: mapputatsu (ma-
+ hutatsu) ‘(to separate) in two parts’, hipparu (hiki° + haru°) ‘to draw’, kodomoppoi
(kodomo° + -poi) ‘childish’, hitorippochi (hitori + -bochi/pochi) ‘all alone’, ho ppe
(probably *ho, ‘cheek’ + expressive suffix -pe) ‘cheek’, ta npopo ‘dandelion’ (the
etymology of (p.61) this word is dubious, perhaps from tanpo ‘swell, bundle’ + -po, a
suffix of obscure value). Thus /p/ never appears at the beginning of independent words.
It is necessary to consider separately the case of mimetics and of familiar expressive

words, including words of Chinese origin which are no longer perceived as such, where
/p/ can appear word-initially. For example puka-puka / pu ka-puka ‘floating’, po npon ‘belly
(child language)’, po chi (a common dog name), pakuru ‘to filch’, peten° ‘head’ or ‘fraud’
(reverse form of Sino-Japanese teppen). It is generally considered that, in the native
words of this type, the ancient bilabial plosive which has otherwise evolved into /h/ (see
section 3.7) has been maintained in the initial because of its expressive value.
In mimetic roots, except for some rare exceptions, /p/ always occurs as a geminate stem
internally. It functions very regularly as the reinforced form of /h/, for example in yappari
‘indeed’, from yahari, or even as that of /b/ in certain cases: sa ba-saba 〉 sappari
(describes a straightforward personality) (Hamano, 1998:195ff.).
In Sino-Japanese words, /p/ exists only as an alternant of /h/ after /Q/ and /N/, as in the
already mentioned examples (see also section 3.7): shuppatsu° (shutsu + hatsu)
‘departure’, or ke npou ‘constitution’ (ken + hou). However, /h/ does not
automatically turn into /p/ when the compound contains more than two morphemes, in
Page 3 of 41
Consonants
other words, when it consists of more than four moras (McCawley, 1968; Kubozono,
2005; Itô and Mester, 1996). We shall revert to this issue in section 3.7.
Furthermore, the contrast between p and pp is never relevant in Yamato, Sino-Japanese,

and mimetic words, since it is impossible to find minimal pairs based on the [pp] / [p]
opposition.
In Western loans, on the other hand, no distributional constraint bears on /p/. It appears
freely in the initial or medial positions as a singleton, or as a geminate word medially, for
instance in po teto {potato}, kyapashitii {capacity}, hi ppii {hippie}. In English loans,
geminate /p/ is generally the reflex of a double p in the orthography, as in hippie.
It is interesting to note that in the first loans from European languages, p was
systematically adapted as a geminate inside words: for example kappa°, from the
Portuguese {capa} ‘rainwear’, or youroppa from Portuguese {Europa} ‘Europe’. The
appearance (or rather, the reappearance, see section 3.7) of a simple /p/ in Modern
Japanese would thus be posterior to the sixteenth century, the date at which these
words were introduced into the language.
So it turns out that it is only in gairaigo and in mimetics that [p] and [h] are fully
contrastive, since [p] does not normally occur as an intervocalic singleton in Yamato and
in Sino-Japanese, and, as we shall see in section 3.7, [h] appears (p.62) intervocalically
only in predictable environments. Even in mimetics, it would seem that h and p can be
considered in most cases as stylistic variants of the same consonant; for instance hu u-
huu and pu u-puu both express the action of blowing strongly through pursed lips, with
the difference that lips are more narrowly pursed in pu u-puu than in hu u-huu. However,
in some cases, there exists a clear difference in meaning between two mimetics differing
only by the presence of h or p; compare for instance hu ka-huka ‘fluffy’ and pu ka-puka
‘lightly floating in the air’ / ‘exhaling a lot of smoke’.
As we shall see in section 3.7, the special distribution pattern of p and its scarcity in the
modern language result in fact from the evolution of an archaic */p/ into the fricative /h/.
However, it seems that a number of */p/’s of the archaic language have not undergone
this process of spirantization. This is the case, in particular, of the geminate /p/, with
expressive connotation, which one encounters in forms like appare ‘admirable’ or
moppara° (or mo ppara) ‘exclusively’, derived from *apare and *mopara. We will return
to the status of /p/ and /b/ in section 3.7 devoted to /h/.
3.3 /t/ And /d/

The alveolar plosive /t/ has three allophones: the apico-alveolar (or apico-dental) [t] before
a,e, and o; two affricate consonants: [ʨ͡] before i and y and [ʦ͡ ] before u.
(1) Realizations of /t/
Page 4 of 41
Consonants
The combinations *[ti] and *[tɯ] do not exist in modern Yamato and Sino-Japanese
words.1 However, the non affricate realization [t] may be found before i and u in some
relatively recent loanwords such as tii baggu [tiːˡbaggɯ] {tea bag}, or fasshon tatuu
[ˡtatɯː] {fashion tattoo}, but only in the speech of certain speakers. In older Western
loans, for example chi imu {team} or tsu aa {tour}, only the affricate realization is
attested.
(p.63) The consonant /d/ displays even more varied realizations: [d] before a,e, and o in
any position, and, inside words, [ʑ] before i and y, and [z] or [dz͡ ] before u. Word-initially,
it is frequently realized as [dʑ͡ ] before i and y, and as [dz͡ ] before u.
(2) Realizations of /d/
a. In word-internal position:
b. In word-initial position:
The combinations *[di] and *[dɯ] do not exist in Yamato and Sino-Japanese words. But in
recent loanwords, a majority of speakers have the pronunciations [di] and [dɯ] to render
the foreign combinations di and du, for example in di naa [ˡdina:ː] {dinner}, duu itto
yuaserufu [dɯ:ːittojɯaˡseɾɯϕɯ] {do it yourself}. It would seem that, all things being
equal, the realizations [di] and [dɯ] are more widespread in these types of recent loans
than their voiceless counterparts [ti] and [tɯ]. This difference no doubt results from the
existence of a neutralization concerning /z/ and /d/ before /i/, since /zi/ and /di/ have the
same phonetic realization, [ʑi] or [dʑ͡ i]. The same applies to /zu/ and /du/, both realized as
[zɯ] or [dz͡ ɯ] (see also sections 3.5 and 3.6 below). On the other hand, /t/ remains always
phonetically distinct from /s/, whatever the following vowel.
The phonological status of /d/ will also be discussed in the chapter devoted to the
phonology of voicing.
3.4 /k/ And /g/

The Plosive Consonants /k/ And /g/ are Always Clearly Velar Before e,a, And o, Even
More So Than the Corresponding Sounds of English Or French for Instance, Which often
Seem Like Palatalized Consonants To Japanese Ears, Especially Before a. This is
Illustrated By the Fact That Certain French Or English Words Like (p.64) {cabaret},
{career}, Or {caramel} Have Been Adapted As kya baree, kya ria, And kyarameru° In
Page 5 of 41
Consonants
Japanese, With A Palatalized k.
Some Speakers Realize /g/ As A Plosive [ɡ] Word-initially But As A Nasal [ŋ] In
Intervocalic Position Under Certain Conditions. Other Speakers Pronounce It As A
Plosive In All Contexts, Or Possibly Sometimes As A Spirant ([ɣ]) Inside Words In
Familiar Register Or Fast Tempo (for Example [kaɣaˡmi] kagami ‘mirror’). the Status of
the Dorso-velar Nasal [ŋ] Poses A Number of Problems Which Will Be Addressed In
Detail In section 3.10.
3.5 /s/ And /z/

The fricative /s/ has two main allophones: [s] before u,e,o, and a, and a pre-dorso-palatal
[ɕ] (or dorso-palatal [ç]) before i and y.
(3) Realizations of /s/
The realization of /s/ as [ɕ] or [ç] before i and y causes a neutralization with /h/, since /h/ is
also often realized as [ɕ] or [ç] in the same context. We will reconsider this question in
section 3.7.
The /z/ consonant has two or even three realizations: [z] before u,e,o, and a inside
words; [ʑ] or [dʑ͡ ] before i and y, and finally, for certain speakers, [dz͡] before u,e,o,a
word-initially or after the mora nasal /N/. Before u, the affricate consonant [dz͡ ] sometimes
occurs even in word-internal position.
(4) Realizations of /z/
a. In word internal position, except after /N/:
b. In word-initial position or after the mora nasal /N/:
(p.65)
The opposition between /d/ and /z/ is neutralized nowadays before u and i, as well as
before y in Standard Japanese in the Yamato and Sino-Japanese strata, although there
exist four different kana symbols to write /du/, /zu/, /di/, and /zi/. This is the vestige of an
opposition which used to be relevant between the four moras [dɯ], [zɯ], [di], and [ʑi].
This neutralization is known as yotsugana no kondô ( つの , lit. ‘four kana merging’) in
Japanese linguistics.
Page 6 of 41
Consonants
(5) ‘Four kana merging’: neutralization of /d/ and /z/ before /i/ and /u/ in the
standard language
kana Kunrei / (Hepburn) Phonetic Phonological

romanization realization transcription
づ du / (zu) [zɯ] or [dz͡ ɯ] /zu/
ず zu / (zu)
ぢ di / (ji) [ʑi] or [dʑ͡ i] /zi/
じ zi / (ji)
The four kana merging seems to have developed around the sixteenth century, the
moment when it begins to be reflected in the writing. It is nowadays more or less
advanced depending on the dialects. The standard dialect, in which there is commonly
confusion between /du/ and /zu/ on the one hand, and /di/ and /zi/ on the other hand, is
called a ‘two-kana dialect’. Several dialects of Kyûshû and Shikoku (prefectures of
Kagoshima, Miyazaki, and Kôchi) maintain a four-way distinction between /du/, /zu/, /di/,
and /zi/ (realized as [dz͡ ɯ], [zɯ], [dʑ͡ i], and [ʑi]), while other dialects, such as the famous
zûzû-ben of Northern Honshû, confuse /du/, /zu/, /di/, and /zi/ under one single
realization, either [ndz͡ ɨ] or [ndz͡ ʉ] (with central vowels, and pre-nasalization of the voiced
obstruents). In these areas, the neutralization also affects the corresponding voiceless
consonants, namely /su/ and /si/ realized as [sʃɨ] or [sʉ], while /tu/ and /ti/ are realized as
[ts͡ ɨ] or [ts͡ ʉ].
In a number of very recent loans, a realization as [dɯ] distinct from that of [zɯ] is
appearing, as illustrated in the already cited example duu itto yuaserufu
[dɯ:ittojɯaˡseɾɯϕɯ] {do it yourself}.
On the other hand, it is important to note that *[si] and *[zi] are not found in any of the
strata, not even in very recent loans where [si] or [zi] of the source language are
adapted under the written forms スィ or ズィ. The oral realizations of these sequences are by
no means faithful to the writing, since スィ is (p.66) pronounced as [ɕi] or [çi] and ズィas [ʑi]
or [dʑ͡ i]. It would be interesting to determine the exact reasons explaining why certain
impossible combinations in the language at a certain time of its history come to be easily
accepted, like [dɯ], whereas others, like [si], resist faithful adaptation.
Of course, the [ɕ] or [ç] realization of /s/ before /i/ can be seen as an instance of
regressive palatal assimilation, but some Japanese scholars, for instance Hamada (1964)
estimate that this could well be the vestige of the earlier phonetic value of /s/, which might
have been realized as a hushing fricative before all vowels and not only before /i/. There
is no doubt that /s/ was hushed before /i/ and /e/ until the seventeenth century, as
reflected in the Jesuits’ romanization of xe for せ (as in Portuguese) for instance in せめ
‘reproach’ romanized as xeme (seme in Modern Japanese). Hoffmann mentions that せ is
pronounced as se,she, or even tse in his 1868 grammar. The [ɕe] realization can still be
heard in a number of dialects, and even marginally in the affected pronunciation of certain
Page 7 of 41
Consonants
Tôkyô words such as in [ˡmiɕete] instead of [ˡmisete] ‘show me’ (seductive, or baby-talk
register). Opinions of Japanese philologists as to what /s/ might have sounded like even
earlier diverge. Mabuchi (1959) proposes that /s/ might have been [∫], Kamei (1970) and
Arisaka (1936, 1955) opt for [ts], and Kobayashi (1981) for [t∫].
As a voiced obstruent, /z/ displays all the properties specific to this type of Japanese
consonant, to which I will revert in Chapter 4. However, note that contrary to /d/, /b/, and
/g/, it is not at all certain that /z/ was prenasalized at an older stage of the language.
3.6 The Phonological Status of Hushing And Affricate Consonants
3.6.1 Sh [ɕ], j [ʑ], and Ch [tɕ͡ ]

Although sh,j, and ch are sometimes transcribed by means of the IPA symbols [∫], [Ʒ],
and [t∫͡], the closest IPA transcription of these sounds is [ɕ], [ʑ], and [tɕ͡] (or [ç], [ʝ], and
[tç͡ ] before i and y), since they are actually alveolo-palatals or pre-dorso-palatals. Note in
addition that they are articulated with no marked lip-rounding, and a flattened tongue
blade.
We have already seen that before i and y, [ɕ] and [ʑ] were allophones of /s/ and /z/, and
[tɕ͡ ], an allophone of /t/.
The sounds [ɕ], [ʑ], and [tɕ͡ ] are frequent, particularly in Sino-Japanese words and in
recent loans, before all five vowels, as the following examples illustrate.
(p.67) (6)
1. ɕ:
kisha [kiˡɕa] ‘train’

shu uri [ˡɕɯːɾi] ‘repair’
bi sho-bisho [ˡbiɕobiɕo] ‘wet’
2. ʑ :
jari° [ʑaɾi] ‘gravel’

ju u [ˡʑɯː], [ˡdʑ͡ ɯː] ‘ten’
joudan [ʑoːˡdaN], [dʑ͡ oːˡdaN] ‘joke’
3. tɕ :
ocha° [otɕ͡ a] ‘tea’

konchuu° [kontɕ͡ ɯː] ‘insect’
cho tto [ˡtɕ͡ otto] ‘a little’
At first glance, it looks as if these consonants constitute true phonemes of their own,
which could stand in opposition with /s/, /z/, and /t/. However, following the Japanese
Page 8 of 41
Consonants
linguistic tradition, it is preferable to consider that they correspond phonologically to

simple consonants followed by the palatal element y: sh [ɕ] = /sy/, j [ʑ] = /zy/ or /dy/, ch
[tɕ͡ ] = /ty/. This is actually the analysis suggested by the Japanese writing system, which
uses the kana し (shi), じ (ji) or ち (chi), ぢ (ji), each followed by や ya, ゆ yu, orよ yo in
reduced size in order to transcribe these units (see section 1.5). Note in passing that the
official romanization system (Kunrei, ), which transcribes si, zi, ti, di, sya, zya, tya,
dya, etc. and not shi, ji, chi, ji, sha, ja, cha, ja, as in the Hepburn system, has been
directly inspired by the spirit of the Japanese kana. It is also much closer to the
postulated phonological representation.
One of the arguments in favour of this analysis comes from the observation that in
Yamato and Sino-Japanese, the sequences *she */sye/,2 *je */zye/ and *che */tye/ do not
exist. This must be considered in relation to the fact that no palatalized consonant can
occur before the vowel e. The combinations *kye, *gye, *nye, *hye, *mye, *rye, etc. are
unattested, just as *ye is also, except in some recent loans. If one supposes the
independent existence of the consonants sh /ɕ/, j /ʑ/, and ch /tɕ͡ /, the absence of the
combinations she /ɕe/, je /ʑe/, and che /tɕ͡ e/ in a large part of the lexicon receives no
explanation. But if one supposes that she,je, and che represent palatalizations of /s/, /z/,
and /t/ before e, these gaps are simply the consequence of the fact that palatalization is
impossible before the vowel e, whatever the nature of the preceding segment (on ye, see
also section 3.11).
Besides, even in Western loans in which she,je, or che would be expected, they are
sometimes realized as se,ze, or chie: for instance shepaado 〉 sepaado ‘sheepdog’
{shepherd}, je neraru 〉 ze neraru {general}, che en 〉 chi en {chain}.
(p.68) In addition, it is necessary to observe that [ɕ] and [ʑ] are also very rare in
Yamato words before a,u, and o. Before a, there also exists an occasional free alternation
between [s] and [ɕ], or [z] and [ʑ] in the Yamato and Sino-Japanese strata. For example,
sa ke / shake ‘salmon’, saboten° / shaboten° ‘cactus’, zari° / jari° ‘gravel’, can be
analysed as cases of alternation between a vowel preceded by a simple consonant (CV)
and a vowel preceded by a palatalized consonant (CyV).
According to this analysis, the consonants /ɕ/, /ʑ/, and /ʨ͡/ (alias /∫/, /Ʒ/ and / t∫͡/ in certain
transcriptions) can be removed from the phonemic inventory.
3.6.2 ts [ts͡]
The question of the affricate consonant [ts͡ ] must be set in different terms from that of /ɕ/,
/ʑ/, and /ʨ͡/. At first glance, this consonant is also a conditioned allophone of /t/ before u.
However, a few occurrences of [ts͡] before vowels other than /u/ can be found. Although
almost all these cases occur in recently borrowed loans like tsetse-bae ‘tsetse fly’, tso isu
{Zeus}, tsaratsusutora {Zarathustra}, it is necessary to mention the existence of certain
Yamato forms with [ts͡] such as otottsan [oˡtotts͡ aᶰ], a variant of otousan ‘dad’. It is
because of this type of example that /ts͡ / (sometimes noted as /c/) is included as a
phoneme in the consonant charts proposed by linguists such as Hattori Shirô or Kindaichi
Haruhiko, who nevertheless do not regard /ɕ/, /ʑ/, and / ʨ͡/ (sh, j, and ch) as phonemic.
Page 9 of 41
Consonants
Moreover, ts [ts͡ ] and its corresponding palatalized version ch [ʨ͡] stand in a curious
relationship with /s/. These two sounds frequently work like a strengthened version of /s/
after gemination,3 i.e. when preceded by /Q/ (see section 5.2), particularly in emotional or
expressive forms, or in baby talk (Hamada 1954:74):
(7)
otou-san 〉 otottsan ‘dad’

chiisai 〉 chicchai ‘small’
de+shiri 〉 de cchiri,decchiri° ‘chubby buttocks’
massugu 〉 mattsugu ‘straight’
shoushuu 〉 sho cchuu ‘often’
Let us also mention the forms -chan, -chama, which are expressive and baby-talk-like
alternants of -san,-sama (personal suffix, ‘Mr, Miss, Mrs’).
(p.69) 3.7 /h/

The fricative /h/ has the following realizations: [h] before a,e, and o, [ç] or [ɕ] before i and
y for many speakers (especially in Tôkyô speech), and [ϕ] before u:
(8) Realizations of /h/
For certain speakers, the opposition between /h/ and /s/ is neutralized before i: hi /hi/ and
shi /si/ are pronounced identically. The opposition is neutralized to the benefit of /h/ in the
Kansai area (/hi/, /si/ = [hi]), to the benefit of /s/ in the Kantô area (/hi/, /si/ = [çi] or [ɕi]), as
shown below. These mergers were already attested at the beginning of the eighteenth
century.
(9)
Kansai Kantô
shima /sima/ [hima] [çima] / [ɕima] ‘island’
hima /hima/ [hima] [çima] / [ɕima] ‘(free) time’
The change in place of articulation of /s/ to [ɕ] before /i/ can be regarded as an
assimilation, since /i/ is a palatal vowel. But there is also a possibility that the coronal
fricative /s/ has actually preserved its original place of articulation before /i/ in the modern
language, since /s/ might have been a hushing, palato-alveolar segment before all vowels
Page 10 of 41
Consonants
at an earlier date, as we have seen in section 3.5.
[h] is very rare intervocalically in Yamato monomorphemes, where it has usually turned
to [pp], [w], or disappeared, as will be seen in more detail in section 3.7.2. It occurs for
instance in ha ha ‘mum’, ho ho ‘cheek’, ahureru ‘to overflow’.
The voiceless bilabial fricative ϕ, one of the allophones of /h/ before u, has been recently
phonemized in newly introduced loans, so that it now occurs contrastively before vowels
other than /u/, for example fa mirii [ˡϕamiɾiː] {family}. So pairs such as hi tto [ˡhitto] {hit},
and fi tto [ˡϕitto] {fit} which stand in opposition in this stratum can now be found. In older
loans dating back to the beginning or the middle of the twentieth century, [h] appears
regularly as the adaptation of {f} in Western loans before vowels other than [ɯ]: te lehon
[ˡteɾehoᶰ] {telephone} (see also section 3.14).
3.7.1 Phonological Correspondence Between h,b, And p

(p.70) A number of alternations involving h,p, and b deserve special consideration.
Geminated /h/ is rendered as [pp], not *[hh]. Consider for example the emphatic or
expressive forms in (10a), as well as the Sino-Japanese compounds in (10b) which involve
the CV 〉 /Q/ process already discussed in section 2.5:
(10) /hh/ 〉 [pp]
a. Emphatic or expressive forms
yahari4 〉 yappari ‘actually’

suki + hara 〉 suki-ppara° ‘empty stomach’
su + hadaka° 〉 su-ppadaka ‘naked’
b. Sino-Japanese compounds
shutsu + hatsu 〉 shup-patsu° ‘departure’

ichi + hon 〉 i p-pon ‘one long object’
(compare with ichi + kai 〉 ik-kai ‘one time’)
As we will see below, this can be accounted for diachronically, since modern h derives
from *p.
It is only in some recent loans, for example go hho {(van) Gogh}, ba hha {Bach}, and in a
limited number of Sino-Japanese or mixed compounds, for example ju hhari ‘ten
stitches’ (Sino-Japanese + Yamato, Lawrence, 1999), zehhuchou° ‘(to feel) awful’
(Sino-Japanese), that occurrences of [hh] or [ϕϕ] can be found.
In Sino-Japanese words, h turns into p after the mora obstruent /Q/ and the mora nasal
/N/ (see also the examples in 14). For instance, the Sino-Japanese morpheme ‘hair’ is
pronounced hatsu or patsu depending on the phonetic environment.
Page 11 of 41
Consonants
(11) h → p / /Q/ or /N/
hatsu ‘hair’ (Sino-Japanese)

/haku/ + /hatu/ hakuhatsu° ‘white hair’
/kiN/ + /hatu/ kinpatsu° ‘blond (golden) hair’
/ketu/ + /hatu/ keppatsu° ‘tied hair’
The same alternation can occasionally occur in Yamato words, for instance na n(i) ‘what’ +
hito° ‘person’ becomes nan-pito° ‘what person’ (obsolete, nan-bito° is more common)
and suki- ‘empty’ + hara ‘belly’ becomes suki-ppara° ‘empty stomach’.
(p.71) [h] can nonetheless follow the mora nasal /N/ in some cases. For instance, the
Sino-Japanese morpheme meaning ‘brush, pencil’ occurs as pitsu in enpitsu° ‘pencil’ but
as hitsu in mannenhitsu ‘fountain pen’. In both cases, the preceding morpheme ends in
/N/. McCawley (1968:78) states that in native and Sino-Japanese, the distribution between
[p] and [h / ϕ] can be predicted on the basis of the syntactic information. He claims that
[p] occurs after the mora nasal when the Nh combination is in the ‘innermost layer of
compounding’, as the following examples with hitsu ‘pencil’, hun ‘powder’, and hatsu
‘emit’ illustrate:
(12)
enpitsu° /eN/ + /hitu/ ((en)(pitsu)) ‘pencil’
mannenhitsu /maN/ + /neN/ + /hitu/ (((man)(nen))(hitsu))

‘fountain pen’
denpunshitsu /den/ + /hun/ + /situ/ (((den)(pun))(shitsu))

‘starchy matter’
shinhatsumei /siN/ + /hatsu/ + /mei/ ((shin)((hatsu)(mei)))

‘new invention’
Page 12 of 41
Consonants
This is what has led McCawley (1968:77–78) to posit /p/ rather than /h/ in the underlying
inventory of the native and Sino-Japanese strata, with a rule converting /p/ into [h] word-
initially and under the morphological conditions stated above, even though a few
exceptions exist, especially in numeral compounds.
(p.72) While it is true that McCawley’s analysis provides a satisfactory and simple
account of most of the facts pertaining to the h / p alternation, it is not followed by most
native Japanese scholars, who, as Shibatani (1990) observes, find it very counter-
intuitive. Note that there are still some cases which must be handled as exceptions.
McCawley cites a dozen words or so which are not transparent compounds but still
contain an intervocalic h, such as ahiru° ‘duck’, ha ha ‘mother’ (see also section 3.2.). As
mentioned in section 3.2. one can also find some word-initial or intervocalic p’s such as
peten° ‘head’, ta npopo ‘dandelion’, or pakuru ‘to filch’ in words of Yamato or Sino-
Japanese origin (most are slang or familiar words). Another interesting example is the
Sino-Japanese word chapatsu° ‘hair dyed brown’, whose second morpheme is ‘hair’,
already cited in example (11) above. In chapatsu°,p is neither geminated nor preceded
by /N/. This word seems to be an analogical formation recently coined out of kinpatsu°
‘blond hair’ to refer to brown dyed hair, a new hairstyle colour which became fashionable
in Japan in the 1990s.
A possible synchronic explanation of the fact that /h/ is rendered by /p/ after the special
segments /Q/ and /N/ is the following: /N/ and /Q/ have no place of articulation of their own
underlyingly. In order to receive full surface phonetic content before a consonant, these
two elements require that the following segment should have a fully specified place of
articulation, to which they will be able to link. However, the laryngeal consonant /h/ does
not fulfil this condition, since it can be regarded as placeless at the phonological and
phonetic level (Ladefoged, 1982, 1990)5. This is arguably why the labial /p/ has been
maintained after /Q/ and /N/. Another explanation would be that fricatives do not undergo
gemination as easily as other consonants, in particular stops. A fricative occurring in a
position or context in which gemination is likely to occur sometimes undergoes a change
in manner of articulation, by a process which can be regarded as strengthening. The
fricative takes the feature [−continuous], that is, it turns into a stop. As previously
mentioned (see footnote 3, this chapter) the other voiceless fricative of Japanese, the
alveolar /s/, displays the same tendency to occlusion in the event of gemination.
In Modern Japanese, /h/ also works as the voiceless counterpart of /b/ morpho-
phonologically, despite the fact that this role should be assigned to /p/ at the strictly
phonetic, articulatory level. This is first manifest in the writing, since the kana denoting
ba,bi,bu, be, and bo are derived from those denoting ha, hi, hu, he, and ho by addition of
Page 13 of 41
Consonants
the diacritics used to indicate obstruent voicing (the nigoriten, or dakuten , example
13), in the same way as the moras starting with d,z, and g are derived from those starting
with t,s, and k (see section 1.5). (p.73) The CV combinations starting with the voiceless
bilabial stop p also take as a basis the kana starting with h, but they use a special symbol,
the handakuten ( ), a small circle, which has appeared relatively recently in the history
of the Japanese writing system, around the sixteenth century according to Okumura
(1972).
(13) Kana writing of p and b
h : b ha hi hu he ho はひふへほ : ba bi bu be bo ばびぶべぼ
h : p ha hi hu he ho はひふへほ : pa pi pu pe po ぱぴぷぺぽ
The phonological relationship between h and b is also manifest in the rendaku (

‘sequential voicing’, see Chapter 4). The stop /b/ functions like the voiced alternant of /h/,
for example in tabi + hito° 〉 tabi-bito° ‘traveller’ (‘travel’ + ‘person’).
The special link between h,b, and p is further evidenced by a number of cases of
alternation between these three consonants. Such alternations are mainly observed in the
mimetic vocabulary and in numeral compounds of the form Numeral + Specifier. They are
conditioned either by the phonetic environment (thus /h/ becomes /p/ or /b/ after /N/), by
application of the rendaku rule or of post-nasal voicing (see Chapter 4), or by register
considerations in the case of mimetics. In some Yamato compounds, [pp] also works like a
marker of compounding, and thus fulfils a role somewhat similar to rendaku. Compare for
instance the three free alternant forms sukihara° / sukibara° / sukippara° ‘empty
stomach’ in the examples below.6
(14) Alternation between /h/, /p/ and /b/
a. Mimetic words
ha ra-hara(-to) pa ra-para(-to) bara-bara°(-ni)

‘fluttering, scattering’ (with various stylistic nuances)
b. Sino-Japanese compounds
-hon : i p-pon ni -hon sa n-bon

(specifier of long ‘1 long object’ ‘2 long ‘3 long
objects): objects’ objects’
hu : hu-sai ni n-pu jou-bu°
‘man’ ‘married ‘labourer’ ‘robust’
couple’
(p.74)
c. Yamato compounds
Page 14 of 41
Consonants
ha mushi-ba° de -ppa/ de-ba (っ)

‘tooth’ ‘decayed tooth’ ‘buck teeth’
suki-hara° き suki-bara° き suki-ppara° きっ
‘empty stomach’ (suki- ‘empty’ + hara ‘belly, stomach’)
The h / p / b alternation in numeral compounds is rather irregular. Some specifiers

starting with /h/ occur with [b] after the numeral san° ‘3’ and the interrogative na ni / nan
‘how many’ but keep [h] after yo n ‘4’, for instance ha i ‘cups’ or hiki ‘small animal’, while
others have [p] after ‘3’, ‘4’, and ‘how many’, for instance hu n ‘minute’, ho u ‘direction’,
and so on (see (15)). All of them have [p] after ichi ‘1’, roku ‘6’, hachi ‘8’, ju u ‘10’, by
application of the CV 〉 /Q/ rule presented in Chapter 2.5.
So there are two sets of /h/ starting specifiers. Those which have allomorphs in [h], [b],
and [p] and those which have allomorphs in [h] and [p] only, as the following examples
show. According to Otaka (2009), the second type is more numerous than the first, but
specifiers of the first type are of more frequent use.
(15)
h / p / b h / p / *b
ha i ‘cup’ hu n ‘minute’
‘x cup(s)’ ‘x minute(s)’
1 ichi i p-pai i p-pun
2 ni ni -hai ni -hun
3 san° sa n-bai sa n-pun
4 yo n yo n-hai yo n-pun
5 go go-hai° go -hun
6 roku ro p-pai / roku-hai ro p-pun
7 na na nana-hai nana-hun
8 hachi ha p-pai / hachi-hai ha p-pun / hachi-hun
9 kyu u kyu u-hai kyu u-hun
10 ju u ju p-pai / jip-pai ju p-pun / jip-pun
‘how many’ na n(i) na n-bai na n-pun
(p.75)
3.7.2 Diachronic Development of /h/

The articulatory variety of /h/ in the modern language actually reflects the complex and
lively history of this consonant in Japanese. The series of evolutions it has undergone is
known as ha-gyô tenkoon (ハ ) in Japanese linguistics, literally ‘phonetic change of the
sounds of the h-row’.
Page 15 of 41
Consonants
In Yamato and Sino-Japanese words, h is generally considered as going back to a labial

plosive *p (Ueda, 1898; Hashimoto Sh., 1928; Martin, 1987). Archaic Japanese (or pre-
archaic Japanese according to some scholars) would thus have had no laryngeal fricative,
which explains why, in the oldest loans from Chinese, an original /h/ is regularly adapted
as the velar /k/ in Japanese, while /p/ is adapted by means of the sound transcribed by the
kana currently denoting the moras starting with the /h/ consonant (ha-gyô ハ ). These facts
point to the conclusion that the initial consonant of this graphemic series was not [h] but
some other sound, most probably [p]. Some examples are given in the chart below. For
the sake of comparison, I also provide the corresponding Sino-Korean forms, which very
regularly maintain the original initial of the Chinese lexemes.
(16) *h and *p in Old Chinese, and their reflexes in Sino-Japanese and Sino-Korean
Old Chinese Modern Sino-Japanese Modern Sino-Korean

– ch *h : jp k
‘blood’ *huet ketsu hjɔl
‘China’ *han kan han
‘perfume’ *hıaŋ kou, kyou hjaŋ
– ch *p : jp h
‘root’ *pսən hon pon
‘wave’ *puar ha pʰa
‘cloth’ *pag hu, ho pʰo
Loans made by Ainu from Japanese are also revealing of the original value of modern /h/.
For example, the Ainu word potoki ‘Buddha’, a very old loan from Japanese, corresponds
to the Modern Japanese form hotoke / hotoke°. Since Ainu does possess a glottal fricative
[h] in its system, the logical interpretation is that the Japanese word in question began
with *p, not with *h, at the time of borrowing. In addition, in several modern Ryûkyû
dialects, /p/ corresponds very regularly to Japanese /h/: jp hi / rk pi ‘fire’, jp hatake° / rk
pataki ‘field’.
It is most probable that */p/ had evolved into a bilabial fricative as early as the Nara
period (eighth century, Hashimoto Sh., 1928; Hamada, 1954), except in mimetic words.
The bilabial realization ϕ has been maintained up to now before /u/ in Standard Japanese
and even before all the vowels in certain dialects of the Tôhoku area and in the
prefectures of Nagano, Shimane, and Nagasaki.
(p.76) Around the tenth century, ϕ is thought to have turned into w intervocalically
(18a), until its complete disappearance, except before /a/ where it gave rise to [w]. Word-
initially, it is maintained (18b) and remains labial at least till the seventeenth century. This
is attested by the notations of Portuguese and Spanish missionaries, who transcribe fodo
‘degree’, fafa or faua ‘mother’, feike monogatari ‘Tale of the Heike’ (today hodo°, ha ha,
Page 16 of 41
Consonants
he ike). A riddle of the Muromachi period provides us with another invaluable indication
(Komatsu, 1981, also cited in Martin, 1987). To the question ‘What is it that meets twice
for mum, but not even once for dad?’ ( ニハアフテ、ニハモアハズ), the expected answer was
kuchibiru° ‘the lips’. We thus know that ‘mum’ (Modern Japanese ha ha) was pronounced
with a bilabial, presumably as [ϕaϕa] or [ϕawa] at that time, and that in turn it was
undoubtedly *[papa] at an even earlier time.
This consonant pursues its transformation, evolving into a segment with a laryngeal place
of articulation, i.e. [h], except before /u/ where it is still nowadays a bilabial.7 Recall also
that before /i/, one often encounters a (pre-)dorso-palatal [ɕ] or [ç]. Such a palatal
realization of /h/ already existed at least at the beginning of the eighteenth century.
This series of evolutions is recapitulated in (17) (following the proposal by my colleague

Takayama Tomoaki, in Labrune and Takayama, 2004):
(17) Diachronic evolution of h
Other hypotheses have been proposed. Let us mention in particular that of Hayashi Ch.
(1992), who considers that archaic */p/ was maintained as [p] until the ninth century at
least, and that the *p 〉 ϕ 〉 w evolution occurred subsequently, and at once. Another
approach is that of Hamano (2000), who assumes that the evolution of */p/ towards /w/ in
intervocalic position has gone through the following stages: *p 〉 *b 〉 *β 〉 *w.
(p.77) Here is a series of examples illustrating the evolution of */p/ towards /h/:
(18)
a. Intervocalically (V V)
*[kapa] 〉 [kaϕa] 〉 [kaˡwa] /kawa/ kawa ‘river’

*[sipo] 〉 [siϕo] 〉 [siwo] 〉 [ɕiˡo] /sio/ shio ‘salt’
*[ipe] 〉 [iϕe] 〉 [iwe] 〉 [ije]8 〉 [iˡe] /ie/ ie ‘house’
b. Word-initially (# )
*[paɾɯ] 〉 [ϕaɾɯ] 〉 [ˡhaɾɯ] /haru/ ha ru ‘spring’

*[pikaɾi] 〉 [ϕikaɾi] 〉 [hikaˡɾi] (/ [çikaˡɾi]) /hikari/ hikari ‘light’
Page 17 of 41
Consonants
*[pɯne] 〉 [ˡϕɯne] /hune/ hu ne ‘boat’
The semi-consonant w having supposedly disappeared before e,o, and then i between the
tenth and thirteenth centuries, many sequences of two consecutive vowels VV in the
modern language correspond in fact to a former VwV sequence, which in turn
represents evolutions from VpV forms (see section 3.12).
This series of changes explains why the glottal fricative h has almost disappeared from the
Yamato lexicon intervocalically. It exists only in words resulting from a reduplication such
as ha ha ‘mother’, hoho ‘cheek’, hata-hata° (hata-hata, hata-hata) ‘sandfish’, ha ra-hara
‘flutteringly’,9 and in transparent compounds such as a sa-hi ‘morning sun’, shira-hata°
‘white flag’, and shira-ho° ‘white sail’. These two types of words have maintained an
internal h in order to ensure morphological transparency and paradigmatic identity of the
morpheme, and to preserve the reduplicative character of the compound in the first
case. Intervocalic [h] is also found in a couple of other exceptional cases such as ahureru
‘to overflow’ (presumably a back formation from abureru) or ahou ‘idiot’ (Kansai dialect).
3.8 /m/
The bilabial nasal /m/ is realized as [m] in all positions.
A rather peculiar segment realized as a moraic m (IPA [m̩]) is sometimes heard in the
speech of older speakers, or in certain forms of the traditional Japanese theatre. This
nasal is found in words starting with um-, like ume° ‘plum tree’, uma ‘horse’, umareru° ‘to
be born’, which are phonetically realized as [m̩me], [m̩ˡma], [m̩maɾeɾɯ], or even, if we
follow Vance (1987:39), as [ʔmːˆ ^ma] for uma. This seems to be the vestige of an archaic
pronunciation, which must have (p.78) been much more widespread formerly since it is
well attested in documents of the Heian period.
3.9 /n/
The nasal consonant /n/ is apico-dental or apico-alveolar. A majority of Japanese
phoneticians mention a palatal realization [ɲ] before the vowel i and the glide y, but it
seems to me that palatalization is not very marked.
The sequence nV frequently undergoes vowel deletion and subsequent transformation of

/n/ into /N/ (the moraic nasal) in familiar speech: kodomo° no toki 〉
kodomontoki,kodomontoki ‘during childhood’, ju ugonichi 〉 juugonchi ‘the fifteenth
(day)’ (Akinaga, 2008).
3.10 The Status of the Velar Nasal [ŋ]

The phonological status of the dorso-velar nasal consonant [ŋ] (bidakuon or ga-gyô
bion が ) in Tôkyô Japanese has been a much debated issue. The question is: is [ŋ] a
mere alternant of /g/ or is it phonemically different from /g/?
Let us first observe that this sound does not possess its own symbol in the two kana
syllabaries. It is transcribed with the g kana series: が, ガ = [ga] or [ŋa], ぎ, ギ = [gi] or [ŋi],
Page 18 of 41
Consonants
ぐ, グ = [gu] or [ŋս], げ, ゲ = [ge] or [ŋe], ご, ゴ = [go] or [ŋo].10 Second, the sound [ŋ] is
regarded as a disappearing segment even in the dialects in which it is attested, as we will
see in more detail below (Kindaichi, 1942; Inoue, 1971; Hibiya, 1999), although it
continues to be, even now, held as a distinctive mark of ‘beautiful Japanese’. But ŋ is
absent in the speech of many speakers, where it is replaced by [g] or [ɣ]. So, in many
Japanese dialects, the [g] / [ŋ] alternation is simply not relevant (see the Japan Linguistic
Atlas for information concerning the dialectal distribution and phonetic realizations of
/g/11 ). The study by Hibiya (1999) clearly demonstrates that there is a clear pattern of
age stratification in the use of [ŋ], which drops off as age diminishes. Note however that
the velar nasal [ŋ] is thought to have appeared relatively recently in Japanese, probably
around the eighteenth century. It is generally assumed to be the result of a phonemic
split out of the /g/ consonant, which has spread from central Japan (including the Tôkyô
and Kyôto areas), to outer regions (Inoue, 1971).
(p.79) Next, in the speech of speakers who realize [ŋ], the distribution of this sound is
peculiar: [ŋ] does not appear at the beginning of independent morphemes (19a). It is
found only word-internally (19b).
(19)
a.
goma° [goma] *[ŋoma] ‘sesame’

geta° [geta] *[ŋeta] ‘geta (Japanese clogs)’
b.
kagami [kaŋa¹mi] *[kaga¹mi] ‘mirror’

nigeru [ni¹ŋeɾɯ] *[ni¹geɾɯ] ‘to flee’
These examples could suggest that [g] and [ŋ] stand in complementary distribution in the
speech of speakers who possess the two sounds, the velar stop appearing in word-initial
position, the velar nasal in word-medial position. It would then be tempting to regard [ŋ]
as a simple alternant of /g/ (as an idiolectal, sociolectal, or regional variant), with the
following allophonic rule:
(20)
However, once again, things are not so simple. A number of small facts make the issue
somewhat more complicated.
First, there exist some cases of semantically relevant contrasts between [g] and [ŋ], as in
Page 19 of 41
Consonants
the oft-cited pairs do ku-ŋa (doku-ŋa) ‘poison fang’ vs. doku-ga° (doku-ga) ‘Oriental
tussock moth’, da igo ‘fifth’ vs. da iŋo ‘ghee’, ju ugo ‘fifteen’ vs. ju uŋo ‘home
front’. But such examples of minimal pairs remain rare, and, for certain linguists, they do
not justify the existence of an additional phoneme in the system of Japanese.
Second, the conjunction ga (‘but’) can be realized as [ŋa] even at the beginning of a
sentence. One may also hear the sound [ŋ] at the beginning of the words gurai°
‘approximately’, or gotoshi° ‘identical’ (NHK, 1998) when they appear utterance-initially
as independent morphemes, even though they rarely do so.
Lastly, the [ɡ] / [ŋ] alternation is not systematic. A non-initial /g/ is never realized as [ŋ] in
mimetic or expressive reduplicated words (21), at the beginning of the morpheme go (or
go°) ‘five’ (22), after the polite prefix o- (23), and in recent loans (24.a) except if a velar
nasal already exists in the source form (24.b):
(p.80) (21) Reduplicated mimetic and expressive words
gunya-gunya° [gɯnjagɯnja] *[gɯnjaŋɯnja] ‘flabby’

geji-geji° (geji-geji) [geʑigeʑi] *[geʑiŋeʑi] ‘millipede’
However, reduplications expressing plurality or intensity undergo velar nasalization,

such as kuni-guni [kɯ¹niŋɯni] ‘countries’. But variation does exist, and descriptions are
often contradictory. For example the adjective gyougyoushii ‘ostentatious’, which
consists of the reduplication of a Sino-Japanese morpheme followed by the Yamato suffix -
shii, should be realized as [gjoːgjoː¹ɕii], not *[gjoːŋjoː¹ɕii] according to Kamei (1956). Yet
NHK (1998) indicates the pronunciation [gjoːŋjoː¹ɕii].
(22) Morpheme go ‘five’
ju ugo [¹ʑɯːgo] *[¹ʑɯːŋo] ‘fifteen’
Note that in the fully lexicalized word juugo-ya° [ʑɯːŋoja] ‘full moon night (‘night of the
fifteenth day’)’, where the morpheme go has lost its numeral meaning, the nasal, rather
than the stop, appears.
(23) Polite prefix o-
o- + genki [o¹geŋki] *[o¹ŋeŋki] ‘health’

o- + guai° [ogɯai] *[oŋɯai] ‘(your) health’
(24) Recent loans
a. Containing [g] in the source word
Page 20 of 41
Consonants
ra gubii [¹ɾagɯbiː] *[¹ɾaŋɯbiː] ‘rugby’
b. Containing [ŋ] in the source word
ha ngaa [¹haŋgaː] or [¹haŋŋaː] {hanger}
The word igirisu° [iŋiɾisɯ] ‘England’, from the Portuguese {Inglês} probably belongs to
the case depicted in (24b).
As for compounds belonging to types other than those which have just been mentioned,
the situation is somewhat murky. Three cases exist: certain words regularly display [ŋ],
for example hiyori° + geta° 〉 hiyori-ŋeta [hijo¹ɾiŋeta] ‘geta for dry days’; others always
maintain [g], as shi ro + goma° 〉 shiro-goma° [ɕiɾogoma] ‘white sesame’; a third group
licenses the two realizations [g] or [ŋ], for example keshi + gomu° [keɕiŋomɯ] or
[keɕigomɯ] ‘eraser’. Curiously enough, in compound words whose second member
starts with /g/ and which fulfil the conditions for rendaku application (see Chapter 4), /g/
may not undergo nasalization; for example ku ro + goma° (‘black’ + ‘sesame’) becomes
kuro-goma° [kɯɾogoma] rather than *[kɯɾoŋoma] ‘black sesame’ (Kamei, 1956).
Hibiya (1999) states that when the second element of a compound is a single Sino-
Japanese bound root starting with /g/, the root initial /g/ is realized as [ŋ], as (p.81) in ho
go ‘protection’ [¹hoŋo]. It is important to note that descriptions such as dictionaries or
articles do not necessarily concord with each other. The reader should consult Vance
(1987) who provides a very rich list of examples and a review of the Japanese literature
concerning the velar nasal issue. Vance (1987) also remarks that in three-character Sino-
Japanese compounds of the shape (X)+(YZ), /g/ seems not to be nasalized at the
beginning of Y as in the example /dai+gen.soku°/ dai-gensoku ‘major principle’ (with [ɡ]).
On the other hand, in words of the shape (XY)+(Z), we seem to get [ŋ] after the major
division, for instance /zi.dai°+ geki/ jidai-ŋeki ‘period drama’.
According to Akamatsu (1997:130), the fluctuation between [ɡ] and [ŋ] can be accounted
for by two main causes: demographic movements and speakers’ age. According to this
scholar, the generalization of the [ɡ] realization in internal position results from the
influence of non-standard regional pronunciations in which [ŋ] does not exist, due to the
migrations after the end of World War II. It would have then spread among younger
speakers. As already mentioned, [ŋ] occurs less in the speech of the younger generation.
Yuzawa and Matsuzaki (2004:20) show that [ŋ] has practically disappeared in the speech
of Tôkyô speakers under forty, whereas it occurs in the speech of nearly all speakers
over eighty. According to Hibiya (1999), analyses of real-time data have all indicated that
native speakers of the Tôkyô Yamanote dialect had word-internal [ŋ] until at least the late
nineteenth century, and so the decline of [ŋ] must have started in the early twentieth
century. It is heading towards completion in the younger generation of today.
Since the status of the velar nasal constitutes one of the most debated issues in the
phonology of Japanese, it seems interesting to present in more detail two extremely
Page 21 of 41
Consonants
different and complementary approaches, each quite representative of its time and
tradition: the treatment proposed by Itô Junko and Armin Mester, published in 1997,
which belongs to the most recent Western theoretical prospect, and that by Kindaichi
Haruhiko, published in Japan in 1942, which draws meticulous attention to the data and
variation in a sociological dimension, and which is very characteristic of the Japanese
linguistic tradition.
3.10.1 Itô And Mester’s Treatment (1997)

In their 1997 paper, the linguists Itô Junko and Armin Mester proposed a novel analysis
of the phonology of ŋ in Japanese. They observe that the two following cases are
observed in compounds whose second member begins with g: (i) possible variation
between [g] and [ŋ] at the compound juncture; (ii) no variation, only [ŋ] is possible. The
first type is that of words whose second component starts with an underlying /g/ in
isolation (examples (25)), the second type concerns words which start with an underlying
/k/ and have undergone (p.82) rendaku voicing (examples (26)).12 As we shall see in the
next chapter, rendaku is a morpho-phonological process of Japanese which requires the
initial consonant of the second member of a compound to become voiced under certain
conditions.
(25) /g/ → [ŋ] or [g] in compounding
niwa° + geta° 〉 niwa-geta° / niwa-ŋeta° ‘garden clogs’

shima° + gara° 〉 shima-gara° / shima-ŋara° ‘striped pattern’
(26) /k/ → [ŋ], *[g] in compounding
yuki + kuni° 〉 yuki-ŋuni *yuki-guni ‘snow country’

ori + kami 〉 ori-ŋami *ori-gami ‘origami’
The difference shown in the above examples between an underlying /g/ and a /g/ induced
by the rendaku of an underlying /k/ is quite puzzling. Itô and Mester (1997) proposed a
particularly elegant analysis of these facts. Their proposal accounts for, on the one hand,
the asymmetry between niwa-geta° / niwa-ŋeta° in (25) and yuki-ŋuni / *yuki-guni in (26),
but also, on the other hand, for the impossibility of velar nasalization occurring in
reduplicated mimetic words as shown above in (18), as well as for the existence of
minimal pairs such as doku-ga and doku-ŋa.
The analysis by Itô and Mester (1997) is cast within the framework of Optimality Theory.
Four main constraints are used:
(27)
Page 22 of 41
Consonants
*[ŋ ŋ is prohibited initially in a prosodic word.

*g Voiced dorsal obstruents are prohibited.
IDENTLS Lexical–Surface correspondents are identically specified for [nasal].
IDENTSS The bound form of a stem is segmentally identical with its
corresponding free form.
To account for the rendaku-induced [ŋ]’s, as in yuki-ŋuni for example, it is necessary, in

addition, to refer to the SEQVOI constraint (sequential voicing, i.e. rendaku), which
stipulates that the second component of a compound starts with a voiced consonant. This
constraint is the highest in the hierarchy together with *[ŋ. The relative hierarchy
between these two constraints remains unspecified.
(p.83) The analysis supposes that the two constraints *g and IDENTSS are unranked, or
freely ranked, with respect to each other, thus providing an account of the observed
variation. Free-ranking of these two constraints entails the derivation of two possible
outputs in cases where the second component starts with [g] in isolation. Recourse to
two different hierarchies, or to one hierarchy with two constraints left unranked with
respect to each other, constitutes a traditional treatment of variation in Optimality Theory.
The hierarchy (adapted) is as follows:
(28)
The choice between hierarchies (a) and (b) is left open by the grammar.
Compare now the tableaux for niwa° + geta° → niwa-ŋeta° and yuki + kuni° → yuki-ŋuni,
both with the nasal, with respect to the two hierarchies (accent will not be indicated in the
tableaux).
(29) niwa + geta ‘garden clogs’
Page 23 of 41
Consonants
(p.84)
(30) yuki + kuni ‘snow country’
[g] is allowed in internal position in niwa-geta (29a) because it satisfies the IDENTSS
constraint which is ranked higher than *g. In other words, niwa-geta is possible because
/g/ is identical to the autonomous surface form geta, whereas *yuki-guni in (30a or b) is
not because /g/ does not appear in the autonomous surface form kuni. In (29b), niwa-
ŋeta is the best candidate because it does not violate *g, which outranks IDENTSS in the
alternative hierarchy.
Page 24 of 41
Consonants
In this model, the possible variation between niwa-geta and niwa-ŋeta thus results from
the free-ranking of *g and IDENTSS. However, in yuki + kuni (30), the two different
ranking scenarios do not make any difference to the output. With either ranking, the same
candidate yuki-ŋuni is selected. Candidate *yuki-guni cannot emerge as the winner since
the /g/ consonant does not reflect a segment present in the autonomous form of the
lexeme kuni (violation of IDENTSS). Consequently, under both hierarchies, the winner is
yuki-ŋuni.
Lastly, to account for the case of reduplicated mimetic stems, in which /g/ occurring at the
beginning of the second part is never nasalized, Itô and Mester adopt McCarthy and
Prince (1995)’s analysis, calling upon a well-known constraint in Optimality Theory, the
constraint IDENTBASE-REDUPLICANT, which they place above *g.
(p.85) IDENTBASE-REDUPLICANT requires identity between the base and the

reduplicant, so a surface form like *gara-ŋara is not optimal. Neither is *ŋara-ŋara
because, according to the proposed hierarchy, prohibition to have a velar nasal ŋ at the
initial of a word is ranked very high. Under these conditions, gara-gara is the optimal
candidate.
The analysis by Itô and Mester is appealing in many respects. It has strong explanatory
capacity, and makes it possible to achieve a unified treatment of seemingly disparate facts,
in particular with regard to the obligatoriness of ŋ resulting from the rendaku of [k]. In
addition, the general principles which govern their analysis seem independent of the
formal framework they use, which still reinforces the interest of their approach, and is
likely to give it lasting value, irrespective of the evolutions of the theory. The g / ŋ
alternation basically comes down to a paradigmatic uniformity issue.
However, this analysis is not without its defects, in particular concerning the data. The
main criticism which one can formulate has to do with the empirical basis of the study. It
explicitly presents the patterns of optional or obligatory alternations as systematic,
implying that the data are firmly established and indisputable. However, this is far from
being so because exceptions are easy to find. For example, according to authors as
normative as Amanuma et al. (1989) or NHK (1998), who deal precisely with the variety
of Japanese adopted by Itô and Mester, certain compound words whose second
component starts with /g/ in the free base form, such as hiyori° + geta° ‘geta for dry
days’ or shi ro + goma° ‘white sesame’, have only one possible realization, the first with
the nasal [hijo¹ɾiŋeta], the second with the stop [ɕiɾogoma]. In addition, words containing a
rendaku frequently present variation between g and ŋ for speakers who do possess the
velar nasal, but the conditions which determine the variation remain unclear (Kamei,
1956; Kindaichi, 1942). It would also have been relevant to refer to the results of
Kindaichi’s survey, which we will present below, since it is precisely the inter- and intra-
speakers’ variations investigated by Kindaichi that Itô and Mester are interested in. In
sum, the data do not seem representative enough of the Japanese language at the end of
the twentieth century, and it does not either seem to be the variety described in
previous studies, for example Kindaichi’s. Interestingly, the study is supposed to concern
the variety of language spoken by the ‘older residents of the Yamanote area’ of Tôkyô,
Page 25 of 41
Consonants
‘which forms the basis for the modern standard language’ and ‘is reflected in standard
pronunciation dictionaries’. This is precisely the variety of language, taken from that very
same Yamanote area, which Kindaichi Haruhiko investigated in his 1942 study.13
(p.86) 3.10.2 Kindaichi’s Treatment (1942)

During the winter of 1941, the linguist Kindaichi Haruhiko conducted a sociolinguistic
survey in a junior high school of the Suginami district of Tôkyô, located in the Yamanote
area, which forms the basis of Standard Japanese. The survey consisted of the reading
out loud of a list of 13 words, comprising an internal velar likely to be realized as [g] or
[ŋ], by seventy 15- and 16-year-old students, born and raised in the capital.
Firstly, the results of this investigation provide evidence for the extreme variety of the
realizations among the speakers. Three different groups can be distinguished: speakers
who realize all the /g/’s as [g] (21 speakers), speakers who realize them as [ŋ] except in
the word ju ugo ‘fifteen’ (20 speakers), and speakers who present one or the other
realization (29 speakers).
Secondly, the survey establishes that the use of [g] or [ŋ] also depends on the words:
some lexemes are more frequently realized with [g] or [ŋ] than others. Thus ka igun ‘the
navy’ presents the [g] realization in 70% of cases, while ama-gasa (or ama-gasa°)
‘umbrella’ exhibits the nasal [ŋ] in more than 67% of the examples. But all the words
tested do not present a clear and constant tendency. There is only one word for which all
seventy speakers provide a uniform pronunciation: ju ugo ‘fifteen’ with [g]. Also, as
Kindaichi observes, ka igun ‘the navy’ and chu ugi ‘fidelity’, the two words most
frequently realized with [g] are Sino-Japanese lexemes. This suggests that this class of
word is less prone to velar nasalization. It will be noted that ama-gasa / ama-gasa°
‘umbrella’ and ha-gaki° ‘postcard’, which have the highest rate of [ŋ] realization, are
Yamato compound lexemes which contain rendaku. However, one should nevertheless
note that more than 30% of the speakers, including speakers who do have [ŋ] in some
other lexemes, do not nasalize the rendaku compounds ama-gasa (ama-gasa°) and ha-
gaki° at the time of the investigation in 1941. Moreover, Kindaichi observes that gi and
gu are less often nasalized than ga,ge, and go. He also raises the possibility that the [g]
realization occurs more frequently in the second mora than in the third mora of a word.
Lastly, his work supports the thesis that women tend to be less conservative than men
socio-phonetically, since they utter fewer [ŋ]’s than men.
We are thus clearly dealing with a variation phenomenon whose conditioning, both
internal and external, is complex. In addition to the strong inter-speaker diversity, quite
remarkable within such a homogeneous group as far as age and (p.87) sociocultural
background are concerned, the survey brings forth several other internal factors likely
to favour the appearance of the velar nazalisation: the lexical stratum of a given word, its
length, its degree of autonomy, the degree of lexicalization in the case of compounds, the
morphological status of the velar as a word boundary, the nature of the vowel which
follows the velar, as well as the position of the mora containing the velar in the word.
Page 26 of 41
Consonants
Noting, at the time of the investigation, that almost all the subjects over thirty have [ŋ] in
internal position, while this allophone was less widespread in the younger subjects’
realizations, Kindaichi predicted the progressive disappearance of [ŋ] to the benefit of [g]
in all contexts. Posterior works, in particular the linguistic survey whose results are
described by Inoue (1993), have regularly confirmed this prediction ever since. The
progressive disappearance of [ŋ] not only in Tôkyô but also in other areas of Japan where
this sound used to be widespread is under way.
Kindaichi also saw a structural reason for the disappearance of [ŋ]. According to him, the
existence of a velar nasal intervocalically in Japanese is a vestige of the time when all the
word-internal voiced stops were prenasalized. Thus formerly one had [mb], [nd], and
[ŋg] where the modern language generally has [b], [d], and [ŋ] ~ [g] (see also Chapter 4).
When prenasalized consonants began to undergo reduction to one single segment, [mb]
and [nd] passed to /b/ and /d/. Things went differently for [ŋg] because that sound
started to phonologize into /ŋ/ in a number of dialects. The reason is that since the velar
nasal /ŋ/ did not exist as a phoneme in the system, such phonologization of [ŋ] to /ŋ/ made
it possible to maintain or to create the following new oppositions: m/b, n/d, and ŋ/g, where
ŋ filled a gap. The reason why this did not happen in the case of [mb] and [nd], which did
not become /m/ and /n/ but rather /b/ and /d/, is because labial and alveolar nasals already
existed as established phonemes. The reduction of [mb] and [nd] to /m/ and /n/ would
thus have involved the loss of many oppositions; for example *kambe ‘wall’ becoming
*kame would have merged with kame ‘tortoise’, while kabe was not in competition with
any existing form. On the other hand, the evolution of kaŋge ‘shade’ to kaŋe was not likely
to create confusion since ŋ did not exist as a contrastive unit.
During the twentieth century, according to Kindaichi (1942), the tendency is that of [ŋ]
evolving back to /g/, on the model of the evolutions [mb] → /b/ and [nd] → /d/. The
disappearance of [ŋ] would thus constitute the final stage of a diachronic process of major
importance in the history of the Japanese language, the evolution of which can be traced
over several centuries: that of the denasalization of the prenasalized voiced obstruents, a
topic which will be addressed again in the next chapter.
(p.88) 3.11 /y/ (and Palatalization)

The dorso-palatal glide /y/ (IPA [j]) functionally stands as a consonant in the moraic units
ya, yu, yo. It can also appear after another consonant in the form of a palatalization. Every
Japanese consonant excluding the glides y and w has a palatalized counterpart known as
yôon . However, it is especially after the consonants /k/, /g/, /t/, /s/, /z/, and /h/, and in
Sino-Japanese words that palatalizations are most frequent. After /r/, palatalization is
generally rare, and impossible in the mimetic stratum (see section 3.13 below).
As seen in section 2.8, the presence of y does not change the prosodic weight of a given
rhythmic unit: kyu counts as one mora, just as ku,yu, or u do. The palatal element must be
considered as attached to the first position of the prosodeme (onset), as represented in
(31):
(31) Palatalized consonant
Page 27 of 41
Consonants
Palatalization is always noted y in phonological and Kunrei transcription. The Hepburn

romanization is less consistent because it renders the palatalized combinations /sy/, /ty/,
/zy/ as sh,ch, and j, as already indicated.
The element y only occurs before the vowels /a/, /u/, /o/. The combination of y with /i/ is
impossible in all lexical strata: *yi, *kyi, *myi, etc. do not exist. It seems that this has
always been so in Japanese.
The glide y is also absent before /e/, except in some very recent loans or in certain
dialects, in particular in the Kyûshû island. Recall though that she /sye/ [ɕe] occurs in
certain speakers’ speech instead of se [se] but in a non-distinctive way: for example she
/sye/ [ɕe] ‘back’ (standard se, see footnote 2, this chapter). But /ye/ did exist in the
archaic language. Phonetically, /e/ and /ye/ were distinguished until the middle of the ninth
century approximately before merging to /ye/ in a majority of dialects, as observed by
the European missionaries of the sixteenth century: yedo (for edo°, the ancient name of
Tôkyô), ye, modern e ‘image’. It is only from the seventeenth century on that the [je]
pronunciation of ye started changing into [e].
The two impossible combinations */yi/ and */ye/ concern vowels which have a palatal
articulation in common with the dorso-palatal glide /y/. The same type of (p.89)
restriction applies to /w/, the dorso-velar glide, which does not appear before the back
(or velar) vowels /u/ and /o/.
Consonantal palatalization is not original in Japanese. It is generally considered to have

appeared in the language under the influence of Chinese loans. Indeed, palatalizations are
particularly frequent in Sino-Japanese morphemes. They occasionally occur in Yamato
words like kyo u ‘today’, but this is always the result of a secondary development,
already mentioned in section 2.7.2. In the familiar register, palatalizations are also
frequent and generally not recorded in writing. We find sore ha° 〉 sorya° ‘this + Topic
marker’, itte ha° 〉 itcha° (/ittya/) ‘going + Topic marker’ (in these two examples,
remember that the topic marker ha is actually pronounced wa), to iu (actually realized as
to yuu°) 〉 chuu° (/tyuu/) ‘so called’ (all of Yamato origin). The appearance of the palatal
element is explained by the presence of the front vowel /e/ in the first two examples, and
by that of /i/ or /y/ in the last one. Palatalizations are also widespread in mimetics (see
below), as in hunya-hunya° ‘flabby’, as well as in recent Western loans. Note also the
frequent insertion of y after /k/ or /g/ in Western words like kyapashitii {capacity}, or
kyarameru° {caramel}, that reflects the place of articulation of the velar stops /k/ and /g/
in English and other European languages, which are more phonetically fronted than the
Japanese equivalents.
Page 28 of 41
Consonants
There exist some CyV / CV (almost) free alternations, such as kyu u ~ ku ‘nine’, sha ke
/syake/ ~ sa ke /sake/ ‘salmon’. The alternations shu ~ shi and ju ~ ji (/syu/ ~ /si/, /zyu/ ~
/zi/), as in Shinjuku° / Shinjiku° (place name), already discussed in section 2.4, are very
frequent (this phenomenon is termed chokuonka ? ? ). Lexicalized alternations of the
sort yV / i, such as yuku° / iku° ‘to go’, yuu° / iu° ‘to say’, yo i / i i ‘good’, can also be
found.
In a number of cases, palatalization is secondary and works like a phonaestheme

associated with the connotations of childishness, instability, uncoordinated movement,
diversity, lack of elegance, excessive energy. This is particularly so in mimetic words
(Hamano, 1998, originally published in 1986). Thus beside po ta-pota, which suggests an
idea of dropping, one finds po cha-pocha (/potya-potya/), which refers to splashing. This
type of palatalization normally occurs on the rightmost coronal consonant of a bimoraic
mimetic root (32a). If the rightmost consonant is not a coronal, then the initial consonant
will undergo palatalization, whatever its place of articulation (32b). Although Alderete and
Kochetov (2009) show that, because of the paucity of relevant examples, there is no real
empirical basis supporting the assumption that palatalization affects the leftmost of two
non-coronals (pyo ko-pyoko / *po kyo-pokyo in 32b) or the rightmost of two coronals (do
sha-dosha / *dyo sa-dyosa 32a), the overall argument put forward by Hamano (1998) and
subsequently by Mester and Itô (p.90) (1989) remains valid and in conformity with
native speakers’ intuitions.14 The only exception to this rule is /r/ (32c), which cannot
palatalize when it occurs in the second mora of the root, an issue which will be further
discussed in section 3.13.
(32)
a. C2 of mimetic root = coronal
ka ka *kyasakyasa ‘dry objects scratching each

sakasa syakasya other’
do do *dyosadyosa ‘something falling heavily’
sadosa syadosya
b. C2 of mimetic root = non-coronal
po kopoko pyo kopyoko *pokyopokyo ‘hopping around’

za buzabu zya buzyabu *zabyuzabyu ‘splashing’
c. C2 of mimetic root = /r/
no ronoro nyo ronyoro *noryonoryo ‘wriggly and curving movement’

ko rokoro kyo rokyoro *koryokoryo ‘looking around restlessly’
Such palatalizations of an expressive nature occur also sporadically elsewhere: thus, the
personal suffix -san becomes -chan (/tyaN/) after children’s names and diminutives (note
also in this example the passage of /s/ to /t/, which can be interpreted as consonant
Page 29 of 41
Consonants
strengthening, see section 3.6). Overall palatalization is also characteristic of baby talk.
3.12 /w/
The labiovelar glide /w/ is slightly less rounded than its English counterpart (for instance
in way). Its phonetic realization is between that of the symbols [ɰ] and [w] of the IPA. It
can be regarded as the semi-vocalic version of the vowel /u/ [ɯ] from the articulatory
point of view, but functionally, it behaves as a consonant.
In the modern language, /w/ occurs only before /a/. The combinations /wi/, /we/, and
/wo/15 have all existed at earlier stages of the language but have disappeared in Modern
Standard Japanese. However, they are making a timid come-back in Western loans. On
the other hand, */wu/ never existed. [o] o and [wo] wo merged around the year 1000 to
wo, a pronunciation which was still heard by the first European missionaries in the
fifteenth and sixteenth centuries. They wrote uo or vo for お (read o in Modern Japanese).
The group [wi] wi (p.91) reduced to [i] i around the thirteenth century. As for [we] we,
it first merged with ye around the thirteenth century. Thereafter ye evolved to modern
[e] (see the preceding section). The first Western romanizations of the Muromachi period
write coye for the modern form ko e ‘voice’, which in turn comes from an earlier form
kowe. The existence of w in this word is also visible in the allomorphic form kowa- for koe,
which appears in a number of compounds such as kowa-iro° ‘quality of voice’ (on the
alternation e/a in compounds, see section 2.3).
The reasons for these changes, which involved a significant decrease in the sound
possibilities of the language, can first be found in the low functional load of the oppositions
at hand, i.e. the small number of minimal pairs involving an opposition between V and wV.
Since the moras consisting solely of a vowel occurred only at the beginning of
independent simplex words in Archaic Yamato Japanese, this was the only position in
which the loss of the glide was likely to create a new homophony. In Sino-Japanese words,
the sequences ye,we,wi, and wo were too rare to cause any problematic opposition
losses. The disappearance of ye, we, wi, and wo also illustrates a strong trend in the
phonology of Japanese (see the charts in section 3.15 for statistical data): maximal contrast
is favoured between the two constituents of the basic prosodeme (the mora), so that the
statistically most frequent CV combinations mostly involve a voiceless stop + vowel, or a
nasal + vowel (a noteworthy exception is the combination r + V, which will be addressed
in the following section). So it is hardly a surprise that the combinations between a semi-
consonant and a vowel articulatorily close to it (w + u / o and y + i / e, as well as w + e)
were the first to disappear.
Moras of the CwV shape (called gôyôon in the traditional terminology) existed up
until recently in certain Sino-Japanese words, for example okwashi お ‘cake’, gwaikoku°
‘foreign country’. They reflect the presence of a labial glide in the Chinese original
forms. Old Chinese accepted /w/ after a large variety of consonants, but, apart from a
small number of exceptions attested in documents of the Heian period, it is only after the
velar consonants /k/ and /g/ that /w/ could be found in Japanese. Although the
combinations /kwa/, /gwa/, /kwe/, /gwe/, /kwi/, and /gwi/ all existed, only /kwa/ and /gwa/
have been maintained until the middle or end of the nineteenth century, and still exist
Page 30 of 41
Consonants
nowadays in certain dialects, mainly in the Tôhoku or Kyûshû areas.
The tendency towards w-lenition seems to continue in the modern language: /w/ is often
deleted before /a/ in very familiar speech. Thus one will often hear maaru° instead of
mawaru° ‘to turn’ (see also the remark in footnote 5, Chapter 2), bi a for bi wa ‘medlar
tree’, akannai for wakaranai ‘not to understand’, korya° for kore wa° ‘this’, or atashi°
for watashi° ‘I’ (but there exists a pragmatic difference between the two forms of this
pronoun; watashi° is rather neutral whereas atashi° has a girly connotation). More
generally, this phenomenon seems to pertain to an overall weakening process of the labial
(p.92) articulations in Japanese. Recall that */p/ has evolved to /h/ or to zero, that /w/ has
disappeared before /i/, /o/, and /e/, that the labiovelar /kw/ and /gw/ have reduced to /k/
and /g/, and that /u/ is only very slightly rounded. Even the nasal /m/ is no exception to
this: in the familiar language, it sometimes undergoes total deletion in fast tempo speech,
for example suimasen for sumimasen ‘excuse me’. One might wonder whether the
causes for this tendency to delabialization are not cultural rather than properly linguistic.
The progressive disappearance or near disappearance of labials in the phonological
system might be related to a search for a certain immobility or facial impassibility. Hagège
and Haudricourt (1978) refer to a cause of a similar nature to explain the absence of
labials in the Iroquoian adult language. Note, in addition, that lip protrusion or
exaggerated labialization in the articulation is culturally codified as an expression of anger
in Japanese. Conversely, delabialization is connoted as a mark of calm and self-control,
which are martial qualities eminently praised in the traditional culture.
3.13 /r/
The prototypic realization of the only Japanese liquid /r/ is [ɾ], the apico-alveolar flap.
According to Matsuno (1971), [ɾ] should be considered the neutral realization of the
rhotic in the language because its articulation is central compared to other variants.
However, /r/ displays a large number of social, geographical, or combinatorial variants.
Outside [ɾ], the following phonetic (social or regional) realizations are attested: [l], [ᶩ], [r],
[rː], [d], [ɽ], [ɮ].
The apico-alveolar lateral [l] is a common variant, frequent before palatalized vowels
(rya,ryu,ryo) and in young women’s speech (Ohnishi 1987; Tsuzuki and Lee 1992).
Retroflex [ᶩ] is also encountered under the same conditions. The short and long apical
trills, [r] and [rː] are socially marked variants, characteristic of colloquial or even vulgar
Tôkyô male Japanese. For instance, street thugs and yakuzas (gangsters) are easily
recognized by their strongly trilled r, at least in the movies, where it is one of their
conventional attributes. The higher the number of trills, the more socially marked as
vulgar the speech will be. The voiced alveolar stop [d] is a combinatorial variant which is
frequent word-initially in certain dialects, or in children’s speech. It can also occur word-
internally. The retroflex [ɽ] might be encountered initially before /u/, or intervocalically in
sequences such as /ere/, /ara/, /uru/, /oro/ (Tsuzuki and Lee, 1992). The voiced lateral
fricative [ɮ] is a combinatorial variant occurring before the high vowels /i/ and /u/. It is
also the most common realization of /r/ in some Ryûkyûan dialects. Phonetically, /r/ is also
by far the shortest of all Japanese consonants (Sagisaka and Tohkura; 1984, Kurematsu,
Page 31 of 41
Consonants
1997).
(p.93) In addition to its phonetic diversity, Japanese /r/ stands out as a segment
exhibiting many idiosyncratic peculiarities, which make it unique in the phonological
system of the language. In Archaic and Old Japanese, /r/ does not appear at the beginning
of independent Yamato morphemes. Nevertheless, although absent at the beginning of
Yamato lexical words, /r/ is paradoxically the most frequent (or second most frequent
depending on the counting method) of all consonants inside Yamato words (including
mimetics), in Old Japanese, and in Modern Japanese alike (see Labrune, 1993, and
Labrune, forthcoming, for a detailed presentation of various statistical data, and Figures
3.1, 3.2, 3.3, and 3.4 at the end of this chapter).
In Old Japanese, two /r/’s never co-occur within a single word, that is, there can be only
one /r/ per word. This co-occurrence restriction is especially remarkable in verb stems.
Whereas -ru is the most frequent verbal inflectional ending in Old Japanese (Yoshida K.,
1976:87, 101), it is not attested after roots which already contain an /r/ (Kuginuki, 1982).
Thus, while kaheru,inoru,musaboru and so on are well-formed and attested Old Japanese
verbs, forms with more than one /r/ such as *kiroru, *aramaru, or *somoriru are
impossible, and are indeed unattested, but for one exception (hiroru ‘to spread, to
widen’).16 The same type of co-occurrence restriction is also operative in Old Japanese
nouns and other parts of speech.
Moreover, the distribution of /r/ within words is peculiar, since /r/’s are more likely to
occur late in words. Kuginuki (1982) establishes that out of 614 words of the archaic
language containing a rhotic, 543 occurrences (88.4%) of those /r/’s appear in the last
mora of the word. In other words, the closer to the end of the word, the higher the
probability for finding /r/. For Kuginuki (1982), such a distributional pattern makes sense
if one supposes that /r/ developed relatively late in the history of Japanese. His hypothesis
is that /r/ was originally added to the phonemic inventory in order to increase the length
of words, which were mostly one or two mora long in pre-archaic Japanese. Japanese
being a suffixing language, these newly added r-moras are expectedly most frequent at
the end of words.
Another remarkable feature of /r/ in Old Japanese is that it stands in complementary

distribution with the zero consonant since moras made of a single vowel never appear
word-internally, and r does not word-initially.
In a well-known paper dealing with palatal prosody in Japanese mimetics in relation to

feature predictability and underspecification in Modern Japanese (p.94) (see examples
(32) above), Mester and Itô (1989) claim that /r/ is the unmarked sonorant of Japanese
and that it is underspecified for the feature [Coronal].
Building on Kuginuki’s insight and on Mester and Itô (1989)’s proposal that /r/ is actually
the unmarked sonorant of the system, I have argued in Labrune (1993, forthcoming) that
/r/ primarily developed in Proto-Japanese as a default, epenthetic consonant in the
intervocalic position by virtue of an ‘Emergence of the Unmarked’ mechanism (McCarthy
Page 32 of 41
Consonants
and Prince, 1994) and that the conditions of its development bear on its present-day
characteristics.
The evidence suggesting that the phonological behaviour of /r/ is that of a phonologically
inert, transparent consonant that lacks phonological content is the following.
First, as already discussed in section 3.11, /r/ fails to undergo palatalization, an important
phonological process which occurs in a systematic fashion in the mimetic stratum
according to Hamano (1998).
Recall that /r/ behaves in an exceptional manner in mimetics because it cannot be

palatalized when it occurs in the second mora, so that no ronoro (32c) does not yield *no
ryonoryo but nyo ronyoro. Moreover, the presence of /r/ in the root does not block the
palatalization of a non-coronal in the first mora. So /r/ actually behaves like a non-coronal
with regard to the palatalization process depicted in (32).
Note that palatalized /r/’s can be encountered in lexical strata other than mimetics, so that
/ryV/ is not an impossible sequence in Japanese. However, the difference between
mimetic palatalization and non-mimetic palatalization results from the fact that in mimetics
palatalization acts as a ‘feature-sized morpheme’ (Mester and Itô, 1989) which can be
productively attached to a root under the conditions stated above.
The second process to which /r/ is transparent in Japanese mimetics, and also in other
strata of the lexicon, is gemination. For instance, in -ri suffixed mimetic adverbs (Kuroda,
1967; Mester and Itô, 1989 citing a personal communication by Poser), when the second
consonant of a C1 VC2V root is voiceless, suffixation of the adverbial ending -ri may cause
C2 to undergo total gemination in cases where it is a voiceless obstruent (33a), or partial
gemination (i.e. prenasalization) in cases where it is a voiced obstruent or a sonorant
(33b). However, in cases where the consonant in question is /r/, neither gemination nor
prenasalization can normally occur (33c).
(33)
a. C2 = [−voiced] → gemination
Base -ri adverbial

bata battari ‘with a bang’
kaki kakkiri ‘exactly’
goso gossori ‘entirely’
(p.95)
b. C2 = [+voiced] → prenasalization
gena gennari ‘to satiety’

shimi shinmiri ‘intimately’
boya bon’yari ‘absent-mindedly’
Page 33 of 41
Consonants
yawa yanwari ‘gently’

koga kongari ‘to be nicely roasted brown’
maji manjiri ‘sleepless’
c. C2 = /r/ → nothing happens
koro korori ?*korrori ?*konrori ‘without effort’
Fully geminated /r/’s are also unattested outside the mimetic lexicon in the Yamato and
Sino-Japanese strata, and they are only marginally reported in Western borrowings and
in some recent mimetic derivatives (Schourup and Tamori, 1992:137). As for the moraic
nasal /N/ + /r/ sequence, it does occur in Sino-Japanese and Western borrowings, and in a
few mimetic forms. However, in such cases, it represents a recent development and is to
be phonologically analysed as a combination of two distinct segments rather than the
result of a prenasalization process in the strict sense.
According to Mester and Itô (1989), total gemination of /r/ is impossible because it
violates the Nasal Coda Condition requiring all voiced sonorant codas to be nasal. Even if
one does not adhere to a syllabic analysis of Japanese but to a moraic one according to
which /N/ stands as an autonomous prosodeme (mora) as we do in this book, the basic
interpretation by Itô and Mester remains valid: /N/ needs to acquire its place articulation
features from the following consonant. So partial gemination (= /N/ insertion) is impossible
because /r/ is underspecified, and thus has no distinguishable parts available for separate
linkage. Schourup and Tamori (1992) have criticized this analysis, arguing that the non-
occurrence of palatalization with /r/ in mimetics is best explained by articulatory difficulty.
However, one can object that articulatory factors alone cannot account for the many
other properties of /r/ in Japanese.
Thirdly, /r/ is the most unstable of all Japanese consonants, both diachronically and
synchronically. Throughout the history of Japanese, rV moras have frequently undergone
syncope (de aru 〉 da (copula), karite 〉 kate, ka te ‘provisions’; see also Kishida (1984) for
additional examples) or unexpected addition (paragoge, kabu° 〉 kabura° ‘turnip’, shippo /
shippori (accent unknown) ‘tail’ [dialectal]; see also Labrune (1998b)). As a recent trend
of Tôkyô Japanese, it has also been observed that rV sequences frequently turn into /N/
or /Q/, for instance wakaranai 〉 wakannai ‘not to understand’, sou suru to 〉 sousutto
‘doing this’ (Akinaga, 2008). Moreover, unlike most other consonants, /r/ is never the
cause of an assimilation process in Japanese, that is, there are no cases where a (p.96)
consonant would assimilate to /r/, whereas /r/ frequently assimilates to a surrounding
segment (Labrune, forthcoming).
The liquid also plays a central role in the morphology. It is crucially involved in the verbal
flexion, where it can be regarded as epenthetic (de Chene, 1985; Labrune, 1996), as the
following data illustrate:
(34) Verbal inflexion
Page 34 of 41
Consonants
Consonant final base Vowel final base

‘to write’ ‘to see’
basic form ka k-u mi -r-u
negative kak-anai mi -nai
hypothetic ka k-eba mi -r-eba
nominal ka k-i (kak-i) mi
polite kak-i-masu mi-masu
passive kak-areru mi-r-areru
imperative ka k-e mi -ro
Observe that, before vowel-initial endings (-u,-anai,-eba, etc.), no consonant surfaces

when the base ends in a consonant, contrary to what happens after a vowel-ending base.
According to de Chene (1985) and Mester and Itô (1989), consonant-stem suffixes
display the basic form of the suffixes, and initial /r/ in vocalic-stem suffixes is epenthetic.
What is significant here is that the surfacing consonant is precisely /r/. The verbal
morphology of Old Japanese provides further arguments for analysing /r/ as an
epenthetic consonant (Labrune, 1996).
Furthermore, /r/ is also extremely frequent at the beginning of several other nominal,
adjectival, and verbal suffixes in the pre-modern language, such as -ra (plural,
directional), -ra / -ro (adverbial), -raka (adjectival ending), -raku (nominalizer), -ru
(passive, potential), -ri (adverbial), -ri (aspectual auxiliary), -re (deictic suffix), -reru
(passive, potential, honorific auxiliary), -ro (imperative suffix), and -ro (a particle of
obscure function in Archaic Japanese). Most of these r-beginning morphemes are attested
in the archaic language, and are still widely used in Contemporary Japanese.
The behaviour of /r/ in the phonetics, phonology, and morpho-phonology of Japanese is

thus characteristic of what any theory of phonology, whether structuralist, generativist,
or OT-ist, would recognize as an unmarked, default segment.
3.14 New Consonants

A number of new phonic possibilities have recently developed in the Japanese language
due to the influence of borrowing. Two different types of new (p.97) consonants can be
distinguished: those which result from the phonologization of sounds already existing in
the language but with no phonemic status, like f [ϕ] or v [β], and those which represent
new phonotactic combinations, that is, a sound which is already granted phonemic status
in the native or Sino-Japanese system has come to acquire new combinatorial possibilities.
For instance /y/ y, /w/ w, /ty/ ch, /zy/ j, /sy/ sh now combine with e, /d/ d now occurs
before i or u as [d], /w/ w occurs after other consonants (kwa,gwo), and ts before vowels
other than u, as shown in the following chart.
(35)
Page 35 of 41
Consonants
she, je, che: [ɕe]シェ [ʑe]ジェ [ʨ͡e]チェ

tsa, tsi, tse, tso: [ʦ͡ a]ツァ [ʦ͡ i]ツィ [ʦ͡ e]ツェ [ʦ͡ o]ツォ
ti, di, tu, du: [ti]ティ [di]ディ [tɯ]トゥ [dɯ]ドゥ
tyu, dyu: [tjɯ]テュ [djɯ] デュ
fa, fi, fe, fyu, fo: [ϕa]ファ [ϕi]フィ [ϕe]フェ [ϕjɯ]フュ [ϕo]フォ
va, vi, vu, vyu, ve, vo: [βa]ヴァ [βi]ヴィ [βɯ]ヴ [βjɯ]ヴュ [βe]ヴェ [βo]ヴォ
ye: [je]イェ
wi, we, wo: [wi]ウィ [we]ウェ [wo]ウォ
kwa, kwi, kwe, kwo: [kwa]クァ [kwi]クィ [kwe]クェ [kwo]クォ
gwa, gwi, gwe, gwo: [gwa]グァ [gwi]グィ [gwe]グェ [gwo]グォ
Concerning the three last series beginning with w-,kw-, and gw-, note that they are not
always realized as one mora. One often hears the bimoraic sequences [ɯi], [ɯe] for wi
and we, and [kɯa], [gɯi], etc. for kwa,gwi, etc.
Most of the above combinations are only found in loanwords which have been borrowed
in the past twenty or thirty years, and in the speech of speakers whose sociolinguistic
profile is urban, educated, and feminine (Inoue, 2002). It is important to make a clear
distinction between the katakana transcription, which may contain sequences such as
those cited above (or even other sequences), and the actual pronunciation, which is often
more conservative, and does not always faithfully reflect the kana spelling. Some people
(mainly older speakers) write pa atii パーティー {party} but actually (p.98) pronounce
[¹paːtɕ͡ iː] pa achii or [¹paːteː] pa atee, thus avoiding the new phonic combination [ti].
The following chart provides some examples of recently borrowed words containing new
phonetic possibilities (underlined).
(36)
f fa mirii [¹ϕ amiɾiː] ファミリー {family}

ti ti ssyu [¹tiɕɕɯ] ティッシュ {tissue}
tsi tsiiru [¹ts͡iːrɯ] ツィール {Ziel} (German ‘goal’)
she shepaado [ɕ e¹paːdo] シェパード {shepherd (dog)}
kwo kwo otsu [¹kwoːts͡ ɯ], [¹kɯoːts͡ ɯ] クォーツ {quartz}
v vinteeji [βin¹teːʑi] ヴィンテージ {vintage}
ye yerusaremu [jeɾɯ¹saɾemɯ] イェルサレム {Jerusalem}
The combinations fa, ti, tsi, she, kwo, and ye in the above examples actually represent a
broadening of the phonotactic possibilities of already existing segments. These either exist
Page 36 of 41
Consonants
as allophones of a given Japanese phoneme (for instance [ϕ] is an allophone of /h/, [ʦ] an
allophone of /t/), or as true phonemes with limited distribution in the rest of the lexicon.
This is the case for y, which only occurs before a,u, and o, and for w, which only occurs
before a, in the Yamato and Sino-Japanese strata. In this case, it is the combination of y
and w with vowels other than those admitted in Yamato or Sino-Japanese words which
represents an innovation. It is interesting to note that some of these combinations have
existed at an earlier stage of the Japanese language, or still exist in dialects, for instance
kw,tsa,ye, she.
The sound transcribed as ヴ in kana (v in Hepburn) is never realized as a labio-dental

voiced fricative. Its most common realization seems to be [b]. It is pronounced [β] by
some speakers, a sound which is an occasional variant of /b/ in the intervocalic position
(Kamiyama T., p.c., and Saitô, 1997) for certain speakers in Yamato and Sino-Japanese
words. However, this Japanese [β] seems to be much less fricative than the
corresponding Castillan Spanish sound in lobo for instance.
In sum, we can say that all the ‘newly introduced sounds’ are not really new: they
already exist in the Japanese language. The only innovation is that they are now granted a
phonemic status.
Other sounds still await a proper and non-ambiguous transposition. Let us mention for
instance those consonants which function as a syllable coda in the source language. Both
English fat and Italian fatto are adapted as ファットfa tto, and there is no way to distinguish
them in the Japanese adaptation even in the kana writing. The English interdental
fricatives [θ] and [ð] are not distinguished (p.99) from [s] and [z]. The same is true of l
and r. Bari and Bali are transcribed and pronounced in an identical manner: バリ [¹baɾi].
The pairs si and shi,zi and ji, although they sometimes receive a different orthography in
kana, respectively as スィ and シ, ズィ and ジ, are never phonetically distinguished in the
adapted form of loanwords by Japanese speakers. This is the source of an important
number of homophonies.
Note that older loanwords containing sounds or sound combinations which were
unknown in Japanese used to be adapted under patterns more in line with the native
phonology, as the following examples illustrate:
(37)
{telephone} 〉 te rehon [¹teɾehoᶰ] テレホン rather than *[¹teɾeϕoᶰ]

{visa} 〉 bi za [¹biza] ビザ rather than *[¹βiza]
{ticket} 〉 chiketto [t ͡ɕi¹ketto] チケット rather than *[ti¹ketto]
Today, these words would presumably be adapted, at least orthographically, as テレフォン te

refon [¹teɾeϕoN], ヴィザ vi za [¹βiza], and ティケット ti ketto [¹tiketto]. We do actually find newly
formed compounds containing the word telephone such as terefon shoppingu {telephone
Page 37 of 41
Consonants
shopping} or terefon redii {telephone lady} with a f [ϕ] (Kamiyama Takeki, personal
communication).
3.15 Relative Frequency of Consonants

The charts below provide data concerning the frequency of the consonants in texts, on
the one hand, and in the lexicon, on the other hand, both in Ancient and Modern
Japanese.
For Archaic Yamato Japanese, the data concern textual (token) frequency. They are given
in Figure 3.1 (based on Ôno (1980), Man'yôshû corpus). The
Figure 3.1. Textual frequency (in %) of Archaic Japanese consonants

(adapted from Ôno, 1980)
(p.100) symbol F denotes the ancestor of /h/; ’ denotes the zero consonant (onsetless
vowel).
What we observe is that the alveolar nasal /n/ is the most frequent of all consonants. This
remarkable textual frequency is undoubtedly due to the fact that several grammatical
words among the most widely used contain a /n/ (ni,nari,no, etc.). Note that the four
voiced obstruents /g/, /b/, /d/, and /z/ occupy the last four positions, preceded by the two
semi-consonants /y/ and /w/.
Figure 3.2 provides data about the lexical (type) frequency in word-initial position. We see
that /’/ (the zero consonant) is the most frequent of all, since actually nearly one fourth of
all Japanese archaic words start with the zero consonant, that is, with a vowel.
The bar graph in Figure 3.3 provides data for two-mora Yamato nouns of the modern
language. It gives the lexical (type) frequency (in absolute value), and
Figure 3.2. Lexical frequency (in %) of Archaic Yamato Japanese

consonants for the initial of words (based on the entries of the Nara
Japanese Dictionary, Jidai Betsu Kokugo Daijiten, jôdai-hen, 1967).
Page 38 of 41
Consonants
Figure 3.3. Lexical frequency in absolute value of consonants

according to their position in bimoraic Yamato nouns in the modern
language (KKK, 1984).
(p.101)
Figure 3.4. Textual frequency of Modern Japanese consonants

(according to Imae, 1960, cited by Hayashi O., 1982).
makes a distinction between initial and medial positions in two-mora nouns. The figures for
the glide y are missing because the data provided by the source cannot be exploited. As
we have already remarked in section 3.13, /r/ is the most frequent consonant word-
medially. It is also interesting to observe that there exists a discrepancy in the
distribution according to word position for /g/, /z/, /d/, /h/, /b/, /r/ and /’/ (see the sections
of the relevant consonants for comments).
Finally, Figure 3.4 presents the textual (token) frequencies of the consonants in Modern
Japanese. The counting of the source being based on the kana spelling (except for the
particles ha は and wo を, which are not distinguished from wa わ and o お), the figures for
the zero-initial ’ are misleading because they also include the second parts of long vowels:
for example, one occurrence of /’/ is counted in the sequence kou こう, which has been
interpreted as ko’u. The frequency of /’/ must thus be considered inferior. According to
my own estimation, it actually ranges between 10% and 13%.
Here, (j) denotes palatalization (for example y in tya,kyo), while y represents the initial
segment in the moras ya,yu, and yo. The total does not reach 100% because the
frequencies of the special segments /N/ and /Q/ are missing (see section 5.6).
As already mentioned in section 3.12, and if we set aside the special cases of /’/ and /r/ for
the reasons discussed in section 3.13, Japanese tends to favour ‘strong’ consonants (i.e.
consonants which are most consonantal, that is voiceless obstruents and nasals) in the
onset position of its prosodemes (moras). It thus follows the universal tendency of the
Page 39 of 41
Consonants
world’s languages which tend to obey a principle of maximal sonority contrast between
CV.
Notes:
(1 ) However, it is generally admitted that in Ancient Japanese (and probably also in
Middle Japanese, that is, until the fifteenth or sixteenth centuries), /t/ was realized as a
plosive before all vowels, including /u/ and /i/: [ta], [ti], [tɯ], [te], [to]. The same goes for
/d/.
(2) A few sporadic occurrences of the she /sye/ combination are encountered in Yamato
words, but they are all free variants of se, for instance mi sete / mi shete ‘show me!’.
This alternation must probably be interpreted as dialectal, or as an example of affective
palatalization, see also p. 66.
(3) This should be considered in parallel with the behaviour of the other Japanese fricative
/h/, after /Q/. /h/ automatically becomes /p/ in cases of gemination in Yamato and Sino-
Japanese words. Voiceless fricatives thus seem to resist gemination, and they tend to be
transformed into occlusives when preceded by the gemination segment /Q/. One can thus
posit a correspondence /Qs/ = [tss] ~ /Qh/ = [pp], which would be attributed to a fortition
process. Note also that even in Western loanwords, /h/ and /s/ undergo gemination (as
[hh] and [ss]) less often than the other consonants (Kawagoe and Arai, 2002).
(4) The adverb yahari is a mimetic which contains an intervocalic [h] root internally and is,
as such, exceptional. It is sometimes considered to be etymologically a compound of ya
‘arrow’ and hari ‘tense’ ( り).
(5) Thank you to Kamiyama Takeki for pointing out to me the Ladefoged references.
(6) [pp] is not the only geminate which can mark compounding in parallel to rendaku. A few
examples can also be found with other consonants, for instance korekkiri (kore° + kiri)
‘this and only this’, de cchiri (de + shiri) ‘chubby buttocks’.
(7) It also seems that /h/ had remained a bilabial before /i/ until the nineteenth century in
the Western dialects, according to the descriptions by Aston (1872) and Hoffmann
(1868).
(8) Concerning the evolution we 〉 ye 〉 e, see sections 3.11 and 3.12.
(9) It is interesting to note that in the words ha ha ‘mother’’ and ho ho ‘cheek’ the
intervocalic fricative has first followed the general evolution rule h 〉 w, as indicated by the
Jesuit transcriptions faua for haha, and fou (〈 *howo) for hoho. However, a phonetic
reversal has occurred and the modern forms have reverted to ha ha and ho ho.
(10) Pronunciation dictionaries sometimes transcribe the moras beginning with ŋ as カ゜キ゜ク゜
ケ゜コ゜.
(11 ) An electronic version of the maps of the Linguistic Atlas of Japan (Nihon Gengo
Page 40 of 41
Consonants
Chizu) issued by the Kokuritsu Kokugo Kenkyûjo (National Institute for Japanese
Language and Linguistics) is available at: http://www6.ninjal.ac.jp/laj%20map/04/01/.
(12) For the sake of legibility and consistency, some notational and presentational
adaptations are made in the presentation below.
(13) The speakers targeted by Kindaichi’s survey in 1941 were aged fifteen or sixteen
and were Yamanote residents. They must have reached an age over seventy in 1997, the
date Itô and Mester published their paper, that is, they precisely fall into the category of
speakers that Itô and Mester’s study is supposed to concern.
(14) For analyses of this secondary palatalization, which may be regarded as an

autosegment, and some of its theoretical consequences, see Hamano (1998) and Mester
and Itô (1989).
(15) The object particle を is sometimes romanized as wo, but this is a purely orthographic
convention, that does not reflect the presence of the glide /w/ in the actual realization of
the particle.
(16) This principle is not totally preserved in Modern Japanese (cf. oriru ‘to get down’,
ireru° ‘to insert’), but even in the modern language, most verbs containing two
occurrences of /r/ are compounds involving two stems, or are derived by adjunction of
the suffix -eru, which is an innovation of the Pre-Modern language.
The phonology of consonant voicing

Laurence Labrune

Print ISBN-13: 9780199545834
DOI: 10.1093/acprof:oso/9780199545834.001.0001

Laurence Labrune
DOI:10.1093/acprof:oso/9780199545834.003.0004
Chapter 4 first presents a comprehensive overview of the issue of consonant voicing in

Japanese, which is characterised by a number of specific and rather outstanding
properties, including low frequency, phonotactic restrictions, historical instability and
transparency, failure to undergo gemination and so on. It also considers their historical
development and status in the writing system and offers an original analysis of the internal
structure of the Japanese voiced obstruents within the framework of autosegmental
phonology and feature geometry. This chapter also provides a detailed presentation of
rendaku, one of the most studied and debated issue in Japanese morpho-phonology, and
a section on post-nasal voicing, before concluding that voicing could be regarded as a
supra-segmental feature rather than as a segmental one in Japanese.
Keywords: gemination, rendaku, post-nasal voicing, Japanese voicing, low frequency, phonotactic
restrictions, historical instability, historical transparency
Page 1 of 29
The four voiced obstruents /b/, /d/, /g/, and /z/ (dakuon , literally ‘impure, or muddy,
sounds’ in the traditional Japanese terminology), display several intriguing properties. In
the present chapter, we begin by examining the general phonological properties of voiced
obstruents (section 4.1). We will then proceed to an overview of rendaku (‘sequential
voicing’, section 4.2) and post-nasal voicing (section 4.3), and end with a discussion
concerning the similarities between Japanese voicing and some supra-segmental features
such as tone and accent (section 4.4).
4.1 General Properties of Japanese Voiced Obstruents
4.1.1 Limited Distribution, Low Frequency, And Co-occurrence Restrictions

The voiced obstruents have a limited distribution, especially in Yamato words. They do
not normally occur at the beginning of independent morphemes, except in a few special
cases which will be examined below. Also, voiced obstruents present an extremely low
frequency in Yamato words, as shown by the statistical data provided at the end of the
preceding chapter. Furthermore, two voiced obstruents rarely co-occur within the same
root. Itô and Mester (1986) state that huda° ‘label, tag’ and buta° ‘pig’ are possible
words of Yamato Japanese, but that *buda is not. However, forms such as de bu ‘fatty’ or
gaburi ‘swallowing at once’ do exist. These are certainly rather marked, but not really
more so than buta ‘pig’, whose structure is also rare and highly marked because of its
initial voiced obstruent. So the paucity of bimoraic words containing two voiced
obstruents is more probably a statistical consequence of the fact that words beginning
with a voiced obstruent are rare, and that forms containing a voiced obstruent in the
second mora are far less frequent than those containing an unvoiced obstruent in the
same position.
The voiced obstruents which appear word-initially in Yamato words come from the
following sources. First, one finds a number of cases which result from secondary voicing
of an originally unvoiced obstruent, whose phono-pragmatic function is to introduce a
pejorative or expressive connotation. This kind of initial (p.103) voicing generally signals
the negative character (unpleasant, disgusting, dirty, big, heavy, etc.) of the referent
(Komatsu, 1981). For example sama ‘appearance’ occurs in the form zama to mean
‘messy appearance, plight’, kani° ‘crab’ becomes gani- in the expression gani-mata°
‘bandy legs’, hareru ‘to clear up’ yields bareru ‘to transpire, to be revealed (about a
secret)’, sara-sara° ‘smooth, silky’ becomes zara-zara° ‘rough’. Incidentally, it is
interesting to observe that the first meaning of the term daku,nigori ( ), which has been
inherited from the Chinese phonological tradition, and designates the voiced character of
a sound both in China and Japan, is originally that of ‘dirty’, ‘impure’, or ‘muddy’, that is,
precisely the connotation associated with voiced obstruents when they occur word-
initially in Japanese.
Pairs such as the above are generally lexicalized, and, apart from the mimetic class, it is
rare to find productive and personal derivations like zakana ‘rotten fish’ from sakana°
‘fish’, bakimono ‘dirty footwear’ from hakimono° ‘footwear, shoes’, or Beike for He ike
‘those bastards of Heike’ (Heike is the name of a war clan and the eponymous title of a
Page 2 of 29
twelfth-century epic tale), which are sporadic examples of the Edo period.
There also exist words starting with b,d,g, and z which do have a negative connotation but
whose corresponding voiceless forms do not exist (or no longer exist). Let us cite, for
example, zurui ‘tricky, sly’, gomi ‘rubbish’, bokeru ‘to become senile’, geppu° ‘a burp’,
damasu ‘to cheat, to deceive’ (for additional examples see Wenck, 1987 and Komatsu,
1981). All the words which contain two voiced obstruents within the same Yamato root
are found within this lexical class. They belong to the familiar or slang register, for
example do ji ‘blunder’, or de bu ‘fatty’. One should consider that the two voiced
consonants do not share the same status. The one which appears word-internally is
unmarked and primary, while the one which occurs word-initially is marked and
secondary, that is, derived from an originally unvoiced consonant even if the lexeme
beginning with the corresponding unvoiced obstruent is no longer attested.
In the modern language, there exist some cases of voiced initials in lexemes which do not
carry a pejorative connotation. These are historically derived from a phonetic
transformation involving a nasal, like bu chi ‘whip’ (probably from mu chi) or ba° ‘place’
(maybe from niwa° ‘garden’), or they result from the loss of a high vowel, for example
bara° ‘rose’ (derived from ibara° / ubara°), do ko ‘where’ (from i duku), daku° ‘to hold
in one’s arms’ (from idaku) (Komatsu, 1981; Yanagida, 1985; Kishida, 1984). Lastly, it
should be noted that a good number of the forms starting with a voiced obstruent and
which lack a negative phono-pragmatic connotation are animal or plant names, for instance
bora° ‘mullet (fish)’, dani ‘acarid’, bu na ‘beech’, ga ma (also gama°) ‘toad’ or ‘reed
mace’, buta° ‘pig’.
It is obvious that in Japanese, the functional load of voicing is not as strong as that of other
features. Voicing works more like a prosodic feature than like a (p.104) segmental one,
since a voicing difference (very much like a pitch accent difference) does not seem to
impede proper comprehension in Japanese. I shall return to this interpretation in section
4.4 below.
4.1.2 Failure To Undergo Gemination

The voiced obstruents display another characteristic in Yamato and Sino-Japanese words:
they cannot undergo gemination. This impossibility is absolute in those two strata. Only a
few marginally formed mimetics like zabbun-to ‘with a large splash’ and Western loans
such as ba ggu {bag} or be ddo {bed}, accept the gemination of /b/, /d/, /g/, and /z/. But
even in this last type of example, one will undoubtedly hear more frequently the forms ba
kku or be tto, with devoicing of the geminate (Kawahara, 2006). Note that ba ggu or be
ddo are noteworthy in three respects from the point of view of their phonological
structure: (i) they begin with a voiced obstruent; (ii) they contain two voiced obstruents
within the same morpheme; (iii) they contain a geminated voiced obstruent.
Although in native Japanese, they cannot be geminated in the classical sense of the term,
the voiced obstruents may sometimes undergo a strengthening process which can be
analysed as a sort of gemination, since it consists of the insertion of a mora nasal before
the consonant, in a manner parallel to what occurs with voiceless consonants, which are
Page 3 of 29
consistently geminated in the same context (that is, the special segment /Q/ is inserted
before the consonant). This nasal insertion is sometimes termed ‘partial gemination’. The
functional equivalence and complementary distribution between gemination and insertion
of /N/ is particularly observable and regular in the mimetic lexicon, as depicted in the
examples below. The examples shown here involve -ri suffixation to a bimoraic mimetic
root and gemination or prenasalization of the second consonant of this root.
(1) Insertion of /Q/ (gemination) and /N/ (prenasalization) in mimetics
a. Before a voiceless consonant
root derived adverb

bata battari /baQtari/ ‘suddenly’, ‘with a clash’
kaki kakkiri /kaQkiri/ ‘precisely’
goso gossori /goQsori/ ‘entirely’
b. Before a voiced consonant (except /r/)
zabu zanburi /zaNburi/ ‘with a big splash’

maji manjiri /maNziri/ ‘not (to sleep) a wink’
koga kongari /koNgari/ ‘well roasted’
chima chinmari /tiNmari/ ‘cosily, compactly’
huwa hunwari /huNwari/ ‘in an airy manner’
(p.105)
The complementary distribution of /Q/ and /N/ also occurs with the intensive prefix bu-,
and a few others, but less productively.
(2) Insertion of /Q/ (gemination) and /N/ (prenasalization) in bu- prefixed (intensive)
derivatives
bu + korosu° bukkorosu ‘to kill’

bu + tobasu° buttobasu ‘to strike’
bu + naguru bunnaguru ‘to beat’
bu + toru bundoru ‘to grab’
bu + *dakuru bundakuru ‘to take forcefully’
This Q / N insertion process is also observable in simplex non-mimetic words, for example
in ta da 〉 ta nda ‘only’ or tabi 〉 tanbi ‘time’ (Hamada, 1952).
The complementary distribution of /Q/ and /N/ in the above examples pleads in favour of
an analysis which treats the specification of nasality as pre-existent to /N/ insertion, and
Page 4 of 29
thus views nasality as an inherent, constitutional feature of the phonological structure of

the Japanese voiced obstruents, as we shall see more thoroughly in section 4.1.6.
4.1.3 Instability And Transparency

Another striking characteristic of the Japanese voiced obstruents is their instability. Both
diachronically and synchronically, one observes many shifts from a voiceless consonant to
a voiced one, or from a voiced one to a voiceless one, as in the following examples (the
symbol / indicates variation in synchrony, while 〉 marks variation in diachrony):
(3) Voiced/voiceless obstruents alternations
hota° / bota° / hoda° ‘firewood’

shita-tsuzumi / shita-zutsumi ‘smack of lips’
kurai° / gurai° ‘approximately’
touboku 〉 touhoku° ‘Province of the North-East’
tenga 〉 tenka ‘all the country’
abureru 〉 ahureru ‘to overflow’
kami-gakura 〉 kami-kagura ‘sacred dance’
tsukumu 〉 tsugumu ‘to be silent’
sawaku 〉 sawagu ‘to be noisy’
In classical poetry, there exists a literary device, kakekotoba, consisting of a pun based
on homophony, in which a sequence of sounds was used to suggest more than one
meaning. Interestingly, voicing can be transparent or irrelevant in kakekotoba. For
instance, in poem 423 of the Kokinshû (dating from the beginning of the tenth century),
we find the following kakekotoba. Capitalized letters (p.106) indicate the sounds for
which voicing distinction is not relevant, allowing double interpretation of the meaning.
(4) Voicing transparency in kakekotoba
kubeki hoDo toKi suginure ya

‘the moment it should have come has passed’
Or
‘the nightingale that should have come’
This verse also contains the noun for the nightingale, hototogisu, in the sequence hodo
toki su(ginure), if one ignores the voicing distinctions. Considering the fact that such
poetry was first composed to be sung aloud, it appears that voicing differences were felt
as secondary and did not impede the understanding of the poem.
This type of pun actually still exists in Modern Japanese. In goro awase, a popular
mnemotechnic device for remembering phone numbers, dates, passwords, and so on,
numbers may be read following their different allomorphs corresponding to a number of
Page 5 of 29
phonetic values (native Japanese, Sino-Japanese, or even English, each number has
generally more than four or five possible readings), in order to be used to cue words or
phrases (see Schourup, 2000, for a detailed study). Interestingly, a voiced number can
cue an unvoiced one, and vice versa. For instance, the phone number of a taxi company
is 3563–5151, which reads sa, gorou-san°, koi, koi ‘well, come and pick me up Mr Goro!’
(Goro is a popular male name).1 Here, the digit 5 which normally reads go (in Sino-
Japanese) is used to cue both the mora go in Gorou-san ‘Mr Goro’ and the mora ko in koi
‘come!’. Similarly, in a study of Japanese rap rhymes, Kawahara (2007) reports that
consonants differing only in voicing are frequently treated as similar. It is also interesting
to observe that in Japanese dictionaries, the difference between an unvoiced and a
voiced consonant is not taken into account for the ordering of the headwords. Thus the
entries kara ‘shell’, karada ‘body’, gara ‘design’ will appear in the following order:
kara,gara,karada. The difference between k and g is simply ignored.2
Ohno (2005) also mentions the fact that recent loanwords in the modern language are
sometimes pronounced with different voicing values, for instance amejisuto for
{amethyst}, batominton for {badminton}.
Finally, let us also note the following phenomenon: in a sequence of the shape C1 VC2V,
where C1 is a voiceless consonant and C2 the same consonant in its (p.107) sonorous
version, haplology frequently occurs in Old and Middle Japanese, in the same way that it
occurs between two strictly identical consonants: for example mashiji evolved into maji
‘will not, ought not (auxiliary)’, kuhibisu (*kupibisu) into kubisu° ‘a heel’ (Yamaguchi
Yoshinori, 1988a:203).
4.1.4 Historical Development of Voiced Obstruents

A large number of intervocalic voiced obstruents have developed out of consonant
clusters containing a nasal segment, as illustrated in the following examples:
(5)
10th century 20th century

humite 〉 *hunde 〉 hude° ‘brush, pencil’
kamipe 〉 *kambe 〉 ko ube ‘Kôbe’
sumisuri 〉 *sunzuri 〉 suzuri ‘inkstone’
ikanika 〉 *ikanga 〉 ikaga ‘how’
We also know, from the transcriptions in the Latin alphabet made by European
missionaries during the Muromachi period, and the notations found in Chinese and
Korean materials of the same period, that a vowel preceding a voiced obstruent used to
be realized with a nasalization. Thus we find, in the documents written by the Iberian
Jesuits, romanizations such as Nangasaqui for Nagasaki (place name), vareranga for wa
rera ga (‘we’), and so on. This nasality is noted in an especially regular manner before /g/
Page 6 of 29
and /d/. It presumably disappeared progressively during the Edo period, but note that it
is still occasionally heard in the modern language in the speech of certain speakers of the
Tôkyô dialect, and very regularly in the dialects of the Tôhoku and Tosa (Shikoku) areas.
In the standard language, the g / ŋ alternation, which was discussed in section 3.10, can
be regarded as the last vestige of this once widespread nasalization.
Most linguists agree today in considering that the voiceless/voiced opposition used to
come down to an oral/nasal opposition. According to Yamaguchi Yoshinori (1997), /b/
would have corresponded to [mb], /d/ to [nd], [ndƷ͡], or [ndz͡ ], /g/ to [ŋ], and /z/ to [ndz͡ ],
[ndƷ͡] in the archaic language. Then, prenasalization disappeared in the following order
according to Inoue (1971): /(d)z/ 〉 /b/ 〉 /d/, and finally /g/ (see section 3.10 for details
concerning the velar nasal [ŋ]; see also Yamane-Tanaka, 2005, for a discussion of the
connection between the loss of prenasalized stops and the history of voiced obstruents
within the OT framework).
For Hayata (1980), phonologically voiceless consonants were all phonetically voiced
intervocalically in pre-archaic Japanese, while phonologically voiced consonants were
prenasalized. However, this hypothesis does not reach full consensus among
Japanologists. Hayata also claims that it is under the influence of Chinese loans—where
voiceless consonants occur intervocalically—that the (p.108) system of Japanese might
have undergone restructuring. Intervocalic consonants would have remained voiceless
in Japanese when they were so underlyingly, while voiced consonants would have started
to lose prenasalization. This assumption is not very different from that of Hamada (1960),
who also considers that the voiceless/voiced opposition has developed under the
influence of the Chinese loans, and that prior to the Chinese influence, the opposition was
not relevant in Japanese. Only a phonetic opposition would have existed, the obstruents
being voiced between two vowels, but in a non-distinctive way, in a manner reminiscent
of what occurs in Modern Korean. One of the arguments in favour of this analysis is that
the voiceless/voiced difference was not recorded in the documents of the Heian period
and even after. However, a counter-argument exists, which would contradict the
absence of a distinctive opposition between voiced and voiceless obstruents in Archaic
Japanese. Some of the oldest documents (in particular the Kojiki and the Nihonshoki)
denote very accurately the presence of a voiced obstruent in the passages transcribed
in Japanese, a notation which later, from the Nara period on, became suddenly optional or
erratic, hence a serious problem. The rather sudden disappearance of voicing indications
in the oldest documents actually constitutes one of the most mysterious enigmas in the
history of the Japanese language, as we shall see in the next section.
4.1.5 Representation In the Writing, Past And Present

Unlike the Latin alphabet, the modern kana writing system captures in a quasi iconic
manner the natural link which exists between a voiced and a voiceless obstruent. Voicing
of an obstruent is transcribed, in hiragana and katakana, by the addition of two
superscript diacritic dots (the dakuten ‘voicing dots’) on the right side of the kana
used to denote the mora starting with the corresponding voiceless consonant (see
section 1.5). The written symbol for the voiced sound is thus derived from the voiceless
Page 7 of 29
one, in a uniform manner. This notation actually reflects the morpho-phonological status of
voicing in Japanese obstruents in a rather appropriate manner, because obstruent
voicing often appears, as we shall see, as a secondary, derived property.
However, the notation of voicing in the kana system is rather recent. Voicing has been
indicated in a systematic way in the common orthography since the middle of the
twentieth century only. The dakuten used in the modern kana writing system are actually
the product of a long and complex history. It is interesting to look back at its development
across the centuries because it sheds interesting light on the phonological nature and
perception of the voicing feature in the Japanese language.
(p.109) First, one has to consider the question of the recognition of the voicing
opposition in Japanese documents. Is voicing indicated or not? In the Kojiki (712 AD) and
the Nihonshoki (720 AD), two of the oldest Japanese texts, the voicing opposition is nearly
always precisely and accurately noted in the sections written in Japanese or transcribing
Japanese nouns (these two documents also contain parts written in Chinese). The writers
used man’yôgana (Chinese characters that are used for their phonetic reading in Chinese
or Japanese), the ancestor of kana. For instance, to transcribe the moras ku and gu, they
use the following characters:
(6) Sets of man’yôgana used to transcribe the moras ku and gu in the Kojiki and
the Nihonshoki (eighth century)
ku:
、、、、、、etc.
gu:
、、、 , etc.
This mode of transcription is radically different in spirit from the one used today since
there did not exist a single and uniform diacritic mark to denote voicing. Moreover, this
system does not capture the phonetic and phonological link which exists between a voiced
and voiceless consonant in the system.
In the Man’yôshû, a collection of Japanese poetry whose compilation was completed

around 760 AD, the written distinction between voiced and voiceless consonants
becomes loose, and even absent in a number of poems. For instance, poem 3645 uses
the same character to transcribe the sound ki in oki ‘offing’ and gi in wagimoko ‘my
lady’. In the texts in prose which flourished during the Heian period (794–1185), when
the man’yôgana progressively gave rise to the hiragana by way of simplification, the
voicing distinction disappears completely in documents written in hiragana (see Seeley,
1991 on the history of Japanese writing). One single kana is used to represent both the
voiced and voiceless version of a given CV mora. Posterior texts in katakana such as the
Hôjôki (1212) follow exactly the same principle.
Page 8 of 29
However, we know from the first missionary accounts of the Muromachi period in the
Latin alphabet (for instance Rodriguez, 1604–1608) that the voicing distinction had a
phonemic status in Japanese since it is consistently transcribed in the first documents in
the Latin alphabet. It is only in the Edo period (1604–1868) that a written distinction
between voiced and voiceless obstruents reappears in printed kana documents, by
means of the dakuten, but even in the Edo period the notation of voicing was neither
systematic nor consistent.
To sum up, whereas the voicing distinction is regularly and faithfully transcribed in the
earliest man’yôgana texts, it disappears within a few decades with (p.110) the
completion of the kana systems during the Heian period, before reappearing during the
Edo period several centuries later.
The temporary disappearance of the voicing opposition in the orthography constitutes

one of the most intriguing mysteries in the history of the Japanese language. Whatever
the exact reason for this fact, what we must retain is that voicing was not felt to be as
distinctive as other features for many centuries. For details about the precise
interpretations of these facts, see Labrune (1998a) or Ohno (2005) and the references
cited therein.
Next, we shall look at how voicing was shown in cases where it was. As pointed out above,
different man’yôgana were used in the oldest texts to denote voicing differences. From
the Heian period on, the voicing distinction is denoted only in kanbun style texts, that is,
texts written by the Japanese in Chinese. A practice starting from the twelfth century
makes use of a mark, posted onto a Chinese character, or onto the kana symbol written
next to a Chinese character, in order to indicate that its pronunciation contains a voiced
obstruent. The marks used vary greatly depending on the texts, writers, or regions,
especially in the older texts. They may consist of a small circle, full or empty, or two
circles, sometimes two or three triangles, placed in one of the four corners of the square
which contains a character, or they may affect the kana placed next to the character.
Interestingly, the initial function of such marks was to indicate the tone of the character
(in its Chinese pronunciation) but soon enough the mark also came to denote the voicing
status of the consonant, in conjunction with the tone. Some texts make a marking
distinction between ‘original voicing’ (hondaku ) and ‘new voicing’ (shindaku ), that is,
voicing resulting from rendaku or post-nasal voicing (see sections 4.2 and 4.3 and ). One
also finds a very curious manner of voicing transcription consisting of writing a given kana
under its mirror image to indicate the presence of voicing (cf. Kokugo Gakkai, 1992,
Tsukishima, 1977, or Labrune, 1998a, for reproductions of the original documents). In
the course of time, the so-called dakuten, that is, the two dots placed at the top right
corner of a kana as we know them today emerged as the most common mark.
It is extremely interesting to look back at the evolution of the utilization and value of the
modern dakuten, because it is an established fact that dakuten were originally used for
the indication of tone. In this respect, it is quite paradoxical to observe that they were
used in documents written in Chinese characters, a system that is basically logographic,
Page 9 of 29
whereas they were not used in kana texts, even though kana is a truly phonographic
system. Some scholars have posited that the voicing distinction was consistently made in
older texts because those were written by people who were well trained in Chinese, and
were used to distinguish voiced from unvoiced (Hamada, 1960; Ohno, 2005). Less
educated people had more trouble in making the distinction because it was less
distinctive than it is in Modern Japanese. We must not exclude the possibility that dakuten
were (p.111) primarily developed as an aid to understand and pronounce foreign
languages (Chinese and Sanskrit) in kanbun texts, mostly written by monks. If this is the
case, it could mean that it is only in this type of text that the voiced/voiceless opposition
was felt to be really distinctive, underlying the fact that it was not so in native Japanese
(Yamato Japanese). However, it remains hard to understand why voicing was accurately
transcribed in the Kojiki and the Nihonshoki, the earliest texts in Japanese, except under
the hypothesis that these documents might have been written by non-natives, for
example Chinese or Korean scribes.
4.1.6 The Internal Structure of Voiced Obstruents

In Labrune (1999), I proposed a theoretical treatment of the phonological structure of
voiced obstruents within the representational framework of autosegmental phonology
and feature geometry (Clements and Hume, 1995). The idea is that voiced obstruents
are intrinsically specified for nasality. I argue, first, that voicing on the surface as it
presently occurs in the modern language can be accounted for by the presence of a nasal
specification in the internal structure of the consonant under a Spontaneous Voicing (SV)
node (following the proposal by Piggott, 1992). Second, I claim that voiced obstruents in
Yamato words possess two root nodes under a single skeletal position, and must
therefore be viewed as heavy segments.
(7) Internal structure of voiced obstruents (Labrune, 1999) (Place features are
omitted)
In the older language, voiced obstruents corresponded to two skeletal positions, the first
one being carried out as a prenasalization. In other words, /b, d, g, z/ (surface [mb, nd,
ŋg, nz]) were contour segments. This representation also helps to understand why they
sometimes alternate with voiceless geminates, as in mi gi + kawa 〉 migi-gawa° (〈 migi-
ngawa) / migi-kkawa° ‘right side’.
In the modern language, nasality is now implemented, in a majority of cases, by means of

a voiced feature at the surface level.
Page 10 of 29
(p.112) This approach which regards voiced obstruents as containing a nasal

specification at the level of their infra-segmental structure is confirmed by the evolution
process these consonants have gone through during the history of Japanese (see the
next section). It also accounts for certain modern forms where the presence of an inter-
morphemic n is apparently unexplainable, for example in on-dori° ‘cockerel’, men-dori°
‘hen’ (from o- ‘male’, me- ‘female’, and tori° ‘bird’), kuman-bachi ‘hornet’ (from kuma
‘bear’ and hachi° ‘wasp’). There also exist many doublets of the type ko bu / ko nbu ‘sea
tangle’, togaru / tongaru ‘to be sharp’, to bi / to nbi ‘kite (milvus migrans)’, tabi / tanbi
‘time’. In all such cases, it is not possible to posit the presence of a rendaku, of the particle
no, or of any other morphological element since we are dealing with simplex words. These
modern examples containing a prenasalization are simply vestiges of a former state of the
language. They reflect a once widespread pronunciation of the voiced obstruents as
prenasalized segments, as is still found in a number of modern dialects.
4.2 Rendaku
Voicing also plays an essential role in Japanese morphology because it functions as a
compounding marker, a process known as rendaku ( ), literally ‘sequential voicing’, or
‘connective voicing’. Rendaku consists of a phonological alteration which occurs at
morpheme juncture when two full morphemes enter the formation of a compound word,
resulting in the voicing of the initial obstruent of the second component. It affects /k, s, t,
h/, which become respectively /g, z, d, b/.3 If the second element of the compounding
starts with a segment other than k,s,t,h, that is, an already voiced consonant like b, d, z,
m, n, r, w, y, or by a vowel, no transformation can occur, except for the velar voiced
consonant /g/ which may (but in a non-obligatory way) be nasalized as [ŋ] in the speech of
speakers who have this allophone of /g/ in their system (see section 3.10).
Consider the following examples.
(8)
u mi + kame umi-game° (or umi-game)

‘sea’ + ‘turtle’ ‘sea turtle’
ku ro + satou° kuro-z atou
‘black’ + ‘sugar’ ‘brown sugar’
kaki + tsurai° kaki-zurai
‘writing’ + ‘difficult’ ‘hard to write’
se + hone se-bone°
‘back’ + ‘bone’ ‘backbone’
te + kaki te-gaki°
‘hand’ + ‘writing’ ‘handwriting’
mome° + koto mome-goto°
‘discord’ + ‘thing’ ‘trouble, discord’
Page 11 of 29
(p.113)
As can be seen from these examples, rendaku not only occurs when combining two nouns
but also with other types of combinations such as Noun + Verb (te-gaki°), Adjective +
Noun (kuro-zatou), Verb + Noun (mome-goto°), Verb + Adjective (kaki-zurai), and so
on, except, as we will see below, in Verb + Verb compounds.
The principal problem raised by the Japanese rendaku is due to the unpredictable and
apparently random character of its appearance. Take for instance the oft-cited example of
the two Japanese syllabaries’ names (Martin, 1952). Katakana / katakana (ka ta / kata
‘side’ + kana° ‘letter’) has not undergone rendaku of the second constituent so kana
remains voiceless, whereas the second syllabary, hiragana / hiragana° / hiragana (hi ra
‘flat’ + kana°), presents a rendaku. However, it is often said that nothing in the
combination mode, the origin, or the phonological structure of the morphemes implied
could explain this difference.4 Examples of this type are not hard to find. In addition to
hiragana and katakana, let us mention for instance:
(9)
ashi + kuse ashi-kuse° *ashi-guse°

‘foot’ + ‘habit’ ‘(particular) way of walking’
vs.
kuchi° + kuse kuchi-guse° *kuchi-kuse°
‘mouth’ + ‘habit’ ‘favourite phrase’
yuki + tama yuki-dama° *yuki-tama°
‘snow’ + ‘ball’ ‘snow ball’
vs.
mizu° + tama mizu-tama° *mizu-dama°
‘water’ + ‘ball’ ‘water ball (water drop)’
u mi + tori° umi-dori *umi-tori
‘sea’ + ‘bird’ ‘sea bird’
vs.
niwa° + tori° niwa-tori° *niwa-dori°
‘garden’ + ‘bird’ ‘rooster’
(p.114)
Moreover, many examples illustrate the fact that the two modes of compounding, with
and without rendaku, can exist in parallel for the same compound, without any difference
in meaning or even in usage between the two, as in waru-kuchi / waru-guchi ‘calumny’,
kenkyuu-sho° / kenkyuu-jo° ‘research centre’, nori-tsuke° / nori-zuke° ‘pasting’,
kami-kakushi / kami-gakushi ‘disappearance (of child) (lit. ‘hidden by the gods’)’, kaki-
Page 12 of 29
tome° / kaki-dome° ‘registered mail’.
Many scholars explicitly or implicitly follow the view that rendaku appearance is the
default, elsewhere condition when two lexemes enter compounding, if some specific
blocking factors such as the ones to be presented below in section 4.2.2 are not involved.
Rendaku can thus be viewed as the materialization of a dependency link which exists
between two lexemes on the occurrence of compounding.
Whether rendaku is still productive is a matter of controversy. Ohno (2000) has

conducted an experiment which demonstrates the usage-based conditioning of rendaku
application. He argues that native speakers refer to a semantically and/or phonetically
parallel existing compound in order to determine whether a novel compound must
undergo rendaku or not. If there is no existing parallel rendaku form, the item will not
undergo rendaku in the novel compound. For this author, speakers simply memorize
whether rendaku occurs for individual compounds, and so synchronically rendaku would
just be the lexical residue of a rule which was once productive and automatic. However,
the data presented by Fukuda and Fukuda (1999), cited in Kubozono (2005), tend to
demonstrate that rendaku is still productive, which suggests that more research is still
needed on this question.5
Rendaku has sometimes been regarded as a fossil of the determination particle no or of

some other particle containing either a nasal (such as ni, which marks the agent, the
recipient, the locative, the attributive, among others), or a voiced obstruent (such as de,
locative marker). This would explain why, as we shall see in section 4.2.2, rendaku does
not appear in coordinative (dvandva) compounds, in Object + Verb compounds, or in
mimetics, where there is no syntactic or semantic reason to assume the underlying
presence of a particle like no,ni, or de. This is the analysis proposed by Lyman (cited by
Yamaguchi Yoshinori, 1988b), Vance (1982), and Hirano (1974, cited by Takayama M.,
1992). Thus for example yama-gawa° ‘mountain river’ is a determinative compound which
would derive from yama no kawa ‘mountain’ Det. ‘river’, whereas yama-kawa
‘mountains (and) rivers’, a coordinative compound, is the simple juxtaposition, (p.115)
on the morphological and semantic levels, of yama and kawa, hence the absence of
rendaku.
Incidentally, it is interesting to observe that rendaku sometimes alternates with voiceless

consonant gemination (Takayama T., 1995), as in migi-gawa° / migi-kkawa° ‘right side’, de
-ba / de -ppa ‘protruding teeth’, hitori-go / hitori-kko ‘only child’, kore-kiri / kore-kkiri /
kore-giri° ‘once (and) for all, that’s it’, suki-ppara° / suki-bara° ‘empty stomach’. So in
addition to rendaku, at least two other common compounding markers exist in Japanese:
gemination of the initial consonant of the second member, and, as we shall see in Chapter
7, accentuation.
In spite of its largely random nature, some factors favourable to the application of
rendaku can be identified, as shown in the next section.
4.2.1 Rendaku Triggering Factors
Page 13 of 29
The factors which condition the appearance of rendaku are of various kinds. First, there
exist factors of a lexical nature. Rendaku can be regarded as a compounding marker, but
it is also a manifestation of the degree of lexicalization of a given compound. All things
being equal, the more lexicalized and frequent, the more a compound will be likely to
contain a rendaku. The occurrence of rendaku also varies according to the lexical class,
reflecting the degree of integration of a compound in the lexicon of Japanese. Whereas
rendaku is extremely rare in loans of Western origin, and only occasional in Sino-Japanese
words, it is very frequent in Yamato words. Lexemes of this latter stratum constitute the
very privileged target for the application of rendaku when they occur as the final
component of a compound. According to Rosen (2003), rendaku occurs in about 75% of
Yamato Noun–Noun compounds that present the right phonological conditions to trigger
it. However, and very interestingly, his survey makes it clear that cases of rendaku
blocking actually cluster around particular lexical items rather than being randomly
dispersed. This means that most nouns undergo rendaku in a regular manner, whereas a
small number of nouns exhibit idiosyncratic behaviour with respect to rendaku: they
either block it in a seemingly unpredictable and unconsistent manner in a number of
compounds, or consistently block it under all conditions. Therefore there exists a clear
unbalance as to the number of rendaku-triggering nouns and rendaku-blocking nouns, to
the benefit of the former. We will return to Rosen’s study in the next pages.
Takayama T. (2005) claims that Sino-Japanese words that undergo rendaku are informal or
colloquial expressions which can be categorized as ‘vulgarized Sino-Japanese’. They are
lexemes of high frequency, which refer to concrete and familiar referents, like shashin°
‘photograph, kaisha° ‘a firm’, satou ‘sugar’.
(p.116) (10) Rendaku in Sino-Japanese words
a o + shashin° /syasiN/ ao-jashin /zyasiN/

‘blue’ + ‘photograph’ ‘blueprint’
kabushiki + kaisha° kabushiki-gaisha
‘share, stock’ + ‘firm’ ‘a stock company’
ku ro + satou kuro-zatou
‘black’ + ‘sugar’ ‘brown sugar’
The few cases of rendaku occurring in Western loans are found only in words which are
no longer perceived as foreign, and have totally assimilated to the Yamato class. This is
the case of kappa° ‘overcoat, raincoat’ and ka ruta ‘playing card’, two very early loans
from Portuguese (see Takayama T., 2005 for additional examples). Many speakers analyse
these words as Yamato words, hence the application of rendaku when they occur as the
second constituent of a compound as in ama-gappa ‘raincoat’ and iroha-garuta ‘iroha
card (a game with the Japanese kana)’.
The phonetic environment constitutes one of the other determining factors for the
Page 14 of 29
application of rendaku. Rendaku occurs much more readily when the first element of the
compound ends with the mora nasal /N/. This type of rendaku is traditionally referred to
as shindaku ( literally ‘new voicing’) in Sino-Japanese words.
(11) Rendaku after /N/
ho n + tana° ho n-dana
‘book’ + ‘shelf’ ‘bookshelf’
ha n + hiraki han-biraki° / han-biraki
‘half’ + ‘to open’ ‘half-open’
shi n + suru° shin-zuru
‘trust’ + ‘to do’ ‘to trust’
Sino-Japanese words usually described as containing a rendaku often belong to this

category. However, the examples cited here can be analysed differently. In ho n-
dana,han-biraki° / han-biraki, and shin-zuru (in which the initial morphemes are Sino-
Japanese and the final ones Yamato) the voicing occurring after the mora nasal might not
pertain, strictly speaking, to a rendaku since this type of voicing can appear elsewhere
than at the boundary between two autonomous morphemes. We might be dealing here
with post-nasal voicing (see section 4.3), which, unlike rendaku, has no morphological
function.
(p.117) It also appears that, other things being equal, the length of the compound or of
one of its constituents might also determine the application of rendaku. Rosen (2003) and
Irwin (2009) claim that long compounds prevent the blocking of rendaku under most
conditions. Rosen (2003) shows that a set of words, which he labels as ‘rendaku-
resisters’, can undergo voicing in a number of short–short noun compounds but always
voice in long compounds (see Table 4.1 in the next section). In other words, rendaku-
resisters never block rendaku when they occur in long compounds, but may voice when
preceded by a short noun. (A compound is considered as long when its first component
exceeds two moras.) So there exists a prosodic size threshold which, when exceeded by
a compound, disables the blocking of rendaku by rendaku resisters. Compare for
example the following examples with kusa ‘grass’, a noun which, according to Rosen
(2003), always undergoes rendaku when appearing after a long first member (12a), but
sometimes resists it when the first member is short (12b).
(12) Rendaku and prosodic length (C2 is not a rendaku-blocking noun)
a. C1 = long → rendaku is systematic

hotaru-gusa ‘firefly grass’
enokoro-gusa ‘foxtail grass’
hitsuji-gusa ‘sheep grass’
b. C1 = short → rendaku is non-systematic
Page 15 of 29
mizu-kusa° ‘water grass’

no -gusa ‘wild grasses’
Ohno (2000) also reports an interesting experiment concerning the voicing of the
morpheme ho n ‘book’ in relation to the length of the first component. Ho n undergoes
rendaku only when the first member of the compound has three or more moras, as in
bunko-bon ‘pocket book’, manga-bon° ‘manga book’, karugaru-bon ‘light book’ (the last
item is a novel compound made up for the experiment). These examples can be compared
to e-hon ‘picture book’ or aka-hon° ‘cheap (pulp) fiction’ in which no rendaku occurs. All
of this shows that the size is definitely relevant for rendaku application.
However, long compounds containing more than two constituents do not undergo
rendaku in a systematic way. This is because rendaku also depends on the morpho-
syntactic structure of the compound. According to Otsu (1980), Rendaku happens only
when a potential rendaku segment is on a right branch at the lowest level of a constituent
tree. So one finds minimal pairs like the following (accent omitted):
(p.118) (13) Right-branch condition (Otsu, 1980)
a. nuri + hashi + hako 〉 nuri-hashi-bako

(nuri((hashi)(bako))) (the second element does not undergo
rendaku)
‘lacquered’ + ‘chopstick’ + ‘chopstick box which is lacquered’

‘box’
b. nuri + hashi + hako 〉 nuri-bashi-bako
(((nuri)(bashi))(bako) (both the second and third elements

undergo rendaku)
‘lacquered’ + ‘chopstick’ ‘box for lacquered chopsticks’

+ ‘box’
Observe the meaning difference between the two compounds in (13). The scope of nuri
‘lacquered’ is different in the two examples.
Semantic factors are also relevant. Rendaku appears systematically in reduplicated forms
with a plural or iterative value as depicted in (14).
(14)
Page 16 of 29
hito° + hito° hito-bito

‘person’ + ‘person’ ‘people’
kuni° + kuni° kuni- g uni
‘country’ + ‘country’ ‘various countries’
ka esu + kaesu kaesu-gaesu
‘to repeat’ + ‘to repeat’ ‘repeatedly’
Note however that in reduplications with distributive value, variation occurs: rendaku
may or may not apply: sore-zore (sore-zore) ‘each’ with rendaku,hitori-hitori ‘each
person’, without rendaku (the form with rendaku hitori-bitori is also attested).
Finally, the grammatical class of words also plays a key role. Vance (2005) shows that 80%
of the compounds involving adjectives, be they of the structure Adj + Verb = V
(naga-biku ‘be prolonged’), Adj + Verb = Noun (waka-gaeri ‘rejuvenation’), Adj + Adj =
Adj (usu-gurai° / usu-gurai ‘dim’) or Verb + Adj = Adj (utagai-bukai ‘suspicious’)
undergo rendaku, a percentage comparable to that found by Rosen (2003) for Yamato
Noun + Noun = Noun compounds (p.119) (see above). Contrastively, as we will see
below, Verb + Verb = Verb compounds do not normally exhibit rendaku.
These few remarks are unfortunately far from providing a correct account of all cases of
occurrence of rendaku. As Yamaguchi Yoshinori (1988b) observes, it is easier to
enumerate the conditions of rendaku blocking than to determine the rules which govern
its application, as shown in section 4.2.2.
4.2.2 Rendaku-blocking Factors

The conditions of rendaku blocking depend on various phonological, semantic, syntactic,
and lexical parameters. As we have seen, the lexical origin of the morphemes has
considerable influence on the chances for rendaku to apply. It occurs less often in Sino-
Japanese words than in Yamato words, and never in recent Western loans (the
exceptions only concern older loans, see above). Reduplicated mimetic words also never
undergo rendaku: kira-kira ‘glitteringly’, pera-pera° ‘eloquently’ (*kira-gira,*pera-
bera°6). Neither do numeral compounds made up of a numeral and a specifier like huta-
hako ‘two boxes’ (*huta-bako) or go -hiki ‘five animals’ (*go-biki), but for a few
exceptions such as hitotsu-boshi ‘one star’, and some compounds starting with san°
‘three’ because of the presence of /N/ (such cases resort to post-nasal voicing, see
below).
A general principle blocks the application of rendaku. This principle is that of dissimilation.
Rendaku does not occur when its appearance would be likely to involve the co-
occurrence of two identical segments or of phonologically resembling segments within the
same domain. The most widely known aspect of this phenomenon is without question the
so-called Lyman’s law. Lyman’s law stipulates that rendaku never occurs when the
second component of the compound already contains a voiced obstruent. It is important
Page 17 of 29
to note that rendaku is not blocked when the second component contains a sonorant (a
nasal or the liquid r). Consider the following examples:
(15)
ka mi + tana° kami-dana° *kami-tana°

‘god’ + ‘shelf’ ‘a domestic altar’
but
ka mi + kaze° kami-kaze *kami-gaze
‘god’ + ‘wind’ ‘divine wind’
toki + toki toki-doki / toki-doki° *toki-toki / *toki-toki°
‘moment’ + ‘moment’ ‘sometimes’
but
tabi + tabi tabi-tabi° *tabi-dabi°
‘time’ + ‘time’ ‘often’
mizu° + kuruma° mizu-guruma *mizu-kuruma
‘water’ + ‘wheel’ ‘water wheel’
but
mizu° + kagami mizu-kagami *mizu-gagami
‘water’ + ‘mirror’ ‘reflection in water’
(p.120)
The non-application of rendaku in kamikaze,tabi-tabi°, and mizu-kagami can actually be

regarded as pertaining to the already discussed general law of Yamato roots, which
forbids the co-occurrence of two voiced obstruents within the same morpheme (see
section 4.1.1).
Observe that the presence of a voiced obstruent in the first component, for example in
mizu°, is irrelevant with regard to Lyman’s law. However, some exceptions to this
principle have been reported, especially in personal names ending with the morpheme ta
‘rice field’ (Kubozono, 2003 based on Sugitô, 1965). Rendaku of ta fails to apply when it
is immediately preceded in the first component by a voiced obstruent or by /y/. We thus
have Ku bo-ta,Naga-ta°,M izu-ta, etc. but Yama-da°,Kan-da°,Yasu-da°, etc. (all are
personal names).
The non-application of rendaku in cases where the second component already contains a
voiced obstruent is often described as an absolute principle, but Vance (1987) mentions
a few exceptions. The least disputable examples are compounds containing the noun
hashigo° ‘ladder’, for example nawa + hashigo° 〉 nawa-bashigo ‘rope ladder’, in which
the rendaku of h occurs, in spite of the presence of g in hashigo°. But interestingly
enough, the two voiced obstruents in -bashigo are not contiguous, and moreover,
Page 18 of 29
hashigo° is a lexeme longer than two moras, and, as already mentioned, prosodic length
has been shown to be a definite rendaku-triggering factor.
Other manifestations of the dissimilation principle, which Lyman’s law can be viewed as a
strong example of, are a little less obvious but nevertheless rather special. They appear in
examples such as those presented in (16), taken from Satô H. (1989). Here, the absence
of rendaku seems to be motivated by a dissimilatory principle, in order to avoid repetition
of the same phonological element, which can be either a mora or a segment.
(16)
a. Compounds containing the morpheme hi ‘fire’
tobi° + hi tobi-hi° *tobi-bi°

‘jump’ + ‘fire’ ‘leaping flames’
Compare with
kitsune° + hi kitsune-bi *kitsune-hi
‘fox’ + ‘fire’ ‘a fox (elf) fire’
nokori + hi nokori-bi *nokori-hi
‘remainder’ + ‘embers’
‘fire’
morai° + hi morai-bi° (morai-bi, morai-bi) *morai-hi
‘receiving’ + ‘fire’ ‘fire caught from a neighbouring
house’
(p.121)
b. Compounds containing the morpheme tsukeru ‘to put’
kizu° + tsukeru kizu-tsukeru *kizu-zukeru

‘a wound’ + ‘to put’ ‘to hurt’
Compare with
a to + tsukeru ato-zukeru *ato-tsukeru
‘a trace’ + ‘to put’ ‘to leave a trace’
na° + tsukeru na-zukeru *na-tsukeru
‘name’ + ‘to put’ ‘to name’
i chi + tsukeru ichi-zukeru *ichi-tsukeru
‘position’ + ‘to put’ ‘to locate’
This series of examples shows that compounds of the shape X + hi and X + tsukeru
undergo rendaku regularly, except tobi-hi° and kizu-tsukeru, the two forms which
already contain the moras bi and zu in their first component. This seems to be a robust
tendency, and I could find only one counter example, tabi-bito° ‘traveller’ (tabi + hito°)
Page 19 of 29
with respect to this rule in the list of about 800 rendaku-containing compounds provided
by Itô and Mester (2003, appendix). However, since voiced obstruents are statistically
much less frequent in the overall Japanese lexicon, as the statistics shown at the end of
Chapter 3 indicate, additional research based on a thorough statistical analysis is called
for in order to confirm the reality of this phenomenon.
Satô H. (1989, quoting Kindaichi 1976) also mentions the lexemes hi me ‘princess’ and
himo° ‘string’, which never undergo rendaku. This could also be interpreted as an effect
of the dissimilation principle, since the voicing of h would produce the components *bime,
*bimo, with two labial consonants. However, a number of other lexemes such as tsuchi
‘soil’, shio ‘tide’, kemuri° ‘smoke’, taka° ‘hawk’, are known to resist rendaku even though
voicing of their initial segment would not cause the repetition of similar segments or
features.7 Such words actually belong to the class of nouns that Rosen (2003) labels as
‘rendaku-immune’. Rosen examines exceptions to rendaku voicing that are independent of
Lyman’s law and which are generally ignored or treated as random and (p.122)
unsystematic by the relevant literature. He argues that although rendaku appears
completely unpredictable seen from the perspective of individual nouns, it exhibits a
strong degree of patterning on a global scale. On the basis of a thorough corpus
examination of Noun + Noun compounds, he identifies three types of nouns with respect
to rendaku blocking:
(a) ‘Rendaku-immune nouns’, which never voice under any circumstances (see
also Martin, 1987, Vance, 1987), such as kita° ‘north’, hasi° ‘edge’, shita ‘below’,
ka su ‘dregs’, hima° ‘leisure’, kemuri° ‘smoke’, ka se ‘shackles’, himo° ‘string’,
katachi° ‘shape’, tsuchi ‘earth’, hi me ‘princess’, kamachi° ‘framework’, hama
‘beach’, tsuya° ‘gloss’.
(b) ‘Rendaku resisters’ which robustly resist voicing but only in short–short
compounds (short–short compounds are compounds in which neither member
exceeds two moras).
(c) The elsewhere case of all other nouns, which undergo rendaku voicing in most
compounds.
The difference between rendaku-immune nouns and rendaku resisters thus lies in the
fact that rendaku-immune nouns never undergo voicing, even in long compounds,
whereas rendaku resisters always voice in long compounds, even though they do not
always do so in short ones (see the examples with kusa ‘grass’ cited in (12)).
Rosen’s study shows that cases of rendaku blocking among rendaku resisters actually
cluster around particular lexical items rather than being randomly dispersed. That is,
about half of the cases of rendaku blocking in short–short compounds occur among just
eight nouns, which are: kusa ‘grass’, ha ra ‘field’, kuse ‘habit’, kawa ‘skin’, saki° ‘tip’, ki
‘tree’, ko° ‘child’, te ‘hand’.8 These findings are summarized in table 4.1. We will return to
Rosen’s study, which consitutes an important recent contribution to rendaku, in section
4.2.4 below.
It is sometimes said that Verb + Verb compounds are another type of compound in
Page 20 of 29
which rendaku is supposed to occur less readily. Okumura (1980) gives the pairs seme-
toru ‘to take by assault’ (accent unknown) with no rendaku, and seme-dori ‘being forced
to add extra stones to remove a captured group from the board (a term of the go
jargon)’, its nominal form, which exhibits rendaku. However, according to Vance (2005),
neither the verbal nor the nominal variant undergoes rendaku in about 90% of the
compounds belonging to this type. To (p.123)
Table 4.1 Summary of blocking patterns among Yamato noun–noun

compounds (Table 5 in Rosen 2003)
Classification of second conjunct Total n° of N° of cases Total n° of N° of cases
short of voicing in long of voicing in
compounds short compounds long
compounds compounds
Rendaku-immune (never 49 0 (0%) 14 0 (0%)

undergoes rendaku voicing)
Rendaku-resistant (usually blocks 119 36 (30%) 13 13 (100%)
voicing in short compounds but
never in long compounds)
Unknown type (never undergoes 21 0 (0%) 0 0 (0%)
rendaku but occurs in few
compounds; likely either
rendaku-resistant or rendaku-
immune)
Non-resistant (usually voices in 580 522 (90%) 196 196
compounds) (100%)
give just one example, mi-toosu° / mi-toosu ‘foresee’ and mi-toosi° ‘prospect’, both occur
with no rendaku.9
Another robust constraint of a morpho-syntactic nature is at work. Rendaku tends to

occur more easily when the initial component (a noun) is an instrument, or place
complement, than when it is the object complement of the final component. Compare for
instance the following pairs (Okumura, 1980):
(17)
e + kaki e-kaki
‘picture’ + ‘writing’ ‘a painter (= picture drawing)’ (Object)
vs.
hude° + kaki hude-gaki° (Instrument)
‘brush’ + ‘writing’ ‘writing with a brush’
ya ne + huki° yane-huki (yane-huki°) (Object)
‘roof’ + ‘covering’ ‘roof covering’
vs.
Page 21 of 29
wa ra + huki° wara-buki° (Instrument)

‘straw’ + ‘covering’ ‘thatching with straw’
(p.124)
Here, it is tempting to resort to the etymological hypothesis already mentioned in 4.2,

which analyses rendaku as the remnant of an enclitic particle containing a voice feature,
such as de (Instumental) or ni (Locative).
This factor seems to be still productive in Japanese. For instance, in the novel Neko no
kyaku (2001:90–91),10 by the well-known poet Hiraide Takashi, the narrator makes a
linguistically interesting comment regarding the meaning difference between two possible
readings of the compound written as (which occurs as the title of a lithograph in the
novel and combines inazuma° ‘lightning’ and to ri ‘catching’). Should it be inazuma-
tori, without rendaku, or inazuma-dori, with rendaku ? He comments that inazuma-tori
should be interpreted as inazuma wo toru, ‘to catch lightning’ (inazuma = Object),
whereas inazuma-dori would mean ‘to catch (something) like lightning, or with a
movement resembling that of lightning’ (inazuma de toru,inazuma = Manner).
The semantico-syntactic relationship between the two morphemes of the compound is

relevant in another way. Rendaku generally appears when the first element syntactically
depends on the second. On the other hand, when no hierarchy exists between the two
components, as is typically the case in dvandva compounds (coordinative compounds), as
well as in reduplicated mimetics, rendaku never occurs. Compare the already cited pairs
of compounds yama-kawa ‘mountains (and) rivers’, and yama-gawa° ‘river of the
mountain’, or tsu yu-shimo ‘dew (and) frost’ and tsuyu-jimo° ‘frost formed by the
dew’.11
4.2.3 Correlations Between rendaku And Accent

There exist strong and curious correlations between voicing and accent. For instance, as
just mentioned, it is noteworthy that dvandva compounds neither undergo rendaku nor
receive an accent following the rules of compound accentuation, as we shall see in
Chapter 7. Long compounds, containing more than two constituents and which possess
two accent kernels do not undergo rendaku either (Itô and Mester, 2003). For
Kubozono (2005), the correct generalization is that ‘phonological unification is blocked
between two constituents A and B, if B does not c-command A’, phonological unification
being materialized by rendaku and/or attribution of a compound accent pattern.
Rosen (2003) also observes a parallel between the predictability of accent patterns and
the predictability of rendaku appearance with respect to prosodic (p.125) length in
compound words. As he states, ‘when a compound has at least one constituent that
exceeds two moras, both its pitch accent and the voicing of the rendaku-targeted
obstruent are predictable from the input forms’. So there is a strong correlation between
prosodic size and irregularity regarding accentuation and rendaku application.
In some cases, one observes that the appearance of rendaku might correlate with an
Page 22 of 29
atonic accent pattern because compounds which fail to undergo rendaku are often atonic.
The series of compounds with tori° ‘bird’ offers interesting examples: ko-tori° ‘little bird’,
mizu-tori° ‘waterfowl’, niwa-tori° ‘domestic fowl’, yaki-tori° ‘grilled chicken’ with no
rendaku are all atonic, whereas chi -dori ‘plover’, miyako-dori ‘oyster bird’, yama-dori
‘mountain bird’, watari-dori ‘migrating bird’, all undergo rendaku and are also tonic (this
series of examples comes from Takeuchi, 1999:49).
Another interesting comparison can be seen in the following pairs of examples, where
rendaku-undergoing compounds also happen to be atonic:
(18)
hito-te ‘one hand’ vs. hito-de° ‘another’s hand (lit. human hand)’
(hito- ‘one’ + te ‘hand’ vs. hito° ‘person’ + te ‘hand’)
hito-koe ‘one voice’ vs. hito-goe° ‘human voice’
(hito- ‘one’ + ko e ‘voice’ vs. hito° ‘person’ + ko e ‘voice’)
These correspondences between accent and rendaku are symptomatic of the particular
status of voicing in Japanese. In many respects, voicing can be compared to a supra-
segmental feature such as tone or accent, a point to which we return in the last section of
this chapter.
4.2.4 Some theoretical Proposals Concerning rendaku

For Komatsu (1981), Yamaguchi Yoshinori (1988b), and many other linguists, the main
function of rendaku is that of a compounding marker. Since virtually no root of the Yamato
lexicon starts with a voiced obstruent, the presence of voicing on the normally voiceless
initial consonant of the second component indicates that this consonant is not initial and,
thereby, that the word is no longer independent. A formal interpretation of this is
provided by Itô and Mester (1986) who posit that rendaku consists of a morphological
operation involving the insertion of a [+voice] autosegment at compound juncture. This
autosegment is associated to an unsyllabified segmental position. Itô and Mester
(2003:83) reformulate this by saying that rendaku is a feature-sized morpheme,
expressed as ℛ, and consisting of the specification [+voiced]. ℛ acts as a prefix to the
second member in word compounds, forming a constituent with it.
Rosen (2003) proposes that the capacity for a given noun to undergo rendaku is encoded
in its underlying representation. Recall, as explained before, that Rosen (p.126)
distinguishes three types of nouns: rendaku-immune nouns, rendaku resisters, and
rendaku undergoers. He proposes that the three types be formally specified as follows:
rendaku-immune nouns such as kita° ‘north’ have a [−voice] feature that is linked to the
root node of the initial obstruent, rendaku resisters such as kusa ‘grass’ have a floating
[−voice] feature, and non resisters such as kuchi° ‘mouth’ have an initial obstruent that
is underspecified for voicing, as follows (the capital K represents a velar stop minus its
voicing feature).
Page 23 of 29
(19)
Following Itô and Mester (1986), Rosen assumes that rendaku voicing occurs because of
a junctural morpheme whose underlying form is a floating [+voice] feature. The basic idea
is that in non-resisting nouns, which have no relevant underlying voicing feature, the
[+voice] feature of the junctural morpheme links to the initial obstruent of the second
conjunct in the output. The linked [−voice] feature of rendaku-immune nouns persists in
all compounds, whereas the floating [−voice] feature of rendaku resisters persists only in
short–short compounds, for a reason which has to do with the length of the compound.
As pointed out above (section 4.2.2), a crucial aspect of Rosen’s study and analysis rests
on the empirical observation that rendaku-resisting nouns block rendaku only in short–
short compounds. So, in addition to the specification of the [voice] feature which
distinguishes rendaku-immune nouns and rendaku resisters, Rosen claims that one has to
further take into account the prosodic difference between the two types of compounds,
and integrate this into the analysis through a constraint-based grammar.
Taking for granted that the canonical Japanese feet are bimoraic (see Chapter 6), Rosen
follows Kubozono (1999a) in considering that every foot is entirely contained within the
same morpheme, so that two adjacent monomoraic feet will not form a single bimoraic
foot: each foot will be parsed as a separate degenerate foot. Adopting the commonly
accepted analysis that a prosodic word must maximally consist of two feet in Japanese, it
follows that long compounds consist of a separate prosodic word for each constituent but
that short compounds do not. This is supposedly what explains that rendaku blocking
occurs only in short compounds. For Rosen, ‘the relevant generalisation is that rendaku
voicing is freely permitted to occur in a syllable that is at the left edge of a prosodic
word’. In short compounds, the initial obstruent of the second conjunct is not at the left
edge of a prosodic word, whereas it is in long compounds. Rosen assumes that the
marked [+voice] feature is permitted more freely in the prosodically strong position that
occurs at the left edge of the second conjunct of a long (p.127) compound, i.e. the
beginning of a different prosodic word, following the assumption of Positional Markedness
as developed by Zoll (1998), Smith (2002), etc. In short, the beginning of the second
conjunct coincides with the beginning of a new prosodic word in long compounds but it
does not in short–short compounds. In Rosen’s approach, the lexical prespecification of a
[−voice] feature thus interacts with prosodic length to cause blocking of rendaku voicing
specifically in short–short compounds but not in long compounds.
Rosen also provides a challenging OT analysis of how different constraints interact in

order to select the right candidates (with or without rendaku). Since his treatment is
extremely technical, and cannot easily be summarized, interested readers are invited to
refer directly to the paper itself.
To give a flavour of an OT analysis of rendaku, we will offer here a summary of Itô and
Mester (2003). Contrary to Rosen’s, it does not rely on a difference in specification
between nouns, and thus may appear to be ‘mechanically’ simpler. The overall process of
Page 24 of 29
rendaku appearance is accounted for by the action of four main constraints.
(20) Constraints involved in rendaku (Itô and Mester, 2003:96)
NO-D2 m No two voiced obstruents per morpheme domain

REALIZE- Every morpheme in the input has a nonnull phonological exponent in
M the output (Kurisu, 2001)
IDENT No change in feature specification
NO-D No voiced obstruents
These four constraints are ordered according to the following hierarchy:
(21) Constraint hierarchy in rendaku (Itô and Mester, 2003)
NO-D2 m 〉〉 REALIZE-M 〉〉 IDENT 〉〉 NO-D
Let us see how the analysis works for naga-sode° /naga + ℛ + sode/ ‘long-sleeve’, in (22)
which does not undergo rendaku because of the application of Lyman’s law, and natsu-
zora° /natu + ℛ + sora/ ‘summer sky’, in (23) which does (ℛ stands for the rendaku
morpheme).
(22) naga-sode
/naga + ℛ + sode/ NO-D2 m REALIZE-M IDENT NO-D

naga zode *! * ***
☞ naga sode * **
(p.128)
(23) natsu-zora
/natu + ℛ + sora/ NO-D2 m REALIZE-M IDENT NO-D

☞ natsu zora * *
natsu sora *!
The presentation above is only intended to provide an overview of how the analysis by
Itô and Mester (2003) works. A number of additional constraints (which can be
considered as secondary as far as the core phenomenon of rendaku is concerned), as
well as considerations of a more general relevance with respect to theoretical research in
Optimality Theory, are also involved. In their account, compounds that do not undergo
rendaku are treated as mere exceptions, and the relevance of prosodic length to the
likeliness of rendaku occurring is not considered.
Page 25 of 29
4.2.5 Concluding Remarks on rendaku

Rendaku is sometimes presented as a unitary phenomenon, but it is not. As we have seen
throughout the preceding pages, rendaku is a process located at the intersection of
several fields of grammar—phonology (both segmental and prosodic), morphology,
syntax, semantics, and so on—where various forces come into conflict. The difficulty one
has in trying to get a full picture of rendaku precisely comes from the fact that rendaku
application or non-application depends on factors of many kinds, which concern almost all
dimensions of linguistic analysis, and hence exceptions can always be found.
What we need now are more corpus investigations with a wider coverage, on the model
of those conducted by Rosen (2003), Irwin (2009), or Vance (2005), but which would
target lexemes belonging to all word categories (nouns, verbs, adjectives), of both
Yamato and Sino-Japanese origin. Ideally, such a database would also contain prosodic
information (accent location as well as length), as well as, needless to say, semantic,
morphological, and syntactic parameters.
Psycholinguistic studies are also likely to provide new insights into rendaku, particularly
concerning the productivity of rendaku in the modern language, not forgetting the
manner in which it is acquired by Japanese children.
4.3 Post-nasal Voicing

Post-nasal voicing refers to the phenomenon whereby a voiceless consonant is replaced
by its voiced counterpart after the mora nasal /N/ in Yamato and (p.129) Sino-Japanese
words. Voicing of the nasal /N/ spreads to the following consonant. Typical examples are
kan ‘sick’ + sha ‘person’ 〉 kanja° ‘a patient’, or the well-known cases of the verbal
suffixes -te and -ta, which become -de and -da after a verbal base ending with /N/: ton-de°
‘to fly and…’, ton-da° ‘flew’ from tobu° ‘to fly’.12 In all the examples just cited, none of the
morphemes can occur in isolation. Note that mora nasals do not always cause voicing of a
following voiceless consonant, especially in reduplicated mimetics (kankan° / *kangan° ‘be
in rage’). Post-nasal voicing is often said to apply productively in Yamato words, in an
erratic manner in Sino-Japanese words, and never in Western words, but not a few
exceptions can be found in the Yamato stratum. For instance, we have chanto° ‘properly’,
yutanpo ‘hot water bottle’, ta npopo ‘dandelion’, chi nko ‘small person’ or ‘penis’ (child
language), Ju nko (personal name), chinpira° ‘young hooligan’, which are all native
(Yamato) words, except Ju nko whose first element is Sino-Japanese. Voicing can thus be
seen as contrastive in the post-nasal position even in Yamato words, as Rice (1997, 2005)
claims, even if minimal pairs are hard to find.
Post-nasal voicing must not be confused with rendaku, even though the two processes
involve alternation between a voiced and a voiceless obstruent. The difference lies in the
fact that post-nasal voicing plays no morphological role, unlike rendaku, and that it can
affect non-independent morphemes. So, whereas rendaku can be viewed as a
morphological marker, post-nasal voicing cannot.
It is true however that there are cases where it is not easy to decide whether the
alternation between a voiced and a voiceless consonant is due to rendaku or to post-nasal
Page 26 of 29
voicing. The two phenomena overlap when voicing occurs at morpheme juncture in
rendaku possible contexts when the first component ends with /N/ as in ho n ‘book’ +
tana° ‘shelf’ 〉 ho n-dana ‘bookshelf’.
‘Weak’ instances of post-nasal voicing also occur morpheme-internally, when no

alternation is involved, for instance in kangaeru (*kankaeru) ‘to think’ which is a simplex
verb of Yamato origin. However, examples of this sort are rare.
Itô and Mester (1995a) and Itô, Mester, and Padgett (1995) posit the action of a *NT
constraint in Yamato words to account for post-nasal voicing. *NT prohibits the
occurrence of a voiceless obstruent after the mora nasal, and voicing after a nasal mora
thus becomes a redundant feature (for a view challenging the analysis by Itô, Mester,
and Padgett, 1995, see Rice, 1997). They claim that *NT is only operative in Yamato
words, such as kangaeru ‘to think’, tonde° ‘fly and…’, akanbou° ‘baby’ but not in Sino-
Japanese or Western words, but, as we have seen, counterexamples are commonly
found. So it seems more reasonable to assume, following Ota (2004) who addresses the
question of the learnability of (p.130) the lexical strata in Japanese through the case
study of post-nasal voicing,13 that the distribution of post-nasal voicing is arbitrary in
Japanese from a synchronic point of view.
Rice (1997, 2005) argues that two types of voicing features exist phonologically in
Japanese. The first type, laryngeal voicing, is involved in rendaku, the second type,
sonorant voicing, is involved in post-nasal voicing. According to her, one of the empirical
differences between the two phenomena would be that post-nasal voicing is not blocked
by Lyman’s law morpheme-initially. However, to my knowledge, the example cited by
Rice, hun-jibaru (hun- + shibaru), is the only one that can be put forward to support this
assumption. It is rather surprising that no other example can be found, for instance in the
paradigm involving the morpheme ho n ‘book’ or ‘main’ as a first element. Words such as
*hon-zuji cannot be found (only hon-suji° ‘main thread’ exists), even though most
compounds starting with the morpheme ho n undergo post-nasal voicing, as ho n-dana
‘book shelf’, ho n-bako ‘book box’, and so on.
4.4 Voicing In Japanese, A Supra-segmental Feature?

The considerations above lead me to assume that in many respects voicing in Japanese
behaves more like a supra-segmental feature than like an infra-segmental one (an idea
already expressed by Komatsu, 1981, Kamei, 1997, Labrune, 1998a). As we have seen
throughout this chapter, voicing in Japanese displays a great number of characteristics
which are typically associated with supra-segmental features such as accent, stress, or
tone. For instance, voicing can function like a juncture marker in the case of rendaku. It
also fulfils a culminative function since only one voiced obstruent is authorized within a
simple word, as well as a demarcative function, because voiced obstruents normally do
not occur word-initially. The fact that two voiced obstruents do not normally cohabit
within a simple word constitutes a co-occurrence constraint which is prototypically
characteristic of stress, since a word cannot contain two primary stresses. Voicing is also
highly unstable, as we have seen in a number of examples in 4.1.3. In addition, there are
cases in which the voicing shifts to the preceding mora, in the manner of a stress or
Page 27 of 29
accent, as in kami-gakura, from kami-kagura, an example which appears in the

Vocabulario da Lingoa de Iapam (1604–1608, reprinted 1976, cited by Morita, 1977 and
Yamaguchi Yoshinori, 1988b).
(p.131) In the current Japanese writing system, the transcription of voicing on

obstruents is made by way of diacritics (the so-called dakuten), much like that of a
prosodic feature, thus reflecting its autosegmental nature. Recall also that voicing, like
accent, has been ignored for centuries in the Japanese script (in the same way as stress
is not denoted in the orthography of languages such as English or Russian). The dakuten,
the ‘voicing dots’, have been in systematic use since the turn of the twentieth century
only. Finally, it must be emphasized that for a long time accent and voicing have shared
the same orthographical symbol (Labrune, 1998a), since dakuten were originally used to
denote the presence of an accent on a given mora.
Notes:
(1.) Example found on the web in May 2009 in the Japan Times online, ‘Canny Japanese
playing it by the numbers’, by Mark Schreiber, 13 June 2002.
(2.) The same goes for h,p, and b, whose difference is ignored in the ordering of
dictionary headwords. We will find hari ‘needle’, bari ‘Bali’, pari ‘Paris’, before hariai
‘rivalry’.
(3.) On the h → b transformation, see section 3.7. Note also that the voiced counterpart of
the mora tsu is zu in the romanized transcription of Modern Tôkyô Japanese which has
been adopted in this book (see section 3.3).
(4.) However, the words katakana and hiragana also differ accentually. Katakana is
always tonic, with penultimate or antepenultimate accent (katakana,katakana), while
hiragana is tonic with final or penultimate accent, or atonic (hiragana,hiragana or
hiragana°).
(5.) See also below the example taken from a story by Hiraide Takashi, which illustrates
the fact that native speakers are sensitive to meaning differences expressed by rendaku
even in compounds that they have never heard before.
(6.) The exceptions which may come to mind are in fact non-ideophonic in origin, like
kori-gori, kori-gori° ‘learning at one’s cost’, from koriru ‘to learn by experience’,
taka-daka ‘very high’, from takai ‘high’. Such words also differ from true mimetics by the
accent pattern.
(7.) It would be interesting to check whether there is a correlation between the accent
pattern of these words and the fact that they resist rendaku, given the fact that none of
the examples cited here is accented on the initial.
(8.) A number of other nouns never voice in short–short compounds, for instance kata
‘shoulder’ or shimo ‘frost’, but since each of them occurs only in at most three
Page 28 of 29
compounds, all of them short–short, it is difficult to tell if they are behaving as rendaku-
immune nouns or rendaku-resister nouns.
(9.) Some rare exceptions exist. Vance cites the verb kaeri-zaku ‘bloom again’ (accent
unknown), from kaeru ‘to return’ and saku° ‘to bloom’.
(10.) Hiraide Takashi, 2001, Neko no kyaku (The Guest Cat). Reference is to the
paperback edition by Kawade Shobo Shinsha in 2009.
(11.) Noteworthy enough, dvandva compounds do not follow the general compound
accent rule either (see section 7.3.4), so it might be expected that identical semantic
constraints account for the resistance to compound accent and to rendaku, as observed
by Takeuchi (1999) based on Kubozono (1987).
(12.) The past auxiliaries -te and -ta are also voiced after the moraic front vowel i in the
flexion of -gu ending verbs: to ide ‘sharpening and…’, to ida ‘sharpened’, from to gu ‘to
sharpen’.
(13.) This article offers a good summary of the phenomenon and an interesting discussion
of the approaches which have been adopted in OT to handle such cases of non-uniform
phonology within a language.
Special segments

Laurence Labrune

Print ISBN-13: 9780199545834
DOI: 10.1093/acprof:oso/9780199545834.001.0001
Special segments
Laurence Labrune
DOI:10.1093/acprof:oso/9780199545834.003.0005
The term “special segments” refers to the three moraic segments of Japanese, which
constitute a rather unique feature of the language and have been granted special status
in traditional Japanese analyses, namely the mora nasal /N/, the first part of an obstruent
geminate /Q/ and the second part of a long vowel /R/. The chapter provides a description
of their phonetic realisation, their historical development, and their general phonological
properties in modern Japanese.
Keywords: moraic segments, mora nasal, gemination, obstruent geminates, long vowels, phonetic
realisation, phonological properties
Japanese scholars generally associate under the label ‘special phonemes’ (tokushu onso
), ‘special rhythmic units’ (tokushu haku ), or ‘mora phonemes’ (môra onso モーラ ),
three phonological elements which exhibit special phonological characteristics, the most
significant one lying in the fact that they are moraic. These units will be referred to as
Page 1 of 11
Special segments
‘special segments’ or ‘moraic segments’ hereafter. Special segments are:
– the mora nasal (hatsuon , haneru oto ねる ), noted /N/ in phonological

transcriptions, as in yanda° /yaNda/ [janda] ‘stopped’, sanpo° /saNpo/
[sampo] ‘a walk’, ho n /hoN/ [ hoN] ‘book’.
– the mora obstruent / first part of a geminate (sokuon , tsumaru oto つまる
), noted /Q/ in phonological transcriptions: kitte° /kiQte/ ‘stamp’, kossori
/koQsori/ ‘secretly’, hakken° /haQkeN/ ‘discovery’.
– the second part of a long vowel (chôon , hiku oto く ), noted /R/ in
phonological transcriptions: ho u /hoR/ ‘direction’, aa° /aR/ ‘like that’, ku uki
/kuRki/ ‘air’.
A number of Japanese structuralist phonologists only recognize /N/ and /Q/ as ‘special
moras’, and exclude /R/ from the set.1 Others (for example Jôo, 1977) sometimes
consider that a fourth element should be added to the list: the moraic palatal vowel i,
transcribed as /J/ phonologically. /J/ appears for instance in ka i /kaJ/ ‘shell’, oyoida
/oyoJda/ ‘swam’. This element is also the one which occurs after the vowel /e/ when it is
lengthened, as in sensei /senseJ/ ‘professor’. I assume that there is no need to
distinguish /J/ from the moraic vowel /i/, since there is no minimal pair such as /kai/ and
/kaJ/. Accordingly the above-mentioned examples will be transcribed as ka i /kai/, oyoida
/oyoida/, and sensei /seNsei/ or /seNseR/ in this book (see also section 2.7.1, as well as
Chapter 6, on this issue).
(p.133) In this work, I shall follow the Japanese approach (for instance Hamada, 1949) in
considering that /N/, /Q/, and /R/ are segments different in nature from vowels and
consonants.
Special segments are worth one prosodic unit (mora, haku) in the same way as CV
sequences. The trimoraic forms sanpo° /saNpo/ ‘a walk’, kitte° /kiQte/ ‘stamp’ and ku uki
/kuRki/ ‘air’, which each contain one special segment, are thus perceived as having the
same prosodic length (three moras) as the word sakura° ‘cherry tree’.
The phonological status of /N/, /Q/, and /R/ constitutes without doubt one of the most
intricate issues of Japanese phonology. In particular, it has to be acknowledged that it is
the very existence of these elements which has led some scholars to posit a distinction
between light and heavy syllables in certain recent approaches, or between normal,
independent rhythmic units (jiritsu haku ) and special rhythmic units (tokushu haku
) in traditional approaches. Since the issue of the mora and syllable in Japanese
phonology will be discussed in detail in the next chapter, in close connection with the
status of special segments, the present chapter will keep to a general presentation of /N/,
/Q/, and /R/’s phenomenology, without entering too deeply into the problem of their
phonological nature and structure.
After having presented each of the three special segments from the point of view of their
phonetic and phonological characteristics (sections 5.1 and 5.2, and 5.3), we will have a
look at their origin (section 5.4.), before considering their common properties (section
Page 2 of 11
Special segments
5.5). We shall finally provide some statistical data concerning their frequency (section 5.6).
5.1 /N/ (The Mora Nasal)

/N/, the mora nasal, is a generic nasal contoid, with no definite place of articulation. By
default, in slow speech and before a pause, /N/ is a uvular realized with no dorsal
occlusion and transcribed as [N] (1a). Before oral or nasal labial stops /p/, /b/, /m/, it is
realized as [m] (1b), before the alveolars /t/, /d/, /n/, as [n] (1c), and before the velars /k/
and /g/, as [ŋ] (1d). Before the fricatives /h/, /s/, and /z/, be they palatalized or not,
phoneticians disagree about its place of articulation: /N/ is either realized as the nasalized
version of the preceding vowel (Saitô, 1997, 2003), as a fricative nasal, or even a
nasalized high vowel [ɯ̃] or [ĩ] (Imada, 1981; Hashimoto Sh., 1950) (1e). Before the semi-
consonants /w/ and /y/ and before vowels, the special segment /N/ is phonetically a nasal
vowel whose quality is said to be that of the preceding vowel (1f). According to Akamatsu
(1997:58ff.) it is actually impossible to determine the exact quality of /N/ before vowels,
semi-consonants, and fricatives. In addition, before the liquid /r/, things are quite (p.134)
fuzzy. It would seem that /N/ is uttered as some kind of [ɯ̃], as the nasalized version of
the preceding vowel, or as a nasal which would be at the same time alveolar and uvular
(1g). Moreover, the presence of /N/ generally involves nasalization of the preceding
vowel (not noted in the examples below). To conclude, the descriptions of phoneticians
diverge considerably as to the exact articulatory nature of /N/.
(1) Phonetic realization of /N/
a. Before a pause
ho n
[ˡhoᶰ] /hoN/ ‘book’
b. Before [p], [b], [m]
kinpatsu͡°
[kimpaʦ͡ ɯ] /kiNpatu/ ‘blond’
bi nbou
[ˡbimboː] /biNboR/ ‘poor’
ge nmai
[ˡgemmai] /geNmai/ ‘whole rice’
c. Before [t], [d], [n]
honten°
[honteN] /hoNteN/ ‘main shop’
tonda°
[tonda] /toNda/ ‘flew’
sannen°
Page 3 of 11
Special segments
[sanneN] /saNneN/ ‘three years’
d. Before [k], [g], [ŋ]
kankei°
[kaŋkeː] /kaNkei/, /kaNkeR/ ‘relation’
kangae
[kaŋˡɡae], [kaŋˡŋae] /kaNgae/, /kaN ae/ ‘thought (noun)’
e. Before /h/, /s/, /z/
sanhujinka°
[saɯ̃ϕɯʑiŋka], [saãϕɯʑiŋka] /saNhuziNka/ ‘gynaecology’
sensei
[seẽˡseː],[seɯ̃ˡseː], [sez̃ ˡseː] /seNsei/ ‘professor’
kanji°
[kaãʑi], [kaɯ̃ʑi], [kaʑ̃ʑi] /kaNzi/ ‘Chinese character’
(p.135)
f. Before /w/, /y/, and before a vowel
kanwa°
[kaãɰa] /kaNwa/ ‘Sino-Japanese’
sen’you°
[seẽjoː] /seNyoR/ ‘exclusive use’
on’in°
[oĩiᶰ], [oĩʔiᶰ] /oNiN/ ‘phonology’
nan’ou°
[naãoː], [naãʔoː] /naNoR/ ‘Southern Europe’
g. Before /r/
enryo°
[ˡeɯ̃ɾjo], [ˡeẽɾjo], [ˡeN̪ ɾjo] /eNryo/ ‘reserve, discretion’
Another notable fact about /N/ is that it can never be linked to an onset position before a
vowel. It can never be resyllabified either. Thus the word on’in° /oNiN/ ‘phonology’ is
realized as [oĩ.iN] or [oĩ. iN] but never as *[o.niᶰ], *[on.niᶰ] or *[o.ᶰiᶰ].2 This impossibility of
resyllabifying /N/ and its significance for the phonological analysis of the moraic segments
will be discussed again in the next chapter (section 6.2.3).
Finally, it is noteworthy to observe that the graphemes ん and ン for the nasal mora /N/ in
Page 4 of 11
Special segments
the Japanese kana syllabaries are the only letters whose origin is totally unknown,
contrary to all other kana symbols.
5.2 /Q/ (gemination)

/Q/ represents a generic moraic oral obstruent with no specific place of articulation. It
occurs only before another consonant, except in a few special cases, in particular that of
interjections, in which /Q/ can be word-final. However, its function in that case is
expressive and non-distinctive.
The phonetic realization of /Q/ depends on that of the following consonant. /Q/ inherits the
totality of its articulatory features from the following obstruent. It is realized as
unreleased when it corresponds to a stop, so we have [p⌝] before /p/, [b⌝] before /b/, [t⌝]
before /t/, and so forth. Note that /Q/ is always noted by means of the small kana tsu (っ , ッ)
before an obstruent in the Japanese writing system, whatever its effective phonetic
realization.
(p.136) Consonants likely to undergo gemination, that is, likely to be preceded by /Q/ in
Yamato (including mimetics) and in Sino-Japanese are normally limited to the voiceless
obstruents /p, t, k, s/ and their palatalized counterparts (2a). The voiced geminates (bb,
dd, zz, gg) appear only in very recent loanwords, or under particular conditions (see
Chapter 4). When they exist, they tend to undergo devoicing (2b) especially if the word
already contains another voiced obstruent (Kawahara, 2006), as is the case in the
examples provided below. The geminate versions of /r/ and /h/ are almost non-existent,
except in some marginal examples such as recent borrowings or mimetic words (2c); (on
geminate /h/, see also section 3.7, as well as footnote 3, Chapter 3). The glides /w/ and /y/
are never geminated. Finally, recall that in Yamato (including mimetics) and Sino-Japanese
words the geminate counterpart of the fricative /h/ is normally [pp] (see section 3.7).
Nasal geminates do exist, but they are interpreted as /N/ + nasal consonant rather than
as /Q/ + nasal consonant. The kana spelling accordingly is minna ‘all’ みんな /miNna/ rather
than * みっな /miQna/.
(2) Realizations of /Q/
a. Before a voiceless obstruent
kappa° [kap⌝pa] /kaQpa/ ‘kappa (river imp)’

mo tto [ˡmot⌝to] /moQto/ ‘more’
kissaten° [kissateN] /kiQsateN/ ‘coffee shop’
hassha° [haɕɕa] /haQsya/ ‘departure’
matcha° [mat⌝tɕ͡ a] /maQtya/ ‘green matcha tea’
sekken° [sek⌝keN] /seQkeN/ ‘soap’
b. Before a voiced obstruent
ba ggu [ˡbaɡ⌝ɡɯ] /baQgu/ {bag}
Page 5 of 11
Special segments
gu zzu [ˡɡɯzzɯ] /guQzu/ {goods}
c. Before /h/ and /r/ (marginal cases)
ju hhari [ˡʑɯhhaɾi] /zyuQhari/ ‘ten stitches’

ba hha [ˡbahha], [ˡbaχχa] /baQha/ {Bach}
uhhuhhu(to) [ɯϕϕɯϕˡϕɯ] /uhhuhhu/ (onomatopoeia for laugh)
barra bara° [baɾɾabaɾa], [ballabala] /baQrabara/ ‘scatteringly’
Before a pause, /Q/ is realized as a glottal stop [ ]. It almost exclusively appears at the
end of interjections (are’° [aɾe ] /areQ/, an expression of surprise), at the end of
onomatopoeia and ideophones (bata’ [bata ] /bataQ/ ‘bang’), and at the end of adjectives
referring to sensations or feelings used in an interjective manner, in replacement of the -i
ending (for instance atsu’ [a ts͡ ɯ ] ‘it’s hot!’, from atsui ‘hot’).
(p.137) Recall finally the CV / Q alternation which occurs in Sino-Japanese morphemes

such as gaku / gaQ ‘study’, teki / teQ ‘enemy’ or in Yamato words such as do kka for do
ko ka which has already been discussed in section 2.5 in relation to vowel deletion. /Q/
can also occur as a juncture mark at morpheme boundary, for instance migi° ‘right’ +
kawa ‘side’ → migi-kkawa° ‘right side’.
5.3 /R/ (vowel Length)

/R/ is a vocoid segment which can be phonologically defined as a generic vocalic position
with no articulatory specifications of its own. Like /N/ and /Q/, it is moraic. /R/ thus
corresponds to the lengthening of the preceding vowel, from which it inherits all its
quality specifications. (See also 2.7 for additional examples and discussion.)
(3) Realizations of /R/
okaasan [oˡkaːsaN] /okaRsaN/ ‘mother’

oniisan [oˡniːsaN] /oniRsaN/ ‘older brother’
ku uki [ˡkɯːki] /kuRki/ ‘air’
oneesan [oˡneːsaN] /oneRsaN/ ‘older sister’
koukou° [koːkoː] /koRkoR/ ‘high school’
5.4 The Origin of Special Segments

It is generally assumed that special segments first emerged in Japanese between the
ninth and eleventh centuries under the influence of Chinese (Hamada, 1952; Komatsu,
1981), a language in which there existed heavy syllables ending in a consonant or a glide,
and from which Japanese made massive borrowing at the time. Some of the sound
changes which resulted in the development of the so-called special segments are known
as onbin ‘sound change’ in Japanese linguistics (see, among others, Hamada, 1949,
1951, Kishida, 1984, Frellesvig, 1995). Today, special segments are found in all lexical
Page 6 of 11
Special segments
strata and not only in Sino-Japanese words. They are the result of one of the three
following processes: reduction of a CV sequence (4) (see also section 2.7.2), expressive
strengthening (5), and adaptation of foreign sounds (6):
(4) Special segments resulting from the lenition of a CV sequence 3
shini-ta 〉 shinda° /siNda/ ‘died’ (Yamato)

tori-ta 〉 to tta /to Qta/ ‘took’ (Yamato)
kamibe 〉 ko ube /ko Rbe/ ‘Kôbe’ (Yamato)
kehu 〉 kyo u /kyo R/ ‘today’ (Yamato)
gakuki 〉 gakki° /gaQki/ ‘musical instrument’ (Sino-Japanese)
(p.138)
In contemporary colloquial Japanese, /N/ and /Q/ are sometimes the result of the lenition
of a nV or rV sequence, for instance aruite iru no 〉 aruite inno ‘are you walking’ or sou°
suru° to 〉 sou° sutto ‘doing this’.
(5) Special segments marking expressive reinforcement (by addition of a rhythmic

position)
kogari 〉 kon gari /koN gari/ ‘well well roasted’ (mimetic)

onaji° 〉 onnaji° /oNnazi/ ‘quite identical’ (Yamato)
totemo° 〉 tottemo° /toQtemo/ ‘veeeery’ (Yamato)
zutto° 〉 zuutto° /zuRQto/ ‘contiiiiinuously’ (mimetic)
(6) Special segments resulting from the adaptation of foreign sounds
Old Chinese *tong 〉 jp. to u /to R/ ‘East’

Old Chinese *sam 〉 jp. san° /saN/ ‘three’
Old Chinese *niet 〉 jp. (go reading) nichi,nit- /niti/, /niQ/ ‘sun’
English pet 〉 jp. pe tto /pe Qto/ ‘pet’
French Bordeaux 〉 jp. bo rudou /borudoR/ ‘Bordeaux’
In Sino-Japanese words, /Q/ is the reflex of one of the three Old Chinese syllable-final
implosives p,t, and k (known as , rusheng in Chinese, nisshô in Japanese, lit. ‘entering
tone’).
5.5 Properties of Special Segments

Special segments share a number of properties, that distinguish them from ‘regular’
consonants and vowels. First, they lack phonological autonomy. For instance, they can
Page 7 of 11
Special segments
never appear in isolation or at the beginning of a word.4 Second, they are moraic (see
Chapter 6 for examples and discussion). Thirdly, they are, in theory, unaccentable. This
principle, which we shall call the NADM principle (Non-Accentuation of Deficient Moras),
will be discussed in more detail in sections 6.2.4, and 6.3.2, and 7.1.4. and Let us merely
observe that in cases where the accent should fall on a special segment, it shifts to the
preceding mora. (p.139) Thus, according to the traditional account in Japanese
phonology in compound words of the type N + e ki ‘N station’, or N + ka i ‘club of N’,
the accent normally falls on the last mora of the first component, except if it ends in /N/,
/Q/, or /R/, in which case the accent moves to the penultimate mora, as shown in (7):
(7)
na goya + eki nagoya-eki ‘Nagoya station’

na goya + kai nagoya-kai ‘Nagoya club’
but
toukyou° + eki toukyou-eki *toukyou-eki ‘Tôkyô station’
ro ndon + eki rondon-eki *rondon-eki ‘London station’
toukyou° + kai toukyou-kai *toukyou-kai ‘Tôkyô club’
ro ndon + kai rondon-kai *rondon-kai ‘London club’
The same phenomenon occurs in verbs to which the -ta (perfect) auxiliary has been
added. The suffixation of -ta normally causes the accent to strike the antepenultimate
mora in tonic verbs. However, if the antepenultimate consists of a special segment, the
accent moves one mora leftward, striking the pre-antepenultimate mora:
(8)
/tabe/ + /ta/ ta beta /tabeta/ ‘ate’

/atume/ + /ta/ atsumeta /atumeta/ ‘gathered’
but
/roNzi/ + /ta/ ro njita /roNzita / *ronjita ‘argued’
/moRsi/ + /ta/ mo ushita /moRsita/ *moushita ‘said (humbly)’
Let us consider another instance of this phenomenon. In foreign loans, the default accent
normally falls on the antepenultimate mora: o pera /opera/ {opera}, sandoitchi
/saNdoiQti/ {sandwich}, maikurohon /maJkurohoN/ {microphone}, sanhuranshisuko
/saNhuraNsisuko/ {San Francisco}. However, if the antepenultimate mora is /N/, /Q/, or
/R/ (and sometimes /i/), the accent shifts to the pre-antepenultimate mora: disukasshon
/disukaQsyoN/ *disukasshon {discussion}, repoutaa /repoRtaR/ *repoutaa {reporter}
(see Chapter 6 for additional examples and discussion of this process).
Page 8 of 11
Special segments
Exceptions to this principle are rare, but they do exist. They pose a challenge to syllable-
based analyses of Japanese, as we shall see in the next chapter. These exceptions occur in
case several special segments follow one another: for instance, in the words obaasankko
/obaRsaNQko/ ‘child cherished by his/her grandmother’ or cheen-ten /tyeRNteN/ ‘chain
store’, in which there is a succession of /N/ + /Q/, and /R/ + /N/, the special segments /N/
and /R/ are high-pitched, that is, accented. This characteristic will be analysed in the next
chapter.
(p.140) Vance (1987:81) also reports a personal communication from Hamano Shoko
stating that some words may be pronounced with an accent on a special segment in some
marginal cases when an unaccented word is quoted emphatically (see the next chapter,
section 6.2.4 for examples and discussion).
Moreover, it is interesting to observe, as Jôo (1977) does (among others), that the
special segments frequently alternate with each other and with i, in a somewhat loose
manner, as the following examples illustrate:
(9)
shouben /syoRbeN/ ~ shonben /syoNbeN/ ‘urine’

omotta /omoQta/ ~ omouta /omoRta/ (dialectal) ‘thought (verb)’
bo uzu /boRzu/ ~ bo nzu /boNzu/ ‘bonze’
toiya° /toiya/ ~ ton’ya° /toNya/ ‘wholesale dealer’
e cchi /eQti/ ~ e echi /eRti/~e ichi /eiti/ ‘h (name of the letter)’
ki rei /kirei/ ~ ki ree /kireR/ ‘pretty’
In ancient Japanese documents, there are frequent confusions in the kana spelling of /N/,
/Q/, /R/ and the mora /i/. One sometimes appears instead of the other, for instance the
word te nki in Modern Japanese, ‘weather’ is noted as teike in a text dating from the
beginning of Heian, with i /J/ replacing n /N/ (Hamada, 1952), or nobotte° って ‘climbing
and…’ is noted as nobonte in an eleventh-century text, with n /N/ replacing /Q/, the first
part of the geminate (Komatsu, 1981:199). According to Hamada (1949), /N/ and /Q/ had
been written using the same symbol until the Muromachi period. This could lead us to
regard /N/ as a sort of nasalized /Q/.
Kuroda (1967) has proposed an analysis of /Q/ and /N/ in Yamato, mimetics, and Sino-
Japanese within a linear generative framework (Chomsky and Halle, 1968) whereby /Q/
and /N/ are unspecified consonantal segments, the former being [−vocalic, +consonantal,
−voiced, −nasal] and the latter [−vocalic, +consonantal, −voiced, +nasal]. In mimetics
and verbal flexion, Kuroda implicitly assumes that /Q/ and /N/ are one and the same
segment, because they stand in complementary distribution in these two native
subclasses of the Japanese lexicon. In the next chapter (section 6.3), I will present an
analysis of the internal structure of /Q/ and /N/ along the same lines as that of Kuroda, but
Page 9 of 11
Special segments
cast in a multilinear representational framework.
The underspecified status of /N/ and /Q/ in Japanese has also led Itô (1988) to formulate
the Coda Condition. Itô assumes that Japanese /N/ and /Q/ occupy the coda position in
heavy syllables. The Coda Condition states that a coda cannot license place features,
which forbids /N/ and /Q/ to possess their own place (p.141) specifications. Place
specifications are acquired through propagation from the following segment.
5.6 Relative Frequency of Special Segments

According to Imae (1960, cited in Hayashi O., 1982), /N/ and /Q/ respectively account for
4.7% and 2.3% of all Japanese moras in textual (token) frequency. This source does not
provide any data for /R/.
Notes:
(1 ) For instance, Hattori Shirô (1960, and other papers) does not need the unit /R/
because he posits a zero consonant /’/ which enables him to distinguish satooya° ‘foster
parent’ from satouya ‘sugar seller’, which he transcribes respectively as sato’oya and
satooya. In the transcription adopted in this book, these two forms are transcribed as
sato-oya° /satooya/ and satou-ya /satoRya/. The former has two full o’s which follow each
other, the latter has a long o (see section 2.7.3 for a discussion on the representational
difference between the two entities).
(2) In Pre-modern Japanese, the process known as renjô ( ‘liaison’) sometimes occurs
in this type of context. Some vestiges of this remain in the modern language, but they are
totally lexicalized, like tennou ‘emperor’ from te n /teN/ ‘heaven’ and o u /oR/ ‘king’,
rather than ten’ou as one would expect. See Vance (1987:164ff.) for a presentation of this
phenomenon.
(3) This process is known as onbin ( ‘sound change’) only when it applies to Yamato
words.
(4) The only exception in Standard Modern Japanese is the interjection un ‘uh huh’,
realized as [ ɯ̃ː]. One can also mention the marginal, obsolete realizations [m me] for
ume° ‘plum’, and other words beginning with u + nasal (see section 3.8).
Prosodic units

Laurence Labrune

Print ISBN-13: 9780199545834
DOI: 10.1093/acprof:oso/9780199545834.001.0001
Prosodic units
Laurence Labrune
DOI:10.1093/acprof:oso/9780199545834.003.0006
This chapter is devoted to the prosodic units of Japanese, the mora, the syllable, the foot,
the prosodic word, and the prosodic hierarchy. It reviews the evidence demonstrating
the central role played by the mora in Japanese phonology. It then proceeds to a re-
examination of the status of the syllable, which has been argued in a number of recent
works to be an indispensable prosodic unit, alongside the mora and the foot, although it
has been absent from the work of most native Japanese phonologists who have always
been content with the mora. Taking as a basis the author’s extensive research on the
subject, this chapter argues that the syllable is not a relevant unit in the phonology of
Japanese, and it shows how all the phenomena which have been inputed to the action of
the syllable can be accounted for with exclusive reference to the mora and the foot.
Keywords: mora, syllable, foot, prosodic word, prosodic hierarchy
This chapter is devoted to the prosodic units of Japanese, the mora, the syllable, the foot,
Page 1 of 36
Prosodic units
the prosodic word, and the other upper units.
Whereas there is no doubt about the relevance of the mora, the foot, and the prosodic
word in Japanese phonology, things are much less clear with regard to the syllable in the
most usual sense of the term, i.e. a prosodic constituent which can be structurally light
(two-slot syllables, like ka) or heavy (three-slot syllables, like kan,kou [koː], or kai). The
claim I would like to put forward in this chapter is that the syllable actually plays no
relevant role in Tôkyô Japanese and that this language is a mora-counting mora language,
thus rehabilitating the Japanese native linguistic tradition which has long been satisfied
with what corresponds to the mora for the analysis of the various prosodic phenomena of
the language. However, it should not be forgotten that some dialects of Japanese such as
the Aomori or Akita dialects (north of Honshû) or the Kagoshima dialect (south of
Kyûshû) are indisputably syllabic and held as such by the proponents of a moraic analysis
of Tôkyô Japanese, while others, for instance the Kyôto/Ôsaka dialect or the Izu dialect
(south of Tôkyô) are clearly moraic and held as such even by the advocates of a syllabic
analysis of Tôkyô Japanese. Japanese phonologists generally operate a distinction between
‘mora dialects’ (haku hôgen ) and ‘non-mora dialects’, i.e. syllable-based dialects (hi-
haku hôgen , see for instance Hirayama et al., 1993, Satô R., 2002). Actually, it is
mainly Tôkyô Japanese that poses a problem with regard to its classification as a mora or
syllable dialect, being analysed either as only moraic, or as syllabic with moras acting as
subconstituents of syllables.
In section 6.1, we shall review the evidence demonstrating the central role played by the
mora in Japanese phonology. Section 6.2 critically reassesses the role and relevance of the
syllable in the standard language, through a review of the scholarship and reexamination
of the alleged evidence in favour of the syllabic approach. It will be shown that the
relevance of a light/heavy syllabic distinction is extremely difficult to justify on the basis of
the language’s internal evidence. All the phenomena which have been imputed to the
action of the syllable in Tôkyô Japanese can be accounted for by exclusive reference to
the mora and to the foot. In section 6.3, I present a model of the basic prosodic unit
(prosodeme) of Japanese that does not rely on the syllable, in keeping with the (p.143)
traditional Japanese approach. Instead, two different types of prosodemes (= moras) are
distinguished: regular CV prosodemes and weak, or deficient, prosodemes, which lack
one of the two components V or C. Section 6.4 is dedicated to the foot. We will see that
Japanese feet obey a structural constraint that stipulates that they start with a regular
mora. Section 6.5 introduces the other upper levels of the Japanese prosodic hierarchy,
and section 6.6 offers a conclusion and summary.
6.1 The Mora

The positive evidence for the mora as a basic, autonomous prosodic unit in Japanese is
well established. The mora is the unit of rhythm and of prosodic measurement of the
Japanese language. It is the only prosodic unit that has been recognized by the native
linguistic tradition, which calls it haku (Kindaichi Haruhiko, 1972a; Kindaichi and Akinaga
2001), môra モーラ (Hattori Shirô, 19601 ), or sometimes onsetsu (Arisaka Hideyo, 1940;
see for example Kindaichi 1972b for a general discussion about these questions).
Page 2 of 36
Prosodic units
Each articulated mora occupies one rhythmic unit. It is perceived as isochronous to other
moras (for a review of the phonetic research about the Japanese mora as an isochronous
unit, and other questions, see Warner and Arai 2001). Japanese moras may have the
following structure:
(1) Structure of Japanese moras
CV in /sa/, /ko/, /ni/, etc.

CyV (with a palatalized consonant) in /nya/, /kyu/, /tyu/ (chu), /sya/ (sha),
etc.
V in /a/, /i/, /u/, /e/, /o/
/N/ (the mora nasal) in /hoN/ ho n ‘book’
/Q/ (first part of an obstruent in /moQte / mo tte ‘to hold and…’
geminate)
/R/ (second part of a long vowel) in /toR/ to u ‘tower’, /maR/ ma a ‘euh
well’
The word hontou° ‘true’ is therefore of the same phonological length as kaminari
‘thunder’, that is, four moras.
The modern standard language has 103 distinctive moras, as shown in Table 6.1, where
plain moras (chokuon ) stand in opposition to their palatalized counterpart (yôon ).
Table 6.1. The 103 distinctive moras of Modern Standard Japanese in

phonological transcription (consonants that have phonologized only
recently, see Chapter 3, are not included)
Plain moras (V or CV) Palatalized moras (CyV)
あ a い i う u え e お o – – –
か ka き ki く ku け ke こ ko きゃ kya きゅ kyu きょ kyo
が ga ぎ gi ぐ gu げ ge ご go ぎゃ gya ぎゅ gyu ぎょ gyo
さ sa し si す su せ se そ so しゃ sya しゅ syu しょ syo
ざ za じ zi ず zu ぜ ze ぞ zo じゃ zya じゅ zyu じょ zyo
た ta ち ti つ tu て te と to ちゃ tya ちゅ tyu ちょ tyo
だ da ぢ di づ du で de ど do ぢゃ dya ぢゅ dyu ぢょ dyo
な na に ni ぬ nu ね ne の no にゃ nya にゅ nyu にょ nyo
は ha ひ hi ふ hu へ he ほ ho ひゃ hya ひゅ hyu ひょ hyo
ば ba び bi ぶ bu べ be ぼ bo びゃ bya びゅ byu びょ byo
ぱ pa ぴ pi ぷ pu ぺ pe ぽ po ぴゃ pya ぴゅ pyu ぴょ pyo
ま ma み mi む mu め me も mo みゃ mya みゅ myu みょ myo
や ya – ゆ yu – よ yo – – –
Page 3 of 36
Prosodic units
ら ra り ri る ru れ re ろ ro りゃ rya りゅ ryu りょ ryo
わ wa ゐ wi – ゑ we を wo – – –
ん N
っ Q
－ R
The groups in italics correspond to sequences that are no longer distinctive in the
contemporary language but for which there exists a specific kana or kana combination.
(p.144) In Japanese traditional phonology, where the mora is considered to be the only
relevant prosodic unit, the following two types of moras are distinguished:
– Autonomous (or regular) moras (jiritsu haku ), which have the

structure CV, CyV, or V, except for the mora い i made up of the onsetless
vowel /i/ when it occurs word-internally, for instance in daigaku° ‘university’.
– Special moras (tokushu haku ) corresponding to the mora nasal /N/, to
the first part of an obstruent geminate /Q/, to the second part of a long vowel
/R/, in other words, the elements identified as the special segments (see
Chapter 5). The onsetless /i/ い which occurs word-internally (denoted /J/ in
some phonological transcriptions) also belongs to this list.
The mora is the metric unit of Japanese verse in poetry and song. Any mora, be it
autonomous or special, stands as one beat. In order to illustrate this point, let us (p.145)
consider the following haiku by the poet Yosa Buson. A haiku is composed of three
verses. The first verse contains five moras, the second one seven, the last one five.
Ikken no (five moras)

chamise no yanagi (seven moras)
oinikeri (five moras)
‘The willow tree
By the lone tea house
It has grown old’2
The first2 verse, ikken no, occupies five rhythmic beats, or slots: i.k.ke.n.no and not
three, as would prima facie suppose an English or a Chinese ear. Similarly, in
oinikeri,o.i.ni.ke.ri, the vowel i occupies one beat.
The same applies in songs. Each mora, including /R/, /N/, or /Q/, generally corresponds to
a rhythmic unit which is sung over one musical beat, even if exceptions are sometimes
found (Tanaka, 2008).
Page 4 of 36
Prosodic units
The mora is also granted a written status in the kana writing system, so that any mora,
including the second part of a long vowel, the mora nasal, or the first part of a geminate,
occupies one virtual square on the sheet. The four-mora words ikken いっけん ‘one
building’ or hontou° ほんとう ‘true’ thus fit in the same writing space as the words okinawa°
おきなわ ‘Okinawa’ or kaminari° かみなり ‘thunder’, namely four virtual squares.3 The only
exception to this principle is (p.146) that of palatalized consonants. CyV clusters,
although being only one mora long, occupy two squares.
The mora also functions as the major prosodic unit in a number of language games, for
example the babibu language analysed by Haraguchi (1991), which consists of inserting a
b + V sequence after each mora. The word sakura° ‘cherry tree’ is thus coded as saba-
kubu-raba, the word nippon° as nibi-tsubu-pobo-nbu.4 These examples clearly illustrate
the fact that sakura and nippon are respectively made up of three and four distinct and
independent units of identical phonological weight. The popular shiritori game (literally
‘buttock taking’) consists of finding a word starting with the final mora of the word given
by the preceding player (Katada, 1990), as in the following example: hutsuu° ‘normal’→
uguisu ‘bush warbler’→ surume° ‘dry squid’→ meirei° ‘order’→ iruka° ‘dolphin’.5 Here
again, we note that the final vowel length of hutsuu°, or the moraic /i/ of meirei° are
treated as single, autonomous units.
The mora also constitutes the phonetic support of pitch accent, in the sense that two
moras belonging to what can be considered a heavy syllable can be articulated in two
different registers. For example, in the trimoraic word kyo uto HLL ‘Kyôto’, the pitch fall
occurs after the initial mora kyo, that is, before the second part of the long vowel. Note
that in some cases, even the second part of what could be interpreted as a long syllable
by advocates of the syllabic analysis can receive an accent in Tôkyô Japanese (examples
are given below, see also section 5.5). Moreover, as Uwano (2003) reports, there also
exist dialects closely linked to the Tôkyô variety in which we find a contrast in surface
melodies between a ‘heavy syllable’ accented on the first part and a ‘heavy syllable’
accented on the second part, as ka i ‘paddle’ and kai ‘shell’.
Several psycholinguistics studies, in particular those of Otake et al. (1993), have

established that it is the mora, rather than the phoneme or the syllable, which constitutes
the basic unit of processing, production, and perception in Japanese. Such studies have
shown, for example, that Japanese listeners detect the target unit mo in the stimulus
word monka as fast as they do in monaka, which means that monka and monaka begin
with the same prosodic unit and that they should (p.147) consequently be segmented
as mo.n.ka and mo.na.ka, that is, under a mora-based segmentation procedure rather
than as mon.ka and mo.na.ka, under a syllable-based segmentation procedure. This
contrasts radically with the responses made by listeners whose native language is
syllabic, and which are asked to perform the same type of task (Mehler et al., 1981). Thus
the response time of French listeners is faster for palace than for palmier when the
target pa is presented. This result is interpreted as indicating that the French words
palace and palmier do not start with the same prosodic unit, whereas the Japanese
monka and monaka do.
Page 5 of 36
Prosodic units
Finally, in speech errors, special moras (including the mora /i/) show a frequent tendency
to replace each other or to be copied elsewhere in a given utterance (Kubozono 1989,
1996).
(2) Speech errors
Target Realization (accents unknown)

ku ubo middowei 〉 kubbo middowei ‘Aircraft Carrier Midway’
ju ugo paa sento 〉 juugo pansento ‘fifteen per cent’
be ichuu kankei° 〉 beichuu kaikei ‘US – China relations’
Note that in the above examples a vocoid element (/i/ or /R/) can be replaced by a contoid
element (/N/ or /Q/), and vice versa. This, according to Kubozono (1989, 1996), shows
that post-nuclear vowels have the same status as post-nuclear consonants, a
phenomenon that is not generally observed in English for instance. The explanation for
this is that the unit of coding of the two languages is different: it is the mora in Japanese,
the syllable in English. In other words, Japanese moras, including special segments,
behave as independent constituents with respect to speech errors. Kubozono (1985,
cited in Kubozono 1995c, 1996) also reports that CVV or CVC sequences are replaced
by CVCV sequences more frequently than by CV sequences. This suggests that what can
be analysed as a heavy syllable is phonologically equivalent to a sequence of two short
syllables. Post-nuclear elements such as /N/, /R/, or /Q/ serve as one prosodic unit just as
CV sequences do. Kubozono (1995a, c) also demonstrates that Japanese speakers tend
to segment words on a mora-based pattern rather than on a syllable-based pattern in
blends.
6.2 The Syllable?
6.2.1 Preliminaries
It is only recently that the syllable has been assumed by a number of modern
phonologists to be a fundamental unit that cannot be dispensed with in the analysis
(p.148) of Standard Japanese alongside the mora,6 following James McCawley, who
claims, in his 1968 thesis, that ‘the prosodic unit of Japanese is the syllable and not the
mora’.
McCawley (1968) posits two syllable types in Japanese: short syllables and long syllables,
and claims that Japanese is a ‘mora-counting syllable language’ (p. 134). McCawley’s
position is in the spirit of Trubetzkoy’s (1939) who considers that there exist two types of
languages: mora-counting languages, which also have syllables, and syllable-counting
languages, which may not have moras. Japanese is supposed to belong to the first type.
With practically no exceptions, McCawley’s generativist and post-generativist successors

trained in the Western linguistic tradition (from linear-SPE-type phonologists to OT ones)
make explicit or implicit reference to the syllable in their accounts of Japanese phonology,
Page 6 of 36
Prosodic units
the terms ‘light’ and ‘heavy’ being generally preferred to those of ‘short’ and ‘long’ to
refer to syllable weight differences. One exception is Higurashi (1983), who adopts a
strictly moraic-based analysis of Japanese phonology. For the majority of the studies
which have tackled the question, the issue is not so much that of the existence of the
syllable itself: this point is taken for granted, since the syllable is granted universal status
in modern phonological theory. Rather, it is the existence of the heavy syllable CVC or
CVV, with three constituents, which is at stake.
After reviewing the various models of the Japanese syllable which have been proposed in
the literature, we shall carry out a critical examination of the main arguments which are
supposed to support the existence of heavy syllables in the language. It will be shown that
there exists no direct indisputable empirical evidence of the phonological reality of the
syllable,7 and that the phenomena which have been regarded as such can receive an
alternative account, by referring exclusively to other prosodic constituents, namely the
mora and the foot.
6.2.2 Previous Proposals Concerning the Structure of the Japanese Syllable

As already stated, for McCawley and his successors, Japanese possesses light syllables,
which are of the following structure: (C)V or CyV, and heavy syllables, (p.149) which
are (C)VC or CyVC, or (C)VV or CyVV. Super-heavy syllables such as sa in (CVVN)
{sign} or zuut in zu utto (CVRQ) ‘continuously’ must be added to this inventory, but
they are far less common.
The details of the exact organization of these various types of Japanese syllables may vary
according to different scholars, but they all share a common basic structure with three
positions.
The moraic framework as developed by Hyman (2003 [1985]) or Hayes (1989), for
instance, considers heavy syllables as having a ternary structure. They are made up of an
onset which has no moraic weight, of a nucleus which is worth one mora, and of a final
element V or C, also worth one mora, as shown in the representation in (3a) below. It has
sometimes been proposed, to account for the Japanese case, that the onset is attached to
the mora containing the nucleus, as in (3b), and not directly to the syllable (Kubozono
1989, 1995a, inspired by Hyman 2003 [1985]). Other works (Terao 1992; Kubozono
1994, 1998b) have gone as far as to dissociate the mora and the syllable (3c), each being
directly and independently associated with the segments, according to a model in which
the mora is no longer a syllable-dependent constituent. The choice of dissociating the
mora from the syllable is rather unorthodox, and not properly discussed by its
proponents. Clearly, it constitutes an attempt to bypass the extremely thorny problem of
how the mora and the syllable are supposed to interact with each other in Japanese.
Lastly, Haraguchi (2003) has proposed that the syllable is structured with a core and a
coda, the core itself being divided into an onset and a nucleus, as in (3d). According to
him, and rather surprisingly, this is supposed to be the unmarked option of syllable
structure, while the onset–rhyme model would be a marked one.
Page 7 of 36
Prosodic units
(3)
The models of the Japanese syllable presented in (3) are seemingly very different but
they all share a common characteristic: none of them recognizes the existence of a rhyme
constituent (consisting of a nucleus and a coda), contrary to most current general
approaches to syllable organization. This choice is revealing of a strong particularity of
Japanese. It is justified by massive evidence showing that the (p.150) rhyme does not
exist in Japanese as a subconstituent, as we shall see below8 (see also the discussion in
Vance, 2008:120).
Clearly, whatever the theoretical framework, the discrepancies in the conceptions and
representation of the Japanese syllable significantly differ from those of most current
models of phonology. The fact that a number of leading phonologists have come to adopt a
rather unorthodox view of the Japanese syllable is, by itself, revealing of the existence of
a major problem with regard to the prosodic structure of Japanese.
6.2.3 Absence of Positive Evidence for the Syllable Or for A Heavy/light Syllabic Distinction
This section presents different strands of evidence which point to the non-relevance of
the syllable itself, or of the syllabic constituents (rhyme or coda) which would support the
opposition between light and heavy syllables.
– Lack of psycholinguistic evidence
First, one has to acknowledge the fact that not a single one of the many psycholinguistic
studies that have been conducted in the last decades has been able to establish the
cognitive reality of the syllable in Japanese, whereas many works have established the
central role of the mora and the foot both at the perceptual and cognitive levels (see the
papers in Otake and Cutler, 1996, and subsequent research by the authors therein).
– Absence of phonetic clues for the existence of a rhyme-like constituent
Considerable evidence points to the non-existence of a rhyme constituent in Japanese

would-be syllables. Segments which can be interpreted as belonging to the rhyme of a
heavy syllable do not behave as expected of rhymes or of codas, on the basis of what can
be observed in other languages.
First, the vowels which occur in presumably closed syllables of the shape CVC are not
phonetically shorter than those that occur in presumably open CV syllables (Homma
1981). Quite the opposite: vowels can even be longer in allegedly closed syllables. Han
Page 8 of 36
Prosodic units
(1994) and Idemaru (2005) have shown that in (p.151) Japanese a vowel preceding a
geminate consonant in /CVQCV/ is longer than a vowel preceding a singleton in /CVCV/.
Kubozono (1999a:34) states that, ‘in Japanese, vowel duration is independent of the
difference in syllable structure […]. These exceptional temporal patterns shown by
Japanese can be properly accounted for if the mora is posited as a unit of temporal
organization in the language’ (see also Sagisaka, 2009 and the works cited therein). This
situation contrasts radically with what is observed in languages with closed syllables, like
English for instance, in which phenomena of temporal compensation exist between the
nucleus and the coda (Maddieson, 1985), and in which vowels are shorter before
geminate consonants than before single ones. This provides evidence against the
existence of a rhyme component in the Japanese syllable symmetric to the onset, and
authorizes us to consider that CVC sequences must be divided as CV/C just as CVCV is
divided as CV/CV (see also Kubozono 1995c, already cited above). The fact that the first
vowel is longer in /CVQCV/ (three moras) than in /CVCV/ (two moras) can be explained as
an effect of temporal compensation within the bimoraic foot.
In addition, Campbell and Sagisaka (1991), cited by Kubozono (1995c), show that in
Japanese it is between the onset and the nucleus that effects of temporal compensation
occur. The duration of a nuclear vowel is inversely proportional to the intrinsic duration
of the consonant preceding it. This causes the same onset consonant to be shorter before
/a/ than before /i/ and /u/, the two shortest vowels in Japanese.
Even within analytical frameworks which do not recognize the rhyme, such as those
mentioned above, one would expect phonetic assimilations or temporal compensation to
occur at least as frequently between segments in position 1 and 2 as between segments
in position 2 and 3 within a single putative ‘syllable’. However, this does not happen in
Japanese, suggesting that a unit such as the heavy syllable is not relevant phonologically.
One also observes that in Japanese, the link between C and V is phonetically tighter than
the link between V and C or V1 and V2 in a would-be heavy syllable. Many restrictions
hold between an onset consonant and the following vowel, but there are no such
restrictions between the vowel and the following would-be tautosyllabic consonant
(Kubozono, 2006a). To take just one example among many, /t/ becomes [t͡s] before /u/,
͡
and [tɕ] before /i/ (see section 3.3). If heavy syllables exist, one would expect that
phonetic assimilations similar to those occurring between two distinct syllabic components
(the onset and the nucleus) would be even more frequent within a unique syllabic
component, i.e. a branching nucleus in the case of a V1V2 sequence, or a rhyme in the
case of VC. However, this is not the case. According to Kubozono (2006a), this shows
that the coda consonants are phonologically independent of the V preceding them.
(p.152) The only obvious case of a would-be coda to nucleus featural assimilation could
be that of the mora nasal /N/. Recall that vowels placed before /N/ can be phonetically
nasalized in Japanese; for instance ho n ‘book’ tends to be realized as [¹hõN].9 However,
feature assimilation involving /N/ extends beyond the strict domain of the putative
syllable, since /N/ also undergoes place assimilation from the following consonant, as in
Page 9 of 36
Prosodic units
sanpo° /saNpo°/ ‘a walk’ realized as [sãmpo], so this phenomenon cannot be taken as

evidence of the relevance of a prosodic domain such as the syllable (and not even as
relevance of the foot).10
– Absence of onset optimization
Another convincing argument against the existence of heavy syllables at the phonological
level can be found in the behaviour of the mora nasal /N/. In Japanese, a VNV sequence is
never syllabified as V.NV (two moras) but as V.N.V (three moras). This is in contradiction
with what is generally assumed in phonology, as stated by Golston and van der Hulst
(1999):
It is a widely observed fact that a sequence of a closed syllable followed by a

syllable that starts with a vowel is empirically unattested. The traditional view that
assumes that linearly organized strings form the input for syllabification explains
this by saying that a string VCV is universally syllabified as V.CV. This is due to the
constraints that Prince and Smolensky (1993) call Onset and NoCoda.
However, counterexamples of this principle can be commonly found in Japanese.

Consider the following contrasts:
(4)
a. an’i µµµ /aNi/ [ˡaᶰ.i] ‘ease’

b. ani µµ /ani/ [ˡani] ‘older brother’
c. anni µµµ /aNni/ [ˡanni] ‘implicitly’
Here, (4)a corresponds exactly to the type which is supposed to be ‘empirically

unattested’ if analysed through the syllabic mirror.
The moraic /i/ behaves similarly to the mora nasal with respect to onset non-optimization.
In the word baiorin° /baioriN/ ‘violin’, the palatal does not group with the vowel o which
follows (*ba.yo.ri.n) in the standard variety of the (p.153) language. It keeps its moraic
status: ba.i.o.ri.n. This word is thus realized in five moras, not in four.
This constitutes evidence that /N/ and /i/ are not syllable margins (codas) but prosodic
units in their own right.
– Non-coincidence of foot and syllable boundaries
Another series of arguments against the syllable was raised by Poser (1990). As Poser
observes in the conclusion of his 1990 paper, the ‘fact that the Japanese foot consists of
morae rather than syllables points to the independence of the mora as a phonological
constituent. It also poses a problem for advocates of the position that morae are
subconstituents of syllables, since the boundaries of feet, composed of morae, need not
Page 10 of 36
Prosodic units
coincide with syllable boundaries’. Kubozono (1995a) also hints at the possibility of
eliminating the syllable to the benefit of the foot in Japanese at the very end of his book.
Examples of the mismatch between foot and syllable are easy to find. For instance, the
truncated form of rimouto kontorouru {remote control}, rimo-kon° would be rimou-kon
(rimô-kon in Hepburn romanization) if foot and syllable boundaries coincided. This
mismatch is rather annoying because it calls into question the ‘strict layer hypothesis’
(Nespor and Vogel, 1986), according to which each prosodic constituent is included in
totality in the immediately higher-ranked constituent.
Could it be that, if foot boundaries and syllable boundaries do not coincide, it is because
syllables do not exist, and thus presumed syllable boundaries are therefore of no
relevance? As we shall see below, in the model proposed here, moras are directly linked
to feet, with no syllable layer in between, so that coinciding boundaries are only mora and
foot boundaries.
– Speech errors
The examination of speech errors finally provides extremely interesting clues. According
to Kubozono (1989, 1995c, 2006a), what corresponds to a ‘heavy syllable’ tends to be
replaced by a succession of two ‘light syllables’ or by a ‘heavy syllable’ more often than
by a ‘light syllable’ in slips of the tongue, for example in Kyo uno Kikujirou (proper name)
〉 Kikuno Kikujirou and ju ugo paa sento ‘fifteen per cent’ 〉 juugo pansento. This shows
that the unit syllable does not have cognitive reality, since one would otherwise expect a
heavy syllable to alternate with a light syllable more often than with two light syllables. On
the contrary, an analysis of this phenomenon based on the mora or the foot provides a
better understanding of the data: a mora is replaced by a mora, a foot is replaced by a
foot. For further evidence, see Terao (1992, 2002) who shows that the fundamental unit
on which speech errors operate is, in Japanese, the mora rather than the syllable.
(p.154) 6.2.4 A Reexamination of the Alleged Evidence In Favour of the Syllable

Let us now review some of the internal evidence which is generally presented in favour
of the recognition of a light (i.e. monomoraic) vs. heavy (bimoraic) syllable opposition in
Japanese. The phenomena that will be put under scrutiny are the following: initial
lowering, unaccentedness of /N/, /Q/, /R/, /i/, accentuation of foreign toponyms,
accentuation of the enclitic particle no, accentuation patterns of compound personal
names whose second member is -tarou (see also Labrune, 2012, for additional
arguments and discussion). As we shall see, in each case an alternative, syllable-free
approach to the data, which is either as convincing as or more convincing than the
syllable-based approach is possible, or the linguistic data are ambiguous or incomplete.
– Initial dissimilation (initial lowering)
Initial dissimilation, or initial lowering, is a phenomenon which is generally accounted for

through reference to the syllable (see for instance Haraguchi, 1999). As we shall see in
the next chapter, there exists in Japanese a principle stipulating that any word which does
Page 11 of 36
Prosodic units
not bear an initial accent lexically begins with a Low–High pitch sequence. However, this
principle applies only optionally in words beginning with a CV + special mora sequence, in
other words, with a putative heavy syllable. Rather, such words can start with a High–
High pattern (Hattori, 1954). Compare for instance:
(5)
kokusai° LHHH ‘international’

kousai° LHHH or HHHH ‘exchange’
At first sight, the correct description of this phenomenon forces one to refer to heavy
syllables: initial dissimilation supposedly does not apply when the word starts with a heavy
syllable. However, this is nothing more than an ad hoc statement. Compare the two
following formulations:
(a) Initial dissimilation does not occur when the word starts with a heavy syllable.
(b) Initial dissimilation does not occur when the second mora of the word is a
deficient mora.
The statement in (a) does not have more explanatory power than the one in (b), nor is it
simpler.
Moreover, even if one follows the syllabic analysis of Japanese, the formulation of the
process at hand is not fully adequate, and can therefore not be held as definite evidence
for the action of the syllable, because it appears that only a subset of putative heavy
syllables, namely those ending in /R/ and /N/, are concerned by this phenomenon. When
they end in /Q/ or /i/ (the two other arguably possible syllabic (p.155) codas of
Japanese), the phenomenology is different. Unaccented words beginning with CVi behave
like words beginning with CVCV, for instance koikuchi° ‘strongly flavored’ is uttered
LHHH while unaccented words beginning with CVQ are pronounced with a sequence of
two low tones, for instance gakkou° LLHH ‘school’ (Haraguchi, 1977, Tanaka, 2008). So
the issue is actually a little more complicated and controversial than it appears to be at
first sight. The intuitive analysis I propose of this phenomenon is the following.
Although the rise in pitch which occurs on the second mora of a word is not an accent in
the phonological sense of the term, it does actually constitute a prosodic phenomenon
with relative relevance, in particular because of its demarcative function, a status which is
actually comparable to that of accent on the functional level. One can thus regard initial
dissimilation as a mirror image of accent, some sort of accent echo, in other words a
secondary accent. Initial dissimilation is characterized by a pitch rise between two moras,
while accent consists of a pitch fall between two moras. And, just as accent does not
normally occur after a special mora, initial dissimilation does not normally occur before a
special mora. In both cases, the near impossibility of pitch change occurs because of the
inherent structural weakness of special moras, which prevents them from being the locus
of pitch change (from high to low or from low to high). The adopted strategy is identical in
Page 12 of 36
Prosodic units
both cases: the pitch change (fall in the case of accent, rise in the case of initial
dissimilation) is moved one position leftward. In the case of initial dissimilation, there is no
leftmost mora available to receive a low pitch, so the initial mora is realized with a high
pitch. There is therefore no need to refer to the syllable in order to account for initial
dissimilation.
– Unaccentedness of /N/, /Q/, /R/, and moraic /i/ (/J/)
The phenomenon most frequently invoked in favour of a ternary conception of the

Japanese syllable comes from accentual phonology. It involves the prosodic status of the
special moras /N/, /Q/, /R/ and that of the moraic vowel /i/, in other words of the elements
which, in some other languages, are likely to constitute the second part of a heavy
syllable. These elements have been presented in detail in the preceding chapter. Recall
that, first of all, they cannot in principle bear accent and have the property of causing a
left shift of the accent when they occupy a prosodic position likely to receive accent
(section 5.5), a mechanism that I propose naming NADM principle (Non-Accentuation of
Deficient Moras) and which will be further developed in 6.3.2 as well as in Chapter 7. For
the moment, let us consider the case of loanwords of Western origin in which a default
accent is assigned to the antepenultimate mora, as presented in section 5.5. The
traditional approach (Kindaichi and Akinaga, 2001) states the following rule: the accent falls
on the antepenultimate mora; if the antepenultimate mora consists of a special mora (/N/,
/Q/, /R/, or moraic /i/), the accent moves to the preceding mora, striking the pre- (p.156)
antepenultimate. Compare for instance a.na.ku.ro.ni.zu.mu µµµµµµµ {anachronism} and
di.su.ka.s.sho.n µµµµµµ rather than *di.su.ka.s.sho.n µµµµµµ {discussion} (dots are
used to denote mora boundaries). Note that this traditional approach, which is widely
followed by Japanese scholars of the philological mainstream, does not refer to the
syllable to account for such accentual patterns.
McCawley (1968) proposes a different analysis: for him, the examples above support the
claim that the syllable rather than the mora is the prosodic unit of Japanese, because in a
heavy syllable only the first mora can be accented. Following his approach, disukasshon
has four syllables: di-su-kas-shon (hyphens denote syllable boundaries). The default
accent rule of Japanese, which applies to foreign loans, is accordingly reformulated as
follows: accent is placed on the syllable containing the third from last mora. Japanese
would thus be categorized as a ‘mora-counting syllable language’ whose prosodic unit is
supposed to be the syllable and not the mora, even though McCawley also recognizes
that the mora plays a major role in Japanese as a ‘unit of phonological distance’.
At this point, let us first observe that, as Uwano (2003:74) remarks, the mora approach is
just as explanatory as the syllabic one.
The first objection to McCawley’s analysis is that if the syllable is indeed the prosodic unit
in Japanese, how can we explain that, in an accented heavy syllable, the third component
(the coda according to the syllabic approach) is never of the same melodic height as the
onset and nucleus? In words such as kyo uto ‘Kyôto’ or nihon ‘Japan’, the fall in pitch that
marks the location of the accent occurs before a special mora (before the second part of
Page 13 of 36
Prosodic units
the long vowel in kyo↓uto and before the mora nasal in niho↓n). The categorization of
Japanese as a syllable-accenting language thus appears inappropriate, since only a
subpart of the syllable is likely to carry the pitch. It is therefore not empirically true to
state that the syllable is accented, because only the first part of it bears a high pitch. This
situation is radically different from what occurs in syllabic languages like English where it
is the whole syllable which bears the phonetic manifestation of accent. As remarked by
Hyman (2003[1985]:96), in pitch accent languages it is the unit corresponding to the mora
that receives the accent, but in the standard formulation of the syllable; this
generalization cannot be captured. So the most straightforward and simple analysis is to
assume that Japanese counts moras and accentuates moras, since pitches coincide with
moras, but that a certain category of moras cannot receive the accent.
Moreover, McCawley’s conception of the mechanisms of Japanese accent appears even

more inadequate in cases where several special moras follow each other. It sometimes
happens that /N/, /Q/, /R/, or /i/ receive the accent. In words like obaasankko
/obaRsaNQko/ ‘child cherished by his grandmother’ or cheenten /tyeRNteN/ ‘chain
store’, the accent does not fall on what would constitute, in McCawley’s model, the
accentual peak of the syllable but on the first special mora after the nucleus, here /N/ and
/R/ respectively.
(p.157) Higurashi (1983) reports another case in which the presumed second part of a
heavy syllable is accented. When the pre-accenting recessive enclitic particle shika ‘only’
is combined with an unaccented noun, the final mora of the noun receives an accent,
including /N/, /R/, or the moraic /i/. Let us look at the following examples:
(6) Accentual pattern of atonic nouns + shika (from Higurashi 1983:35)
miyako° + shika miyako-shika /miyako sika/

‘only the capital’
kouen° + shika kouen-shika /koReN sika/ *kouen-shika
‘only the park’
tekkyou° + shika tekkyou-shika /teQkyoR sika/ *tekkyou-shika
‘only the iron bridge’
Not only is it not the entire syllable that receives the accent but it is not even what
constitutes the heart of the syllable, i.e. its nucleus, that does. Such cases are rather
marginal, but it is obvious that they cannot be accounted for within McCawley’s
approach, thus crucially weakening the overall syllabic analysis of Japanese. Vance
(1987:81, quoting a p.c. by Hamano Shoko) also mentions interesting cases in emphatic
speech where an atonic word ending in /N/ can receive the accent on /N/ before the
quotative particle to. So for instance in kouban to itta ‘I said POLICE BOX’, kouban ‘police
box’ normally atonic (kouban°) is pronounced kouban with an accent on the final /N/.
Hamano (1998:32) also says that in some ‘dramatic uses’ of mimetic words, a special
Page 14 of 36
Prosodic units
mora can be accented as in paanto or pinpin-to.
Finally, one has to take into account the fact that some dialects of Japanese belonging to
the same dialect subfamily as Tôkyô Japanese, and very closely related to it, allow special
moras to receive an accent in a very general fashion. Thus, in the Izu dialect,11 there
exists a possible contrast between to u HL ‘political party’ and tou (too) LH ‘ten’, or
between ka i ‘a paddle’ and kai ‘shell’ (Uwano, 2003). As Uwano correctly observes,
this type of contrast would be totally impossible in a syllable-based language like English.
– Accentuation of foreign toponyms
Other pieces of evidence supposed to prove the relevance of the heavy syllable have
been presented since McCawley. Prima facie, one of the most convincing examples is
provided by Kubozono (1996). Examining the assignment of accent in foreign toponyms (a
class representative of accent assignment processes in loanwords in general), Kubozono
observes that although the default accentuation (p.158) rule stated above applies quite
generally in accented loans, whatever formulation of the rule one adopts—be it à la
Kindaichi and Akinaga or à la McCawley—it nevertheless encounters a number of
exceptions. Such exceptions are accented on the fourth or fifth mora from the end, as in
mo nburan /moNburaN/ ‘Mont Blanc’, a mazon /amazoN/ ‘the Amazon’, pi renee
/pireneR/ ‘the Pyrenees’, rather than *monburan, *amazon, *pirenee as predicted both
by Kindaichi and Akinaga and McCawley’s analyses. Taking such examples into
consideration, Kubozono posits the following accentuation rule: the accent falls on the
penultimate syllable if it is heavy, on the antepenultimate otherwise. However, we shall
see in the following chapter (section 7.2.5) how the present approach to the structure of
the foot and mora provides a simpler account of these crucial examples, without referring
to heavy and light syllable types.
– Accentual behaviour of no
Another example worthy of interest is examined by Miyake (1943), Martin (1952), Vance
(1987, 2008), Kubozono (1999a), Haraguchi (1999), Uwano (2003), among others. It
involves the determination particle no, whose accentual behaviour is rather peculiar.
When no occurs after a noun which bears a final accent, that noun sometimes undergoes
de-accentuation. Consider the following examples.
(7)
a. yama + no yama-no° ‘of the mountain’

otoko + no otoko-no° ‘of the man’
b. nihon + no nihon-no° ‘of Japan’
kinou + no kinou-no° ‘of yesterday’
c. kokoro + no kokoro-no ‘of the heart’
d. me + no me -no ‘of the eye’
Page 15 of 36
Prosodic units
e. do u + no do u-no ‘of bronze’
te n + no te n-no ‘of the sky’

ka i + no ka i-no ‘of the shell’
According to Kubozono, an adequate description of these examples can be made through

explicit reference to the syllable, and should be formulated as follows: words longer than
one syllable such as yama,otoko in (a), nihon and kinou in (b), accented on the final
syllable become unaccented when followed by the no particle, while monosyllabic words
(whose only syllable is either heavy or light and bears an accent) like me, do u, te n or ka
i maintain the original accent. This phenomenon might at first sight appear to be quite
convincing as a justification for the relevance of heavy syllables. However, on the one
hand, many apparent exceptions can be found (see Vance, 1987:82, 2008:156–157 for
lists and additional references). For instance, takusan ‘much’ is not de-accented before
no: takusan + no yields takusan-no and not *takusan-no°. On the other hand, some
(p.159) speakers do not realize (7b) as predicted by Kubozono. This is probably why,
according to Vance (1987) and Uwano (2003) accent deletion before no should not be
taken as reliable evidence for syllable structure. Moreover, it seems that words such as
nihon or kinou that undergo de-accentuation before no are actually lexical exceptions
(Takayama Tomoaki, p.c. 2009) belonging to a close and numerically limited set, while the
takusan type, which does not yield de-accentuation, is more likely to be the general,
default type.
Clearly, the issue is controversial and does not reach consensus. The data would need
further investigation to determine which pattern is regular: the one involving
deaccentuation or the one involving no deaccentuation.
An alternative approach to these data consists of positing that the words belonging to
types (7b) and (7e), which end in a special mora and bear a surface accent on the
penultimate mora, actually have an underlying accent on the final mora, as follows:
/nihoN/, /kinoR/, /doR/, /teN/, /kai/ (see also sections 7.2.3 and 7.2.5). This analysis is also
proposed by Uwano (2003:75). The underlyingly final accent is shifted one mora leftward
by virtue of the NADM principle stipulating that special moras cannot bear an accent at
the phonetic level, except in a few special cases such as the examples with shika above, as
well as the obaasankko example previously mentioned. This analysis accounts very easily
for exceptions like takusan, because takusan is accented on the mora sa at the
underlying level, as well as at the surface level, contrary to the words in (7b) and (7e),
which actually carry a final accent at the underlying level. This is why they behave exactly
like yama and otoko in (7a) rather than like kokoro in (7c). In do u, te n, or ka i, it is the
initial mora of the word which, on the surface, receives the accent. In this case, de-
accentuation does not occur, and the accent is maintained on the initial mora, a
phenomenon which has to do with the fact that the beginning of a word constitutes a
privileged position, where contrasts are more often preserved and neutralizations
avoided (Beckman J., 1999).
Page 16 of 36
Prosodic units
– Personal compound names whose second member is -tarou
Let us now examine the following examples, also cited by Kubozono (1999a) as another
piece of evidence for the syllable. Consider the following personal names (all from
Kubozono 1999a: 46, romanization adapted; when no accent information is provided, it
means that the word does not have its proper accent pattern because it never occurs in
isolation):
(8) Accentuation of personal names ending in -tarou
a.
ki + tarou kitarou°
ne° + tarou netarou°
b.
kin + tarou kintarou°

kyuu + tarou kyuutarou°
(p.160)
c.
momo° + tarou momotarou

kane° + tarou kanetarou
d.
chikara + tarou chikaratarou

karee° + tarou kareetarou
urutoraman + tarou urutoramantarou
According to Kubozono (1999a), the accent behaviour of the compounds with -tarou is
predictable on the basis of the syllable structure of the first member. When the first
member is monosyllabic, the compound is unaccented. Otherwise, it is accented. Note
that -tarou exhibits two different accent patterns when combined with a bimoraic first
member: a monosyllabic N1 (8b) yields an unaccented compound just like a monomoraic
N1 (8a), whereas forms with a bisyllabic N1 receive the accent on the first member (8c)
but only if it is also bimoraic.
However, the data in (8) can receive an alternative account. First, as Kubozono (1999a)
mentions, personal compound names with -jirou behave in a different way, in so far as
they do not trigger the same accent behaviour. The first member kin is accented, and
thus behaves like kane rather than like ko when combined with -jirou. We find Ki njirou
(not Kinjirou°) and Kanejirou vs. Kojirou°. This shows that the conditioning of the accent
rule is lexical rather than strictly phonological, since not all suffixes activate the rule which
Page 17 of 36
Prosodic units
is supposed to refer to the syllable: -tarou does, but -jirou does not. Second, examination
of the examples in (8) shows that trimoraic bisyllabic first members behave like trimoraic
trisyllabic ones (8d). The syllabic approach fails to capture the fact that momotarou and
kareetarou, whose first members are both bisyllabic, do not receive the same
accentuation. This suggests that it is not the number of syllables that is crucial but
something else. One can achieve a correct descriptive account of these data by referring
simply to the foot and to its constituents, i.e. moras. The rule can accordingly be
reformulated as follows: when the first member is equivalent to a monomoraic foot or to a
bimoraic foot ending in a special mora, the compound is atonic (8a, b). When it is equal to a
bimoraic foot ending in a regular mora, the accent is placed on the final mora of N1 (8c).
Finally, when the first member contains more than one foot made up of two regular
moras, the accent is placed on the initial mora of -ta rou.
To conclude, unlike the mora, the relevance of the syllable appears highly questionable.
First, the models of the Japanese syllable which have been proposed in the literature
reflect the existence of a problem concerning the articulation between the mora and the
syllable with regard to the foot. Second, the lack of traditional evidence in favour of the
syllable or of its constituents is particularly blatant. Finally, the linguistic data which have
been claimed to attest the relevance of the syllable are either not sufficiently
documented, or they are ambiguous and can receive an alternative, syllable-free
account.
(p.161) In the next section, I shall propose a new model of the basic Japanese prosodic
unit that does not rely on the syllable.
6.3 for A Strictly Binary Model of the Basic Prosodic Unit In Japanese
The facts discussed so far show the inadequacy of a three-position syllabic model (onset–
nucleus–coda) in the phonology of Japanese. This calls for another analysis of the basic
prosodic unit of Japanese, that I will temporarily call a prosodeme, following Trubetzkoy’s
term.
The ideas that are developed here are inspired by Larry Hyman’s theory of phonological
weight (2003 [1985]), which argues that the universal phonological anchor tier consists of
weight units, or beats, that correspond to moras, and that the syllable is not a universal
constituent but a language-particular construct built out of the weight units.
On the basis of the evidence reviewed so far, I claim that the basic prosodic unit of
Japanese is maximally binary, i.e. with two positions: position 1, the onset (ideally
represented by C) and position 2, the nucleus (ideally represented by V), as shown
below.12
(9) Structure of the basic prosodeme
Page 18 of 36
Prosodic units
The dots represent linear positions corresponding to the skeleton as conceived in

multilinear phonology, that is, some sort of temporal projection of each segmental unit
within a prosodeme, which allows the encoding and representation of quantity.
Accordingly, geminate consonants and long vowels are considered to be one single
segmental unit (actually a set of distinctive features) associated with two skeletal
positions, whereas affricates or some types of diphthongs correspond to two segments
(i.e. two sets of distinctive features) linked to one position.
In the model argued for here, there is no need for a more complex structure such as the
heavy syllable, that is, for a three-position prosodic unit. All the elements which can be
considered to belong to the third position within a syllable in competing approaches are to
be analysed as so many distinct prosodemes.
(p.162) Further, and most importantly, some prosodemes are to be considered as

structurally incomplete, in the sense that they contain an empty position, either the onset
or the nucleus. These units will be called ‘deficient’, ‘weak’, or ‘degenerate’. Japanese is
thus characterized by the fact that it contains two types of prosodemes: regular and
deficient.
Deficient prosodemes may belong to the following four types:
– Prosodemes containing only one nuclear vowel.

– Prosodemes containing /N/, /Q/, /R/ (the so-called ‘special moras’ of the
Japanese linguistic tradition).
– Prosodemes containing a devoiced vowel.
– Prosodemes containing an epenthetic vowel.
Note that the units recognized as deficient prosodemes here do not correspond exactly
to the special moras of Japanese traditional phonology (see Chapter 5 as well as section
6.1 above), since they also comprise onsetless vowels other than /i/, moras containing a
devoiced vowel, as well as those containing an epenthetic vowel.
Let us now examine the structure and representation of the deficient prosodemes.
(10) Deficient prosodeme made up of a vowel (with empty onset)
The representation in (10) is that of onsetless prosodemes whose nucleus is filled by a

vocoid, for instance e in kangaeru ‘to think’, or i in ka i ‘shell’.
The three so-called special moras /N/, /Q/, and /R/ are deficient moras which, I assume,
have the representations in (11):
(11) Prosodemes made up of a special mora
Page 19 of 36
Prosodic units
(p.163) /R/ consists of a [−consonantal] segmental specification, while /Q/ is

[+consonantal].13 /N/ is also [+consonantal] but note that it contains an additional
segmental specification in comparison to /Q/, the nasality feature. In fact, except for the
[±consonantal] feature specification, which simply encodes their vocoid or contoid nature,
/R/ and /Q/ have no segmental specification at the underlying level. Note also that the /R/
features are linked to the second position, but the /N/ and /Q/ features to the first one.
Let us now examine two other types of deficient moras: prosodemes containing a
devoiced vowel (already discussed in section 2.6) and prosodemes containing an
epenthetic vowel:
(12) Prosodemes made up of a devoiced vowel
In the case of devoiced vowels, a V segment present in the underlying form is

phonetically deleted on the surface, but the structural position initially associated with the
nucleus (V) is not. The nucleus is thus left empty, but it nonetheless remains as a position
at the phonological level14 (possibly with a [−consonantal] feature), thus yielding a
deficient unit. The proposal to include the moras with a devoiced vowel in the set of the
‘special moras’ was also made by Akinaga (1968), but her proposal has not been followed
up in subsequent studies.
(13) Prosodemes containing an epenthetic vowel
The exact phonological status of epenthetic vowels in phonological representations and

derivations constitutes a major problem for phonological theory, which still awaits a
proper solution. Nevertheless, whatever the theoretical framework, an epenthetic vowel
comes down to an empty position at the most underlying level which receives (p.164)
default phonetic realization at a later stage. In that sense, epenthetic vowels can be
considered to be the opposite of devoiced vowels, as seen in (12), because devoiced
vowels correspond to a position initially filled which loses its phonetic content.
Page 20 of 36
Prosodic units
Onsetless prosodemes, those containing epenthetic and devoiced vowels, as well as

special moras thus display strong representational similarity in the approach proposed
here: they all contain an empty position at some level of the representation. This is very
desirable, because these four types of moras share a common empirical characteristic
which has not received sufficient attention in previous studies: they are not readily
accentable. We have already mentioned the fact that /N/, /Q/, and /R/ are not, under
normal conditions, able to receive an accent, but it is also a well-established fact in the
phonology of Japanese that onsetless vowels, epenthetic vowels, and devoiced vowels
share the same property, although not as systematically as /N/, /Q/, and /R/ (a point to
which we will come back below). The following examples illustrate this fact:
(14)
a. Accent shift caused by the presence of a devoiced vowel (see also section
2.6) in conservative Tôkyô Japanese
expected form realized form

hu kaku [ˡFɯkakɯ] 〉 hukaku [Fɯ̥ˡkakɯ] ‘deeply’
ki sha [ˡkiɕa] 〉 kisha [ki̥ˡɕa] ‘train’
b. Accent shift caused by the presence of an onsetless vowel expected form

realized form
keizai-ryoku 〉 keizai-ryoku ‘economical power’

kangaeru 15 〉 kangaeru ‘to think’
anaunsu 〉 anaunsu {announce}
c. Accent shift caused by the presence of an epenthetic vowel in a CV mora

expected form realized form
amusuterudamu 〉 amusuterudamu ‘Amsterdam’

andesu-kai 〉 andesu-kai ‘Andes Club’
That moras containing an15epenthetic vowel do not behave in the same way as moras
containing an underlying vowel, but rather like special moras, is further demonstrated by
the fact that four-mora loanwords that end in a sequence of two CV moras are mostly
unaccented (90%) if their final vowel is underlying, (p.165) whereas the ratio goes
down to 30% if the final vowel is epenthetic (Kubozono, 2006b), a percentage which is
close to that displayed by words ending in a supposedly heavy syllable, that is, words
ending in a special mora. So not only are epenthetic vowels difficult to accentuate, their
presence also causes word accent computation to differ significantly when they occur
word-finally.
Note that the occasional ability of onsetless moras to repel the accent constitutes a
serious problem for phonology because it is commonly admitted that onsets do not
Page 21 of 36
Prosodic units
contribute to prosodic weight. However, the data under examination contradict this
assumption if analysed through the syllabic mirror.16 In the present approach, this can be
accounted for by the fact that structurally, onsetless moras are phonological objects that
contain a position with no segmental specification, in the same way as moras containing an
epenthetic or a devoiced vowel. At the phonetic level, this corresponds to a lack of
acoustic prominence, not suitable for receiving an accent. The syllabic framework does
not allow us to capture onsetless, devoiced, and epenthetic vowels and would-be third-
position syllable constituents (i.e. codas) as entities belonging to the same category,
whereas the model developed here does, because all four elements are conceived as
containing an empty slot, which explains their relative weakness or transparency.
Actually, the relevance of the onset for prosodic weight is already attested in the
Man’yôshû, a compilation of poems dating back to the eighth century. The meter of
Japanese poetry is fixed, and it is based on the number of moras per verse. The verses of
tanka, the most representive genre of Japanese classical poetry, are composed
respectively of 5–7–5–7–7 moras. Quite interestingly, this meter is not always respected.
This notably happens when the verse contains an onsetless vowel, as the following two
examples taken from the Man’yôshû illustrate:
(15)
– 8 moras instead of 7:
imada sakazukeru ‘does not bloom yet’ (poem 2123)
– 9 moras instead of 7
tori age mahe ni oki ‘taking (it) and putting (it) in the front’ (poem 4129)
Such hypermetric poetic licences are called jiamari ( り). It is important to note that the
onsetless vowel can occur at the beginning of the verse, as in the first example, thus
excluding the hypothesis of syneresis.
I claim that this phenomenon can be accounted for along the same lines of analysis as the
one I adopt for modern Tôkyô Japanese, that is, syllables did not exist in eighth-century
Japanese, and the basic prosodic unit consisted of a two-slot element equivalent to a
mora. When one of the two slots is empty, the unit is (p.166) considered to be deficient,
and presents a type of phonological behaviour which is not that of full units. In the verses
above, it is clear that deficient moras may not always count as one rhythmic unit as full
moras do.
In the above representations, empty positions generally represent the vestige of

segmental material which has been disassociated, either from a diachronic or a
synchronic perspective. From the diachronic point of view, as already mentioned in the
previous chapter, special moras /N/, /Q/, and /R/ are known to result from consonant or
vowel loss in Yamato Japanese, a process which has left an empty structural position in
the underlying representation. Interestingly, internal onsetless vowels in simplex words
Page 22 of 36
Prosodic units
are also the result of a similar process. Archaic Japanese did not accept onsetless vowels
morpheme-internally, so that all the modern occurrences of internal V 1 V 2 sequences are
due to the loss of a consonant in native words. For instance, modern ko e ‘voice’ comes
from kowe, a u ‘to meet’ from afu 〈 *apu.
6.3.1 Japanese As A Syllable-less Language

The reader will have noticed that the model of the Japanese prosodeme proposed here
corresponds exactly to the mora, which leads us to consider that the mora is the core
prosodeme of Japanese. I assume that moras are structured as in (16).
(16) Structure of Japanese moras
Recall that in the case of special moras, one of the two positions, C or V, is left empty at
some level of the analysis.
Taking into account the phenomena reviewed thus far, three other lines of analysis of the
prosodic structure of Japanese ought to be considered. It is important to examine them
in order to justify the claim made here. These options appear in (17a–c), while (17d)
represents the analysis which we adopt here (the dots representing segment position are
omitted).
(17)
(p.167) The first representation (17a) consists of positing that moras and (light)
syllables, although isomorphic, coexist in Japanese, and that they are organized
hierarchically. This option is undoubtedly the most cautious one. In this representation,
however, there is a redundancy between the mora and the syllable, so that the principle
of Occam’s razor leads us to adopt one of the three remaining possibilities, (17b), (17c),
or (17d).
The second option, (17b), appears to be the least challenging one for the model which
actually dominates in current phonological theory. However, the facts examined
throughout this chapter did not provide any evidence that there would be any advantage
in distinguishing an onset directly attached to the syllable, whereas the nucleus would be
associated with a mora. On the contrary: we have seen that there exists in Japanese a
Page 23 of 36
Prosodic units
remarkable cohesion between the element C and the element V of a prosodeme. Such a
close cohesion is not reported for truly ‘syllabic’ languages. In addition to the facts
already presented, one can also mention data taken from stuttering speech which can be
taken as evidence that (17b) is not the right representation. According to Ujihira and
Kubozono (1994) and Kubozono (2006a), the portion repeated by Japanese stutterers at
the beginning of words starting with a consonant is a CV unit in 88.6% of the cases (for
instance, na-na-na/nde ‘why’). Only in 1.2% of the cases is it a C unit. This, as Kubozono
observes, differs from English, for instance, where initial segmentation generally occurs
before the vowel (n-n-n-n/ever). If one assumes that it is not the syllable and its
constituents which are the units of prosodic encoding but rather a prosodeme
corresponding to the mora, these data can be accounted for quite naturally. Also, recall
that, as stated earlier, the temporal adjustment between C and V is much more important
in Japanese than it appears to be in syllabic languages. We can interpret this fact as a
consequence of the strictly binary and symetric structure of the Japanese prosodeme,
and of the strong solidarity and equality of status which exists between its two
components. Since (17b) hardly reflects this cohesion and equality between a C and a V
belonging to the same prosodeme, it won’t be adopted here.
The elimination of (17a) and (17b) leaves us with two possibilities: (17c) and (17d). (17c)
considers the syllable as the only necessary constituent, but limited to a maximally binary
structure, more or less in the fashion of Lowenstamm (1996) or Scheer (2004) who
defend the idea that all syllables, in all languages, consist fundamentally of a CV structure.
If we adopted this position, the references to the mora could be simply replaced by
references to the syllable. This choice would have the advantage of not questioning the
very largely followed postulate according to which all languages have syllables, and which
views the mora as an optional constituent. However, it seems desirable to maintain a
conceptual distinction between the mora and the light syllable, in other words, to ensure
that these two terms are not understood as different denominations of the same entity.
This is why we shall adopt the hypothesis in (17d). The justification for this stand is that
(p.168) Tôkyô Japanese prosodemes have a number of specific properties which, as
explored in the preceding pages, make it phonologically different from syllable-based
languages.
6.3.2 The Non-Accentuation of Deficient Moras Principle (NADM)

We have seen that on the structural and functional levels, two types of moras have to be
distinguished: full moras which have a CV structure, and weak or deficient moras in
which one of the two positions is empty. The main empirical difference between full and
deficient moras is that deficient moras are not able to work as proper accent kernels
under normal and general conditions. From now on, this inability of deficient moras to
bear the accent will be referred to as the NADM principle or constraint:
(18)
Non-Accentuation of Deficient Moras (NADM)

Deficient moras must not be accented
Page 24 of 36
Prosodic units
Examples of the NADM principle application have been presented on many occasions
throughout this book, for instance in 5.5 and 6.2.4, and will be seen again in sections 7.1.4,
7.2.3, and 7.3.2.
This principle is nothing but a reformulation of the well-known OT constraint PEAK

PROMINENCE (Prince and Smolensky, 1993:39) which stipulates that more prominent
elements make better prosodic peaks within a foot.
Let us take for instance a sequence such as /hoN/ ho n ‘book’, which, in a syllabic
approach, would be interpreted as a heavy syllable. In our model, this is simply a foot
made up of a regular mora, /ho/, followed by a deficient mora, /N/. Further, we shall
assume, as argued in section 6.2.4(see also sections 7.2.3 and 7.3.2) that /hoN/ is
underlyingly accented on the final mora /N/, and that the accent moves to /ho/ at the
surface level: [ˡhoN]. This process can be accounted for in an extremely simple manner
by the interaction of the following three constraints: NADM, MAX(ACCENT) and
FAITHIO(ACCENT).
NADM dominates MAX(ACCENT), the constraint which imposes that an accent present in
the input has a correspondent in the output, and MAX(ACCENT) dominates
FAITHIO(ACCENT) which demands that an accent present in the input be preserved in
the same location in the output. The ranking between the three constraints accounts for
the accentual shift to the initial mora of the morpheme, as shown in (19)
(19) Interaction of MAX(ACCENT), NADM and FAITHIO(ACCENT)
a. Hierarchy: NADM 〉〉 MAX(ACCENT) 〉〉 FAITHIO(ACCENT)

(p.169) b. Tableau
/hoN/ NADM MAX(ACCENT) FAITHIO(ACCENT)

☞a. hon *
b. hon *!
c. hon° *! *
In (19b), the candidate in b., which infringes the dominant constraint NADM, is the worst
one. Of the two remaining possibilities, a. with initial accent, and c., without accent, a. is
selected as the optimal output because, contrary to c., it preserves the accent present in
the input, although in a location which is not the original one. Note that the constraint
MAX(ACCENT) does not govern positional faithfulness but the preservation in the output
of an element present in the input, contrary to FAITHIO(ACCENT) (see also section
7.3.2).
Dialects such as the Izu dialect which allow special moras to be accented at the surface
level can be assumed to have a different hierarchy, ranking NADM below
FAITHIO(ACCENT).
Page 25 of 36
Prosodic units
However, some deficient moras may, under certain conditions, carry the accent even in
Tôkyô Japanese. Variation is frequently observed, especially in cases of onsetless and
epenthetic vowels, which are sometimes accented. The mora nasal /N/ can also receive
the accent in certain cases, as in obaasankko ‘child cherished by his/her grandmother’,
as well as /R/ (see also the examples in (6) above), whereas /Q/, for instance, never can.
Reliable and phonetically controlled data which would inform us about the factors which
condition this variation, especially concerning the accentuation of onsetless and epenthetic
vowels, is still needed. In a constraint-based approach, we would be led to posit that
there exists a constraint placed higher in the hierarchy than NADM, and that this higher
constraint forces an accent to be put on a deficient mora. I leave the issue of the exact
nature of this constraint for future investigation. Nonetheless, the fact that all deficient
moras do not always behave exactly in the same manner is indisputable and will lead us to
the proposal that Japanese moras are arranged along a scale, as will be shown now.
6.3.3 Hierarchy of Japanese Prosodemes (moras)

Because the opposition between full and deficient moras is not dichotomous, the
conception of the Japanese prosodeme developed above needs some further
refinements. It seems reasonable to assume that Japanese moras have to be classified
along a scalar hierarchy according to their acoustic prominence. This is because, as just
stated, the NADM principle is neither absolute nor free of variation. The general scale of
Japanese moras according to their relative capacity to receive the accent is as follows
(here ●’s denote empty positions):
(p.170) CV 〉〉 ●V 〉〉 C●
I propose that these three major mora types be further subclassified as follows:17
●i, CV epenthetic, and CV devoiced seem to occupy the same position in the hierarchy, in
between ●u and /R/, hence the use of the brace.
The relative capacity of a mora to receive the accent is conditioned by its intrinsic
phonetic prominence and acoustic energy, as well as by the number of filled structural
positions it contains. Any consonant followed by a high vowel (for instance mi,ku), the
mora nasal, the first part of a geminate, as well as voiceless consonants when followed by
a devocalized nucleus vowel, on the one hand, are quite unsurprisingly characterized by
their relative lack of prominence. On the other hand, onsetless vowels can also be
considered as relatively weak, in any case weaker than CV moras. This may be more
unexpected at first sight, but, following Burzio (1994:158) one can assume that onsets
contribute to acoustic energy, so onsetless units are prosodically weaker. As for moras
containing an epenthetic vowel, I assume that their lack of prominence is phonological (or
Page 26 of 36
Prosodic units
representational) rather than phonetic, even if the exact nature of their phonological
representation is an issue which must be further investigated. Moreover, among
onsetless moras, it is also clear that ●a and ●o are not ‘as weak’ as ●i, ●u or even ●e.
In addition to its phonological structure, the intrinsic sonority of the mora thus
unquestionably conditions its capacity to receive the accent.18 However, I argue that the
relationship between accent, sonority, and structure is not direct. Rather, it is mediated
by reference to the foot, as we shall see now.
6.4 The Foot

The canonical Japanese foot (hutto フット or kyaku ) is bimoraic. It can have the following
structures:
(p.171) (20) Japanese feet = (µµ)
(C)VCV ka ta ‘shape’, iru° ‘to exist’

(C)VV hae° ‘fly’, o u [ˡoː] ‘king’
(C)VC ho n ‘book’, kit(to)° ‘certainly’
Prevocalic consonants can be palatalized (i.e. Cy). In (C)VC, the final C equals the mora
nasal (/N/) or the first part of a geminate (/Q/). In (C)VV, the second V can be equivalent
to /R/.
However, monomoraic feet (sometimes called degenerate feet) are not uncommon in
Japanese. Kubozono (1999a) argues that every foot must be entirely contained within the
same morpheme, so that two adjacent monomoraic feet will not form a single bimoraic
foot in polymorphemic lexemes: F (µ) + F (µ) = F (µ)F (µ) rather than F (µµ). Each foot is
supposedly parsed as a separate degenerate foot. Similarly, trimoraic feet (or heavy feet)
can also be found. For instance, Tanaka (2008:203) argues, on the basis of an empirical
study of foreign words accentuation, that trimoraic feet must be posited in certain cases
where a sequence of two regular moras is followed by /Q/. He proposes the following
respective footing for the words to mato {tomato}, biniiru {vinyl}, and so netto
{sonnet}: (toma)to, bi(nii)ru, but (soneQ)to.19
Although recognition of the foot in Japanese phonology is rather recent, dating back to
Poser’s seminal 1990 work, the evidence for its relevance is massive and
uncontroversial. Poser (1990) provides a number of foot-based phenomena which
demonstrate the significant role of the bimoraic—rather than disyllabic—foot in Japanese
morpho-phonology through data taken from geisha / bargirl client names, kinship terms,
rustic girls’ names, ren’yôkei reduplication, mimetic bases, the secret language of
jazzmen, and compound accentuation.
From Poser’s list, let us present the well-known case of hypocoristic formation. We will
then introduce two other foot-based phenomena: women’s secret language (nyôbô
kotoba) derivatives, and compound loans truncations.
Page 27 of 36
Prosodic units
Hypocoristics are made by adding the suffix -chan to a bimoraic foot containing segmental
material obtained from the base (the complete name). When the name is longer than five
moras, two-foot-long hypocoristic derivatives can also be found. The important point to
note is that one-, three-, or five- mora-long patterns are prohibited. This provides robust
evidence that it is at the foot level that hypocoristic derivation operates.
(21) Hypocoristic formation
Full name hypocoristics

mi dori mi i-chan, mit-chan, mido-chan *mi-chan
ta rou ta ro-chan, taa-chan *ta-chan
kazuhiko ka zu-chan *ka-chan *kazuhi-chan
kenzaburou° ke n-chan, kenzabu-chan *ke-chan, *kenza-chan
*kenzaburo-chan
(p.172)
Derivatives used in the feminine language and originating from nyôbô kotoba ( ), an
ancient secret language used by court ladies, are also based on the foot. Nyôbô kotoba
forms are obtained by truncation of the base to foot-size, and addition of the polite prefix
o-, as shown in (22). Note that the truncated base can be equivalent to two light syllables,
or a heavy syllable, in a syllable-based approach. But it can never consist of two heavy
syllables, thus demonstrating once again that Japanese feet are made up of moras rather
than syllables.
(22) Nyôbô kotoba
base derived form (o- + µµ)

satsuma imo° o-satsu ‘sweet potato’
juubako° o-juu ‘stackable box’
dengaku° o-den ‘oden (Japanese hotchpotch)’
neshouben o-nesho ‘bed wetting’
The foot is also used as the basic unit in compound abbreviated loanwords (Labrune
2007), which are normally built by extracting one bimoraic foot (generally the initial one)
out of each component of the original form, yielding a two-foot-long derivative.
(23) Compound abbreviated loanwords
base truncated form

dejitaru kamera deji-kame° {digital camera}
Page 28 of 36
Prosodic units
rimouto kontorouru rimo-kon° {remote control}
enjin sutoppu en-suto° {engine stop} ‘motor stalling’

patorouru kaa pato-kaa {patrol car} ‘police patrol car’
The foot is also the prosodic domain within which certain accent shifts occur. Whenever
several vowels in the environment of a devoiced vowel are likely to receive accent as a
consequence of a NADM accent shift, the basic principle seems to be that the accent will
remain within the same foot (Tanaka, 2001). Another study by Tanaka (1998, 2008)
highlights the fact that in certain cases, evaluation of the relative sonority of the vowels in
order to determine accent placement occurs within the head foot. Accent then falls on the
most sonorous vowel in the foot (see section 7.3.1 (35) for examples about the
correlation between vowel sonority and accent placement).
Finally, in their experimental study of Japanese speech rhythm, Ayusawa et al. (1998)
have also found that a high proportion of the units of rhythm consist of two moras, that is,
one foot.
(p.173) I assume that Japanese feet obey a well-formedness constraint, which enforces
that they start with a mora of relative prominence, preferably a full CV mora, i.e. the
canonical prosodeme of the language. So the inability of deficient moras to be tonic does
not come directly from their weak sonority, nor from their structural incompleteness, but
from their position within feet. Let us review again the structure of the canonical Japanese
foot. Recall that it can have the following shapes: (C)VCV, (C)VV, or (C)VC. Since the level
of the syllable is considered as not relevant in the prosodic hierarchy of Japanese, feet
are directly made up of moras, as in (24).
(24) Structure of Japanese feet
In the prosodic model of Japanese proposed here, the role usually devoted to the heavy
syllable is entirely taken up by the foot. Supposedly heavy syllables of the shape CVC or
CVV are simply reinterpreted as feet containing a deficient mora in the second position,
i.e. CVC● or CV●V.
Another important characteristic of Tôkyô Japanese feet is that they are trochaic, that is,
initially headed (Shinohara, 2000; Labrune, 2002; Ota, 2003).
I further propose that a well-formedness constraint determines the shape of the

Japanese foot:20
Page 29 of 36
Prosodic units
(25) Foot well-formedness constraint
*F[m A foot must start with a full mora
In other words, feet must be properly headed by a full mora, and their head is aligned to
the left (trochees). A full mora is a mora whose structure corresponds to one of those
standing at the left end of the hierarchy in 6.3.3.21
(p.174) All this means that deficient moras are not, except in very special or secondary
cases, in a position to stand as foot heads and to receive the accent. This is obviously so in
the case of /N/, /Q/, and /R/, even if exceptions can be found (the already oft-cited
obaasankko). It is less obviously so in cases of onsetless vowels. The *F[m constraint
predicts that feet that start with a vowel cannot be accented on the initial mora. Yet, such
feet do exist in Japanese, which contradict this prediction. However, it seems that such
feet are in the majority word-initial, as in i to ‘thread’, a mazon ‘the Amazon’, for instance.
Remember that onsetless vowels have an intermediary status in the deficientness
gradation, and it seems that when they occur word-initially, they are more able to stand
as accent nucleus than in the word-internal position. Interestingly, a statistical survey
carried out on a corpus made up of 211 toponyms of Western origin,22 a class of words
in which the accent is normally attributed according to the general, default principles that
govern Japanese accentuation, establishes that vowel-initial words are less frequently
accented on the initial mora than consonant-initial words, in a statistically significant
manner: 52% of #V- words are initially accented, as against 70% of #CV- words, all other
things being equal. This shows that a constraint preventing feet from starting with a
deficient mora does exist, even if this constraint is not top-ranked in the hierarchy, which
explains that onsetless initial vowels—which, as stated in section 6.3.3, stand in between
truly full moras and truly deficient ones—can still be found, especially at the beginning of
words, a position which commonly allows exceptional phonological configurations to occur.
While a number of issues certainly remain, which require additional empirical and
theoretical investigation, I believe that the above general framework of analysis of the
Japanese mora and foot is correct and that it provides a satisfactory account of the
phonology of Japanese. The role played by *F[m in the phonology of Japanese will be
further exemplified through the case study of the accentuation of foreign toponyms in the
following chapter (section 7.2.5).
6.5 The Prosodic Word And the Upper Levels of the Prosodic Hierarchy
The foot is dominated by the prosodic word (go or inritsugo ), which is the domain
of accentual rules (see Chapter 7) and of a number of phonotactic and morpho-
phonological rules or constraints such as h lenition (section 3.7), non co-occurrence of
voiced obstruents (section 4.1.1), rendaku (section 4.2). In Japanese, the prosodic word
corresponds to a lexical word with its ‘satellites’, i.e. the (p.175) enclitic particles which
may follow it. In some cases, it also contains a prefix such as o- or go- (both honorific
markers).
Page 30 of 36
Prosodic units
There is no consensus as to the number and nature of the prosodic units above the
prosodic word in the literature. The most widespread view, which we will follow in this
book, posits two major categories between the prosodic word and the utterance: the
minor phonological phrase and the major phonological phrase (McCawley, 1968; Selkirk
and Tateishi, 1988; Kubozono, 1993b) also known respectively as accentual phrase and
intermediate phrase (Pierrehumbert and Beckman, 1988) or the accentual phrase and
the intonation phrase (Venditti et al., 2009). If, as this book claims, Japanese does not have
phonological syllables, there is no need to recognize the existence of a syllabic level in its
prosodic hierarchy. Instead, the mora constitutes a prosodic constituent in its own right
directly linked to the foot, which has the full capacity to license segments. Accordingly, the
prosodic hierarchy of Japanese may be represented as follows.
(26) Japanese prosodic hierarchy
However, some scholars posit only one corresponding category, the phonological phrase,
which covers the two subtypes of phonological phrases (Labrune, 2006; Itô and Mester,
2008).
The minor phonological phrase (akusento-ku アクセント ) generally contains two prosodic
words representing an [Adj + Noun + Particle] phrase, such as oishii gohan ga ‘delicious
rice + Subject particle’, or [Noun no Noun] ‘Noun of Noun’, in a phrase like A tsuko no
gohan ‘Atsuko’s rice’. It has at most one pitch accent (one HL pitch change, see Chapter
7). The major phonological phrase (chô akusento-ku アクセント ) may contain two minor
phrases, for instance A tsuko no oishii gohan ga ‘Atsuko’s delicious rice + Subject’. It is
the domain (p.176) of the application of catathesis, or downstep (Poser, 1984;
Pierrehumbert and Beckman, 1988).
The utterance (bunsetsu ) is the domain of declination and of final raising or lowering,
which may, for instance, differentiate questions from statements. Exemplifications of how
intonation operates at the levels of the minor and major phonological phrases will be
provided in the next chapter (sections 5.1 and 7.1.5, and 7.4), after the accentual system
has been introduced.
6.6 Summary And Concluding Remarks

The mass of facts examined throughout this chapter shows that the mora, a rhythmic and
Page 31 of 36
Prosodic units
temporal unit of segment grouping, and the foot, a domain of mora grouping and
organization, are sufficient for the understanding and analysis of Japanese phonology at
the lowest levels of prosodic organization. Two types of moras have been distinguished:
regular moras, which are ideally made up of a consonant + a full vowel, and deficient
moras, which lack one of the two full C or V elements in their prototypical shape.
However, the difference between full and deficient moras is gradual, some moras being
‘more’ deficient than others, which accounts for the fact that some deficient moras
sometimes behave like regular ones, whereas others, most typically /Q/ for instance, are
always treated as deficient.
The main advantage of the approach retained here is that it enables us to capture under
the same phonological category a set of objects (namely onsetless vowels, epenthetic
vowels, voiceless vowels, the mora nasal, the first part of a geminate, and the second part
of a long vowel) which share a number of characteristics in Japanese phonology, but that a
syllabic analysis fails to capture as similar entities. In the present approach, all normally
syllable-linked phenomena, and, in particular, references to the heavy syllable, are
replaced by references to the foot and/or mora. We shall see in the next chapter how this
approach can help provide a better account of some classical issues of Japanese
phonology, through the formal analysis of loanword accentuation and compound word
accentuation, two problems which have been claimed by the proponents of the syllabic
analysis to prove the relevance and necessity of the syllable.
The proposal that Tôkyô Japanese makes no use of syllables contrasts sharply with the
dominant current in mainstream phonology, which assumes that the syllable is not an
optional unit in human languages but a universal constituent whose existence cannot be
called into question.23 However, it should be (p.177) emphasized that this is in
accordance with the traditional analysis of the Japanese kokugogaku school of thought,
which recognizes only what corresponds to the mora for the analysis of Japanese, with a
distinction between autonomous and special moras.
Notes:
(1.) In addition to the mora (haku), Hattori (1960) also refers to the syllable (onsetsu) in
one of his papers, but this appears in a footnote, and is made in a rather polemical tone in
response to a comment by Kindaichi Haruhiko, a linguist with whom Hattori often stood in
opposition.
(2.) Translation: Ueda Makoto, 1998, The Path of Flowering Thorn: The Life and Poetry
of Yosa Buson, Stanford: Stanford University Press. I thank my friend and colleague
Michel Vieillard-Baron for finding this English translation of Yosa Buson’s haiku for me.
(3.) It is legitimate to question the relationship between the writing system and the
phonological system in Japanese, and to consider the extent to which the two systems
influence each other. Thus one might wonder for example if it is because they are written
by means of autonomous and specific graphic units that the mora nasal, or the first part of
a geminate, are treated as one unit in the phonological system, or is it the other way
round: is it because each corresponds to an autonomous phonological unit, the mora, that
Page 32 of 36
Prosodic units
they possess a specific and autonomous graphic sign? In other words, is it phonology
which conditions writing, or writing which conditions phonology?
My position is that it is the phonological structure of the Japanese language which has
determined the kana graphic system and its viability. The kana system is derived from
Chinese characters. Had it evolved in accordance with the logic and spirit of the Chinese
graphic system, it would have produced a true syllabary in Japanese, i.e. a system which
uses one graphic sign (and not two as is currently the case) to transcribe a heavy syllable
such as kan,sou, or kai. In Chinese, where morphemes are monosyllabic, any syllable, be
it heavy or light, occupies one graphic virtual box. This is not how things work in Japanese,
however: each kana sign transcribes one mora in one graphic virtual box, not one
syllable. This evolution was by no means induced by the Chinese graphic system, so one
can reasonably assume that if Japanese has followed ‘the path of the mora’, it is because
of its phonological structure, not the reverse.
This does not mean that the graphic system could not, in return, contribute to
consolidating and to establishing more firmly the phonological representation of Japanese
as a mora language, in the same manner as the use of the alphabet, based on the
phoneme, unquestionably contributes to the development of a phonemic conscience in
literate speakers (Morais, 1994).
(4.) The rendering of the first part of the geminate p by tsu っcan be explained by the kana
spelling にっぽん, since the letter that denotes the sound tsu is also used for the
transcription of gemination.
(5.) Kubozono (1993a) studies the case of a child aged 4 years and 9 months who knows
how to play shiritori, but ignores the special moras which occur word-finally. Instead, she
utters a word beginning with the penultimate mora. For instance, she produces cha-iro
after rika-chan,takoyaki after bataa, or kemushi after tokei. Obviously, special moras
pose a problem for very young children. It is noteworthy that the child studied by
Kubozono does not use the final (special) mora of the words which are submitted to her
but the last full mora, totally ignoring the special mora. Had she uttered chanpon after
rika-chan, taa-chan after bataa, or keiretsu after tokei we would have had strong
evidence that this child language is syllable-based, but it is not the case.
(6.) Actually, the first Western linguist to operate a distinction between long and short
syllables in Kyôto Japanese, on the model of Latin, is the Portuguese missionary
Rodriguez, in his Arte da lingoa de Iapam published in 1604–1608.
(7.) Of course, the fact that there does not exist any proof of the existence of an object is
not enough to constitute the absolute certainty that this object does not exist. As Hyman
(2003 [1985]) points out, ‘it is of course logically impossible to prove that a language does
not have syllables, since it may be the case that it has them but does not show obvious
evidence of it’.
(8.) The branching rhyme approach which posits that a heavy syllable consists of an onset
Page 33 of 36
Prosodic units
and a rhyme, itself divided into a nucleus and coda, has sometimes been adopted, for
instance by Yoshida S. (1990, 1991), whose work is cast within the framework of
government phonology. Although Yoshida argues that heavy syllables must be analysed
as dissyllabic sequences with two nuclei (in his model, kuukou° /kuRkoR/ ‘airport’ or
konpon° /koNpoN/ ‘base’ consist of four syllables), he does not give up the idea that
certain Japanese syllables have branching rhymes: according to him, this would be true of
sequences where two special moras (special segments) follow each other, as in ho ntte
/hoNQte/ ‘book + Citative’. In ho ntte, the first part of the geminate is considered to be
associated with the rhyme of the syllable whose nucleus is /N/.
(9.) I thank a participant to the Berlin Conference ‘32. Jahrestagung Deutsche Gesellschaft
für Sprachwissenschaft’, Humboldt University in Berlin (February 2010) for bringing this
point to my attention.
(10.) Moreover, in the speech of certain Tôkyô conservative speakers, I have heard
nasalized vowels before a voiced obstruent, for instance in ka zu ‘number’ uttered as
[¹kãd͡ zɯ], showing that vowel nasalization does not occur within the domain of the
syllable, since it is not even questionable that the affricate [d͡ z] may belong to the same
syllable as the preceding /a/ in ka zu. However, to my knowledge, this phenomenon has
never been fully reported nor instrumentally verified for Tôkyô Japanese, so I only
mention it in passing.
(11.) The Izu dialect belongs to the Tôkyô type family of dialects. Izu is located 100 km
southwest of Tôkyô.
(12.) For the sake of convenience, I use the terms ‘onset’ and ‘nucleus’ to refer to,
respectively, the first and second position constituents of the prosodeme.
(13.) The feature [consonant] is defined here following Chomsky and Halle (1968). It
indicates a sound which is produced with a radical obstruction in the mid-sagittal area of
the vocal tract.
(14.) See Hyman (2003 [1985]) and Kager (1997) for arguments in favour of the
theoretical existence of such vowels in languages.
(15.) Accent shifts caused by the presence of an onsetless vowel are rather inconsistent.
Accent variation occurs with certain words but not with others, for instance ‘to think’ can
be realized as kangaeru or kangaeru, and so does kotaeru / kotaeru ‘answer’ and
totonoeru / totonoeru ‘to arrange’, but other verbs containing the same ae sequence do
not seem to accept variation. So haeru ‘to grow’ (*ha eru) is accented on the penultimate
mora with no possibility of accent shift, while ka eru ‘to return’ is only accented on the
antepenultimate (*kaeru), in violation of the verb accent rule which requires tonic verbs
to be accented on the penultimate mora.
(16.) A small number of languages, including English, in which the structure of the onset is
argued to be prosodically relevant have been reported (see Kelly, 2004, Davis, 1988,
Page 34 of 36
Prosodic units
Everett, and Everett, 1984). See also Topintzi (2010) for an extensive study.
(17.) Tanaka (2008) also proposes that the special moras should be arranged along a
hierarchy: /J/ 〉 /R/ 〉 /N/ 〉 /Q/ because they do not have a uniform action with respect to
accent location in certain categories of words. In particular, his study shows that /Q/ does
not behave like the other three special moras.
(18.) The relationship between sonority and accent in Japanese has been brought to light
by Tanaka (1998, 2008) and Yokotani (1997). It is also attested for other languages
(Hayes 1995; Kenstowicz 1994b).
(19.) Notice that so netto is not accented on the antepenultimate mora ne, as would be
expected if Q behaved exactly like the other special moras of Japanese, but on the pre-
antepenultimate mora so.
(20.) I would like to thank Marc Plénat for first suggesting to me the existence of a
constraint such as *F[m, and for providing many valuable comments on the issue of
moras and feet interaction. The *F[m constraint in (25) is actually a cover constraint for a
set of three different constraints (see Labrune, 2012).
(21.) As stated in section 6.3.3, the border between full and deficient moras is somewhat
uncertain. It falls somewhere in between Cu and ●e, depending on a variety of factors
which remain to be discovered.
(22.) This list comes from one of the appendices of NHK (1998).
(23.) But note that a small number of scholars have questioned the universality of the
syllable, for instance Kohler (1966), Hyman (2003 [1985]), and Auer (1994). Hyman
(2003 [1985]) claims that the syllable is not a universal constituent but a language-
particular construct built out of weight units (which correspond to moras). According to
him, some languages simply do not construct syllables. See Labrune (2012), for a more
developed discussion.
Accent

Laurence Labrune

Print ISBN 13: 9780199545834
DOI: 10.1093/acprof:oso/9780199545834.001.0001
Accent
Laurence Labrune
DOI:10.1093/acprof:oso/9780199545834.003.0007
Tôkyô Japanese has been described as a pitch accent system which contains tonic and atonic words,and has been
extensively described and analysed both in and outside Japan. After presenting the basic mechanism of present day
Tôkyô Japanese accentuation and offering a summary of two theoretical treatments of it, the chapter enters into the
details of each word type accentuation: simplex vs. compound, native vs. Sino Japanese vs. Western, nouns vs. verbs
and i adjectives, numeral compounds, and so on. A new formal account within the framework optimality theory of the
accentuation of Western loans and of that of compound nouns is offered. After exploring the dialectal and sociological
variation of accent, the chapter concludes with a discussion on the status of the Japanese word prosodic system from
a typological point of view, since a number of features of this system could be interpreted as tonal rather than
accentual.
Keywords: accent p tch accent system s mp ex words compound words S no apanese words Western words numera compunds
opt ma ty theo y ton c words aton c words
Because of its complexity and rich phenomenology, as well as the typological variety exhibited across dialects, the issue
of the Japanese accent (akusento アクセント) poses a challenge to linguistic description and analysis.
We will begin this chapter with a presentation of the general principles of the Tôkyô Japanese accent (section 7.1),
before examining in turn (sections 7.2,7.3, and 7.4), the question of simplex words accentuation, compound words
accentuation, and that of the phonological phrase. In section 7.5, we will broaden our perspective in order to get a
glimpse at individual and dialectal variation. Section 7.6 will allow us to reflect on the nature of the Japanese accent from
a typological point of view. It will be followed, in section 7.7, by a review of the various accent studies which have been
carried out in Japan and in the West from the Edo period until the present time.
The study of the Tôkyô Japanese accentual system occupies a particular position within theoretical phonology because
this original prosodic system provides an interesting and challenging ground for the testing of prosodic models which
have often been elaborated for ‘stress’ languages like English or ‘tone’ languages like Chinese, i.e. prosodic systems
seemingly quite different from that of Japanese. Such testing is thus all the more invaluable to judge the validity of
theoretical principles because the Japanese accent system is without question among the best documented in the
Page 1 of 55
Accent
world. Moreover, and this is, in my view, a crucial point, rare enough in the field of linguistics outside non Indo
European languages to be worth mentioning, the descriptions of Japanese that are available are, in their majority, by
linguists working on their own native tongue within a variety of frameworks, including a largely indigenous approach.
It might have been necessary to begin a chapter devoted to the Japanese accent system by a reflection on what an
accent language is, and on how it should be defined in contrast with stress and tone languages. However, for clarity of
exposition, this discussion is postponed until section 7.6, after the presentation of the details of the so called Japanese
‘accent’ have been introduced, but some readers might prefer to read section 7.6 first.
(p.179) 7.1 General Principles of Tôkyô Japanese Accentuation
7.1.1 Bas c Mechan sms

The accent of Tôkyô Japanese consists of a distinctive lexical pitch accent (sometimes called musical accent). It is
marked phonetically by the change from a high pitched mora (noted H henceforth) to a low pitched mora (L). The last
mora of the word carrying the H tone before the drop towards L is regarded as the accented mora of the word, its
prosodic peak, or accent kernel (akusento kaku アクセント ).
The lexicon is divided into tonic words (accented type, kihuku shiki ) and atonic words (unaccented type, heiban
shiki ). A tonic word contains an HL sequence (in other words, a pitch drop), an atonic word does not. In this book,
accented moras appear in bold characters while unaccented words are followed 1 by the symbol °. In Standard
Japanese, there exists in theory for any given word, n + 1 accent possibilities, n being equal to the number of full
moras of the word. Thus, for a three mora word having a CVCVCV structure, the following four prosodic patterns are
all possible at the lexical level:
(1) Accent possibilities for three mora words
CVCVCV na mida ‘tear’

CVCVCV kokoro ‘heart’
CVCVCV kagami ‘mirror’
CVCVCV° (atonic) sakura° ‘cherry tree’
In Japanese terminology, the na mida type, with initial accent, is known as atamadaka gata ‘high head pattern’ ( ),
the kokoro type, with internal accent, is a nakadaka gata ‘high middle pattern’ ( ), and kagami, with final accent, is
termed odaka gata ‘high tail pattern’ ( ). The last word, sakura°, is a heiban gata ‘flat pattern’ ( ).
The surface prosodic pattern of a word is then governed by the two following principles:
*HLH, or Adjacency Principle: two high pitches (H) on the surface can cohabit in a word only if they are adjacent.
That is, several H’s cannot be separated by one or more L’s. There is thus only one high plateau in a word. Here
accent fulfils a culminative function.
(p.180) *#HH and *#LL, or Initial Dissimilation Principle (also known as initial lowering): the first and second moras
in a word always have different heights. A word thus starts obligatorily, either by HL if the first mora of the word
carries the accent, or by LH in all other cases. The only exception seems to be if the second mora of the word consists
of a special segment (see section 6.2.4), in which case the word may start with HH rather than with LH.2 The presence
of the LH melody on the first two moras is dissimilatory in nature and always predictable. The accent here fulfils a
demarcative function (see Warner, et al., 2010, an experimental study which shows confirmation of this traditional
formulation of initial dissimilation as a demarcative cue). As we will see below in this subsection, as well as in section 7.5,
it is actually at the level of the phonological phrase rather than at the level of the word that initial dissimilation occurs
but it makes no difference here since we are talking of single lexemes.
From the application of the two above mentioned principles, all the moras which occur before the accented mora carry
a high pitch at the phonetic level in words other than those bearing an initial accent, except the first mora, which is low.
Consequently, surface prosodic patterns such as (2a and b) are impossible within a prosodic word in Standard
Japanese because they violate the two principles stated above.
(2) Impossible surface prosodic patterns
a. Violations of *HLH:
Page 2 of 55
Accent
*HLH *LHLH *HHLH etc

b. Violations of *#HH and *#LL:
*LL *HH *HHH *LLL
*HHL *LLH *HHHL *LLLH etc
The actually occurring surface patterns are given in (3). The lexical word and the following particle, which constitute a
prosodic word, are separated by a hyphen. We shall limit ourselves here to examples from one to five moras (actual
examples are provided in Table 7.2 below):
(p.181) (3) Possible prosodic patterns
a. Tonic words (containing an HL sequence)
H L HL L HLL L HLLL L HLLLL L

LH L LHH L LHHH L LHHHH L
LHL L LHLL L LHLLL L
LHHL L LHHLL L
LHHHL L
b. Atonic words (containing no HL sequence)
L H LH H LHH H LHHH H LHHHH H
An atonic word, as in (3b), is a word which does not contain any shift from H to L. Following the principles stated
above, the prosodic pattern of an atonic word will start inevitably with a LH melody (under the principle of initial
dissimilation), followed by as many H’s as the word contains moras, with no significant downfall towards L.
The accent of Japanese bears all the properties of a distinctive accent, and indeed, it is. However, the importance of
the distinctive function should not be exaggerated, since only 14% of the total number of minimal pairs in the language
are distinguished by a difference in the accent pattern according to Sibata and Sibata (1990), quoted by Kubozono
(2001a). The principal function of the Japanese accent thus consists of organizing the linguistic units within the
utterance linearly, and, by means of intonation, of structuring morpho syntactic and semantic information. The location
of the accented mora is indeed that from which the whole intonation structure will be built.
Although the accent of Standard Japanese is by and large lexical, there exist, in certain areas of the lexicon, some
rather regular principles which determine the accent pattern of words. There are notably two such areas: the accent
of loanwords of Western origin and that of compound nouns. These will be examined in sections 7.2.4 and 7.3.1 4
respectively. As we shall see, in both cases, one observes a tendency towards the emergence of a default accent on
the antepenultimate mora.
The presentation which has been made in the preceding pages represents the most consensual and widespread
description of the accent of Tôkyô Japanese. But other views exist. For instance, Kubozono (2006b, 2008) proposes a
different approach to the accent pattern system of nouns in Tôkyô Japanese by positing that Tôkyô Japanese is not a
multiple pattern system but a two system pattern. For him, the major opposition is between the accented and the
unaccented types (this is in agreement with the standard analysis). But, within the accented type, Kubozono proposes
that there exists a default accent pattern which is assigned by rule (the antepenultimate rule), and a number of lexical
exceptions which must be (p.182) memorized. This class of lexical exceptions is made of nouns which bear an accent
on a mora other than the antepenultimate.
7.1.2 Aton c Words (and the r D fference From F na Accented Words)

Unaccented words and final accented words are generally described as having the same phonetic realization when
uttered in isolation. The difference between the two is supposed to be perceivable only when the word in question is
followed by an accentually neutral element, for instance an enclitic particle such as ga (Subject) or wo (Object), whose
presence after the final mora will allow the realization of the HL downshift, which is the only truly reliable clue of the
presence of a final accent.
In order to illustrate this point, let us consider the examples hana ‘flower’ (accented) and hana° ‘nose’ (unaccented) in
Figures 7.1 and 7.2. In both cases,
Page 3 of 55
Accent
Figure 7.1. Accent curve (F0) of hana ga ‘flower’ (0.30s)
Figure 7.2. Accent curve (F0) of hana ga° ‘nose’ (0.27s.)
(p.183) the final mora na is realized with a high pitch by speakers of Tôkyô Japanese when the words occur in
isolation (that is, with no following particle). The phonetic realization of both hana ‘flower’ and hana° ‘nose’ is thus
supposed to be identical at the perception level (but see the remark below). It is only when hana and hana° are
followed by an enclitic particle, for example ga, that the prosodic difference between the two can be fully materialized.
In hana ga ‘flower’, a significant fall of register occurs on the mora ga, after the final na. In hana ga° ‘nose’, there is no
significant prosodic fall after the final mora na, as can be seen in the instrumental analyses presented in the two figures.
(The two words were uttered by a male speaker of Standard Japanese.)
However, and very interestingly, there actually exists a slight difference in the accentual curve of hana and hana°
before the particle ga: the frequency of na in hana ‘flower’ is higher than that of hana° ‘nose’. In reality, this difference
also occurs when the words are uttered in isolation. This type of data is in contradiction with the assumption by
Haraguchi, presented further in 7.1.5, according to which a final accented word and an unaccented word are
supposed to receive exactly the same autosegmental representation at the surface level because, if it were the case,
no phonetic difference would be carried out. As we shall see below, the model developed by Pierrehumbert and
Beckman (1988), on the other hand, correctly predicts that such a difference between hana and hana° may be
realized.
The issue of whether final accented words and unaccented words are different both at the phonetic level and at the
phonological one or at either one of these two levels is actually a controversial one. The mainstream opinion has it that
the two types of words are not phonetically different, but many scholars have argued that they are (Sakuma, 1929;
Uwano, 1977; Pierrehumbert and Beckman, 1988; Warner, 1997; see the latter for a comparison of the competing
theories and a review of the literature on the subject). Experimental results are inconsistent because they generally
show that a difference is made by some speakers but not by other speakers, which suggests that there indeed exists
a difference at the latent level between final accented and unaccented words, in accordance with Pierrehumbert and
Beckman’s theory. Moreover, it has been demonstrated (Sugitô, 1982; Vance, 1995) that the phonetic difference,
when it actually exists is not perceptible by most speakers, and that one should thus regard it as linguistically
irrelevant at the perception level.
The experimental study by Warner (1997) further establishes that unaccented noun phrases have a lower f0 curve
than accented noun phrases.
The existence of an atonic word class actually constitutes the most puzzling fact of the Japanese word prosodic system.
It is also a challenge for the (p.184) categorization of Japanese as an accent language, an issue to which we shall turn
to in section 7.6.
Statistical data drawn from various sources (see, among them, Table 7.1) show that the atonic pattern is the most
frequent one. According to Hayashi O. (1982), about half of all words (regardless of their category or length) are
unaccented. Of course, the proportion varies according to the lexical strata, the prosodic structure, and the
morphological structure of lexemes, but the correlation between word length and accent pattern is quite evident in
Page 4 of 55
Accent
Japanese. For instance, the data in Table 7.1 show that 65.7% of quadrimoraic nominal words of Yamato and Sino
Japanese are atonic as against 52.6% of trimoraic words. This contrasts with nouns up to two moras in length, which
favour the initial accented pattern (Uwano, 2003:81). Even Western loanwords, which are less frequently atonic than
words belonging to other strata, show a higher percentage of the atonic pattern for four mora long words than for
shorter or longer words (Kubozono, 2006b).
There also exists a clear tendency to convert tonic nouns into atonic ones in contemporary Tôkyô Japanese. For
example, densha° ‘train’, kareshi° ‘boyfriend’, and houki° ‘broom’ are realized more and more as atonic words
instead of de nsha, ka reshi, and ho uki, which represent the conservative realization. The atonic pattern thus seems
to have become the most productive one.
Moreover, high frequency words, as well as words whose referent conveys a feeling of proximity, familiarity, or
attachment to the speaker are the quickest to undergo de accentuation, in particular among younger speakers. For
instance, someone who is very fond of guitar playing will tend to pronounce the word gitaa° ‘guitar’ as unaccented,
instead of gi taa, the more conventional form of the word. It is well known that in languages most frequent words
regularly undergo phonetic reduction, either in the form of abbreviation (apocope, etc.), articulatory simplification, or
prosodic simplification, like the example of de accentuation in Japanese.
De accentuation of a tonic word can thus function as a mark of social identity. In order to mark his/her belonging to a
given group, to show that he/she is well integrated in the field denoted by the word, a speaker may de accentuate a
word, in a similar way to what one observes in the use of clipped forms in certain social groups in English or French.
Yet, the sociolinguistic dimension does not suffice to account for the phenomenon of de accentuation in all its aspects.
This is because, as already stated, all things being equal, three and four mora long words show a higher percentage
of the atonic pattern than shorter or longer words, whatever their frequency or familiarity connotation.
One might wonder whether the generalization of the atonic pattern in present day Japanese is to be regarded as some
sort of fashionable, temporary phenomenon, or whether it is the manifestation of a major general evolution of the
(p.185) language towards prosodic simplification, whose final step would be the loss of the phonological accent, as it is
the case in certain modern dialects in the Tôhoku or Kyûshû area (see section 7.5). The question remains open.
A possible (but only partial) explanation of this preference for the unaccented pattern could be that the accentual
levelling of three and four mora words is practically never likely to involve a loss of lexical opposition. Indeed, pairs of
words longer than three moras which are homophonous except for the accent pattern are extremely rare, especially
in the Yamato and Western strata. In addition, three and four mora long accented nouns undergo de accentuation
more often than longer nouns. This is probably because words longer than four moras are likely to be compounds in
Japanese, and as we will see later in this chapter (section 7.3), the presence of an accent as well as its location play an
essential morphological role in compounds, since accent location is frequently determined in relation to the internal
boundary. This probably accounts for the fact that longer compound words are less likely to lose their accent than
simplex words.
Yet, the reason for this special link between atonicity and quadrimoraicity is very puzzling and poorly understood. It
might be necessary to invoke frequency effects and analogical factors, as I did in Labrune (2006:180). Because the
atonic pattern is statistically most frequent with four mora words in the Japanese lexicon, it tends to become more and
more productive, leading this pattern to apply to recent loans, inter alia. But Tanaka (2008:216) presents what is, I
think, the best explanation for the generalization of the atonic accent pattern in four mora long words. He observes
that since the primary function of the accent in Japanese is the culminative function, that is, to mark the presence of a
word in a sequence, de accentuation can be interpreted as a loss of this function. This, at first sight, appears to be
totally contradictory with the role and nature of the accent in Japanese. However, since four mora is the most
frequent word length for Japanese lexemes, that is, the unmarked, most expected, and default word length, word
identification and demarcation is not so dependent on accent for quadrimoraic words. Quadrimoraicity constituting the
first cue for word recognition and identification, the presence of an accent becomes redundant for four mora long
words and can be suppressed with no functional loss. The same type of reasoning can be applied to three mora long
words, which constitute the second most frequent word length in Japanese. The fact that familiar and frequently used
words undergo accent loss in a similar manner reinforces this interpretation, since familiar and frequent words are
also more readily recognized than less familiar ones. This line of explanation is particularly appealing, especially in
consideration of the fact that de accentuation does not affect the initial dissimilation principle, which assumes the
demarcative function by signalling the beginning of a word or phrase. It can thus be argued that, if de accentuation is
possible in Japanese, it is thanks to the initial dissimilation mechanism. Should the trend (p.186) towards de
accentuation affect the entire lexicon of Tôkyô Japanese, the language would turn into a fixed accent (non distinctive)
Page 5 of 55
Accent
system with initial accent, initial dissimilation now assuming the role of an accent.
7.1.3 Frequency of the Accent Patterns

Examination of the frequency of the various accent patterns within the Japanese lexicon is instructive, and provides a
new type of data likely to shed new and complementary light on the observations of a more traditional nature. Table
7.1 provides some statistical information concerning nominal words of Yamato or Sino Japanese origin, according to
Sibata (1994). Figures 0, −1, −2 and so forth indicate the position of the accented mora starting from the end of the
word. 0 thus marks atonic words, −1, finally accented words, −2, penultimate accented words, and so forth.
The global results show that the atonic words (type 0) are most numerous: they represent almost half of the total, with
47.3% (see the comments in the previous section). Then we have the antepenultimate pattern (type 3), which amounts
to 26.1%. However, these results should not be taken in too general a fashion since there are sometimes important
differences if the lexical strata, the phonological structure, and the morphological structure are taken into account. So,
for instance, Kubozono (2006b) has found that, whereas 71% of Yamato nouns are unaccented, only 7% of loanwords
and 51% of kango are.
Closer examination of the data in Table 7.1 highlights the existence of other curious correlations between the accentual
pattern and the length of the lexemes.
One notes that one mora words are divided roughly evenly between the atonic and tonic patterns, with a slight
preference for the latter (57.6%).
For bimoraic words, it is the penultimate pattern (in other words, the initial accent pattern) that is most widespread. It
accounts for 56.3% of the words, well ahead of the unaccented pattern (23.1%). In three and four mora words, the
atonic pattern is by far the most frequent, while in words longer than five moras, on the other hand, the penultimate
and antepenultimate tonic patterns are most widespread.
It is clear thus that there exists a tendency to atonicity for three and four mora words. Because of their massive
presence in the lexicon almost 61% of the words of the corpus tri and quadrimoraic lexemes are responsible for
the supremacy of the atonic pattern in the overall figures. Besides, the atonic pattern is also spreading among certain
speakers, as mentioned in section 7.1.2.
One also realizes that the frequency of the initial accent falls brutally in words longer than four moras. Therefore, the
term ‘initial accent’ (atamadaka gata) which is of general use among Japanese linguists does not seem suitable. It is
preferable to talk about final accent for one mora words, of penultimate accent for two mora words, and
antepenultimate accent for three mora words. Indeed, the accent in Japanese is undoubtedly attributed from the end
of the word, except in some notable exceptions belonging to the class of the numeral and dvandva compounds, as we
will see later in this chapter.
(p.187)
Table 7.1. Location of accent in nominal Yamato and Sino-Japanese words (according to Sibata,
1994), in relation to length of lexemes
Accent Length 0 −1 −2 −3 −4 −5 −6 −7 −8 −9 TOTAL %
1 µ 42 57 99 0.2
42.4% 57.6% 100%
2 µ 446 396 1085 1927
3.6
23.1% 20.6% 56.3% 100%
3 µ 5789 857 868 3483 10997
20.8
52.6% 7.8% 7.9% 31.7% 100%
4 µ 13932 999 2000 2832 1446 21209
40.1
65.7% 4.7% 9.4% 13.4% 6.8% 100%
5 µ 2898 375 633 4547 1061 75 9589
18.1
30.2% 3.9% 6.6% 47.4% 11.1% 0.8% 100%
Page 6 of 55
Accent
6 µ 1571 108 329 1822 2406 26 99 6361

12.0
24.7% 1.7% 5.2% 28.6% 37.8% 0.4% 1.6% 100%
7 µ 134 12 97 955 397 8 12 106 1721
3.3
7.8% 0.7% 5.6% 55.5% 23.1% 0.5% 0.7% 6.2% 100%
8 µ 139 6 16 111 444 1 9 2 21 749
1.4
18.6% 0.8% 2.1% 14.8% 59.3% 0.1% 1.2% 0.3% 2.8% 100%
9 µ 50 11 7 41 51 1 0 0 1 1 163
0.3
30.7% 6.7% 4.3% 25.2% 31.1% 0.6% 0.6% 0.6% 100%
⋝10µ 11 1 2 22 48 84
0.2
13.1% 1.2% 2.4% 26.2% 57.1% 100%
TOTAL 25012 2822 5037 13813 5850 111 120 108 22 1 52896 100%
47.3% 5.3% 9.5% 26.1% 11.1% 0.2% 0.2% 0.2% 0.04% 0.001% 100%
(p.188) A rather surprising result emerges with long words. One observes a clear difference between five and
seven mora long lexemes, which are mainly accented on the antepenultimate mora, and six , eight , and ten mora long
ones, in which pre antepenultimate accent is the most common. Nine mora words are divided into about three almost
equal accent types: atonic, with a tonic penult, and with a tonic antepenultimate. I do not have any satisfactory
explanation concerning these clear differences regarding the dissimilarities of the dominant patterns in long words.
However, the fact that words with an odd number of moras (5, 7) differ from those with an even number of moras (6,
8) allows us to suppose that the difference could be explained by the action of constraints pertaining to foot formation
and alignment. Roughly speaking, it could mean that there exists a tendency to favour an even number of moras
before the accent, but further investigation is needed on that question.
In order to correctly interpret those figures, one should also take into account the fact that nearly all the Japanese
words longer than four moras are compounds (with the exception of Western loanwords), and that compounds
undergo accent rules that are different from those of simplex words.
7.1.4 Accent Sh ft By V rtue of the NADM Pr nc p e

A number of phonological factors may cause a shift of the lexical accent at the surface level. The main source for accent
shift is the presence of one of the deficient moras /R/, /Q/, and /N/ which causes the application of the NADM principle
presented in sections 5.5. and 6.3.2. Recall that /R/, /Q/, and /N/ cannot receive the accent in Tôkyô Japanese. If an
accent nucleus is to be put on one of these units by application of one of the general accent assignment rules, it is
automatically moved to an adjacent full mora, generally the left one. In the same way, the other deficient moras
containing a voiceless vowel (see section 2.6) are also de accented in conservative Tôkyô speech, while those
consisting of a single vowel or an epenthetic vowel only sometimes are (4b, c). (For a theoretical treatment of this
phenomenon, see Chapter 6.)
In the following examples, expected patterns marked as * do not occur. Those which are not starred are attested
alongside the ‘alternative patterns’.
(4) Accent shift due to the NADM principle
a. Obligatory accent shift due to the presence of a special segment
Expected pattern Actual pattern

*nihon kai 〉 nihon kai ‘sea of Japan’
*chichuu kai 〉 chichuu kai ‘the Mediterranean Sea’
(p.189)
b. Accent shift due to the presence of a voiceless vowel (frequent, but not obligatory)
Expected pattern Alternative pattern

hu kaku Φɯkakɯ 〉 hukaku Φɯ̥ kakɯ ‘deeply’
Page 7 of 55
Accent
ki sha kiɕa 〉 kisha ki̥ ɕa ‘train’
aidokusha aido kɯɕa 〉 aidokusha ai dokɯ̥ɕa ‘regular reader’
c. Accent shift due to the presence of a single vowel (obligatory or optional shift, depending on vowel
quality (?); variation is frequently observed)
Expected pattern Actual or alternative pattern

*keizai ryoku 〉 keizai ryoku ‘economic power’
*kaetta 〉 ka etta ‘returned’
*donau gawa 〉 donau gawa ‘the Danube’
kangaeru 〉 kangaeru ‘to think’
anaunsu 〉 anaunsu {announce}
As we shall see in the following pages, extra phonological factors of semantic, lexical, syntactic, etymological, or
morphological nature interact with purely phonological mechanisms in order to determine accent location, to say
nothing of sociolinguistic variation, in particular dialectal, which will be discussed in section 7.5.
7.1.5 An Overv ew of Two theoret ca Treatments of the Japanese Accent

At this point I introduce two outstanding formal treatments of the Japanese accent. Both have exerted important
influence on subsequent accentual studies. The first was proposed by Haraguchi (1977), the second by
Pierrehumbert and Beckman (1988).
(a) Haraguchi (1977)
Haraguchi Shôsuke has presented a formal treatment of the Japanese accent within the framework of autosegmental
phonology (Goldsmith, 1976) which has exerted unquestionable influence both on the development of autosegmental
theory and on the comprehension of the Japanese accent as a whole. Autosegmental phonology assumes pitch accent
to be a tone mark located on a separate tier, and that it is realized on the surface when associated with some tone
bearing unit such as the syllable or the mora. What we have thus far called ‘accent’ is assumed to be the result of the
association of a HL melody to a specific tone bearing unit within the word, lexically specified in the case of nouns. In
tonic words, the accented mora is associated with the H tone of the HL melody, while the low tone remains
unassociated. Atonic words are seen as lexically unspecified.
In tonic words, there exists, at the lexical level, one mora marked as accented. It is this mora which is associated with
the H of the melody, followed by the L (p.190) which remains unassociated. The other moras are not specified as H
or L. If the word is atonic, i.e. not specified as accented in the lexicon, a default HL melody will be associated with the
last mora of the word, as in the miyako° example below.
(5) Lexical representation of words within the autosegmental framework (Haraguchi 1977)3
A series of rules and association conventions, the ‘tone association conventions’, the ‘initial lowering rule’, and the
‘contour tone simplification rule’ (of which I shall leave aside the detailed presentation) applies, so that in the end all
surface vowels are associated with an H or L tone, as shown below.
(6) Surface representation after the application of the tone association conventions, the initial lowering rule, and
the contour tone simplification rule
Page 8 of 55
Accent
Haraguchi’s assumption is that at the surface level, all moras receive an H or L specification.
(b) Pierrehumbert and Beckman (1988)
Janet Pierrehumbert and Mary Beckman have proposed an analysis in which the Japanese accent is taken as a tonal
melody associated with a mora at the lexical level. Their work follows the laboratory phonology approach, which
consists in exploring, with experimental tools, the relationship between the phonetic realization and the phonological
representation. It is both theoretically and model (p.191) oriented. The book published in 1988 aims at describing
the surface tonal patterns of Japanese by identifying the phonetic mechanisms that control the interpretation of
phonological representations, in other words, it aims to uncover the phonetic rules that govern the surface tonal
realization of utterances. It is not directly a study about the lexical accent but a study on the tonal, in the sense of
‘intonational’ structure. Their purpose is ‘to develop a theoretically well defined and empirically motivated system of
surface representation, in order to derive the quantitative characteristics of the F0 contour’. Accordingly, their study
is to be considered as a theory of the interaction between word accent and prosodic structures of a higher level. The
experiments thus shed new light on the status and nature of the lexical accent because, in a language such as
Japanese, the lexical accents constitute by and large the representational input to which the phonetic rules that interest
Pierrehumbert and Beckman are supposed to apply.
Pierrehumbert and Beckman call into question Haraguchi’s analysis based on the following two theoretical postulates
(see above). First, accent is conceived and represented as a mark associated with a vowel within the word. This mark
is later reinterpreted as an HL melody through the application of rules. Second, all vowels are linked to a tonal
specification at the surface level.
Pierrehumbert and Beckman draw a distinction between three types of tones, based on the level of the prosodic
hierarchy at which the tones operate. At the lexical level, there is first an accent tone which consists of an HL melody.
The H tone is phonologically associated with the accented mora of the word, in the spirit of Haraguchi’s model.
At the level of the accentual phrase (also known as minor phrase or phonological phrase, see Chapter 6) such as akai
seetaa wa (‘red’ + ‘sweater’ + Topic), there is, on the one hand, an isolated H tone (phrasal H), which is inserted and
linked with the second mora, except if the phrase starts with an initial accented lexeme. On the other hand, we have a
boundary tone L%, which is inserted at the end of each phrase. The accentual phrase is thus delimitated by an H
associated with the second mora of the word and an L% in final position.
Finally, at the utterance level, one has an L% boundary tone, inserted at the beginning of the phrase as a whole. It will
be ‘strong’ and associated or ‘weak’ and unassociated, depending on whether the word starts with a light or heavy
initial ‘syllable’ (in Pierrehumbert and Beckman’s model) or on whether the second mora is special (in our model).
In this model, tone spreading is rejected, so that no tones other than those that are described here are claimed to
exist in the surface phonological representations of utterances of Tôkyô Japanese. The result is that there are fewer
tones than tone bearing units. Accentual phrases are thus unspecified for tone even at the surface level, as
represented in (7). The main innovation of this model, in comparison with a majority of the preceding analyses, lies in
the fact that (p.192) Japanese is treated as a tone language (but note that the tonal perspective goes back to Poser,
1984) and, especially, that an explicit formal distinction is made between tones assigned at the lexical level (HL) and
those assigned at the level of the accentual phrase (H and L%) and at the level of the utterance (L%). Such a
distinction between word level and phrase level accent assignment was already made by the Japanese phonetician
Kawakami (1957) who analyses the initial HL melody as resorting to the accentual phrase ( ku) rather than to the
lexical accent in the strict sense.
(7) Representation of surface tone patterns for the words yama za kura, ka geboushi,tou mo
rokoshi,moushikomi°, and murasakiiro° in isolation (adapted from Pierrehumbert and Beckman, 1988:14)
Page 9 of 55
Accent
Pierrehumbert and Beckman’s hypothesis in favour of tonal underspecification is based on robust phonetic
experimental data, showing in particular that the F0 contour over the prosodic units following the phrasal H in
unaccented phrases (p.193) (for example moushikomi° or murasakiiro°) is not identical to that occurring after a
lexical HL. The most plausible explanation for this fact is thus that there are, at the surface phonological level, fewer
tones than tone bearing units likely to condition the slope of the F0 contour. In the absence of tonal specifications,
phonetic mechanisms determine the transitions between the phrasal H on the second mora and the final L%.
7.2 Accent of Simplex Words

Japanese words almost always occur with affixes: enclitic particles such as case markers in the case of nouns,
auxiliaries or enclitic particles in the case of verbs and adjectives. However, in so far as these affixes are deprived of
lexical autonomy, formations built upon single words consisting of a Noun + one or several enclitics, or a Verb/Adj +
one or several auxiliaries or enclitics, constitute single prosodic words (or, extended word structure in the terms of
Higurashi, 1983). We thus call ‘simplex word’ any autonomous linguistic form that possesses at most one accent peak
and that is not decomposable into smaller elements likely to have lexical autonomy. Following this definition, yama ga
‘mountain + Subject’, yama-bakari or yama bakari ‘only the mountain’, kaeri masu ‘to return + Polite’ are simplex
words, while yama michi ‘mountain lane’ or kaeri-tsuku ‘to return’, which have one accent peak but are made up of
two autonomous elements (yama,michi°, ka eru, tsu ku), are considered as compounds.
7.2.1 Yamato Nouns

All the accent possibilities described above (in 3a and b) are attested for simplex Yamato nouns. A Yamato noun may
be tonic or atonic. If it is tonic, any of its full moras has the ability to carry the accent kernel. Examples are presented in
Table 7.2, where each noun is followed by the accentually neutral particle ga (subject marker).
However, not all particles are accentually neutral like ga. Some have the ability to modify the position of the lexical
accent of the preceding noun by way of complex mechanisms, as the examples in Table 7.3 illustrate. The term ‘particle’
is used to designate what is called joshi ( ) in traditional Japanese grammar, a category that actually includes
elements of various linguistic types (Japanese so called particles belong to one of the following types: casual, adverbial,
final, and connective).
Two major types of particles need to be distinguished with respect to their accentual behaviour: dominant and
recessive. Prosodically dominant particles, (p.194)
Table 7.2. Accent of simplex Yamato nouns 4

1 mora 2 moras 3 moras 4 moras
Page 10 of 55
Accent
atonic L H hi ga° LH H hana ga° LHH H sakura ga° ‘cherry LHHH H tomodachi ga° ‘friend’
nouns ‘day’ ‘nose’ tree’
tonic H L hi ga HL L ne ko ga ‘cat’ HLL L ka rasu ga ‘raven’ HLLL L ko omori ga ‘bat’

nouns ‘fire’
LH L hana-ga LHH L kagami-ga ‘mirror’ LHHH L imouto-ga ‘younger
‘flower’ sister’
LHL L kokoro ga ‘heart’ LHLL L kudamono ga ‘fruit’
LHHL L mizuumi ga ‘lake’
like dake° ‘only’ or ba kari ‘solely’, may impose their own accent pattern on the noun phrase, while recessive particles,
like yo ri ‘from’ and shika° ‘only’, do not. The presence of a dominant particle causes the accent of the noun to be
deleted, and the accent pattern of the particle determines the accent pattern of the phrase, including cases where the
particle is atonic, as in the example with dake°. The accent of recessive particles only shows up in the surface when the
preceding noun is unaccented. Otherwise it is deleted and only the noun’s accent is realized on the surface as with yo
ri. Note that some particles are pre accenting, that is, they cause the appearance of an accent on the last mora of the
noun which precedes them, for instance shika.
The particle ga° is prosodically inert: the noun to which it is attached always keeps its inherent accent. All the
monomoraic particles but for no and ne belong to this class, as well as some bimoraic particles such as kara° ‘from’ or
hodo° ‘just’ (type 1 in Table 7.3).
The particle yo ri ‘from’ (3 in Table 7.3) is recessive, initially accented. That is, its inherent accent surfaces only after an
atonic noun like hana° ‘nose’, but is deleted after tonic nouns. A number of other bimoraic or trimoraic particles, for
example ma de ‘until’,nado ‘and the like’, nomi ‘merely’, demo ‘even’, datte ‘even’,ka shira ‘(I wonder) whether’,
koso ‘precisely this’, behave in the same manner. Shika ‘only’ is a recessive pre accenting particle for certain speakers
(4b in Table 7.3), but a recessive atonic particle for others (4a).
(p.195)
Table 7.3. Accentual effect of particles

1. N + ga 2. N + no 3. N + yo 4. N + shika 5. N + dake Atonic 6. N + ba kari Tonic
Atonic Special ri Initially Atonic dominant (a) or dominant (a) or
Recessive accented recessive (a) recessive (b) recessive (b)
Recessive or
preaccenting
recessive (b)
atonic hana° hana ga° hana no° hana yori a. hana dake° hana bakari
nouns ‘nose’ hana
shika°
b.
hana
shika
tonic ha‘tooth’ ha ga ha no ha yori ha shika a. ha dake° a. ha bakari

nouns b. ha dake b. ha bakari
hana hana ga hana no° hana yori hana shika a. hana a. hana bakari
‘flower’ dake° b. hana bakari
b. hana
dake
ne ko ‘cat’ ne ko ga ne ko no ne ko yori ne ko shika a. neko a. neko bakari

dake° b. ne ko bakari
b. ne ko
dake
kokoro kokoro ga kokoro no kokoro kokoro shika a. kokoro a. kokoro bakari

‘heart’ yori dake° b. kokoro bakari
b. kokoro
dake
Page 11 of 55
Accent
kudamono kudamono kudamono kudamono kudamono a. a.

‘fruit’ ga no yori shika kudamono kudamono bakari
dake° b. kudamono
b. bakari
kudamono
dake
(p.196) Dake° ‘only’ behaves as a dominant particle for certain speakers (5a), that is, it totally inhibits the accent
pattern of the preceding noun by imposing its own atonic pattern. However, for other speakers, it is a recessive atonic
particle (5b). Bakari ‘solely’ behaves either like a dominant particle with initial accent (6a), or like a particle with
recessive accent (6b), on the model of yo ri.
It is also necessary to posit an exceptional class comprising the genitive particle no (2 in Table 7.3) which behaves in a
special way accentwise. As we saw in section 6.2.4, it causes some finally accented nouns which are at least two mora
long, like ha na ‘flower’, to become atonic, but remains neutral with the other patterns including one mora long
words like ha ‘tooth’ which preserve their intrinsic accent, as they do after ga. A fair number of exceptions exists,
principally nouns ending with special segments, toponyms, and numerals (see 6.2.4.).
It has to be noted that variation is not rare: a given particle such as dake° or ba kari behaves as recessive in the
speech of some speakers but as dominant in the speech of others.
7.2.2 Verbs And i Adject ves

Verbs and i ending adjectives display fewer accent patterns than nouns. They are either tonic or atonic, and, in the
former case, there exists only one single pattern, whatever the length of the base, as shown in Tables 7.4 and 7.5
below.
Among verbs, the two patterns are about equally represented. Among adjectives, on the other hand, the tonic pattern
is much more frequent. Akinaga (2002) notes that many verbs and adjectives belonging traditionally to the atonic class
currently show a tendency to become tonic. Note, in passing, that this trend contradicts the one observed for nouns,
since nouns clearly tend to become unaccented under certain conditions (sections 7.1.2 and 7.1.3).
The accentuation of verbs and i adjectives follows principles that differ from those of nouns. As stated above, the
possible accent patterns are fewer than they are for nouns. However, this does not mean greater simplicity at the
level of the linguistic mechanisms at work. Quite the contrary, the principles which govern verb and adjective
accentuation appear to be extremely complex.
The complexity is first due to the fact that it is difficult to determine which mora of the verb or adjective serves, at the
lexical level, as the accent kernel because the position of accent varies throughout the paradigm of a given verb or
adjective. For example, in the verb taberu ‘to eat’ the accented mora occupies no less than three different locations at
the surface level, depending on the form of the paradigm that is taken by the verb and the affixes attached to it. We
thus have ta be ru / ta bete / tabe ma su (respectively final form / connective form / polite form). In contrast to nouns,
verbs and adjectives never occur in basic form, without any affix. Under these conditions, it is hard to determine
where the (p.197) accent lies at the most abstract level. In consideration of the fact that tonic verbs and adjectives
bear an accent on the penultimate mora at the final form (the sh shi or dictionary form), as in ta be ru,shira be ru,a
tsu i,ure shi i, it would be tempting to consider that the penultimate mora is the accent nucleus at the most abstract
level. However, there is actually no particular reason to regard the final form as basic, since it is neither shorter nor
more frequent than any other form. It is not morphologically simpler either because it includes a suffix, u/ ru in the
case of verbs, i in the case of adjectives.5
Kubozono (2008) makes the interesting claim that the same compound rule which assigns an accent on the last mora of
a C1 in nouns with a monomoraic C2 also applies to verbs and adjectives, whose final vowel or mora can be seen as a
suffix. However, with suffixes other than u, ru, and i which mark the final (or dictionary) form of verbs and adjectives,
this claim seems hard to follow because the accent patterns then appear to be very different in nature from what they
are in compound nouns, as Tables 7.4 and 7.5 show.
A categorization of verbal and adjectival affixes (such as auxiliary, particles, etc.) following that which has been adopted
for nouns, which would distinguish between recessive and dominant suffixes, also appears as inadequate and non
operational in the case of verbs and adjectives. Consider for example the volition suffix tai: its accent behaviour would
lead us to categorize it as dominant after tonic verbs but as recessive after atonic verbs. The hypothetical suffix eba is
even more problematic: it causes an accent to appear on the third to last mora in tonic verbs (shira be reba) and on
the fifth to last mora in tonic adjectives (u re shikereba), but on the penultimate mora of atonic verbs (ki re ba,koro be
Page 12 of 55
Accent
ba) and fourth to last mora of atonic adjectives (tsume ta kereba).
Under these conditions, it is more appropriate to consider that verbs and i adjectives are specified as tonic or atonic
(or +accent / −accent ) lexically, and that specific rules then come to determine which mora must receive the accent,
depending on the affixes. In this approach, there is thus, within a given verb or i adjective, no mora that must, in
essence, be considered as the accented mora at the underlying level. A number of linguists have subscribed to this
view, for instance McCawley (1968) and Poser (1984). The problem, however (but is it really a problem?), is that
nouns, on the one hand, and verbs and adjectives, on the other hand, are to be represented in a different fashion in
the grammar.6
(p.198)
Table 7.4. Accent of verbs

a. Atonic verbs
‘to wear’ ‘to fall’ ‘to arrange’ Accent pattern

Final form kiru° korobu° naraberu° Atonic
Suspensive form kite° koronde° narabete° Atonic
Negative form kinai° korobanai° narabenai° Atonic
Polite form kimasu korobimasu narabemasu Penultimate
Hypothetic form kireba korobeba narabereba Penultimate
Volitive form kitai° korobitai° narabetai° Atonic
b. Tonic verbs
‘to cut’ ‘to eat’ ‘to check’ Accent pattern

Final form ki ru taberu shiraberu Penultimate
Suspensive form ki tte ta bete shirabete Antepenultimate
Negative form kiranai tabenai shirabenai Antepenultimate
Polite form kirimasu tabemasu shirabemasu Penultimate
Hypothetic form ki reba tabereba shirabereba Antepenultimate
Volitive form kiritai tabetai shirabetai Penultimate
Two arguments in favour of this analysis can be offered. First, as Table 7.4 shows, some verbal affixes such as tai
(volitive), present seemingly contradictory behaviour. When attached to a tonic verb, tai receives the accent on the ta
mora. When added to an atonic verb, the whole verbal form is atonic. In other words, the suffix tai does not carry any
intrinsic accent of its own, since, if it did, this accent would inevitably surface with atonic bases, but it nevertheless
acquires an accent on the ta mora after tonic verb bases. This suggests that the accent of a tonic verb is not a priori
associated with any particular mora of the stem at the lexical level. In addition, verbs resulting from the same
etymological root always belong to the same prosodic type +accent or −accent . Thus pairs of transitive/intransitive
verbs always display the same accent pattern: tonic or atonic, for instance kimaru° (intr.)/kimeru° (tr.) ‘to decide’,
tsuzuku° (intr.)/tsuzukeru° (tr.) ‘to continue’, de ru (intr.)/da su (tr.) ‘to go out’, ka wa ku (intr.)/kawa ka su (tr.) ‘to
dry’, ki re ru (intr.)/ki ru (tr.) ‘to cut’, and many other examples. (p.199)
Table 7.5. Accent of -i adjectives
a. Atonic adjectives
‘thick’ ‘cold’ Accent pattern

Final form atsui° tsumetai° Atonic
Suspensive form atsukute tsumetakute Antepenultimate
Adverbial form atsuku° tsumetaku° Atonic
Polite form atsui desu tsumetai desu Pre antepenultimate
Hypothetic form atsukereba tsumetakereba Pre antepenultimate
b. Tonic adjectives
‘hot’ ‘happy’ Accent pattern

Final form atsui ureshii Penultimate
Page 13 of 55
Accent
Suspensive form a tsukute ureshikute Pre antepenultimate
Adverbial form a tsuku ureshiku Antepenultimate

Polite form atsui desu ureshii desu Pre antepenultimate
Hypothetic form a tsukereba ureshikereba Pre pre antepenultimate (−5)
Note that the accented mora is not necessarily the same one in the two members of a pair, as ka wa ku/kawa ka su
and ki re ru/ ki ru illustrate. The point is that no verb pair of this type associates a tonic verb to an atonic one. The
assumption that a verb root is specified as tonic or atonic, and that the accented mora is then determined according to
the suffixes added to the base, better accounts for this regular correspondence between members of a
transitive/intransitive pair. Consequently, which accent pattern should apply has then to be specified for each auxiliary
or suffix.
Tonic verbs and adjectives are always tonic while atonic ones may surface as tonic or atonic. Note also that denominal
verbs derived from a Sino Japanese noun containing the dummy verb suru° preserve the original accent pattern of
the noun: we thus have benkyou° 〉 benkyou suru° ‘to study’, an na i 〉 an na i suru ‘to guide’, ku rou 〉 ku rou suru ‘to
have a hard time’, and so on.
7.2.3 S no Japanese Lexemes Correspond ng To A S ng e Ch nese Character

As stated in section 1.6.2, Sino Japanese morphemes are one or two mora long, with their second mora (when there is
one) corresponding to /N/, /R/, /i/, /ku/, /tu/, /ki/, or /ti/.
A majority of Sino Japanese lexemes corresponding to a single character (i.e Sino Japanese monomorphemes which can
be used autonomously) are finally accented: for example e ‘image’, shi ‘death’, ji ‘letter’, bi ‘beauty’, ki ku (p.200)
‘chrysanthemum’, ji tsu ‘truth’, shi ki ‘ceremony’. One can consider that when the lexeme ends in /N/, /R/ or an
onsetless /i/, the underlying final accent is moved one position leftward (thus onto the initial mora) at the surface level
by application of the NADM principle. We thus have as in /teN/ 〉 te n ‘sky’, /ai/ 〉 a i ‘love’, /toR/ 〉 to u ‘political party’.7
The assumption that monographemic Sino Japanese lexemes ending in /N/, /R/, or moraic /i/ are finally accented at the
underlying level is supported by the following evidence. First, almost all the compound nouns which exhibit an atonic
accent pattern result from the combination of a C1 and a C2 of Yamato or Sino Japanese origin bearing final accent (8a),
or of Sino Japanese origin with a surface initial accent ending in a special segment (8b):
(8)
a. C2 = Yamato or Sino Japanese word with final accent
kodomo° + heya 〉 kodomo beya° ‘children’s room’

mi dori + iro 〉 midori iro° ‘green colour’
kao° + yaku 〉 kao yaku° ‘influential person’
b. C2 = Sino Japanese word with initial accent (at the surface)
nihon + hu u 〉 nihon huu° ‘Japanese manner’
denwa° + se n 〉 denwa sen° ‘telephone line’
If one considers that hu u, and se n in (8b) are actually oxytonic at the phonological level (i.e. /huR/, /seN/), these
morphemes can be considered as belonging to the same class as heya ‘room’, iro ‘colour’, or yaku ‘role’. They all end
with an accented mora underlyingly and cause the atonicity of a compound noun when they occur as C2.
Furthermore, Sino Japanese morphemes such as ka n ‘building’, ka i ‘sea’, jo u ‘castle’, shu u ‘province’, which yield
the appearance of an accent on the final mora of the first component when they undergo compounding behave exactly
like the Yamato words of the u ma type (9b) rather than those of the ne ko type (9c). As will be seen in section 7.3.1.,
the ne ko type is the type which preserves the accent in its original position in compounding. Compare, for example:
(9)
a.
ariake° + ka i 〉 ariake kai ‘sea of Ariake’

na goya + jou 〉 nagoya jou ‘Nagoya castle’
Page 14 of 55
Accent
b.
abare° + uma 〉 abare uma ‘restive horse’
c.
perusha + neko 〉 perusha neko ‘Persian cat’
The compounds in (9a) all have an accent on the final mora of the C1, just like the type displayed in (9b). This
constitutes an additional argument for considering (p.201) that ka n, ka i, jo u, shu u, etc. are underlyingly just like
u ma, with an inherent accent on their final mora: /kaN/, /kai/, /zyoR/, /syuR/.
Lastly, recall that we also saw in 6.2.4 that some words ending with a defective mora and bearing an accent on their
penultimate mora are de accented when followed by the determinative particle no, in exactly the same way as oxytonic
words are. This constitutes evidence that it is the final mora, rather than the penultimate mora, that carries the
underlying accent in this type of word.
An interesting consequence of this analysis is that a majority of Sino Japanese morphemes can now be regarded as
finally accented. The final accent pattern can therefore be considered to be the default accent pattern for the
monomorphemes belonging to this stratum.
The rest of the Sino Japanese lexemes are generally atonic: i° ‘stomach’, cha° ‘tea’, shi° ‘poem’, taku° ‘residence’,
toku° ‘favour’, shitsu° ‘quality’, teki° ‘enemy’, etc., or sometimes with initial accent: se ki ‘place, seat’, ba tsu
‘punishment’. Once again, we observe that words of greater frequency and familiar use tend to become atonic.
7.2.4 Western Loans

Whereas the accent of simplex words is lexically determined and cannot, in theory, be predicted by rules, it is to a
certain extent predictable in the case of Western loans, as demonstrated by Kubozono (1996, 2006b and a number of
other papers). Words belonging to this lexical class frequently have an accent on the antepenultimate mora (10a) if it is
a regular mora, i.e. a CV mora, or on the pre antepenultimate mora if the antepenultimate mora is deficient (10b), by
virtue of the NADM principle. If the word is short, the accent is placed on the first mora (10c).
(10) Accent of Western loans
a. Antepenultimate accent
chokoreeto {chocolate}
piramiddo {pyramid}
ri zumu {rhythm}
oosutoraria {Australia}
b. Pre antepenultimate accent (antepenultimate mora = deficient mora)
nabigeetaa {navigator}
washinton {Washington}
ahuganisutan {Afghanistan}
risaikuru {recycle}
u uman {woman}
(p.202)
c. Penultimate accent (short words)
ka a {car}
gya ru {gal}
However, quadrimoraic loans as well as integrated loans of high frequency are often unaccented, for instance
amerika° {America}, huransu° {France}, rittoru° {litre}, botan° {botão}, oopuningu° {opening}. This is especially
true (at 90%) of four mora long words if they end in a sequence of two regular moras (amerika°), whereas the ratio
goes down to about 30% if their final mora is deficient (ending with a special mora or an epenthetic vowel): serekuto
Page 15 of 55
Accent
{select}, pi renee {the Pyrenées} (Kubozono, 2006b).
According to Kubozono (2006b) loanwords do not crucially differ from native words in accent structure and
preference. What appears to be an accentual difference between the two types of words stems largely from their
difference in phonological structure and in the nature of vowels, since loanwords contain plenty of deficient moras
(including moras containing an epenthetic vowel).
Lastly, a few loanwords have retained the accent in the location in which it occurs in the source language, as me
ntenansu {maintenance} or kon sa rutanto {consultant}, except in cases where the original accent is final. Indeed,
there does not exist in Japanese any loanword finally accented at the surface level.
7.2.5 A Constra nt based Account of the Accent of Western Loans

The accent of Standard Japanese is lexical, but, as we have seen in the preceding sections, the emergence of a default
accent is observable in a number of word classes belonging to different etymological strata, including native Japanese
or Sino Japanese lexemes. However, the default accent pattern is most widely productive in Western loanwords.
In this section, we offer an analysis of the Tôkyô Japanese default accent pattern within the framework of Optimality
Theory, based on the ideas developed in this paper, and building on the study of the accentuation of foreign toponyms
by Kubozono (1996). Foreign toponyms constitute an exemplary case of a lexical class in which the default accent is
implemented. These examples also offer an ideal theoretical testground because their study is based on a statistical
approach, on the one hand, and because, on the other hand, they relate to a representative, well circumscribed
corpus of foreign place names where accent is attributed by means of a default process. In addition, an examination of
the accentual patterns of foreign toponyms is particularly interesting because these data have been taken as evidence
for the relevance of the heavy syllable (p.203) in Japanese by several scholars (Katayama, 1995; Kubozono, 1996,
2006b; Tanaka, 2008).
Therefore one of the purposes of the present analysis is also to demonstrate that one can achieve a satisfactory
treatment of this classical issue in Japanese phonology, which has been taken as evidence for the necessity of the
syllable (Kubozono, 1996), without the syllable. As will be shown, only the mora and the feet are necessary to account
for the data. The analysis which is proposed here is no less simple and natural than syllable based analyses. In addition,
and more crucially, we will see that it accounts for a greater number of facts, since it allows a regular treatment of the
accent pattern of words such as pi renee ‘the Pyrenees’ and se negaru ‘Senegal’.
The data
Kubozono (1996) observes that in foreign toponyms longer than two moras, default accent is generally placed on the
antepenultimate mora (11a). In cases where the antepenultimate mora contains /N/, /Q/, or /R/, the accent shifts one
position leftward and is placed on the pre antepenultimate mora, all of this being in accordance with the general
principles of Japanese accentuation (11b). If the word is two moras long, the accent falls on the initial mora (11c). This
accent rule, it is important to note, is actually that which generally applies to common nouns of Western origin, as
established by Tanomura (1999). Note that, like Kubozono, we shall only consider here the case of accented nouns,
leaving aside that of unaccented ones.
(11) Default accent
a. On antepenultimate mora
meraneshia ‘Melanesia’
isutanbuuru ‘Istanbul’
waiomingu ‘Wyoming’
mo nako ‘Monaco’
de kan ‘Deccan’
b. On pre antepenultimate mora
ro ndon ‘London’
washinton ‘Washington’
kentakkii ‘Kentucky’
ejinbara ‘Edinburgh’
Page 16 of 55
Accent
c. On penultimate mora (bimoraic words)
pa ri ‘Paris’
do n ‘the Don’
(p.204)
However, and this is what makes this set of data particularly interesting, a number of words do not conform to these
principles. The examples in (12), which all end in a deficient mora, receive pre antepenultimate accent instead of the
expected antepenultimate accent.
(12) Pre antepenultimate accent
pi renee *pirenee ‘the Pyrenees’

te heran *teheran ‘Teheran’
se negaru *senegaru ‘Senegal’ (the final u is epenthetic)
Kubozono (1996:74), according to whom the syllable plays an active role in the accentuation of loanwords, proposes
the following generalization:
If the word ends with a heavy syllable (or a light syllable containing an epenthetic /u/) accent is placed on the
initial syllable, whether it is light or heavy … . If the word ends with a light syllable and the penultimate syllable is
heavy, accent is placed on the heavy syllable … . Note that in this case, it does not matter whether the final light
syllable contains an epenthetic vowel /u/ or not.
Although this formulation seems perfectly correct on the descriptive level, provided that one accepts the idea that a
distinction between light and heavy syllables is relevant in Japanese phonology, it has no explanatory power. Why
should the presence of a heavy syllable word finally have such an accentual effect on the pre antepenultimate syllable?
I argue that a better, sounder explanation for this data is possible and that this account can be made with no direct or
indirect reference to the syllable.
I will now propose a formal treatment of this data within the framework of Optimality Theory. This treatment will take as
a basis the analysis of default accent assignment in Japanese adaptations of French words proposed by Shinohara
(2000). (For other OT treatments, see Kubozono, 1996 or Tanaka, 2008.)
OT constraints as proposed by Shinohara (2000)
The French words studied by Shinohara (2000) obey accentuation principles that are identical to those displayed by
foreign toponyms (except for a few exceptions), hence the relevance of her work for the present study.
The constraints proposed by Shinohara (2000) in order to account for the default accent pattern are the following:
(p.205) (13) Constraints (adapted from Shinohara, 2000)
HEAD LEFT: Feet are trochaic.

ALIGN (F, R, PRWD, R): Align the right edge of every foot with the right edge of a prosodic word (PrWd).
NONFINALITY: No prosodic head (accented foot or accented syllable) of PrWd is final in PrWd.
FOOTBINARITY: Feet are binary at some level of analysis.
PARSE SYLLABLE: Parse every syllable into a foot.
An additional constraint prohibits an epenthetic vowel to be the head of a foot:
(14)
*v (epenthetic vowel) A non prominent nucleus cannot be the head of a foot
Due to space limitations, I will not take up here all the details of Shinohara’s (2000) analysis, which is rather complex
Page 17 of 55
Accent
and implements two concurrent hierarchies of the above constraints. Note that her analysis makes explicit reference
to the syllable, through the constraints NONFINALITY and PARSE SYLLABLE. However a major shortcoming of
Shinohara’s analysis is that it cannot provide any explanation for the initial accent of a word like se negaru.
In the pages that follow, I shall propose another line of analysis which preserves some of the proposals formulated by
Shinohara, which have the advantage of being very standard in the treatment of accent within OT. My intention is thus
to use Shinohara’s proposal as a basis for a syllable free analysis of default accent assignment in Japanese, and to
propose a simple and natural explanation of the accent of Western loan words including senegaru.
A syllable free account
The general ideas that underlie my analysis are the following: feet are preferably binary and trochaic and, within a
word, the head foot is located on the rightmost edge, but it must not be final. In addition, no foot can start with a
deficient mora. The interaction between these principles gives an account of the position of accent in foreign place
names, but also, more generally, in the majority of Western loans.
In the lines which follow, the symbol µ corresponds to the mora, whatever its structure, while the symbol m
corresponds to a weak or deficient mora.
(p.206) A first group of relevant constraints is the following:
(15)
* F m : No foot starts with a deficient mora.

NONFINALITY (FOOT) (= NONF(π)): The head foot must not be final.
PARSEMORA (= PARSEM): Moras are parsed into a foot.
ALIGNRIGHT (=ALIGNR): The accented mora is rightmost in the prosodic word.
In the category of deficient moras falling under the scope of the *F m constraint, we find the elements located at the
bottom of the hierarchy provided in Chapter 6, whose phonological structure is of the type C• or •V, namely, the first
part of a geminate obstruent /Q/, the mora nasal /N/, the second part of a long vowel /R/, onsetless moras made up of
the vowels /i/ and /u/, and finally, some8 moras containing an epenthetic vowel that behave as if their nuclei were
empty.
Note that the constraint NONFINALITY(FOOT) concerns the head foot exclusively. PARSEMORAS stipulates that no
mora is left outside a foot. The action of this constraint guarantees an exhaustive organization of the feet. Finally,
ALIGNRIGHT is different from the constraint ALIGN(F, R, PRWD, R) as proposed by Shinohara (2000). ALIGNRIGHT
stipulates that the accent be as close as possible to the end of the prosodic word. Since, in theory, no Western loan
longer than three moras has its accent located on one of the last two moras, everything points to ALIGNRIGHT being
located below NONFINALITY(FOOT), which prohibits the accented foot from being final. Note that ALIGNRIGHT is a
gradient constraint. For this reason, each mora located between the accented mora and the rightmost edge of the
word counts as one violation.
A second group of constraints, FOOT=BINARY and FOOT=TROCHEE, relates to foot structure:
(16)
FOOT=BINARY (F=BIN): Feet are binary under moraic analysis.

FOOT=TROCHEE (F=T): The head foot is trochaic.
FOOT=BINARY ensures that feet are made up of two moras, no more and no less, while FOOT=TROCHEE
guarantees that the head foot, i.e. the accented foot, is trochaic. The case of the word pa ri {Paris}, which will be
examined just below, will show that it is necessary to place PARSEMORAS above FOOT=BINARY in the hierarchy. This
is because otherwise, the unaccented form *pari° would turn out to be the optimal output, a totally undesirable
result.
(p.207) With regard to FOOT=BINARY and FOOT=TROCHEE, we know that these two constraints are both placed
above ALIGNRIGHT because, as we will see, they are the very two constraints which play a role in the choice of pi
Page 18 of 55
Accent
renee rather than *pirenee as the optimal output of /pireneR/, or of se negaru rather than *senegaru for /senegaru/
(see tableaux (18) and (19)). However, we have no evidence to determine the relative subranking of FOOT=BINARY
and FOOT=TROCHEE.
There is no decisive argument to determine the position of NONFINALITY(FOOT) in the general constraint ranking
either, except that it dominates ALIGNRIGHT. I thus choose to place NONFINALITY(FOOT) above all the other
constraints, with *F m at the top of the hierarchy. However, this choice has no direct effect on the analyses, and
NONFINALITY(FOOT) could just as easily be placed anywhere else above ALIGNRIGHT.
These considerations lead to the following hierarchy.
(17) *F m, NONF(π) 〉〉 PARSEMORAS 〉〉 F=T, F=BIN 〉〉 ALIGNRIGHT
With the exception of *F m, the constraints which have been introduced here, namely FOOT=TROCHEE,
NONFINALITY(FOOT), FOOT=BINARY, and PARSEMORAS (an adaptation of PARSESYLLABLES), are identical to
standard constraints found in the OT analyses of the Japanese accent based on the idea that the syllable plays a crucial
role in the phonology of the language (see in particular Kubozono, 1997, Shinohara, 2000, 2002, Shirose et al., 1997,
and others). The main innovation in the treatment proposed here thus lies in the introduction of the *F m constraint.
Even though this constraint does not play a decision role in every example that will be examined, resorting to *F m
brings simplification to the analysis and avoids using ambivalent constraints referring indifferently, and in an ad hoc
manner, to the foot and/or syllable (Shinohara, 2000) or to the syllable and/or mora (Kubozono, 1997), while allowing
us to provide a principled account of the general cases of default accent assignment.
In the following tableaux, the candidates containing only unfooted moras will not be examined, since such candidates
would remain accentless (and, as stated at the beginning of the section, the issue of atonicity will not be dealt with
here). However, it is not without significance to observe that the candidates in question would, in any event, never be
in a position to emerge as optimal candidates.
The tableaux
Let us now proceed to the examination of the tableaux. We will start with the two examples which are of particular
interest to us: those comprising a deficient mora in final position, like pi renee ‘the Pyrenees’ and se negaru ‘Senegal’,
and which appear irregular within the traditional framework because they carry an accent on (p.208) the pre
antepenultimate mora rather than on the antepenultimate. As we will see, such words turn out to be perfectly regular
if one adopts the constraints and the hierarchy proposed here.
(18) pi renee ‘the Pyrenees’
Candidate a., (pire)(nee), with pre antepenultimate accent, wins over the others because it is the only candidate that
satisfies all the constraints except the lowest ranked ALIGNRIGHT. Yet, let us consider candidates d., e., g., and h.
which present an antepenultimate accent, as expected in the traditional approach. The form in d. (pire)(nee) contains a
non trochaic foot, which is enough to eliminate it with respect to a. The form in e. (pi)(rene)(e) contains a foot beginning
with a deficient mora e (/R/), making it the worst candidate of the series (the foot in question being, in addition, non
binary). Candidate g., pi(rene)e is excluded because, despite its trochaic foot, and the fact that no foot starting with a
deficient mora is constructed, it has two unbounded moras, in violation of PARSEMORAS, which make it a candidate
worse than (pire)(nee). Finally, h. (pi)(rene)e contains a non binary foot, in addition to one unparsed mora.
Candidate f. (pire)nee with initial accent is another serious competitor to a. However, the presence of two moras left
unparsed is fatal in comparison with a (pire)(nee).
Page 19 of 55
Accent
The same type of reasoning applies to candidates d., e., f., g. and h. of the input /senegaru/ whose final u is epenthetic.
Recall that the mora ru belongs to the category of deficient moras just like the final vowel length in /pireneR/.
(p.209) (19) se negaru (with final epenthetic /u/) ‘Senegal’
Let us now consider the general case, with antepenultimate accent, of words which do not contain a deficient mora in
their final or penultimate position, such as baruserona ‘Barcelona’.
(20) baruserona ‘Barcelona’
Candidates b., c., and d. are the least optimal because they violate *F m. The form e. is excluded because it violates
FOOT=TROCHAIC. F, g., and h. leave one or several unbounded moras. Under these conditions, the victorious
candidate can only be candidate a. (baru)(sero)(na). Note that f. would be optimal if the order between
FOOT=BINARY and PARSEMORAS were inverted (this, as already stated, remains a possible option), but this change
would not affect in any way the location of the accent in the output.
(p.210) The word burunji /buruNzi/ ‘Burundi’ also constitutes a case traditionally recognized as regular. It is also in
our analysis. Contrary to baru se rona, it contains a special mora in penultimate position, and it is precisely for this
reason that it deserves special interest.
(21) bu ru nji ‘Burundi’
Candidate a. (bu)(run)(ji) wins over all the other candidates because it is the only form which respects the four higher
ranked constraints. The candidates h. bu(run)ji and g. (bu)(run)ji also display an antepenultimate accent, but the
former also has two non binary feet, while the latter has a non binary foot and an unparsed mora. They are thus
Page 20 of 55
Accent
disqualified in comparison with a.
Let us see now how the form ro ndon ‘London’ is selected. It is interesting to note that words such as ha aremu
‘Harlem’ or a ndesu ‘the Andes’, which end with an epenthetic vowel and contain a special segment in second position,
behave exactly as ro ndon since they are indeed accented on the initial.9 This confirms the correctness of positing a
‘deficient mora’ category broader than the traditional ‘special mora’ category.
(p.211) (22) ro ndon ‘London’
The candidate a. (ron)(don) respects all the constraints higher than ALIGNRIGHT, and thus wins over all the other
candidates.
Let us now consider three mora words, which are all accented on the antepenultimate mora. Two subtypes can be
distinguished: those that do not contain any deficient mora, like mo nako ‘Monaco’, and those that do, either in final
position, like de kan ‘Deccan’, or in medial position, like me kka ‘Mecca’. All these examples can be accounted for in a
straightforward manner.
(23) mo nako ‘Monaco’
Candidate a. (mona)(ko) emerges as the optimal output.
The next example, de kan, contains a deficient mora in final position, just like pi renee or se negaru, which makes it
particularly worthy of interest.
(p.212) (24) de kan ‘Deccan’
Tableau (24) shows that candidate a. (de)(kan) is the best candidate.
In the case of me kka, the result of the tableau is also as expected. Candidate a. (mek)(ka) emerges as victorious.
(25) me kka ‘Mecca’
Page 21 of 55
Accent
Let us close with the examination of the bimoraic word pa ri, whose accent is initial. According to the hierarchy that I
propose, candidate c. (pa)(ri) is optimal. Here, once more, the other possible hierarchies would produce an optimal
output with initial accent. Of all the alternatives considered here, the candidate with initial accent emerges as the
winner, a desirable result.
(26) pa ri ‘Paris’
(p.213) However, notice that the unaccented, prosodically unstructured form pari (pari°, which does not figure in
the tableau) could appear here as a better candidate than (pa)(ri) if PARSEMORAS had been placed below
FOOT=BINARY, as tableaux (27) and (28) show. If, as in (27), PARSEMORAS is ranked below FOOT=BINARY,
candidate a. pari° becomes the best candidate, a totally undesirable result. This is what justifies, in the present
analysis, the ranking of PARSEMORAS above the constraints FOOT=TROCHEE and FOOT=BINARY.
(27)
(28)
The above analysis accounts in a simple and natural way for the assignment of the accent in foreign toponyms, including
quadrimoraic forms with initial accent like pi renee, te heran and se negaru, which resist existing analyses, with no
need to assume any opposition between heavy and light syllables.
Although all the examples reviewed above are place names of foreign origin, the present analysis applies more
generally to common nouns borrowed from Western languages, since they follow the same accentuation principles.10
(p.214) 7.2.6 Other Types of S mp ex Words

The remainder of simplex words is divided into a multitude of specific subcases, for which general principles of accent
assignment can sometimes be posited. I shall rapidly mention here some of these categories. More detailed
descriptions can be found in the references already cited.
Adjectives ending in na form a heterogeneous class from the point of view of accentuation. The pattern of the source
word (in most cases a noun) from which the na adjective is derived generally applies to the adjective. Thus we have o
oki 〉 o oki na (Yamato) ‘large’, shi npuru 〉 shi npuru na (Western) {simple}, da me 〉 da me na (mixed Sino Japanese
+ Yamato) ‘vain, no good’. However, na adjectives whose base ends in ka (originally, an adjectival suffix) are
automatically accented on the antepenultimate mora of the base: shi zuka na ‘quiet’, ko ma yaka na ‘detailed’.
Page 22 of 55
Accent
Interrogative words are all initially accented: na ni ‘what’, do ko ‘where’, i kutsu ‘how many’, do chira ‘in which
direction’, etc., but the corresponding deictics are generally atonic: koko° ‘here’, are° ‘that’, sono° ‘this’, asoko° ‘over
there’, sonna° ‘such’, etc.
Adverbs show a tendency to receive a penultimate accent when they are bimoraic, and an antepenultimate accent
when trimoraic or longer: ma da ‘not yet’, mu shiro ‘rather’, mochiron ‘of course’, shi ba raku ‘for one moment’, yu
rai ‘in the beginning’ (but note that as a noun yurai° ‘the origin’ is atonic). A fair number of four mora adverbs are
atonic: ainiku° ‘unfortunately’, hanahada° ‘extremely’, kekkyoku° ‘finally’.
Simplex mimetic words (onomatopoeias and ideophones) bear a final or penultimate accent when they end with ri (to),
and a penultimate accent when they end in n (to) or t (to): korori( to) / korori( to) ‘rolling’, katan( to) ‘with a clatter’,
gyorot( to) ‘glaring goggle eyed’.
Two and three mora truncated words are regularly accented on the initial: sha mi (〈 shamisen°) ‘shamisen’, te ro (〈
tero ri zumu) ‘terrorism’, a nime (〈 ani me eshon) ‘cartoon’. This rule also applies to hypocoristic truncations: ma sa,
ma sa kun 〈 masaki (male first name), na o, na o chan 〈 naomi° (female first name). Four mora long abbreviations are
generally atonic: rihabiri° (〈 rihabiri te eshon) ‘rehabilitation’, masukomi° (〈 masukomyunikeeshon) ‘mass media’.
It is also interesting to observe that modification of the grammatical category of a word can materialize through
modification of its accent pattern. For instance, nouns with temporal or quantitative reference accented on the final
mora such as (p.215) ashita ‘tomorrow’, hutatsu ‘two’ are unaccented when used as adverbs: ashita°,hutatsu°
(Kindaichi and Akinaga, 2001; Uwano, 2003). Semantic factors also come into play: thus the suffix sei is pre accenting
with the meaning ‘pupil, student’ (ichinen sei ‘first year pupil’, yuutou sei ‘brilliant pupil’), but atonic in the
botanical sense (ichinen sei° ‘annual plant’, tanen sei° ‘perennial plant’, Satô H., 1989).
7.3 Accent of Compound Words

The accentuation of compound words constitutes one of the thorniest and most interesting issues within the domain of
Japanese accentology, both from the descriptive and theoretical point of view, and as such it has been the focus of a
number of descriptive or analytical studies. Whereas the accent pattern of a compound, whatever it is, seems very
generally predictable from that of its components, except when the second constituent is short, the factors that
condition compound accentuation are numerous and varied, and their interactions extremely difficult to capture.
Contrary to what occurs in a language like English, where the position of the primary accent in a compound can be
analysed as the projection of the primary accent of one of its members (chris tmas + ca ke 〉 chris tmas cake, fea
ther + pi llow 〉 feather pi llow), things work very differently in Tôkyô Japanese. In this language, the accentual
algorithm of the compound is entirely recomputed, so that accent may fall on a mora which was not accented in the
simplex form of the lexeme to which it belongs. For example, the combination of yu ki ‘snow’ + oto ko ‘man’ gives yuki
o toko ‘snowman, yeti’, with an accent on o, and deletion of the original accents on ki and ko (neither *yu ki otoko nor
*yuki oto ko is attested). In Japanese, an analysis which would directly project the accent of the head member of the
compound cannot be conducted, except, as we shall see, in the two following notable cases: when the length of the
second component is equal or superior to five moras, or when the compound has a coordinative meaning (dvandva
compounds).
The main parameters which may determine the accent of a Japanese compound word are the following:
The nature of the morphological and syntactic relation between the two constituents.
The size of the compound.
The size of each constituent of the compound.
The grammatical category of the constituents (noun, verb, etc.).
The intrinsic accent of each constituent.
(p.216) The lexical stratum of the constituents (Yamato, Sino Japanese, Western).
The degree of sonority of the vowels in the head foot.
The degree of lexicalization of the compound also constitutes a major parameter. For example, it is very unlikely that
ancient, strongly lexicalized compounds whose compositional nature is rather opaque, such as kaya° ‘mosquito net’, i
do ‘a well’, tamago (or tamago°) ‘egg’, amado ‘shutter’, namae° ‘name’, or ha nabi ‘fireworks’, are to be treated in
the same way as formations such as kokuritsu daigaku ‘national university’, yuki otoko ‘snowman, yeti’ or orenji iro°
‘the colour orange’, whose compositional nature is much more transparent. Unfortunately, most studies do not
provide a clear definition of what a ‘compound word’ is, and this inaccuracy accounts for much of the differences, and,
even, the apparent contradictions, which one observes in the analyses of various scholars. Thus Kindaichi and Akinaga
(2001), which adopts a strictly etymological criterion to define what a compound is and thus treats kaya°, i
Page 23 of 55
Accent
do,tamago,amado,namae°, and ha nabi as compounds states a general rule that is exactly the opposite to that
proposed by Tanaka and Kubozono (1999). According to Kindaichi and Akinaga (2001, page 13 of the appendix), the
accent of such compounds is determined by and large by its first member, whereas Tanaka and Kubozono (1999)
considers that only the second element is decisive for the attribution of the accent in compounds.
In the present section, the focus will be on transparent compounds whose global meaning is easily deducible from that
of their components, which both exist independently in the modern language. Accordingly, kaya°, i do, and so on are
regarded as non compound words.
We will successively review the accent of modifier head structured nominal compounds, of dvandva compounds
(equipollents, or coordinative compounds) and mimetic compounds. We will then examine the case of fixed Sino
Japanese compounds made up of two Chinese characters, as well as that of compound verbs. We will finally have a look
at numeral compounds, with or without a specifier. For additional descriptive or theoretical treatments of compound
accent, the reader can refer to McCawley (1968), Higurashi (1983), Poser (1990), Kubozono (1993b, 2006b, 2008),
Uwano (1997, 2003), Satô H. (1989), Shinohara (2002), Tanaka (2008), and to the appendices of the two accent
dictionaries (NHK, 1998; Kindaichi and Akinaga, 2001). Most of the examples provided here are drawn from these
sources.
7.3.1 Compound Nouns W th A [mod f er head] Structure Conta n ng On y One Accent Nuc eus
In compound nouns having a modifier head structure, for example yama inu° (‘mountain’ + ‘dog’) ‘wild dog, coyote’,
or den wa ki (‘telephone’ + ‘machine’) (p.217) ‘telephone’, the accent pattern depends principally, in theory (in a
manner that will be examined below) on the size and original accent pattern of the second constituent (henceforth C2),
which is syntactically the most important one. However, it seems that the accent pattern of the first constituent (C1)
also plays a role in certain cases. For instance, nearly all quadrimoraic compounds containing hito° ‘human being’ as
their C1 are atonic,11 whatever the accent pattern of the C2, as the examples below illustrate:
(29)
hito° + naka (‘middle’) 〉 hito naka° ‘in the company of people’

hito° + kaki (‘fence’) 〉 hito gaki° ‘a row of people’
hito° + nami° (‘ordinary’) 〉 hito nami° ‘average’
As we will see later, the accent of numeral compounds is also largely determined by the accent pattern of the C1.
The general rules presented below apply to modifier head compound nouns whose components are Yamato,
Western, or Sino Japanese, with the exception of fixed Sino Japanese compounds made up of two characters such as
benkyou° ‘study’, shinju° ‘pearl’, or a kumu ‘nightmare’, which will be examined in section 7.3.6.
One has to distinguish three main categories, according to whether C2 is short (one or two moras), long (three or four
moras), or extra long (five moras and longer). The shorter the word, the more irregular its accent pattern in
compounding is likely to be.
(i) Short C2
If C2 is one or two mora long, we encounter three types of patterns:
the accent of the compound falls on the last mora of the first member (30). If it is a deficient mora,
consisting of a special segment or single vowel, the accent might move one position leftward, according to
the NADM principle. This is the most productive pattern according to Kubozono (1995b, 2008).
the accent of the compound falls on the initial mora of the second member (31). C2’s affected by this rule
amount to a small number. All of them are initially accented, and are mainly of Yamato or Western origin,
with only very few Sino Japanese words. They never end with a deficient mora.
the compound is atonic (32). Almost all C2’s that fall under this rule have a final accent in their
independent form, and are of Yamato or Sino Japanese origin.
Note that a given lexeme can belong to two different categories. For instance kuni° ( guni) is both pre accenting and
de accenting.
(p.218) (30) Pre accenting C2
Page 24 of 55
Accent
ka buto + mushi° 〉 kabuto mushi ‘beetle’

(‘helmet’ + ‘insect’)
abare° + uma 〉 abare uma ‘restive horse’
(‘unruly’ + ‘horse’)
ni ngyo + hi me 〉 ningyo hime ‘Little Mermaid’
(‘mermaid’ + ‘princess’)
hukuoka + shi 〉 hukuoka shi ‘city of Fukuoka’
(Fukuoka + ‘city’)
denwa° + ki 〉 denwa ki ‘telephone’
(‘telephone’ + ‘machine’)
Here is a list of some of the most frequent pre accenting lexemes (when the lexeme is not used independently, no
accentual information is given). They are mostly finally accented (see also McCawley, 1968: appendix I, NHK, 1998, for
lists):
machi ‘town’, kawa (/ gawa) ‘river’, dake ‘mount’, u ri ‘melon’, su ( zu) ‘vinegar’, sushi ( zushi) ‘sushi’,
uta ‘song’, kami ( gami) ‘paper’, mochi° ‘rice cake’, kuni° ( guni) ‘country’, toshi ( doshi) ‘year’, hi me
‘princess’, uma ‘horse’, jin ‘person’ (except after certain nouns ending with a mora nasal as in nihon jin), in
‘member’, in ‘institute’, e ki ‘station’, en ‘park’, on ‘sound’, kai ‘association’, kai ‘world’, gai ‘street’,
gaku ‘study’, kan ‘feeling’, kan ‘building’, ki ‘chronicle’, ki ‘period’, ki ‘récipient’, ki ‘machine’, gou
‘number’, koku ‘country’, shi ‘Mr’, shiki ‘style’, shitsu ‘room’, sha ‘society’, shuu ‘collection’, shou
‘ministry’, shoku ‘colour’, sei ‘pupil’, zoku ‘tribe’, hi ‘expenses’, ki ‘machine’, byou ‘disease’, bu
‘part’, ryoku ‘power’.
(31) C2 maintaining the original initial accent in the compound
uroko° / uroko + kumo 〉 uroko gu mo ‘a cirrocumulus’

(‘scale’ + ‘cloud’)
garasu° + mado 〉 garasu mado ‘glass window’
(‘glass’ + ‘window’)
ka rasu + mugi 〉 karasu mugi ‘oats’
(‘crow’ + ‘wheat’)
Most of the nouns which maintain the accent in its original location when used as a C2 are initially accented. The most
frequent are: ku mo ( gumo) ‘cloud’, ne ko ‘cat’, a me ‘rain’, shi ru ( jiru) ‘juice’, hu ne ( bune) ‘boat’, ka sa
( gasa) ‘umbrella’, o bi ‘belt’, ko e ( goe) ‘voice’, mu ko ‘son in law’, tsu ru ( zuru) ‘crane’, mu gi ‘wheat’,
ma e ‘front’. Some nouns of (p.219) Western origin behave in the same manner, for example ga su ガス ‘gaz’, pi za ピザ
‘pizza’, as well as the Sino Japanese derivative morpheme shu gi ‘ ism ’ (shakai shugi ‘socialism’, keishiki shugi
‘formalism’).
(32) De accenting C2
kodomo° + heya 〉 kodomo beya° ‘children’s room’

(‘child’ + ‘room’)
orenji + iro 〉 orenji iro° ‘colour orange’
(‘orange’ + ‘colour’)
otoko + te 〉 otoko de° ‘man’s help’
(‘man’ + ‘hand’)
nihon + huu (/huR/)12 〉 nihon huu° ‘Japanese way’
(‘Japan’ + ‘manner’)
Page 25 of 55
Accent
Among the most frequent de accenting lexemes, we find iro ‘colour’, mura ‘village’, kumi ( gumi) ‘group’, inu
‘dog’, kao° ( gao) ‘face’, tera ( dera) ‘temple’, tama ( dama) ‘ball’, kami ‘hair’, kuni° ( guni) ‘country’, kata (
gata) ‘pattern’, gawa° ‘side’, te ‘hand’, ba° ‘place’, me ‘eye’, ka ‘section’, ka ‘section’, tou ‘party’, ka
‘agent’, ka ‘ ation’, kyou ‘religion’, go ‘language’, jou ‘place’, tai ‘body’, chuu ‘while’, hu ‘woman’,
hu u /huR/ ‘manner’, you ‘usage’, ryuu ‘current’. They are mostly final accented and a small number of them
are atonic.
A few words yield final accentuation of the compound, such as kaze° ‘wind’, mo no ‘thing’, or mo no ‘person’. We thus
have: minami ka ze ‘southern wind’, hitori mo no ‘single person’. However, there always exists a variant which follows
one of the regular patterns: minami kaze°, hitori mono°.
Finally, it is interesting to observe that four mora compounds resulting from the combination of two bimoraic Yamato
nouns are in their majority atonic (Kubozono and Fujiura, 2004). This is especially the case when the head noun has a
final lexical accent such as u ma ‘horse’ or i ro ‘colour’.
Kubozono (1997, 2008) proposes an analysis which assumes that only de accenting short morphemes (such as those
shown in 32 above) are lexically marked, that is, specified in the lexicon with respect to their accent behaviour in
compounds. So initial accenting and pre accenting patterns are predictable by rule on the basis of their own lexical
accent pattern. Kubozono thus posits the following two basic principles as a generalization of the pre accenting and
initial accenting patterns (the romanization and terminology have been adapted):
(p.220) (33) Kubozono’s generalization
a. Keep the accent of C2 as compound accent except when it is on the very final mora:
ne ko ‘cat’: perusha neko ‘Persian cat’
b. Otherwise, put a compound accent on the final mora of C1:
inu ‘dog’: akita inu ‘Akita dog’
mushi° ‘bug’: kabuto mushi ‘beetle’
(ii) Long C2
When the C2 is three or four mora long, the pre accenting and de accenting types such as those seen above in (30)
and (32) do not exist. The general principle is that the compound will receive an accent on the first mora of the C2,
except when C2 has an accent that is neither initial nor final in isolation, in which case the accent will generally be
preserved in the same position within the compound, particularly (according to Shinohara, 2002) if it is quadrimoraic
with an antepenultimate accent.
However, when C2 has a penultimate accent in isolation, the situation is rather confusing. Tanaka and Kubozono (1999)
observe that in such a case, the compound may be accented on the initial mora of C2 (34e).
(34)
a. Long C2 with an initial accent
kogata° + kamera 〉 kogata kamera ‘small camera’

(‘small size’ + ‘camera’)
shimaguni + konjou 〉 shimaguni konjou ‘insularism’
(‘island country’ + ‘disposition’)
b. Atonic long C2
yuki + daruma° 〉 yuki daruma ‘snowman’

(‘snow’ + ‘Dharma (doll)’)
kuchi° + yakusoku° 〉 kuchi yakusoku ‘verbal promise’
(‘mouth’ + ‘promise’)
c. Long C2 with a final accent
yuki + otoko 〉 yuki otoko ‘yeti’

(‘snow’ + ‘man’)
nuka +yorokobi 〉 nuka yorokobi ‘vain joy’
Page 26 of 55
Accent
(‘rice bran’ + ‘joy’)
d. Quadrimoraic C2 with an antepenultimate accent
kami + hikouki 〉 kami hikouki ‘paper plane’

(‘paper’ + ‘plane’)
wa ka + murasaki 〉 waka murasaki ‘light purple’
(‘young’ + ‘purple’)
e njin + sutoppu 〉 enjin sutoppu ‘stalling of a motor’
({engine} + {stop})
(p.221)
e. Long C2 with a penultimate accent
hidari° + uchiwa(‘left’ + ‘fan’) 〉 hidari uchiwa/ hidari uchiwa ‘to live in ease’
yude° + tamago(‘boil’ + ‘egg’) 〉 yude tamago/ yude tamago ‘boiled egg’
onna + kokoro / kokoro 〉 onna gokoro ‘woman’s heart’
(‘woman’ + ‘heart’)
denki + nokogiri / nokogiri13 〉 denki nokogiri ‘power saw’
(‘electricity’ + saw’)
The examples presented in (34e) are actually very debatable. This is because the majority of compounds cited in the
literature in order to exemplify the cases at hand (for example Tanaka, 2001) concern the following categories: (i) their
C2 is a noun with two possible accent patterns such as noko gi ri / nokogi ri ‘saw’, kami so ri / kamiso ri ‘razor’, name
ku ji / namekuji° ‘slug’, hoo be ni / hoobeni° ‘cheek rouge’, ko ko ro / koko ro ‘heart’; (ii) their C2 is a noun which
must be analysed as deverbal, with an original final accent which has undergone an accentual shift due to the NADM
principle (machiga e 〉 machi ga e ‘error’, kanga e 〉 kan ga e ‘thought’); (iii) the compound admits two accent patterns
(hidari° + u chi wa 〉 hidari u chi wa / hidari u chiwa ‘to live at ease’). Non ambiguous examples of compounds which
illustrate the fate of C2’s with a penultimate accent are actually hard to find.
An additional and secondary factor conditions accent placement: the sonority of vowels. Tanaka (2008:157ff.)
demonstrates that vowel sonority plays a role in the attribution of the accent when the C2 is four mora long (except if it
is of Western origin and if the third mora of the C2 is a deficient mora). So, when the first vowel in the C2 has a lower
degree of sonority than the second vowel, the compound is more frequently accented on the antepenultimate mora
than on the pre antepenultimate one. The two following sets of examples illustrate this phenomenon:
(p.222) (35) Correlation between vowel sonority and accent (Tanaka, 2008)
a. V1 is higher in sonority than V2 in C2 ( → regularly accented compound)
a 〉 o oka° + yadokari° 〉 oka yadokari ‘land hermit crab’

e 〉 i kokusan° + benibana° 〉 kokusan benibana ‘safflower produced in Japan’
b. V1 is lower in sonority than V2 in C2 ( → irregularly accented compound)
i 〈 o ie + shiroari° 〉 ie shiroari ‘house termite’

e 〈 a reitou° + edamame° 〉 reitou edamame ‘frozen green soybeans’
(iii) Extra long C2
When the C2 is five mora long or longer, the rules are considerably simpler. The accent of the second member is
preserved even when it is final. If it is atonic, the compound will also be.
(36)
shi donii + orinpikku 〉 shidonii orinpikku ‘Sydney Olympic Games’

({Sydney} + {Olympic})
isoppu + monogatari 〉 isoppu monogatari ‘Aesop’s fables’
Page 27 of 55
Accent
({Aesop} + ‘story’)
chihou + saibansho 〉 chihou saibansho ‘regional tribunal’
(‘region’ + ‘tribunal’)
minami° + kariforunia° 〉 minami kariforunia° ‘Southern California’
(‘south’ + {California})
In the above pages, we have reviewed the accentuation of compound nouns. Although the phenomenology seems
incredibly complex, the following general trends can be established: the number of exceptions to the accent rules
decreases according to the length of the second component; the compound most generally carries the accent on the
second member (except for some compounds with short C2); a final accent is generally avoided (except in extra long
compounds); a certain convergence in favour of the antepenultimate pattern can be observed, resulting either from
the privileged maintenance of the original accent of C2, or from the attribution of a default accent. Finally, all things
being equal, it seems that if the accent of C2 can be preserved in its original location in the compound, it will be,
provided this does not infringe some of the other basic principles.
7.3.2 A Constra nt based Account of Compound Noun Accentuat on

Let us now move to the formal and theoretical analysis of the mechanisms reviewed above. We shall take as a starting
point the work by Kubozono (1997) who presents an OT treatment of these data. After summarizing (p.223)
Kubozono’s paper, which brings new light to this complex problem and has greatly contributed to its reconsideration,
we will see what treatment we can propose of the same facts.
As we have just seen, the length of C2 is traditionally regarded as the main parameter in order to determine the
prosodic pattern of a compound. It is precisely to the issue of this length parameter that Kubozono (1995b, 1997)
makes a first significant contribution, by proposing that the same principles apply to the accentuation of short and long
C2 compounds, and that it is thus irrelevant to distinguish between these two classes. On this basis, the accent
principles can be reformulated according to the following generalization:
(37) Kubozono’s generalization (slightly adapted)
A short or long C2 accent is parsed in compounds except when it is final.

If C2 is atonic or oxytonic, a default compound accent emerges on the rightmost non final foot.
There exists a class of exceptions among the words whose C2 is short (equivalent to a one or two
mora foot). They yield unaccented compounds.
However, as Kubozono observes, there exists in fact a second class of exceptions among short C2’s (the ‘Little
Mermaid’, ningyo hime type, see below). It will be noted moreover that Kubozono does not address the issue of
compounds with an extra long C2. They thus constitute, de facto, a special class with regard to his analysis. There is
also, and this should not be forgotten, a whole list of words whose accent is idiosyncratic, as well as many cases of
accentual variation.
The distinction between short and long C2’s being now irrelevant, only the de accenting type must be specified as
such lexically, and considered as exceptional. This contrasts with former analyses which consider (implicitly or
explicitly) that there exist three different classes of short C2’s. This descriptive simplification constitutes the first
important contribution of this work: one now has a general case, which can be accounted for in a unified way, and a
number of exceptions, which should be treated as such.
The second important contribution lies in the theoretical treatment that is proposed, on the basis of this new
formulation of the problem. This treatment will be summarized here. We shall then proceed to a reanalysis of the data,
which, I believe, improves on that originally proposed, while being in accordance with the conception of the Japanese
prosodic units and the role of the mora and the foot in the phonology of Japanese defended in the present book.
OT treatment: Kubozono (1997)’s analysis
According to Kubozono, the principles highlighted in (37) can be accounted for by the interaction between a small
number of constraints, defined as follows:
(p.224) (38) Kubozono’s constraints (1997, adapted)
Page 28 of 55
Accent
OCP: No more than one prominence peak (i.e. word accent) is allowed in a
single PrWd.
PARSE(ACCENT): Parse the lexical accent of the C2 in compound nouns.

NONFINALITY(µ): The head mora, i.e. the accented mora, is not final in PrWd.
NONFINALITY(σ): The head syllable, i.e. the accented syllable, is not final in PrWd.
NONFINALITY(π): The head foot, i.e. the accented foot, is not final in PrWd.
EDGEMOSTNESS/RIGHTMOSTNESS: A peak of prominence lies at the right edge of the Word.
As pointed out by Kubozono, an essential aspect of his analysis lies in the fact that the NONFINALITY constraint is
decomposed into three independent subparts: NONFINALITY(µ), NONFINALITY(σ), and NONFINALITY(π). Recall,
in connection with this issue, that Kubozono is a scholar for whom the distinction between mora and syllable is held as
essential in Japanese, and who regards the analysis of compound accentuation as a definite argument in favour of such
a distinction.
Kubozono assumes that the constraints OCP and NONFINALITY(µ) are undominated. The hierarchy that he
proposes is given in (39):
(39) Constraints hierarchy (Kubozono, 1997, adapted)
OCP, NONFINALITY(µ, σ) 〉〉 PARSE (ACCENT) 〉〉 NONFINALITY (π) 〉〉 EDGEMOSTNESS
The undominated constraint OCP will be omitted in the following discussion and tableaux. The following four tableaux
illustrate the action of the constraints and their hierarchy according to Kubozono’s analysis. Only the candidates that
respect the constraint OCP appear here. The examples we will consider are the following (we continue to use our
notation conventions throughout):
(40)
pe rusha + neko 〉 perusha neko (accent is kept in C2)

ne bada + shu u 〉 nebada shuu (C2 is pre accenting)
abare° + uma 〉 abare-uma (C2 is oxytonic and pre accenting)
ka buto + mushi° 〉 kabuto mushi (C2 is atonic and pre accenting)
(p.225)
(41) pe rusha + neko 〉 perusha neko ‘Persian cat’
/perusya/+/neko/ NONF(µ, σ) PARSEA NONF(π) EDGEMOST

☞ a. perusha) (neko) * σ #
b. perusha) (neko) *! σσ #
c. perusha) (neko) µ, σ ! * *
In (41), the first candidate, perusha neko is optimal because it maintains the non final accent of ne ko, thus respecting
the higher ranked constraints NONFINALITY (µ, σ) and PARSEACCENT.
The following tableau illustrates the case where the C2 is assumed by Kubozono to be a monosyllable.
(42) ne bada + shuu 〉 nebada shuu ‘the State of Nevada’
/nebada/+/syuu/ NONF(µ, σ) PARSEA NONF(π) EDGEMOST

a. nebada) (shuu) σ ! *
☞ b. nebada) (shuu) * σ #
c. nebada) (shuu) * σ σ σ # !
Page 29 of 55
Accent
Here, candidate a. cannot win because it violates NONFINALITY(σ). It is thus candidate b. which, although violating
PARSEACCENT, is optimal because it incurs fewer violations to EDGEMOSTNESS than c.
In tableau (43), C2 uma is bimoraic (as well as bisyllabic in Kubozono’s terms). Once again, the candidate with an
antepenultimate accent, in c., emerges as victorious.
(43) abare° + uma 〉 abare uma ‘restive horse’
/abare°/+/uma/ NONF(µ, σ) PARSEA NONF(π) EDGEMOST

a. abare) (uma) µ, σ ! *
b. abare) (uma) * *! σ #
☞ c. abare) (uma) * σσ #
Finally, we consider the case where the C2 is atonic, with no lexical accent to parse.
(p.226) (44) ka buto + mushi° 〉 kabuto mushi ‘beetle’
/kabuto/+/musi°/ NONF(µ, σ) PARSEA NONF(π) EDGEMOST

a. kabuto) (mushi) µ, σ ! *
b. kabuto) (mushi) *! σ #
☞ c. kabuto) (mushi) σ σ #
d. kabuto) (mushi) σ σ σ σ # !
Here too, the candidate with an antepenultimate accent, kabuto mushi in c., is selected.
Apart from the class of de accenting nouns, excluded from the scope of the analysis from the beginning because they
are regarded as marked exceptions in the lexicon, and that of extra long compounds which are not discussed in the
article, this treatment generates two types of exceptions. The first type consists of the so called ‘Little Mermaid’
pattern (as exemplified by nin gyo + hime 〉 ningyo hime ‘Little Mermaid’ or yoyaku° + seki 〉 yoyaku seki ‘reserved
seat’), which are words in which the initial accent of the C2 is not faithfully parsed, contrary to ne ko in perusha neko.
The second type includes cases ending with a supposedly bimoraic ‘syllable’, such as sunakku + baa 〉 sunakku baa
‘snack bar’ or eiga° + fan 〉 eiga fan ‘movie fan’, which preserve the accent of C2 on the final putative ‘syllable’, in
violation of NONFINALITY(σ), and thus contradict the analysis proposed for nebada shuu. We will not enter the
details of the extensions that Kubozono gives to his analysis in order to provide an account of these exceptions and of
the strong inter speaker variation observed in the accent pattern of some compounds. We shall merely observe that
he analyses these exceptions as resulting from a ‘minimal deviation’ from his standard constraint hierarchy. This is a
classic approach to exception and variation in OT. However, although such an approach correctly succeeds in
formalizing the exceptional character of the words under consideration, it fails to explain why it should be so.
I claim that a better analysis of these data can be proposed. It remains based on the descriptive generalization of
Kubozono, but ba a and fa n no longer appear as exceptions. This reanalysis builds on the conception of the Japanese
lower prosodic units as consisting only of the mora and foot, with no reference to the syllable, presented in Chapter 6.
(p.227)
My analysis
As mentioned above, Kubozono’s analysis rests on the assumption that the difference between mora and syllable is
relevant in Japanese. It is precisely this assumption that we call into question. The explanation that I propose in the
following lines rests on the distinction between the mora and the foot. I claim, on the one hand, that a simpler and more
coherent account can be provided, and, on the other hand, that we can get rid of one class of exceptions, the
sunakku baa and eiga fan type. Indeed, these words, which pose a problem in the original analysis, turn out to be
perfectly regular in our approach.
The reason why Kubozono calls upon the constraint NONFINALITY(σ) is because it is essential in the analysis to
account for one character Sino Japanese morphemes ending in a deficient mora like shu u , in nebada shuu. When
they occur as the second constituent of a compound, these morphemes require an accent to be put on the last mora
of the first component: they thus do not preserve their apparent lexical accent after compounding, contrary to the ne
Page 30 of 55
Accent
ko type. For Kubozono, as seen before, the difference between shu u and ne ko lies in the fact that shu u is a
bimoraic monosyllable while ne ko is a bimoraic bisyllable. The behaviour of the words fa n or ba a which, despite the
fact that they are also bimoraic monosyllables like shu u, behave like ne ko rather than like shu u in keeping their
accent on the moras fa and ba in compounding (sunakku baa,eiga fan) leads Kubozono to treat them as exceptions.
However, one can assume that it is not the number of moras or syllables that is at stake here but some other kind of
difference.
In reality, the difference between the shu u type and the ba a type is to be captured at some other level. I assume
that there exists an accentual difference at the underlying level between these two lexemes. Morphemes like shuu are
accented on their last mora at the lexical level: /syuR/ (see also sections 6.2.4 and 7.3.6 for additional evidence),
whereas baa or fan are accented on the first one: /baR/, /faN/. One thus distinguishes, at the lexical level, the two
following types for words consisting of one foot whose second mora is deficient (that is, which consist in a heavy
syllable following Kubozono’s analysis):
(45) Underlyingly final accent
a. Sino Japanese morphemes
/syuR/ shuu ‘state’ /tyuR/ chuu ‘middle’

/toR/ tou ‘party’ /hoN/ hon ‘book’
/yoR/ you ‘use’ /ryuR/ ryuu ‘stream’
etc.
b. Western morphemes
/piN/ pin {pin} /paN/ pan {paõ}

/maN/ man {man} etc.
(p.228)
(46) Underlyingly initial accent (Western morphemes)
/baR/ baa {bar} /faN/ fan {fan}

/kiR/ kii {key} etc.
Morphemes in (46) are all of relatively recent Western origin. Morphemes in (45) are mostly Sino Japanese, but it is
worth noting that they also include a small number of ancient loans from Western languages, such as pin ‘pin’, pan
‘bread’, or man ‘man’. Interestingly, there exist some Western words like ka a ‘car’ or ti i ‘tea’ that behave like the
words in (45) or (46) depending on speakers. One thus has patorooru kaa or patorooru kaa {patrol car}, and
remon tii or remon tii {lemon tea}. Noteworthy enough, these words obviously belong to a class of loans which are
neither very old nor very recent. So, the phonological difference in the localization of the lexical accent that we see
actually reflects a difference in the dating of the borrowing, be it from Chinese or from a Western language. Older
loans generally bear an underlyingly final accent while more recent ones tend to bear an underlyingly initial accent.
Words such as ka a or ti i represent an intermediate stage, which is why they allow accentual variation in
compounding.
The surface pattern shu u corresponding to the underlying form /syuR/ and that of the other words of the same class
is accounted for by the action of the NADM principle, which prohibits accentuation of deficient moras (see section
6.3.2).
Note that PEAKPROMINENCE, the constraint of which the NADM principle is a reformulation (see section 6.3.2) is
needed to correctly derive the output form of monographemic Sino Japanese morphemes finishing with a deficient
mora, but since it is not directly active here, it will not be mentioned in the tableaux below.
As seen in section 7.2.3, it is thus not because it consists of a ‘heavy syllable’ that shu u behaves differently from ne ko
in compounding but because it actually possesses a final accent at the input level, exactly like uma, examined in (43).
This question being set, the NONFINALITY(σ) constraint proposed by Kubozono is no longer necessary.
I thus propose to revise the list of constraints as follows:
Page 31 of 55
Accent
(47)
OCP: No more than one accent peak in PrWd.

FAITHIO(HEAD The accent kernel of the head noun occupies the same position in the input and in
ACCENT): the output.
NONFINALITY(µ): The accented mora must not be final.
NONFINALITY(π): The accented foot must not be final.
ALIGNRIGHT: The accent lies at the right edge of the word.
Except for NONFINALITY(σ), which has now become superfluous, the constraints proposed here are essentially the
same as those introduced by Kubozono (1997). We adopt the constraint FAITHIO(HEADACCENT) in replacement
(p.229) of PARSEACCENT, for the sake of clarity. FAITHIO(HEADACCENT) requires that the accent occupy exactly
the same position in the output as in the input. Moreover, ALIGNRIGHT, already used for the analysis of the default
accent in Western loans in section 7.2.5, henceforth replaces EDGEMOSTNESS/RIGHTMOSTNESS.
The analysis by Kubozono (1997) does not address the issue of compounds with a long C2. This class of compounds
was the subject of a previous work by the same author (Kubozono, 1995b), but the treatment suggested for
compound nouns with long C2 in the 1995 paper is independent of the one proposed for short C2 in 1997. This is why
it is essential to look for a unified approach, which can account for the two types: compounds with a short C2 and
compounds with a long C2. I will retain from the 1995 paper the constraint given in (48). I shall also assume that this
constraint is at work both for long and short C2 compounds:
(48)
ALIGNCA:
Align the accent with the boundary between C1 and C2.
ALIGNCA
stipulates that the accent is aligned with the boundary between C1 and C2, either on the last mora of C1 or on
the first mora of C2.
In addition, I propose that a constraint imposing the realization of an accent in the compound, the ACCENT constraint,
be ranked above all the other constraints. Due to the action of this constraint, atonic candidates are systematically
eliminated. Like Kubozono, I consider that atonic compound nouns constitute a closed word class, limited to forms
containing a short C2, which it is preferable to treat as exceptional.
(49)
ACCENT: Compounds must have an accent.
In the following tableaux, the ACCENT constraint does not figure, nor do the atonic candidates that fatally violate it.
Mutatis mutandis, the hierarchy in (50) remains that of Kubozono, but without the NONFINALITY(σ) constraint. Note
that it is also fully compatible with our analysis of foreign place names developed in section 7.2.5.
(50) Final constraint hierarchy
ACCENT 〉〉 NONF(µ) 〉〉 FAITHIO(A) 〉〉 NONF(π) 〉〉 ALIGNCA 〉〉 ALIGNRIGHT
Let us now examine how five of the examples presented earlier in this chapter, that is, perusha neko, abare uma,
kabuto mushi, nebada shuu, and sunakku baa, can be handled following this new proposal.
Let us start with perusha neko, abare uma, and kabuto mushi.
(p.230) (51) pe rusha + neko 〉 perusha neko ‘Persian cat’
Page 32 of 55
Accent
/perusya/+/neko/ NONF(µ) FAITHIO(A) NONF(π) ALIGNCA ALIGNR

☞ a. perusha) (neko) * *
b. perusha) (neko) *! * * *
c. perusha) (neko) *! **
(52) abare° + uma 〉 abare uma ‘restive horse’
/abare°/ + /uma/ NonF(µ) FAITHIO(A) NONF(π) ALIGNCA ALIGNR

a. abare) (uma) *! * *
b. abare) (uma) * *! *
☞c. abare) (uma) * **
(53) ka buto + mushi° 〉 kabuto mushi ‘beetle’
/kabuto/ + /musi°/ NONF(µ) FAITHIO(A) NONF(π) ALIGNCA ALIGNR

a. kabuto) (mushi) *! ****
b. kabuto) (mushi) *! * *
c. kabuto) (mushi) *! *
☞ d. kabuto) (mushi) **
In each of the three cases above, the expected form is correctly selected. These examples do not pose any particular
problem because they do not include a prosodic unit likely to be analysed as a heavy syllable. In this respect, the
following example, nebada shuu, appears as crucial. As we have stated, the morpheme shu u carries an initial accent at
the surface level, but we now assume that it actually contains an accent on its final mora at the lexical level, which
constitutes the input of tableau (54).
(p.231) (54) ne bada + shuu (/syuR/) 〉 nebada shuu ‘the State of Nevada’
/nebada/ + /syuR/ NONF(µ) FAITHIO(A) NONF(π) ALIGNCA ALIGNR

a. nebada) (shuu) * *! *
b. nebada) (shuu) *! * *
☞ c. nebada) (shuu) * **
The candidate in c. is victorious, even though it does not maintain the accent in its original position. It is actually non
optimal to maintain the final accent of the input, as candidate b. does, because this violates the higher ranked constraint
NONFINALITY(µ). The nebada shuu case is actually identical to that of abare uma in (52).
Finally, let us see what happens with sunakku baa {snack bar}, an example which is problematic in Kubozono’s
analysis. Recall that, as seen above, the word ba a has its underlying accent on the initial mora ba in the present
approach, contrary to shu u which has an underlyingly final accent: /baR/versus /syuR/.
(55) sunakku + ba a 〉 sunakku baa ‘snack bar’
/sunaQku/ + /baR/ NONF(µ) FAITHIO(A) NONF(π) ALIGNCA ALIGNR

☞ a. sunakku) (baa) * *
b. sunakku) (baa) * ! **
c. sunakku) (baa) * ! * * *
The candidate in c. displays a fatal infringement to NONFINALITY(µ), and is eliminated. So is b. which violates
FAITHIO(HEADACCENT). It is thus the form in a., whose non final lexical accent is faithfully parsed, which emerges as
Page 33 of 55
Accent
optimal.
The analysis proposed here has the advantage of getting rid of what appears, in Kubozono’s analysis, as a class of
exceptions. The accent pattern of ba a can now be distinguished from that of shu u. The former is accented on the
initial mora underlyingly (/baR/), while the latter is on the final mora (/syuR/), hence the difference in their accentual
behaviour when they undergo compounding.
(p.232) There remains one type of lexical exception in the treatment I have proposed, as in Kubozono’s, the ‘Little
Mermaid’ case: ni ngyo + hime yields ningyo hime instead of *ningyo hime, the expected output. As mentioned
above, this type resorts to a closed class, and one should resign oneself to mark it as exceptional at the lexical level at
the present state of research.
One can go one step further and extend the analysis to the class of compounds containing a long C2, such as kogata°
+ kamera 〉 kogata kamera ‘small camera’, shimaguni + konjou 〉 shimaguni konjou ‘insularism’, yuki + daruma° 〉
yuki daruma ‘snowman’, kami + hikouki 〉 kami hikouki ‘paper plane’, kuchi° + yakusoku° 〉 kuchi yakusoku ‘verbal
promise’, nuka +yorokobi 〉 nuka yorokobi ‘premature joy’, wa ka + murasaki 〉 waka murasaki ‘light purple’ (see
section 7.3.1).
We can consider that the regular output for compounds with a long C2 is to preserve the accent in its original position
except when it is final. In this case, or if the C2 is atonic, an accent is placed on the first mora of C2. We are only dealing
here with cases that do not present any variation, be it in the accent of the compound or in that of the noun in C2 (we
thus leave aside those examples likely to present more than one accentual possibility like hidari° + uchiwa 〉 hidari
uchiwa/hidari uchiwa, or de nki + nokogiri/nokogiri 〉 denki nokogiri, previously discussed in section 7.3.1).
In the first example (tableau 56), the candidate kogata kamera in a., in which the non final accent of the trimoraic C2 ka
mera is preserved in its position of origin, wins over all other candidates because those violate one or more than one
of the four higher ranked constraints. The same applies to the compound with a four mora C2 in tableau (57), where it
is the form in a. with an accent on the first mora of ko njou that is optimal.
(56) kogata° + kamera 〉 kogata kamera ‘small size camera’
/kogata°/ + /kamera/ NONF(µ) FAITHIO(A) NONF(π) ALIGNCA ALIGNR

☞ a. kogata) (kame)(ra)/ (ka)(mera) **
b. kogata) (kame)(ra) * ! ***
c. kogata) (kame)(ra) * ! * * *
d. kogata) (ka)(mera) * ! * * *
(p.233)
(57) shimaguni + konjou 〉 shimaguni konjou ‘insularism’
/simaguni/ + /konzyoR/ NONF(µ) FAITHIO(A) NONF(π) ALIGNCA ALIGNR

☞ a. shimaguni) (kon)(jou) ***
b. shimaguni) (kon)(jou) * ! ****
c. shimaguni) (kon)(jou) * ! * * *
The following tableau exemplifies the case of a trimoraic C2 atonic in its independent form, in which there is thus no
original accent to parse.
(58) yuki + daruma° 〉 yuki daruma ‘snowman’
/yuki/ + /daruma°/ NONF(µ) FAITHIO(A) NONF(π) ALIGNCA ALIGNR

☞ a. yuki) (daru)(ma) / (da)(ruma) **
b. yuki) (daru)(ma) *! *
c. yuki) (daru)(ma) * ! * *
Page 34 of 55
Accent
d. yuki) (daru)(ma) *** !
The form yuki daruma in a. will be selected. Its most serious competitors are b. and d., which also respect the three
higher ranked constraints. However, candidate b. fatally violates ALIGNCA, while d. presents one infraction more in
comparison to a. for ALIGNR.
In (59), C2 is a four mora long noun with an original antepenultimate accent. Here, candidate b. wins out because it is
the only one that respects the three highest constraints, and particularly FAITHIO(HEADACCENT), which both the
candidates a. and c. violate.
(59) kami + hikouki 〉 kami hikouki ‘paper plane’
/kami/ + /hikoRki/ NONF(µ) FAITHIO(A) NONF(π) ALIGNCA ALIGNR

a. kami) (hi)(kou)(ki) *! ***
☞ b. kami) (hi)(kou)(ki) * **
c. kami) (hi)(kou)(ki) *! ****
d. kami) (hi)(kou)(ki) *! * * *
(p.234)
Turning now to (60), we can consider the case of a quadrimoraic atonic C2. The candidate in b., kuchi yakusoku, is the
best one, as expected. Its most serious competitors are the forms in a. and e. with an antepenultimate accent. Their
elimination is due to the fact that the accent is too far from the C1 C2 boundary (infringement of ALIGNCA). The same
applies to candidate d., which presents four infringements of ALIGN(RIGHT), whereas b. has only three.
(60) kuchi° + yakusoku° 〉 kuchi yakusoku ‘verbal promise’
/kuti°/ + /yakusoku°/ NONF(µ) FAITHIO(A) NONF(π) ALIGNCA ALIGNR

a. kuchi) (yaku)(soku) *! **
☞ b. kuchi) (yaku)(soku) ***
c. kuchi) (yaku)(soku) * ! * *
d. kuchi) (yaku)(soku) ****!
e. kuchi) (ya)(kuso)(ku) * ! **
In the compound nuka + yorokobi in (61), the original accent of C2 cannot be maintained in the compound since that
would involve fatal violation of NONFINALITY(µ), which is higher in the hierarchy than FAITHIO(HEADACCENT). The
candidate in a. is thus excluded. The correct output must then be selected among the forms which infringe
FAITHIO(HEADACCENT). It is the low ranked constraint ALIGN(RIGHT) that makes the difference between b. and e.,
allowing candidate b. to win over e., because, all other things being equal, b. has one violation less than e. with respect
to ALIGN(RIGHT).
(61) nuka +yorokobi 〉 nuka yorokobi ‘premature joy’
/nuka/ + /yorokobi/ NONF(µ) FAITHIO(A) NONF(π) ALIGNCA ALIGNR

a. nuka) (yoro)(kobi) *! * *
☞ b. nuka) (yoro)(kobi) * ***
c. nuka) (yoro)(kobi) * *! * *
d. nuka) (yoro)(kobi) * *! **
e. nuka) (yoro)(kobi) * **** !
f. nuka) (yo)(roko)(bi) * * ! **
(p.235)
Tableau (62) exemplifies the case of a compound with a quadrimoraic C2 accented on the antepenultimate mora in its
Page 35 of 55
Accent
lexical form. One expects that the accent will be kept in this position, even after compounding. And indeed, this is what
occurs. One of the forms in a., the only ones that respect FAITHIO(HEADACCENT), is selected as the optimal
candidate because all the other possible outputs present fatal violation of this constraint.
(62) waka + murasaki 〉 waka murasaki ‘light purple’
/waka/ + /murasaki/ NONF(µ) FAITHIO(A) NONF(π) ALIGNCA ALIGNR

☞ a. waka) (mura)(saki)/ (mu)(rasa)(ki) * **
b. waka) (mura)(saki) * ! ***
c. waka) (mura)(saki) * ! ****
d. waka) (mura)(saki) * ! * * *
In conclusion, it must be pointed out that the treatment proposed here presents a number of improvements. First,
the recourse to the ambivalent and problematic constraint NONFINALITY(µ, σ) is no longer necessary. Only
NONFIN(µ) and NONFIN(π) are. Second, the sunakku baa and eiga fan types, treated as mere exceptions in the
previous analysis, are no longer exceptional. In third place, our analysis provides a unified account of compounds with
short and long C2’s. And, last but not least, it illustrates the uselessness of resorting to the syllable. Not only is
reference to the mora and to the foot sufficient for handling the examples but it also provides a more satisfying
account of the data, thus confirming the proposals made in Chapter 6.
7.3.3 Compound Nouns Conta n ng Two Accent Nuc e

The accentuation principles which have been presented up until now apply to modifier head structured compounds
which only get one accent nucleus after the compounding process. However, there also exist compounds with two
accent nuclei. Such compounds preserve the accent pattern of each of their components after the compounding
process (Kubozono, 1993b, 1998a; Kubozono, Itô and Mester, 1997).
Such formations involve two or more nouns, which, on the morphological level, do not seem any different from the
constructions previously seen. However, they do not follow the same accent rules. Thus kyu ushuu + na nbu receives
(p.236) the pattern kyu ushuu na nbu ‘Southern Kyûshû’, with two accent nuclei, rather than *kyuushuu na nbu if it
had followed the previously mentioned rules. In the same way, jishin° + soushitsu° yields jishin° soushitsu° ‘loss of
self confidence’ (but jishin so ushitsu also seems to be attested), and o bama + dai to uryou becomes o bama dai to
uryou ‘president Obama’ (*obama dai to uryou). Each member of the compound preserves its original accent nucleus
in its original location, and we thus have two distinct prosodic words. The conditions which govern this prosodic
structuring are not clear: whereas jishin° soushitsu° is treated as a succession of two distinct prosodic words,
kioku so ushitsu ‘loss of memory’ (kioku° + soushitsu°) is treated as only one. Yet, the morphological and
semantic structure of these two compounds is strictly identical.
In a number of such cases, we are no longer dealing with a modifier head structure in the narrow sense but rather
with an appositive like morpho syntactic structure (see the examples cited above: ‘South(ern) Kyûshû’, ‘president
Obama’), which can justify that each member keeps its own accent. Nevertheless, this analysis does not apply to all the
cases concerned, for instance, jishin° soushitsu° cannot be considered to be an appositive construction.
In constructions made up of three nouns, accent differences may reflect differences in the morphological and syntactic
structure of the compounding. Kubozono (1993b) mentions the example ni ho n + buyou° + kyoukai° (‘Japan’ +
‘dance’ + ‘association’), which can be realized with two accented nuclei, ni ho n buyou kyo ukai (with a A B C
structure) or with only one nucleus nihon buyou -kyo ukai ( A B C ), depending on the meaning of the compound:
‘Japanese association of dance’ in the first case, ‘association of Japanese dance’ in the second.
7.3.4 Yamato Dvandva Compounds

In Yamato dvandva (coordinative) compounds, containing two equipollent terms whose order is not fixed by syntactic
factors (none of the two modifies the other one), the first member determines the accentual pattern of the compound.
The accent of the second word is deleted, and the first member maintains its accent in the original location, following a
principle somewhat identical to that governing the accentuation of the phonological syntagm (see section 7.4.). If the
first word is atonic, the compound will also be.
(63)
Page 36 of 55
Accent
natsu + huyu 〉 natsu huyu ‘summer and winter’

ha ru + a ki 〉 ha ru aki ‘spring and autumn’
momo° + kuri 〉 momo kuri° ‘peaches and chestnuts’
(p.237)
7.3.5 Compound M met cs

Compound mimetics mostly occur under the reduplicated form of a monomoraic or bimoraic base, possibly followed
by the particles to or ni, or the copula da (and its pre nominal form na). They also include echo words, which do not
consist of the reduplication of an identical base but rather of the concatenation of two different mimetic bases.
When used in isolation or followed by the particle to, compound mimetics bear an accent on the initial mora: ki ri kiri(
to) ‘diligently’, ko n kon(to) ‘coff coff’, pa ppa( to) ‘puff puff’, chi ra hora( to) ‘scatteringly’.
When followed by da (the copula), na (adjectival ending), or ni (adverbial particle), they are atonic: kan kan ni° ‘red
hot’, tsuru tsuru° da ‘it is soft’.
When to is preceded by /Q/, an accent is placed on the antepenultimate mora: doki dokit to ‘fast beating (heart)’.
Sino Japanese reduplicated mimetics are generally atonic, sometimes paroxytonic: rin rin° / rin rin ‘piercingly (cold)’.
Forms which are, in nature, mimetic, but are etymologically derived from the reduplication of a non mimetic base, are
accented on the penultimate mora when they occur with the particle to: taka daka to ‘high’, and are atonic when they
occur with da or ni: atsu-atsu da° ‘it is very hot’.
7.3.6 Two character F xed S no Japanese Compounds

The use of the term ‘compound’ to refer to Sino Japanese lexemes made up of two Chinese characters, like denwa°
‘telephone’, ka gu ‘piece of furniture’, isha° ‘doctor’, henji ‘answer’, de shi ‘disciple’, which are so
numerous in the Japanese lexicon, is debatable. One can argue that such combinations are not necessarily analysed as
compounds by speakers in their everyday oral use because they are generally semantically and referentially simple.
Besides, their meaning cannot always be deduced from the meaning of each component in a direct and transparent
fashion. Moreover, the constituents themselves are bound morphemes that are generally not used independently.
For example, the two components of denwa° ‘telephone’, den ‘electricity’, and wa ‘to speak’, are never employed
in isolation.14 In fact, many compounds of this type are totally lexicalized and are to be considered as single lexemes.
However, in writing, they contain two identifiable components, each with stable and transparent meaning represented
by one Chinese character. This is the reason why I use the term ‘fixed compound’ to distinguish such formations from
those which are made up of two or more components with true lexical autonomy as in (p.238) the examples seen
above in section 7.3.1. Some scholars also call them ‘Sino Japanese binomials’.
Fixed Sino Japanese compounds written by means of two Chinese characters follow specific accent rules. These rules
are mainly conditioned by the phonological length of the compound, its phonological structure, and by its nominal or
verbal character. However, many exceptions can be found. In certain cases, semantic criteria also have a role to play.
Finally, one should note that some characters impose their own special accent pattern when they appear in the second
position of the compound. This leads to lists which try to enumerate all the special cases. Actually, the accent of Sino
Japanese compounds seems to be a topic rather neglected in the field of accent research. More work is thus
necessary, especially since the area appears to be a very promising field in terms of the wealth of the data, the
theoretical implications, as well as the historical depth.15
Sino Japanese compounds that I call ‘non fixed’, that is, compounds which are made up of three or more characters, of
which one is a derivational suffix like go ‘language’, jin ‘person’, tou ‘party’ (as in nihon go° ‘Japanese
language’, nihon jin ‘Japanese person’, shakai tou° ‘socialist party’) or which represent the combination of two
fixed compounds (like kokuritsu ko uen ‘national’ + ‘park’ = ‘national park’, seishin bunseki ‘spirit’ +
‘analysis’ = ‘psychoanalysis’) will thus not be dealt with here. Such compounds fall under the scope of the general
accent rules described in section 7.3.1.
A thorough analysis of the principles governing the accent of fixed Sino Japanese compounds requires, first of all,
detailed morphological and semantic analysis of the various types of Sino Japanese compounds. Such a study cannot be
carried out here. For this reason I will confine myself to an overview of the most general principles.
Most words of this class are either initially accented or unaccented (Nakada and Hayashi, 1982).
Page 37 of 55
Accent
Bimoraic two character fixed compounds with a (µ)(µ) structure generally bear an initial accent: chi ri ‘geography’,
ka gu ‘piece of furniture’, shu hu ‘housewife’. According to Kindaichi and Akinaga (2001), there exist
approximately 20% of exceptions, among which isha° ‘doctor’, jama° ‘obstacle’, se wa ‘assistance’, de shi
‘disciple’, including a rather high number of old loans related to Buddhism, according to Takeuchi (1999).
(p.239) Kindaichi and Akinaga (2001) states that when the compound is trimoraic, with either a (µµ)(µ) or (µ)(µµ)
morphological structure, it will be attributed an initial accent if it is nominal in nature but will be atonic if it is verbal: ko
kka ‘state, nation’, se ishi ‘official history’, shi gai ‘suburbs’, ko kyou ‘native village’ vs. akka°
‘aggravation’, nyuukyo° ‘moving in’, seishi° ‘stop’, shibou° ‘death’. There too exceptions can be found, like hen
ji ‘answer’, chi tsu jo ‘order’, dou gu ‘tool’.
However, the criteria for determining whether a word is nominal or verbal remain vague. According to Ogawa (2004,
2008), the influence of the noun/verb distinction on the accent pattern of trimoraic Sino Japanese compounds is only
partial, and not limited to this part of the lexicon (it can also be observed in the Yamato and Western strata). Ogawa
presents a statistical study that demonstrates that the prosodic structure of the compound plays the most important
role in determining the accent of Sino Japanese words. Indeed, if one compares the accent patterns of Sino Japanese
compounds having MmM and MMm structure 16 (in syllabic terms, HL and LH structures respectively, where H
represents a bimoraic foot ending in /N/, /R/, or the moraic vowel i, and L, a regular mora), it appears that nearly 80%
of MmM (HL) words are accented on the first mora, while 82 % of MMm (LH) words are atonic. However, since this
calculation does not include words that contain a geminate or that have a bimoraic CVCV structure with one
component like jitsu or se ki whose final u and i is etymologically epenthetic, these results need further examination.
Moreover, the morphological structure also plays a role, since a majority (50%) of MM M trimoraic Sino Japanese
words tend to bear an initial accent, whereas 80% of those having a M MM structure are atonic, as shown by
Ogawa (2003), cited by Tanaka (2008:176). According to Tanaka (2008), the morphological structure is, on the whole,
less determining in kango than it is in wago for accent attribution.
The important role of the prosodic structure of fixed Sino Japanese compounds on accentuation probably explains why
accent has practically no distinctive function with homophonous Sino Japanese two character words, according to
Coyaud (1985). Only a few minimal pairs based on an accentual difference can be found, for instance ko ukou ‘filial
piety’ and koukou° ‘high school’.
In the case of four mora compounds, the same general tendency is observed towards atonicity as that already seen in
quadrimoric words of other lexical groups.We thus have daigaku° ‘university’, kokusai° ‘international’, tankyuu°
‘search’, suigai° ‘flood’, teikoku° ‘empire’, koukou° ‘high school’. However, words in which the accent falls
on the last mora of the first component (with NADM left shift of the accent when relevant) can be found (p.240)
among older and/or strongly lexicalized compounds such as ko ku nai ‘domestic’, sho ku butsu ‘plant’, (ma n
getsu〉) ma ngetsu ‘full moon’, (ka i shaku 〉) ka ishaku ‘interpretation’. Note also that there are a few rare
words with final accent like shouga tsu ‘New Year’s day’.
According to Kindaichi and Akinaga (2001), the occurrence of a penultimate accent frequently results from an accent
shift consecutive to the devoicing of the initial vowel. This factor could account for many apparent exceptions such as
chi ho u (〈 chi hou) ‘area’, or ki ka i (〈 ki kai) ‘machine’.
Moreover, many characters impose a specific accent pattern when they occur in final position, whatever the size of the
first component: the characters tou ‘party’, tai ‘body’, jou ‘place’, sei ‘nature, gender’, ren ‘ream’, wa
‘speech’, etc. cause the compound to be atonic (seitou° ‘political party’, dantai° ‘group’, josei° ‘woman’) while
ou ‘king’, suu ‘number’, ryou ‘fee’, and others involve the appearance of an initial accent on the second
component (koku o u ‘king’, san su u ‘arithmetic’).
When the first character has prefixal status, such as sho ‘various’, i ‘more’, kaku ‘each’, shin ‘neo’, the
compound generally receives initial accent: sho koku ‘various countries’, i nan ‘south of’, ka kui ‘all’, etc.
7.3.7 Compound Verbs

In the conservative variety of speech, the accent of the first member determines the accent pattern of a compound
verb made up of two independent verbs (V1 and V2), following a rather curious principle of accent inversion: if the
first verb is atonic, the compound will be tonic with the accent on the penultimate mora (64); if the first verb is tonic,
the compound will be atonic whatever the pattern of V2 (65).
(64) Atonic V1 〉 tonic V1 V2
Page 38 of 55
Accent
kiru° + kazaru° 〉 ki kazaru ‘to dress up’
(‘to wear’ + ‘to decorate’)

naku° + da su 〉 naki -da su ‘to burst into tears’
(‘to cry’ + ‘to take out’)
(65) Tonic V1 〉 atonic V1 V2
taberu + owaru° 〉 tabe owaru° ‘to finish eating’

(‘to eat’ + ‘to finish’)
hu ru + da su 〉 huri dasu° ‘to begin to rain’
(‘to rain’ + ‘to take out’)
(p.241)
However, in the innovative speech, the current trend is towards the attribution of a tonic pattern to all compound
verbs, even if V1 is tonic: hu ru + da su 〉 huri da su ‘to begin to rain’.
7.3.8 Numera Compounds

The label ‘numeral compounds’ designates two types of compounds comprising a numerical expression. On the one
hand, we have numeral cardinals containing at least two (linguistic) elements, for example sa n juu san ‘33’, se n ni
hyaku hachi juu nana ‘1287’. On the other hand, we have forms comprising a numeral followed by a specifier (this
label also includes measurement terms), like ro k ko ‘6 small objects’, sa n juu san mai ‘33 sheets’, hitotsu boshi ‘1
star’, ichi meetoru ‘1 meter’. We will examine these two types in turn.
The principles that govern the accentuation of this particular type of compound are incontestably among the most
complex of the Japanese language. One outstanding characteristic of this class is that it is sometimes the accent of the
initial component that determines the accent of the compound, contrary to the majority of other non verbal
compounds of Standard Japanese, in particular those of the Yamato stratum, where the prosodic characteristics which
are relevant in accent assignment are those of the final component.
The rules that apply here are thus specific to this class, and notably different from those applying to the common run of
non numeral compound nouns.
The factors, very diverse, which condition the accentuation of numeral compounds, are determined on phonological,
lexical, or even syntactic grounds. These factors are:
the intrinsic accent of the numeral.

the origin of the numeral (Yamato, Sino Japanese, or Western).
the origin of the specifier.
the length of the specifier.
the syntactic status (nominal or adverbial) of the numeral expression.
Inter and intra speaker variation is very frequent, even if, as a whole, speakers generally agree with each other, even
when it comes to the accent pattern of numeral expressions which they have never heard before, as can be noted for
example with the recent diffusion of the new specifier yu uro ‘euro’.
Let us first consider the intrinsic accent of cardinal numerals when they occur without a specifier.
(i) Cardinal numerals (with no specifier)
The series used to form cardinal numerals is that of Sino Japanese numerals. Let us recall however that the forms 4 yo
n and 7 na na, of Yamato origin, are also frequently employed in this series.
(p.242) As will be noted in the following examples, there are a number of variants of some of the numerals. Cardinals
are accented as follows in isolation:
(66) Accent of Sino Japanese numerals
Oxytonic (with leftward accent shift conditioned by NADM principle)
Page 39 of 55
Accent
1 ichi, 2 ni, 4 shi (yon), 5 go, 6 roku, 7 shichi, 8 hachi, 9 kyu u / ku, 10 ju u, 100 hyaku, 1000 se n,
10000 ma n (only in innovating varieties for 5 go and 9 ku)
Atonic:
3 san°, 5 go°, 9 ku° (only in conservative varieties for 5 go° and 9 ku°)
One may consider that with the exception of san° ‘3’, all the Sino Japanese numerals are finally accented underlyingly in
the innovative variety.
Among the numerals of Western origin, only 0 ze ro is commonly employed in compounding.
In compound cardinal numerals, either the original accent of the rightmost element is maintained even when it is final
(67), in a manner which is somewhat reminiscent of what occurs in the accentuation of the major phonological phrase
(see section 7.4) or, in compounds of the series 20, 30, 40, 70, 90, and 300, 400, 700, 900 (examples 68), an accent is
attributed to the leftmost numeral even when that numeral is unaccented in isolation, so for instance san° becomes sa
n in sa n juu ‘30’. This rule does not apply solely to round figures, since 51, or 888, for instance, are also accented on
their rightmost element. However, in cases of leftmost accent preservation or attribution, it seems that long numbers
are treated as two or even more than two prosodic words by some speakers, and may thus contain more than one
accent kernel, as some of the examples in (68) illustrate.
(67) Rightmost accent preservation
13: ju u + san° 〉 ju u san (also juu san)

16: ju u + roku 〉 juu roku
51: go° + juu + ichi 〉 go juu ichi
251: ni + hyaku + go° + juu + ichi 〉 ni hyaku go juu ichi
800: hachi + hyaku 〉 hap pyaku
888: hachi + hyaku + hachi + juu + hachi 〉 hap pyaku hachi juu hachi
(68) Leftmost accent preservation or attribution
20: ni + juu 〉 ni juu

31: san° + juu + ichi 〉 sa n juu ichi (also sa n juu ichi)
75: na na + juu + go° 〉 nana juu go (also nana juu go)
300: san° + hyaku 〉 sa n byaku
311: san° + hyaku + juu + ichi 〉 sa n byaku juu ichi or sa n byaku juu ichi (the latter seems more
common)
426: yo n + hyaku + ni + juu + 〉 yo n hyaku ni juu roku or yo n hyaku ni juu roku (the latter seems
roku more common)
700: na na + hyaku 〉 nana hyaku
(p.243)
(ii) Numeral compounds containing a specifier
Let us now proceed to the examination of numeral compounds containing a simplex numeral expression followed by a
specifier. We will only look at the forms from 1 to 10, and at compounds with a nominal syntactic status, keeping in mind
that in adverbial use they sometimes present an atonic pattern. For example, ringo° mittsu wo katta° (‘bought three
apples’), where mittsu ‘three’, followed by the object particle wo, is tonic, and is distinct from ringo° wo mittsu°
katta°, where mittsu, which modifies the verb, is atonic.
Note also that when the numeral meaning of a compound is weakened, the regular compound nouns accent rules tend
to apply, for example we have hutago° ‘twins’ (literally ‘two children’) rather than *huta go or *hu ta go.
Yamato numerals
The intrinsic accent of simple Yamato numerals likely to appear in compounding (according to Kindaichi and Akinaga,
2001) is as follows. Note that these forms are never used in isolation, except yo n ‘4’ and na na ‘7’.
Page 40 of 55
Accent
(69) Accent of short Yamato numerals
Final accent:
1 hito, 2 huta, 3 mi, 4 yo, 6 mu, 8 ya, 10 to
Penultimate accent:
4 yo n, 5 itsu, 7 nana, 9 kokono
If the two components are of Yamato origin, and the specifier is one or two mora long, it is in theory the last mora of
the numeral that will carry the accent: hi to ri ‘one person’, hu ta hako ‘two boxes’, hito tsu boshi ‘a star’, mit tsu
boshi ‘three stars’, na na tsubu ‘seven grains’, koko no e ‘nine layers’, etc. The names of the days of the month
ending in ka are all atonic, thus irregular: hutsu ka° ‘the 2nd of the month’, mik ka° ‘the 3rd’, yok ka° ‘the 4th’, itsu
ka° ‘the 5th’, etc. but nano ka ‘the 7th’ and kokono ka ‘the 9th’ may also receive a final accent.
Some compounds starting with huta ‘2’ are sometimes exceptional in the conservative varieties in so far as they
receive a final accent, whatever the size of the specifier: huta ri ‘two persons’, huta mi chi (also huta michi°) ‘two
ways’, and so on.
(p.244) Forms with the generic specifier tsu are strongly irregular. They are accented sometimes on the second
mora from the beginning, sometimes on the final mora (70).
(70) Yamato numerals in tsu (‘x things’)
Accent on the second mora:

1 hitotsu, 5 itsutsu, 7 nanatsu, 9 kokonotsu
Accent on the final mora:
2 hutatsu, 3 mittsu, 4 yottsu, 6 muttsu, 8 yattsu
It is interesting to observe that the accentual behaviour of words such as i ro ‘colour’, ta ba ‘bundle’, tsu bu ‘grain’,
or hako° ‘box’ employed as specifiers is completely different from that which they adopt as elements entering into the
formation of endocentric compound nouns. Thus, as we saw in section 7.3.1 (examples 32), within a non numeral
compound i ro causes the compound to de accentuate (mizu iro° ‘colour of water, light blue’) whereas in numeral
compounds, it is pre accenting, and the accent is placed on the numeral (hi to iro,hu ta iro, mi iro,nana iro, etc.). The
same applies to ta ba ‘bundle’ and tsu bu ‘grain’ whose initial lexical accent is generally preserved in a non numeral
compound (satsu ta ba ‘bundle of banknotes’, meshi tsu bu ‘grain of rice’) in the manner of the examples seen above
in (31), whereas in a numeral compound, it is the initial constituent that is accented (hi to taba,hu ta taba, etc. and hi
to tsubu, hu ta tsubu, mi tsubu / sa n tsubu). Lastly, hako° ‘box’, which is unaccented in its lexical form almost always
yields de accentuation of non numeral compounds, but in numeral compounds, it is pre accenting (hi to hako, hu ta
hako, etc.).
The rules which have just been seen also apply when the specifier is a one or two mora long word of Sino Japanese
origin written using one character, such as ma ku ‘act (in theatre)’ (hito maku,hu ta maku/huta ma ku, mi maku/ sa
n maku…).
When the Yamato specifier is longer than two moras, such as hashira°/hashira (a specifier for gods, lit, ‘pillar’), or too
ri / to ori ‘manner’, another rule applies. The general tendency is then to place the accent on the second mora of the
compound, whatever the length of the numeral element. One thus finds an accent either on the last mora of the initial
component (the numeral), or on the first mora of the final component (the specifier): hito hashira,huta hashira / huta
hashira, mi hashira, yo hashira, itsu hashira, etc. Kasane° ‘layer’, which is atonic in isolation but receives initial accent
on ka in numeral compounding (hito ka sane ‘1 layer’), is an exception. This type of accent assignment, starting from
the beginning of the word, is rather unexpected in Tôkyô Japanese, where the accent is in theory attributed from the
end of the word.
(p.245) In short, the general conclusion is that, when the numeral is Yamato, the accent does not depend on the
intrinsic accent of the specifier except in rare cases. Moreover, atonic numeral compounds are almost non existent. It
should be noted that this is fundamentally different from what is observed in cases of non numeral compound words
with a modifier head structure, as previously presented in section 7.3.1, whose accent depends mainly on the final
component, and among which atonic compounds are legion.
Sino Japanese numerals
We have seen above that in isolation Sino Japanese numerals had the following accent pattern: 1 ichi, 2 ni, 3 san°, 4
Page 41 of 55
Accent
shi, 5 go/go°, 6 roku, 7 shichi, 8 hachi, 9 kyuu/ku/ku°, 10 ju u. Recall that 4 yo n and 7 na na, although of Yamato
origin, have slipped into the so called Sino Japanese series, so that one frequently encounters them in place of 4 shi
and 7 shichi. It is actually not uncommon for the numeral and the specifier to be heterogeneous from the point of view
of the lexical stratum.
If the specifier is Yamato and one or two mora long, the tendency is that the numeral receives an accent (ichi kumi, ni
kumi, san kumi ‘1, 2, 3 groups’, ichi wari, ni wari, san wari ‘1, 2, 3 tenth’), except with gata ‘pattern’, which always
yields an atonic compound (ichi gata°,ni gata°, etc.).
If the specifier is Yamato and longer than two moras, the long compound accent rule applies. An accent is placed on the
initial mora of the final component. The word shiai° ‘match’, atonic in its isolated form, thus produces is shiai,ni shiai,
san shiai, yon shiai…‘1, 2, 3, 4 matches’.
Finally, in cases where both the numeral and the specifier are of Sino Japanese origin, the accent patterns are various
and largely dependent on the specifier, but with many complications or exceptions, of which Table 7.6 gives an
overview. One should not, however, be surprised to encounter, in actual usage, an important variation on this type of
data, even within Tôkyô Japanese. One will find in the appendix of NHK (1998) a more or less complete list of specifier
accents, but the diversity of the actual uses is not always reflected.
All the accent possibilities are represented in Table 7.6: accent on the numeral (with the specifiers ji , dai ); accent
on any of the moras of the specifier (satsu , gatsu ); atonic forms (kai ). The compounds with 3, 4, 5, and 9
sometimes behave differently from the other members in the same paradigm of their series. Some remain atonic when
the rest of the series is tonic (ban , nen ). It is rare for a paradigm to be completely regular, even if the usage of the
younger generations presents a tendency towards paradigmatic regularization within the series.
(p.246)
Table 7.6. Accent of compounds made up of a numeral + Sino-Japanese specifier.

‘hour’ ‘book’ ‘animal’ ‘machine’ ‘time’ ‘floor’ ‘year’ ‘time’
‘object’ ‘long ‘month’ ‘ordinal ‘degree’
object’ n°’
1 ichi ji ik is satsu ip piki ichi dai ip ik kai ik kai° ichi ichi nen ichi ichi do ichi do
ko pon gatsu ban
2 ni ji ni ko ni satsu ni-hiki ni dai ni ni kai ni kai° ni gatsu ni- nen ni- ban ni do ni do
hon / ni
gatsu
3 sa n ji sa n satsu san sa n dai san san kai san sa n san nen° san sa n do sa n do
san ko biki bon kai° / gatsu ban°
san
gai°
4 yo ji yon yo n satsu yon yo n dai / yo n kai / yon shi yo nen° yo n yo n do yo n do
ko hiki yo dai° yon yon kai kai° gatsu / ban / / yo do
hon shi yo
gatsu ban°
5 go ji go go satsu go hiki go dai° go go kai go kai° go go nen° go go do go do
ko hon° / go gatsu ban°
hon
6 roku ji roku satsu / roku dai rok kai rok kai° roku roku roku roku do roku do
rok ko roku satsu rop rop pon / gatsu nen ban
piki / roku hiki roku hon
7 shichi ji nana satsu nana nana dai nana kai nana shichi shichi nana nana nana do
nana ko hiki nana hon / kai° gatsu nen / ban do / / shichi
nana kai nana shichi do
nen do
8 hachi ji has satsu hap hachi dai hachi kai hachi hachi hachi hachi hachi hachi
hachi ko / piki / hachi hiki hap pon / / hak kai kai° / gatsu nen ban do do
hak ko hachi hon hak
kai°
9 ku ji kyu u satsu kyuu kyu u dai / kyu u kyuu ku kyu u kyu u kyu u kyu u
kyuu ko hiki ku dai° kai kai° gatsu nen ban do do / ku
kyuu hon do
Page 42 of 55
Accent
10 ju u ji juk jus satsu / jis ju u dai jup juk kai / juk kai° juu ju u nen ju u ju u do ju u do
ko / jik ko satsu jup piki / pon / jip pon jik kai / jik gatsu ban
jip piki kai°
(p.247) The examples in Table 7.6 illustrate fairly general types.

the specifiers ku (‘ward’), hun (‘minute’), bu (‘part’), sai (‘year old’), gou (‘number’),
byou (‘second’) behave like ji (‘hour’) and ko (‘small object’) in the first column.
soku (‘pair of footwear’), and hatsu (‘shot, round’) behave like satsu (‘book’) and hiki (‘small
animal’) in the second column.
dai (‘ th’, ‘generation’), chou (‘sheet’, ‘block’), chou (‘city’), hai (‘glass’), mai (‘leaf, sheet’),
mei (‘person’) behave like dai (‘machine’) and hon (‘long object’) in the third column.
kan (‘volume’), sen (‘a sen (penny)’), tsuu (‘letter’), ten (‘point’) behave like kai (‘time’) in
the fourth column.
kyuu (‘level’), shuu (‘week’), ren (‘series’), tou (‘class’), bai (‘time’) behave like kai (‘floor’)
in the fifth column.
Lastly, one should not forget that a number of specifiers are of Western origin, for example ki ro ‘kilogram’ or
‘kilometre’, paasento ‘per cent’, burokku ‘block’, shiishii ‘CC’, diikee ‘DK’ {dining kitchen} (a specifier for the type
and room number of flats or houses), to n ‘ton’. They combine in theory with a Sino Japanese numeral. The accent
principles which apply are the same as those of extra long compounds, as presented in section 7.3.1: the accent is
maintained in the specifier on the mora which carries the inherent accent (thus kiroguramu keeps its accent on gu:
ichi kiroguramu, ni kiroguramu; paasento has an accent on se: ichi paasento,ni paasento), except for the specifiers ki
ro and to n, which have the common properties of being short and old loans. Numeral compounds containing these
two elements are accented on the numeral: ichi kiro, ni kiro, san kiro, yon kiro, etc; and i t ton, ni ton, san ton, yon
ton, etc.
Only a very short outline of the complexity of the accent patterns encountered in numeral compounds has been
provided. The principles that govern the accent of this particular type of compound word are basically different from
those that apply elsewhere in the language, in particular in modifier head compounds, except when the specifier is
long and/or of Western origin. It is interesting to note that the prosodic prevalence of the initial component, as
observed on several occasions in numeral compounds, is also found in dvandva formations (shi ro kuro ‘black and
white’, see section 7.3.4) or in mimetic compounds (te ki paki ‘brisk’), which, just like numeral compounds (7.3.5), have
a common property of not being endocentric.
To sum up, the basic generalization that can be drawn from the above can be summarized as follows:
when the numeral is Yamato, with the specifier being Yamato, Sino Japanese, or Western, the numeral
determines the accentuation pattern, except when the (p.248) specifier is a long Yamato lexeme, in which
case the accent is generally assigned from the beginning of the word and falls on the second mora.
when the numeral is Sino Japanese, the accent pattern is determined by the specifier, be it Yamato, Sino
Japanese, or Western, except when the specifier is a short Yamato lexeme.
numeral compounds are de accented when employed as adverbials.
7.4 The Accentuation of Phonological Phrases

Up to this point, we have mainly considered the accent at the level of the prosodic word, consisting in a simplex or
compound lexeme to which suffixes may be attached. We will now examine what occurs when prosodic words are
integrated into units of a higher level. As we shall see, the principles which apply are notably different from those in
compounds.
The prosodic hierarchy of Japanese was introduced in Chapter 6. Recall that at the upper levels, it contains the
following layers:
The minor phonological phrase (or accentual phrase) can contain one or several prosodic words, generally not more
than two. It has at most one pitch accent, and its periphery is marked with an H tone at the beginning and a L% tone at
the end (Pierrehumbert and Beckman, 1988:16; see also section 7.1.5).
Page 43 of 55
Accent
The major phonological phrase (or intermediate phrase in Pierrehumbert and Beckman’s terminology) is the domain of
post accent downstep and catathesis. The presence of an accent triggers catathesis, which lowers everything to the
right of the H accent. An intermediate phrase boundary blocks the effect of catathesis (Pierrehumbert and Beckman,
1988:90).
The general principle is that the major phrase, like the prosodic word, contains at most one HL pitch fall. This HL
sequence corresponds to the projection of an accent at the prosodic word level.
If the prosodic words constituting the minor phrase (or accentual phrase) contain more than one inherent lexical
accent, the principle is as follows: the accent located at the left of the syntagm is preserved and all subsequent accents
are removed. If there is no accent, the whole group remains atonic. The general (p.249) accentuation principles that
have been introduced earlier then apply, namely, the principle of adjacency and that of initial dissimilation. The
difference between the accentuation of accentual phrases and that of compound words is that in the latter it is
generally the accent pattern of the last element that determines the accent of the new form (except in dvandva
compounds, mimetic compounds, and in some numeral compounds), whereas in the former prevalence is given to the
initial accent of the group.
In each of the following examples, two simplex lexemes (possibly followed by suffixes), form a minor phrase. We only
consider here the sequences of L and H corresponding to the mapping of accent patterns in the strict sense, not the
surface intonational structure. Below are the four possible combinations of two lexemes in a phrase: tonic + tonic,
atonic + tonic, tonic + atonic, and atonic + atonic.
(71) Two lexemes domain (minor or accentual phrase)
a. tonic + tonic
utsukushii + hana ga 〉 utsukushii hana ga

LHHHL + LH L 〉 LHHHL LL L
‘beautiful’ + ‘flower Subject’ 〉 ‘beautiful flower Subject’
b. tonic + atonic
utsukushii + sakura° ga 〉 utsukushii sakura ga

LHHHL + LHH H 〉 LHHHL LLL L
‘beautiful’ + ‘cherry tree Subject’ 〉 ‘beautiful cherry tree Subject’
c. atonic + tonic
sakura° ga + utsukushii 〉 sakura ga utsukushii

LHH H + LHHHL 〉 LH H HHHHL
‘cherry tree Subject’ + ‘beautiful’ 〉 ‘the cherry tree is beautiful’
d. atonic + atonic
sakura° ga + saku° 〉 sakura ga saku°

LHH H + LH 〉 LHH H HH
‘cherry Subject’ + ‘bloom’ 〉 ‘the cherry tree blooms’
In (71a), the word hana on the right loses its accent; only the accent of utsukushii is maintained as the accent nucleus
of the phrase. In (71b) and (71c), there is each time only one tonic lexeme, utsukushii, whose accent becomes the
accentual peak of the phrase. In the fourth example (71d), there is no tonic lexeme, and the whole phrase remains
atonic. These principles apply in a similar way to groups containing more than one minor phrase:
(p.250) (72) Major phrase (intermediate phrase)
Page 44 of 55
Accent
1.
2. ‘garden Locative’ + ‘bloom’ + ‘flower Subject’ + ‘beautiful’
3. ‘the flower blooming in the garden is beautiful’
4.
5. ‘garden Locative’ + ‘bloom’ + ‘cherry tree Subject’ + ‘beautiful’
6. ‘the cherry tree blooming in the garden is beautiful’
7.
8. ‘garden Locative’ + ‘beautiful’ + ‘flower Subject’ + ‘bloom’
9. ‘a beautiful flower blooms in the garden’
10.
11. ‘spring Det’ + ‘flower Subject’ + ‘beautiful’
12. ‘spring flowers are beautiful’
Focus displacements may involve accent patterns different from those presented here,17 but we will not go into the
details of what actually relates to (p.251) another issue, that of intonation, which will not be treated in this book.
Interested readers can refer to the studies by McCawley (1968), Higurashi (1983), Poser (1984), Kubozono (1987,
1989), Pierrehumbert and Beckman (1988), Koori (1997, 2003), and to Venditti et al. (2008) for a recent synthesis.
It will be noted, finally, that in the preceding examples (71a, b, and c), the initial LH melody is present only at the
beginning of the utterance, and not at the beginning of each lexeme. The prosodic words saku°, hana ga, sakura ga°,
and utsukushii do not start with a L tone, because they do not appear in initial position of the phrase. On the other
hand, niwa°, the first element of the phonological phrase, has a LH melody (but not haru-no in (71d) since this word is
initially accented). This type of data is confirmation of the analysis by Pierrehumbert and Beckman (1988) presented in
section 7.1.5, which postulates that an initial L tone is a boundary tone attributed at the level of the utterance.
7.5 Dialectal And Sociological Variation In Accent

Accent appears as a key phenomenon for the understanding of both dialect relations and historical linguistics, including
the reconstruction of the Proto Japanese language.
The disparities between the Japanese dialects are huge, to the point that mutual comprehension is not always possible.
The prosodic system constitutes one of the loci of variation, as well as the segmental units, the lexicon, and the syntax.
From the point of view of word prosody, Japanese dialects are traditionally classified into several basic types (see Map
2), generally from three to five: the accentless type (mu akusento アクセント), the one pattern type (ikkei akusento アクセン
ト), the Kagoshima type (sometimes called ‘special type’ tokushu shiki akusento アクセント), the Tôkyô type (tôkyô shiki
akusento アクセント), and the Kyôto Ôsaka type (keihan shiki akusento アクセント).
The accentless and one pattern types are sometimes confused. This is because they are both characterized by the
fact that a constant, recognizable prosody is assigned to phrases. However, whereas this melody fulfils a demarcative
function in the one pattern accentual type for example any word is invariably accentuated on the last mora in Miyako
no jô (Miyazaki prefecture) or on the initial mora (p.252)
Page 45 of 55
Accent
Map 2. Geographical distribution of accent types
in Ôzu (Ehime prefecture) according to Akinaga (1986) it does not do so in the case of the accentless type.
Moreover, native speakers of the latter type are said to be incapable of perceiving it. The accentless type extends
over a vast area north of Tôkyô, from the Ibaraki and Tochigi prefectures as far as Sendai, as well as into certain parts
of the Kumamoto prefecture in Kyûshû.
The Kagoshima type is widespread in Western Kyûshû. It operates a distinction between two types of words, tonic
and atonic. The location of the accent is fixed in tonic words and thus always predictable. For instance, it invariably falls
on the penultimate mora in the dialect of the city of Kagoshima.
(p.253) The Tôkyô type is the type that has been described throughout this chapter: it is a system in which the
prosodic pattern of a word is determined by the presence or absence of an accent and by its location within the word.
The Kyôto Ôsaka type is characterized by the fact that, in addition to one accent nucleus that may be located on any of
the moras of the word, each lexeme carries a register (a tonal melody) that determines the tonal pattern of the moras
occurring before the accent nucleus. Thus, in the Kyôto variety, the principle of initial dissimilation does not exist. Two
mora nouns are accented or unaccented, and there exist two registers: a high register (noted H ), and a low register
or rising register according to certain scholars (noted L ), which affects the beginning of the word. In this system,
there is a three way contrast for one mora words 18 (73), a four way contrast for two mora words (74), a six way
contrast for three mora words (75), a seven way contrast for four mora words, and a nine way contrast for five mora
words. Moreover, in this dialect family, deficient moras except /Q/ have the ability to constitute accent nuclei.19 Most
of the data concerning the Kyôto dialect has been taken from Nakai (2002). All the nouns cited below are followed by
the subject marking enclitic particle ga.
(73) Accent patterns in the Kyôto dialect (Kyôto Ôsaka type): underlyingly monomoraic nouns
a. Words with high register
H kaa ga° HH H /ka/ ‘mosquito’

H hi i ga HL L /hi/ ‘day’
b. Words with low register
Lmee ga° LL H (or LH H in certain areas) /me/ ‘eye’
(74) Accent patterns in the Kyôto dialect: bimoraic nouns
H hana ga° HH H ‘nose’

H na tsu ga HL L ‘summer’
(p.254)
Page 46 of 55
Accent
Lhashi ga° LL H ‘chopstick’

Lame ga LH L ‘rain’
(75) Accent patterns in the Kyôto dialect (Kyôto Ôsaka type): trimoraic nouns
H sakura ga° HHH H ‘cherry tree’

H a tama ga HLL L ‘head’
H niwashi ga HHL L ‘gardener’
Lusagi ga° LLL H ‘rabbit’

Lkabuto ga LHL L ‘helmet’
Ldeppa ga LLH L ‘buckteeth’
In two and three mora low register nouns with final accent (L ame, L deppa), the final mora is articulated with a falling
melody when no particle follows. When uttered in isolation with no enclitic attached to it, thus ame for instance is
generally realized with a falling pitch on the second mora in the Kyôto dialect. Some speakers maintain this falling pitch
even when a particle follows, while most others replace it by a high pitch, and put a low pitch on the particle.
It is to be noted that high register words never have an accent on the final mora, and that low register words never
have an accent on the initial mora.
Another major difference between Tôkyô and Kyôto Ôsaka lies in the fact that it is generally the second noun that
determines the accent pattern of a compound in Tôkyô, whereas it is the first noun that does this in Kyôto Ôsaka.
In view of the various dictionaries written and studies carried out by Japanese dialectologists, one should
acknowledge that the criteria for determining whether a dialect belongs to such or such group are not always explicit.
Other more rigorous and explicit accent classifications have been proposed, in particular that of Uwano (1999). Uwano
(1999) proposes that the Japanese accent systems are first divided into two groups, accented and accentless. The
former is then subdivided into multi pattern accents (the number of accentual distinctions increases in proportion to
the length of the lexeme) and N pattern accents (N oppositions exist, independent of the length of the lexeme), as
follows:
(p.255) (76) Classification of Japanese accent systems according to Uwano (1999)
Accented Multi pattern accent With registers Three registers Ibuki

Two registers Kyôto
Without registers Syntagma dependent20 Shizukuishi (Iwate)
Syntagma independent Tôkyô
N pattern accent Three pattern Oki (Shimane)
Two pattern Kagoshima
One pattern Miyako no jô
Accentless Sendai, Kumamoto
Two striking and apparently contradictory facts characterize cross dialectal accent variation. First, it is the extreme
diversity of the systems in use, going from accentless dialects (Sendai, Kumamoto) to dialects with several tonal
melodies, which look rather more tonal than accentual, and are somewhat reminiscent of some Bantu languages (we
shall return to this issue in section 7.6). Second, the systematic nature of the correspondences that exist among the
various dialects, in spite of the surface differences, are remarkable: a given word class with the same accent pattern in
one dialect will generally correspond to one common other pattern in some other dialect in a consistent manner. Thus,
for instance, LH two mora words of the Tôkyô dialect such as hashi ‘bridge’, hana ‘flower’, or yama ‘mountain’
regularly correspond to initially accented lexemes in the Kyôto dialect (ha shi, ha na, ya ma). The general regularity in
the correspondences between dialects can be taken as evidence that the accent (p.256) systems of present day
Page 47 of 55
Accent
Japanese dialects represent different evolutions from a single, common original system of which they are the
descendants (Hattori, 1951; Kindaichi, 1974, 1977). The prevailing view in Japan is that the Kyôto type has remained
the most faithful to the proto system, but it is sometimes assumed that Tôkyô type dialects have better preserved the
original system (Tokugawa, 1972 cited by Shibatani, 1990:213, Ramsey, 1979. There is also a cautious proposal by
Hattori Shirô according to Ramsey, 1982). Yamaguchi Yukihiro (1998) even assumes that accentless dialects reflect
the most ancient stage of the language (see also the next section).
Following the proposal by Kindaichi Haruhiko (1974), based on philological evidence and modern cross dialectal accent
correspondences, monomoraic nouns of Proto Japanese can be divided into three accent classes, bimoraic nouns into
five classes, and trimoraic nouns into six (or seven) classes. Table 7.7 presents some of the modern correspondences
between the dialects of Tôkyô, Kyôto, Ibuki jima, Ôita, Kagoshima, and Miyako no jô for each of the five bimoraic noun
classes.
In the present day Tôkyô dialect, classes 2 and 3, as well as classes 4 and 5 have merged and now display the same
accent pattern. Class 1 is distinguished from classes 2 and 3, on one hand, and from classes 4 and 5, on the other
hand; there are thus, in Tôkyô, three accent possibilities for two mora nouns. In Kyôto and Ôsaka, classes 2 and 3
have merged, but 1, 4, and 5 remain distinct, hence the four accent patterns nowadays in this dialect. In the Shikoku
Island, which belongs to the Kyôto Ôsaka type, classes 1, 4, and 5 each form a distinct type, while 2 and 3 are
confused; the number of accent constrasts thus amounts to four (this type is not represented in Table 7.7). In Ôita, a
Tôkyô type dialect spoken in Kyûshû, classes 1 and 2 have merged, as well as classes 4 and 5, yielding three distinct
accent patterns, partially different from those that exist in Tôkyô. In
Table 7.7. Cross-dialectal accent correspondences for bimoraic nouns for the five Kindaichi
word classes
Classes Tôkyô Kyôto Ibuki jima Ôita Kago shima Miyako no jô
1 hana ga ‘nose’ LH H LH H HH H HH H LH L LL H
2 natsu ga ‘summer’ LH L HL L HL L
3 inu ga ‘dog’ HH L LH L LL H
4 hashi ga ‘chopstick’ HL L LL H LH H HL L
5 ame ga ‘rain’ LH L LH L
(p.257) Kagoshima (Kagoshima type), only two patterns are attested: accented with an accent kernel on the final
syllable, and unaccented. Finally, in certain dialects, for example that of Sendai in the Miyagi prefecture or in Miyako
no jô, Miyazaki prefecture, the difference between the five classes has been completely neutralized. We therefore find
an accentless dialect in Sendai (not represented in Table 7.7), and a one pattern dialect in Miyako no jô.
For a long time, accentologists have believed that no modern dialect had maintained a five way accent opposition for
two mora nouns, similar to the one reconstructed for Archaic Japanese, until the discovery by Wada Minoru, in the
1960s, of the Ibuki jima dialect, spoken on a small island of the Inland Sea (Kagawa prefecture), in which the five
classes were still preserved in a distinct way. It is the most complex word prosodic system among modern dialects.
The preceding lines offer the merest hint of the richness and complexity of cross dialectal comparison and of their
relevance for Japanese linguistics as a whole, as well as for general linguistics. However, even if, as pointed out by
Matsumori (2003), a colossal amount of descriptive work has been accomplished from the 1930s on in the field of
accentual dialectology, much remains to be done. Indeed, scholars have been mainly concerned with the collection of
data from a comparative point of view, with a special emphasis on short, morphologically simple Yamato words.
However, there is much more to be uncovered concerning the accent of longer words, of compounds, of inflected
forms, or pertaining to the lexical classes of Sino Japanese or Western loans in the various modern dialects.
Within the same dialectal community, accent variation is not rare, as some of the examples given throughout this book
illustrate. A number of words display two or even three different accent patterns which may vary according to the
socio cultural profile of the speaker. Accent dictionaries often provide contradictory information about this sort of
linguistic variation, and the actual observation of the empirical facts sometimes also seems in opposition with current
descriptions. Age seems to be one of the determining factors for such diversity within a single dialect, as the examples
below show. One characteristic of younger people’s speech that is commonly mentioned (and sometimes deplored) in
the literature is the spreading of the atonic pattern in most frequently used words, or in words a speaker feels
familiarity with, as discussed in 7.1.2.
(77) Examples of accent variation
Page 48 of 55
Accent
standard variety innovating varieties

kareshi kareshi° ‘boyfriend’
kuma kuma, kuma° ‘bear’
odoroki, odoroki odoroki° ‘astonishment’
oi°, oi oi, oi° ‘old age’
(p.258)
According to Akinaga (2008), initial accented and final accented patterns are declining in contemporary Japanese; for
instance the words akatonbou ‘dragon fly’ and katana ‘blade’ now tend to be realized as akatonbou and katana.
Uwano (2003) states that two and three mora finally accented words tend to become atonic or initially accented: µµ,
µµµ change to µµ°, µµµ°, or µµ, µµµ. Similarly, middle accented trimoraic words turn into atonic or initial accented:
µµµ 〉 µµµ° or µµµ.
In addition, it is necessary to point out the role of vowel devoicing, which, as mentioned previously, is likely to cause
accent shifts within a word (even if this tendency is not as strong as it used to be in the contemporary language). For
this reason, voiceless vowels constitute a major cause of intra and inter speaker accent variation.
7.6 Tone Or Accent? the Japanese Word-prosodic System From the Typological Point of View
Throughout this book, the term ‘accent’ has been used to refer to the word prosodic system of Japanese, thereby
following the practice of most modern Japanese linguists. However, it is time to reflect more thoroughly upon the
essential nature of the so called Japanese ‘accent’ from the typological point of view.
Traditionally, canonical types of ‘accent’ or ‘stress’ languages such as Spanish, English, or Russian are generally
opposed to ‘tone’ languages like Chinese or Yoruba. Some researchers consider that, typologically, two subsets of
accent languages exist, in opposition to tone languages: stress accent languages with an intensity accent (like English)
and pitch accent languages with a musical accent (or restricted tone languages). Following this point of view, Tôkyô
Japanese is regarded as a prototypical example of a pitch accent language, of which it is undoubtedly the best and most
extensively described type.
The very use of the term ‘accent’ to qualify the word prosodic system of (Tôkyô) Japanese suggests that Japanese
might be typologically closer to the accentual type than to the tonal type. Here, it is essential to note that the fact that
Japanese linguists use the term ‘accent’ (akusento) undoubtedly introduces a bias in the discussion (on the use of the
term ‘accent’ by Japanese philologists of the Meiji Period, see the discussion in the following section). However, if one
abstracts away from the common terminology and starts reflecting upon the very nature of the Tôkyô Japanese word
prosodic system, taking into consideration the intricate picture formed by such a constellation and variety of other
dialects that (p.259) often differ only slightly and in a parametric like manner from their neighbouring dialects, the
issue is far more complicated. So the first question is: typologically speaking, should not some Japanese dialects be
classified as tonal rather than as acccentual? Another, more difficult question is: what are the criteria that will allow us
to determine whether a given dialect is tonal whereas a neighbouring one is accentual within the Japanese linguistic
area?
If, like Odden (1999), who, on this point, reflects a rather traditional position, one considers that there exists, on the
one hand, languages with distinctive intensity stress like English or Russian, and, on the other hand, languages with
tones like Chinese, a language such as Tôkyô Japanese, usually described as a pitch accent language, poses a major
problem in so far as its prosodic characteristics hold, some for stress, others for tone. Thus, like stress, the so called
accent of Tôkyô Japanese is privative in nature (a mora is accented or not); it is culminative (there is at most one pitch
drop per word); it fulfils a demarcative function; its domain is the word, rather than a lower prosodic unit such as the
mora or the syllable, unlike tone; its position may be predictable under certain conditions; it is attributed from one of
the margins of the word (see, inter alia, the rule of antepenultimate accent in foreign loans).
In contrast, just like tone, the so called Japanese accent first manifests itself through variation in the fundamental
frequency; it can spread from one mora to another mora: in fact, it can be considered to spread across all the moras
preceding the HL mark (‘accent’ nucleus) but the first one of a given word; one may consider that there exist latent
marks, for example the fact that an atonic word and a word with a tonic final can only be distinguished when an enclitic
particle is attached to them; the so called accent does not affect the quality of the vowels, whereas stress is known to
sometimes modify the length and quality of the accented vowels, as in English or Russian; a full word may not possess
any ‘accent’ (i.e. a HL shift), something that is inconceivable in a prototypical stress accent language where any lexical
polysyllabic word inevitably has an inherent accent; one also encounters in Japanese mechanisms of prosodic polarity
Page 49 of 55
Accent
(or inversion) typical of tone languages, i.e. a lexeme is assigned a tonal pattern opposite its original pattern
(Maddieson, 1978), becoming atonic if it is tonic, and tonic if it is atonic (see for example the accent of compound
verbs, presented in section 7.3.7).
The categorization of a given Japanese dialect as tonal or accentual thus depends on the importance which one chooses
to attach to each criterion. For several decades, Japanese linguists, taking into account the fact that there exists at
most one mora that is relevant to determine the prosodic pattern of a lexical word in Standard Japanese, have
unanimously been in favour of the accent analysis.
(p.260) However, as Odden (1999) points out, the culminative criterion is not inevitably determining. In the Bantu
family, there exist languages which are classified as tone languages but whose lexemes can have at most a single high
tone. In reality, it should be acknowledged that it is precisely for these types of languages that accent analyses are
sometimes called upon; see the discussion in Creissels (1994) or Yip (2002:260) who assumes that ‘“accentual” is a
convenient descriptive term for a particular type of language in which tone is used in a rather limited way, with one (or
perhaps two) tone melodies … . Such languages occupy a transitional ground between pure stress languages and
pure tone languages’. It is thus all a matter of definition. If one adheres to the views of Hyman (2001), who considers
that a tone language is a language in which an indication of pitch enters the lexical realization of at least certain
morphemes of the language, then, Tôkyô (and Kyôto) Japanese is indisputably to be classified among the tone
languages. On the other hand, if, like Clements and Goldsmith (1984:13), one categorizes as an accent language any
language in which a ‘fixed Basic Tone Melody’ (or more rarely, two) can be specified for all words, this melody being
associated with the accented tone bearing units following the principles of autosegmental phonology (through
association of the melody with the vowel marked as accented in the lexicon, and application of well formedness
conditions), then most Japanese dialects are just as indisputably accentual. But whatever the definition and the
authors that I have just mentioned seem to agree on this point it is manifestly apparent that tone and accent are not
disjoint classes. There is no rigid dichotomy, it is more a matter of gradual transition or parametric specification. Yip
(2002) considers Japanese to occupy an intermediate position between accent languages and tone languages. So does
Labrune (2006). Hyman (2009) also agrees that there is no pitch accent prototype but rather that so called pitch
accent systems freely pick and choose various properties from the tone and stress prototypes. He claims that it is
possible to define ‘tone’ and ‘stress’ but that it is impossible to provide an independent definition of ‘pitch accent’. In
his 2001 paper and others, Larry Hyman actually categorizes Japanese as hybrid. This linguist indeed adopts a
parametric approach of the prosodic phenomena. For him, there exist only two prototypes: the tonal type and the
intensity accent type (stress). Each type is characterized by a set of distinct properties. Following this point of view,
so called pitch accent languages represent a mixed type, with prototypical properties that are sometimes typical of
purely tonal systems, sometimes of stress systems. So there is no pitch accent prototype. I shall now adopt this stand.
Moreover, Hyman argues that prosodic systems cannot be treated as a continuum placed along a single dimensional
scale. Hyman thus takes the rather extreme view that no language must be analysed with a pitch accent system
because a tonal analysis is always possible.
(p.261) Cases of evolution of a tonal system towards an accentual system have been reported. Chen (2000)
describes the example of the Chongming dialect (Northern Wu dialect, China), and Goldsmith (1984), that of Tonga
(Bantu). As for the other direction, that is, the shift from a stress system to a system which can be regarded as tonal,
Swedish or Serbo Croat are well known cases. This shows that the difference between accent and tone is not clear cut
and dichotomous but that it is either gradual or parametric (I am not sure that the two configurations are as different
as Hyman claims because it seems to me that gradation can be obtained by means of a parametric specification, but
this discussion is far beyond the scope of this book).
In the case of Japanese, such a non dichotomous, gradual, or parametric approach has as an advantage that the
various Japanese dialects would not be classified in typologically disjoint classes. And it is useful, indeed, to be able to
account for the linguistic continuity 21 that one observes between dialects such as Sendai Japanese (accentless),
Kagoshima Japanese (one pattern accent), and Ibuki jima Japanese (with a five way contrast), which are all closely
genetically linked. One cannot of course be satisfied with a classification which would call the first ‘an accentless
language’ like French or Indonesian, the second ‘an accent language’, a little like Spanish or Italian, and the third a
‘tone language’, like a certain number of Bantu languages. If one language Archaic Japanese can develop in a rather
short period of time (on the scale of language evolution) and on a small, naturally well delimited geographical surface
so many prosodic types seemingly as divergent as those presented by the Japanese dialects, one must posit that there
exists a typological continuity between them. Moreover, as Yamaguchi Yukihiro (1998) and other linguists observe,
their geographic distribution obeys a concentric, circular model: the prosodically most complex systems lie at the
centre, the simplest ones (the accentless pattern and one pattern accent) are found at the periphery in Kyûshû and in
Northern Kantô, that is, in areas that are not contiguous but on the contrary very remote from each other, and even
at the two extreme points of the Japanese linguistic domain. It is thus in a gradual and continuous mode, by successive
Page 50 of 55
Accent
transitions or waves, that one passes from one type to another by spreading outwards, following a centre periphery
model. On the basis of this observation, Yamaguchi Yukihiro (1998) makes the assumption that accentless dialects
actually reflect historically older forms of the language (the language of the Jômon period?), whereas those (p.262) of
central Japan, in particular the more prestigious dialect of the oldest Nara and Heian documents, would have
undergone more recent changes. This quite unorthodox view of central dialects as being more recent than peripheral
ones actually seems to date back to the work by the great folklorist Yanagita Kunio (1875 1962; see Ramsey, 1982).
Two geographically close dialects generally differ only by tiny amounts; but from dialect to dialect, the accumulation of
such tiny differences ends up producing extremely different types. Thus there is not, within the Japanese area, any
real linguistic break from the point of view of word prosodic systems. Any theory of word prosodic systems must be
able to account for this situation. This is why a parametric conception, or property driven approach such as the one
that Hyman (2009) calls for, seems better fitted to capturing the diversity of the Japanese prosodic subtypes and to
understanding their relationships, even if much work remains to be done in order to modelize the prosodic variety
found across the Japanese linguistic domain.
7.7 An Overview of Accent Studies In Japan

The preceding reflections quite naturally lead us to look back at the history of accent studies in Japan, because such an
epistemological reflection sheds novel and interesting light on the current debates concerning the typological
categorization of Japanese from the prosodic point of view. It will also provide us with an opportunity to mention some
essential past references within the discipline.
Accentology constitutes, within Japanese linguistics, one of the key disciplines, and a very significant number of works,
in their majority of a descriptive nature, have been devoted to the study of accent in Japan.
Interest in the prosodic phenomena goes back a long way. For instance, the Kojiki (712 AD), one of the first
documents to be compiled in the Japanese language, already provides some information on accent. A considerable
number of texts of all epochs and of various types also give rough accent information, or sometimes more detailed
remarks and analyses on the prosody of Japanese (for a critical presentation of these documents, see Kindaichi, 1974).
Leading kokugaku scholars of the Edo period (thinkers and philologists of the ‘national studies stream’) such as Keichû
(1640 1701), Motoori Norinaga (1730 1801), and Fujitani Nariakira (1738 1779), to name but a few, have shown
interest in accent and paved the way for prosodic studies. These precursors have left analyses of unquestionable
interest.
Pre Meiji Restoration scholars use the traditional term of Chinese phonetics (read as sei, shou, or kowe in Japanese)
to describe the prosodic system of Japanese words. A majority of them distinguish three different ‘tonalities’ (p.263)
(my translation for ) in the native words (and not four as in Chinese), which are generally (high), (mid, level),
(falling). This recourse to the descriptive categories of the Chinese language does not seem to have posed any
problem. It should be noted that before them, the Iberian missionaries João Rodriguez and Diego Collado, authors of
remarkable works of description of the Japanese language in the seventeenth century, had also distinguished three
different registers. This is due to the fact that the language described by all these authors is mainly that of the Kansai
area (Kyôto, Ôsaka) in which, as stated earlier, there exists a falling pitch at the phonetic level, in addition to the high
and low pitches. Moreover, it is interesting to observe that ancient scholars of the Japanese language, native and
foreigners alike, were all struck by the fact that the same lexeme undergoes drastic modification in the prosodic
pattern when used in a compound, an issue that is still of major concern in current research.
The term akusento, borrowed from English, has been in use since the Meiji period, but it should be noted that it is
also occasionally employed, including at the present time, to refer to the tones of Chinese. As stated earlier, the term
usually employed by Japanese scholars of the Edo period to talk about the Japanese word prosody was sei / shô /
kowe . It would be interesting to know more about the conditions which have governed the terminological change
from sei (shô / kowe) ‘tone’ to akusento ‘accent’, and the reasons which made Japanese philologists prefer to talk
about ‘akusento’ (accent) rather than ‘tone’ in the Meiji period for the description and analysis of their native language.
The term akusento already figures as an entry in the 1875 edition of the Daigenkai dictionary, the Japanese language
dictionary of the Meiji era. From what we know of the history of the linguistic and conceptual loans in Japan, one can
suspect that the choice of the term ‘accent’ may have had something to do with extra linguistic considerations. It is not
impossible that ‘accent’ might have been considered more prestigious than ‘tone’, at a time when Japan was overtly
looking to the West as a civilization model, and trying to enter the circle of the nations that were considered more
‘civilized’ by Meiji thinkers and elites, i.e. Western countries (in accordance with the ‘leaving Asia’ datsu a ron ideology
advocated by Fukuzawa Yukichi). The most prestigious European languages being accent languages, the Meiji
scholars may have sought to differentiate their native language, and hence their nation, from other Asian and African
languages, which were labelled as tone languages. Of course, the typical tone language that Japanese philologists of the
Page 51 of 55
Accent
Meiji period were most familiar with was not Bantu but Chinese, a typologically very distant language with regard to
prosody. They were undoubtedly struck by the prosodic differences which distinguish Chinese from Tôkyô Japanese.
This also undoubtedly justified their preference for the term ‘accent’. However, and to put it in a nutshell, the choice
of the term akusento ‘accent’ might not have been motivated by entirely (p.264) linguistic reasons. This is a
terminological, sociological, and historical problem that deserves further investigation.
Furthermore, and rather surprisingly, the word ‘accent’ was adapted as it is in katakana form as akusento, whereas
the nineteenth century practice was, notoriously, and especially in linguistics, to create semantic calques using Chinese
characters. For instance meishi mei ‘name’ + shi ‘word’ was coined on the Dutch term naam woord. According to
Howland (2002), it is especially after 1880 that the use of Western katakana words becomes widespread. For
instance, he reports a tripling of loanwords in newspaper articles between 1879 and 1887. Since akusento アクセント
(‘accent’) already figures as an entry in the 1875 Daigenkai dictionary, this term appears as a very early borrowing,
anterior to the explosion of katakana loanwords in Japanese.
In such a context, it is not illegitimate to wonder about the extent to which the preference for the term akusento might
well have conditioned the categorization of Japanese among the accent languages rather than among tone languages
from the Meiji period onwards and oriented the subsequent vision and research on Japanese prosody (see the
discussion in the preceding section), whereas many languages whose prosodic systems look typologically close to the
Japanese systems are regarded as tone languages, for instance a number of Bantu languages, and of American native
languages.
The first modern linguistic study of the prosodic system of Tôkyô Japanese was published in 1892 by Yamada Bimyô,
to whom we owe the discovery that only two pitches are relevant in lexemes of the Tôkyô dialect. This position will
nevertheless be challenged later on, at the beginning of the twentieth century, by Sakuma Kanae and Jinbô Kaku.
Yamada uses the term onchô to refer to accent, a term which also has in Japanese the meaning of intonation.
At the beginning of the twentieth century, Sakuma Kanae’s and Jinbô Kaku’s phonetically oriented work, based on
instrumental analyses, dominate the field of Japanese accentology. For these two linguists, the Japanese accent consists
of a pitch with three distinctive tones: high, low, and medium. This approach was criticized by Miyata Kôichi (1927),
who, like Yamada Bimyô, recognizes only two relevant levels of height, a vision which is largely dominant nowadays.
Observe in passing that since Meiji the issue of Japanese accent has been the subject of much dispute and
controversy. The analyses and therefore the linguists who stand behind them have often conflicted: Sakuma Kanae
vs. Yamada Bimyô at the beginning of the twentieth century, Sakuma Kanae vs. Miyata Kôichi, Hattori Shirô vs.
Kindaichi Haruhiko, and Sibata Takesi vs. Kindaichi Haruhiko, in the 1950s are among the best known instances of such
academic strife.
Meanwhile, prosodic descriptions of the various Japanese dialects were developing. Let us mention the works by
Hattori Shirô, Kindaichi Haruhiko, Hirayama (p.265) Teruo, Sibata Takesi, and Willem A. Grootaers (inter alia), as well
as the historical studies based on the analysis of old materials by Kindaichi Haruhiko, Komatsu Hideo, and Akinaga
Kazue.
In a 1954 paper, Hattori Shirô clarified the phonological status of the Japanese accent by making a distinction between
phonetically oriented descriptions (such as those by Sakuma or Jinbô) and phonologically oriented ones. Hattori thus
opened the path towards more formal treatments, cast in various theoretical frameworks, which all have in common
that they only specify, at the most abstract lexical level (in Tôkyô Japanese), one mora at most in a word as the location
of the accent kernel.
Hattori (1954) defines the notion of ‘accent nucleus’ (akusento kaku アクセント ) as the mora after which a prosodic fall
occurs, and poses the question of the Japanese accent in the following terms: ‘does a given word have an accent
nucleus? If it does, on which mora?’. This treatment, which consists in making a distinction between the abstract H
tone marking the lexical accent (at most one within a word) and the H and L tones which are attributed at the phonetic
level, thus seems more abstract compared to that of Kindaichi (1960) for example, who posits the existence of two
tonemes (chôso ) high and low. According to Kindaichi’s analysis, each mora of a lexical word is underlyingly
specified at the phonological level as high or low (Sakuma and Jinbô had already adopted a comparable view, but with
three tonemes: high, low, medium), whereas Hattori argues that only one mora should be specified. Hattori’s
approach thus authorizes a truly phonological approach to the problem, even if as Kindaichi Haruhiko has legitimately
observed his analysis does not easily allow a unified treatment of some other dialects with prosodic systems more
complex than that of Tôkyô.
The year 1958 witnesses the publication of the first modern accent dictionary, the Meikai Nihongo Akusento Jiten (The
Meikai Dictionary of Japanese Accent), under the direction of Kindaichi Haruhiko and Akinaga Kazue. This work will be
Page 52 of 55
Accent
continuously reedited, the last edition to date being that of 2010. Actually, Kindaichi and Akinaga are also the authors
of the other accent dictionary, the Nihongo Hatsuon Akusento Jiten (A Dictionary of Accent and Pronunciation of
Japanese), published by the Nippon Hôsô Kyôkai (NHK), the Japanese Broadcasting Corporation.
Another important step was taken by Kawakami (1957, 1961) who proposes to consider that the initial LH sequence
occurring in a word is attributed at the level of the accent phrase ( ku) and not at that of the lexical word, an idea that
will be taken up by Pierrehumbert and Beckman (1988), among others (whereas for Hattori, the initial LH is regarded
as a non distinctive property of the lexical accent).
(p.266) More recently, Uwano Zendô has proposed in a 1999 article a new typological classification of the Japanese
prosodic systems (see section 7.5). Let us mention also Hayata Teruhiro (1999) whose research from the typological
and contrastive point of view is based on his own analyses of the prosodic systems of various languages of Asia,
including Japanese dialects. Hayata has also worked out a new proposal for the classification of the prosodic systems of
the world’s languages. For the Japanese dialects, Hayata (1999:195) makes an interesting proposal. Roughly speaking,
for him, the dialects of Eastern Japan (Tôkyô type) are accentual in nature, whereas the Kagoshima type (located in
Western Japan) is basically tonal (onchô ). For Hayata, tonal refers to the following property: the location of the pitch
within the word is not relevant; the only relevant distinction is whether or not there is going to be an HL pitch within
the word, which for Hayata makes this type comparable to a tone language, with the difference that in traditional tone
languages such as Mandarin Chinese, the tonal unit is the syllable, whereas it is the word in Kagoshima Japanese. The
interesting point in Hayata’s theory is that he claims that the Kyôto Ôsaka accent type can be analysed as a
combination of the Tôkyô type (accentual) with the Kagoshima type (tonal). So Kyôto Ôsaka would be both tonal and
accentual, and as he persuasively argues, the Kyôto Ôsaka type is located in the centre of Japan, in the zone where
the tonal (Kagoshima/Western) and accentual (Tôkyô/Eastern) zones meet.
Finally, over the past two decades or so, the work by Kubozono Haruo has brought major theoretical contributions
to the understanding of the Japanese accent system, especially with regard to the accent of compounds and that of
Western loanwords.
The overall picture that emerges from this rapid overview is that accent studies in Japan have followed an original
development path, and that the studies which have been devoted to the field are remarkable both in quantity and
quality.
Notes:
(1 ) In certain studies, the accent is noted by the sign placed after the accented mora: namida = naˡmida.
(2) Words beginning with a putative ‘heavy syllable’ (in our approach, a full mora followed by a deficient mora) are
sometimes described (Hattori, 1954; Haraguchi, 1977; Vance, 1987:80) as not obeying the principle of initial
dissimilation. However, this issue is controversial (see also 6.2.4). Haraguchi (1977:34 35) and Tanaka (2008: 210)
further claim that initial dissimilation does not occur when the second mora is Q, so gakkou° /gaQkoR/ ‘school’ is
realized as LLHH. It does not occur either when the second mora is an onsetless vowel, including /i/.
(3) Following the common practice of autosegmental phonology, Haraguchi uses a star to denote the vowels carrying
the accent. For convenience sake, we shall continue to mark the accented moras in bold characters, and to add the
symbol ° to unaccented words.
(4) The lexemes tomodachi°,kudamono,mizuumi, ko omori, and imouto in Table 7.2 are actually compounds from the
point of view of etymology. However, they are no longer perceived as such by modern speakers, which is why they
are cited here. Most Yamato words longer than three moras are actually etymologically compound words.
(5) The analysis presented here is synchronic and description oriented, so any element likely to comute with another
verbal or adjectival suffix is regarded as a verbal or adjectival suffix. For instance, the i type adjective ureshii ‘happy’
is decomposed as ureshi (base) + i (suffix), despite the fact that shi is also a suffix from the point of view of
etymology. In the same manner, a tsukereba ‘if it is hot’ is decomposed as atsu + kereba, and not as atsu kere ba or
atsu k er eba as a diachronic analysis would suggest.
(6) Smith (1998) proposes an OT account of the Japanese accent according to which nouns and verbs/ i adjectives
receive the same type of underlying representation (input). This constitutes a definite advantage over McCawley or
Poser’s analyses. However, in her model, the problem is actually displaced from the representations to the
constraints. It is now the constraints that have to operate a distinction between nouns and verbs/adjectives. Some
constraints are supposed to apply only to nouns. Kubozono (2008) also hints towards a similar approach by arguing
that Tôkyô Japanese has a two accent system for nouns, just as it does for verbs and adjectives. Nouns would only
Page 53 of 55
Accent
differ from verbs and adjectives to the extent to which they admit lexical exceptions (no lexical exceptions exist for
verbs and adjectives, but many do for nouns).
(7) Note that the NADM principle applies to /N/, /R/ and /i/, and not to the other final possible moras /ku/, /tu/, /ki/, or /ti/
which are not considered as epenthetic from a synchronic point of view (see section 2.5).
(8) As previously mentioned, all epenthetic vowels in Western loans do not have the same status; some behave like full
vowels, others like empty vowels.
(9) In contrast, toponyms such as ankara° ‘Ankara’, kinshasa° ‘Kinshasa’, or aidaho° ‘Idaho’ with a special mora in the
second position are generally atonic, exactly like words containing a succession of four regular moras such as
himaraya° ‘the Himalayas’ or minesota° ‘Minnesota’. See Kubozono (1996), who shows that if a quadrimoraic loan
ends in two ‘light syllables’ (i.e. for us, two full moras) and its final vowel is not epenthetic, the word will be atonic in
Japanese.
(10) But, whatever the analysis one adopts, some exceptions remain. If one puts aside, on the one hand, atonic words
whose very existence constitutes a particularly problematic issue in Japanese phonology, and, on the other hand,
lexemes which have kept in Japanese the original position of the accent in the source language (for example in te
rijensu {intelligence}, kyan pe en {campaign}, de za in {design}, ra iburarii {library}, etc.), it is also necessary to take
into account a small number of forms which, to our knowledge, have remained unaccounted for. These forms are all
trimoraic, and they comprise one or two deficient moras, like words of the type ku rasu {class}, do rama {drama}, su
riru {thrill}, ku ran {clan}, with initial accent, which contrast with the type gu re e {grey}, hu ro a {floor}, su ta a
{star}, pu ra u {plow}, with penultimate accent (Kubozono, 1996; Tanomura, 1999; Tanaka, 2008). All these words
share the common characteristic of starting with a mora containing an epenthetic vowel. However, the latter group,
ending in a long vowel or a sequence of two different vowels, carries the accent on the penult, and is thus
distinguished from the former group, which displays an antepenultimate accent. Such a difference shows that, as
claimed in Chapter 6, all deficient moras should not be treated as equal, and that there exists a gradation in mora
deficientness. Some moras are ‘more deficient’ than others.
Moreover, words such as su ki n {skin}, su pa na {spanner}, su pe ru {spell}, which begin with the /su/ sequence
(/u/ is epenthetic) followed by a consonant other than /r/, all bear a penultimate accent, just as the words whose
second mora consists of a single vowel: i e su {yes}, tsu i n {twin}. I leave these issues open for future research.
(11 ) Kubozono and Fujiura (2004) have also shown that 96% of compounds with an atonic C1 are atonic, whereas only
61% are when the C1 is tonic.
(12) Recall that the Sino Japanese morpheme hu u is considered as carrying an accent on its final mora underlyingly
(/huR/). It is phonetically realized with an initial accent because of the application of the NADM principle.
(13) As already stated, kokoro and nokogiri also admit final accentuation (kokoro,nokogiri). Under this condition, the
compounds onna gokoro and denki nokogiri can be regarded as resorting to the cases examined in (34c).
(14) Such formations can be compared to neo classical compounds in European languages, such as geology or
laryngectomy.
(15) According to Nakada and Hayashi (1982:394), Sino Japanese has lost all connection with the original Chinese tones,
but little is known about the development and attribution of accent in Sino Japanese and more study is needed on this
issue.
(16) M represents a full mora, m a deficient mora, and µ any mora.
(17) Thus, in the statement (72d), focus on hana would prevent deletion of the word accent: haru no hana-ga
utsukushii° ‘SPRING FLOWERS are beautiful’.
(18) Note that one mora words are phonetically realized as two moras in this dialect family. So all the monomoraic
nouns of Tôkyô Japanese are realized with prosodic lengthening: hi = hii ‘fire’, na = naa ‘name’, etc.
(19) For descriptions and analyses of accent of the Kyôto and Ôsaka dialects, see Kindaichi (1974, 1977), McCawley
(1977), Haraguchi (1977, 1999), Pierrehumbert and Beckman (1988), Nakai (2001, 2002), Sibata (1955).
Nakai (2001, 2003) provides a presentation of the historical evolution of the Kyôto accent from the Heian period up
until the present day which is both all embracing and well documented.
Page 54 of 55
Accent
(20) In Uwano’s account, ‘syntagma dependent’ refers to the following property: the fall in pitch is a syntactic marker
indicating that the word or syntagma is phrase final. It does not appear otherwise. So in a dialect like that of
Shizukuishi, the fall is not a property of the word but of the phrase final position. As Uwano puts it, it is a mirror image
of the Tôkyô dialect (a syntagma independent dialect): in the Shizukuishi dialect, the rise in pitch indicates the accent
kernel, and the fall marks phrase finality, while in Tôkyô, the fall is the accent kernel, and the rise functions as a phrase
initiality marker.
(21 ) Such linguistic continuity is reminiscent of what one encounters in the Bantu area. Creissels (1994) notes that the
Eastern Bantu languages offer particularly rich material for the study of the various stages of evolution of a tonal
prosodic system towards an accent system, as is illustrated by the Swahili case, a language in which the position of the
word accent is phonologically predictable. Clements and Goldsmith (1984:13) make a similar remark concerning the
typological variety found throughout the Bantu area.
References

Laurence Labrune

Print ISBN-13: 9780199545834
DOI: 10.1093/acprof:oso/9780199545834.001.0001
(p.267) References
Bibliography references:
Akamatsu, Tsutomu (1997) Japanese Phonetics, Theory and Practice, München: Lincom
Europa.
Akinaga, Kazue (1967) ‘Edo akusento kara Tôkyô akusento he [From Edo accent to Tôkyô
accent]’, Kokubungaku Kenkyû 16, reprinted in Akinaga, 1999, 33–52.
—— (1968) ‘Iwayuru tokushu onsetsu (tokushu haku) ni tsuite [About the so-called
special syllables (special moras)]’, Kôza nihongo kyôiku 4: 36–51.
—— (1986) ‘Akusento gaisetsu [Overview of accent]’, in Kôza Hôgengaku 1, Hôgen

gaisetsu, Tôkyô: Kokusho Kankôkai, 97–115.
—— (1999) Tôkyôben akusento no hen’yô [Accentual changes in the Tôkyô dialect],

Tôkyô: Kasama Shoin.
—— (2002) ‘Tôkyôgo no hatsuon to yure [Variation in the pronunciation of Tôkyô

Japanese]’, in Hida Y. and Satô T. (eds.), Gendai Nihongo Kôza 3, Tôkyô: Meiji Shoin, 40–
58.
Page 1 of 25
References
—— (2008) ‚ ‘Tôkyô-ben no 20 seiki (hatsuon ni tsuite [The Tôkyô dialect in the twentieth-
century (pronunciation)]’, Nihongo no Kenkyû 4-1: 211–213.
Alderete, John and Kochetov, Alexei (2009) ‘Japanese mimetic palatalisation revisited:
implications for conflicting directionality’, Phonology 26: 369–388.
Amanuma, Yasushi, Ootsubo, Kazuo, and Mizutani, Osamu (1989) Nihongo Onseigaku
[The phonetics of Japanese] (revised edition), Tôkyô: Kurosio.
Arisaka, Hideyo (1936) ‘Jôdai ni okeru sa-gyô no on’in [The phonology of the s-row sound
in Archaic Japanese]’, reprinted in Arisaka, 1969, 145–159.
—— (1940) On’inron [Phonology], Tôkyô: Sanseidô.
—— (1955) Jôdai on’in kô [Remarks on archaic phonology], Tôkyô: Sanseidô.
—— (1969) Kokugo on’inshi no kenkyû [Research in the history of Japanese phonology],

Tôkyô: Sanseidô.
Aston, William George (1872) Grammar of the Japanese Written Language (1st edition),
London: Luzac and Co., and Yokohama: Lane, Crawford, and Co.
—— (1904) Grammar of the Japanese Written Language (3rd edition), London: Luzac
and Co., and Yokohama: Lane, Crawford, and Co.
Auer, Peter (1994) ‘Einige Argumente gegen die Silbe als universale prosodische
Hauptkategorie [Some arguments against the syllable as a universal major prosodic
category]’, in Ramers, K. H., Vater H., and Wode H. (eds.), Universale Phonologische
Strukturen und Prozesse, Niemeyer: Tübingen, 55-78.
Ayusawa, Takako, Kitazawa, Shigeyoshi, and Toki, Satoshi (1998) ‘A study of Japanese
rhythm by means of TEMAX’, Onsei Kenkyû / Journal of the Phonetic Society of Japan 2–
1: 34–40.
Beckman, Jill N. (1999) Positional Faithfulness: An Optimality Theoretic Treatment of

Phonological Asymmetries, New York and London: Garland Publishing.
Beckman, Mary (1996) ‘When is a syllable not a syllable?’, in Otake, T. and Cutler, A.
(eds.), Phonological Structure and Language Processing, Cross-Linguistic Studies,
Berlin/New York: Mouton de Gruyter, 95–123.
—— and Shoji, Atsuko (1984) ‘Spectral and perceptual evidence for CV coarticulation in
devoiced /si/ and /syu/ in Japanese’, Phonetica 41: 61–71.
Burzio, Luigi (1994) Principles of English Stress (Cambridge Studies in Linguistics 72),
Cambridge: Cambridge University Press.
Campbell, Nick (1999) ‘A study of Japanese speech timing from the syllable perspective’,
Page 2 of 25
References
Journal of the Phonetic Society of Japan 3–2: 29-39.
—— and Sagisaka, Yoshinori (1991) ‘Onsei taimingu ni mirareru môra to onsetsu no eikyô’
[Moraic and syllable-level effects on speech timing], Denshi jôhô tsûshin gakkai onsei
kenkyû/IEICJ Technical Report SP90–107, 35–40.
Chen, Matthew Y. (2000) ‘From tone to accent’, Cahiers de Linguistique d’Asie Orientale,
vol. 29–2, 179–203.
Chomsky, Noam and Halle, Morris (1968) The Sound Pattern of English, New York:
Harper and Row.
Clements, George N. and Goldsmith, John (1984) ‘Introduction’, in Clements, N. and

Goldsmith, J. (eds.), Autosegmental Studies in Bantu Tone, Dortrecht: Foris, 1–17.
—— and Hume, Elisabeth (1995) ‘The internal organization of speech sounds’, in

Goldsmith, J. (ed.), 1995, 245–306.
Coleman, John (1998) Phonological Representations, their Names, Forms and Powers,
Cambridge: Cambridge University Press.
Collado, Diego (1632) Ars Grammaticae Iaponicae Lingua, Rome.
Coyaud, Maurice (1985) L’ambiguïté en japonais écrit [Ambiguity in written Japanese],

Paris: PAF.
Creissels, Denis (1994) Aperçu sur les structures phonologiques des langues négro-
africaines [Overview on the phonological structures of Negro-African languages] (2nd
edition), Grenoble: Ellug (Stendhal University).
Daniels, F. J. (1958) The Sound System of Standard Japanese, Tôkyô: Kenkyûsha.
Davis, Stuart (1988) ‘Syllable onsets as a factor in stress rules’, Phonology 5: 1–19.
De Chene, Brent (1985) ‘r-epenthesis and the Japanese verb’, Papers in Japanese
Linguistics 10: 170–207.
Everett, Daniel L. and Everett, Karen (1984) ‘On the relevance of syllable onsets to
stress placement’, Linguistic Inquiry 15. 705–711.
Faber, Alice and Vance, Timothy J. (2000) ‘More acoustic traces of “deleted” vowels in
Japanese’, in Nakayama, Mineharu and Charles, J. Quinn, Jr, Japanese Korean Linguistics,
vol. 9, CSLI Publications, Stanford, California, 100–113.
Frellesvig, Bjarke (1995) A Case Study in Diachronic Phonology, The Japanese Onbin
Sound Changes, Aarhus: Aarhus University Press.
Fukuda, Suzy E. and Fukuda, Shinji (1999) ‘The operation of rendaku in the Japanese
Page 3 of 25
References
specifically language-impaired: a preliminary investigation’, Folia Phoniatrica et

Logopaedica 51: 36–54.
Goldsmith, John (1976) Autosegmental Phonology, PhD thesis, MIT.
—— (1984) ‘Tone and accent in Tonga’, in Clements, N. and Goldsmith, J. (eds.),

Autosegmental Studies in Bantu Tone, Dortrecht: Foris, 19–51.
—— (1990) Autosegmental and Metrical Phonology, Oxford: Blackwell.
—— (ed.) (1995) The Handbook of Phonological Theory, Cambridge, MA and Oxford:

Blackwell.
Golston, Chris and van der Hulst, Harry (1999) ‘Stricture is structure’, in Hermans, B.
and van Oostendorp, M. (eds.), The Derivational Residue in Phonological Optimality
Theory, Amsterdam: John Benjamins, 153–173.
Hagège, Claude and Haudricourt, André (1978) La phonologie panchronique

[Panchronic phonology], Paris: Presses Universitaires de France.
Hamada, Atsushi (1949) ‘Sokuon to hatsuon [gemination and mora nasal]’, Jinbun Kenkyû
11–12: 91–114.
—— (1951) ‘Chôon [long vowels]’, Jinbun Kenkyû 2–5 and 6, reprinted in Hamada, 1986,
90–144.
—— (1952) ‘Hatsuon to dakuon to no sôkansei no mondai, kodaigo ni okeru dakushiin no

onka [The problem of the correlation between the mora nasal and voicing, the sound
value of voiced obstruents in the ancient language]’, Koku Kokubun 21–3: 18–32.
—— (1954) ‘Hagyôon no mae no sokuon – p-on no hassei [Gemination before h-sounds –

the development of p]’, Kokugogaku 16, reprinted in Hamada, 1983, 71–80.
—— (1960) ‘Rendaku to renjô [rendaku and renjô]’, reprinted in Hamada, 1984, 89–109.
—— (1964) ‘Yôon [palatalization]’, Kokugo Kokubun 33–5, reprinted in Hamada, 1984, 56–
71.
—— (1983) Zoku Chôsen shiryô ni yoru Nihongo kenkyû [Research in Japanese

linguistics on the basis of Korean materials, 2], Kyôto: Rinsen Shoten.
—— (1984) Nihongo no shiteki kenkyû [Research on the history of Japanese], Kyôto:

Rinsen Shoten.
—— (1986) Kokugoshi no shomondai [Some problems in the history of Japanese], Tôkyô:

Waizumi Shoin.
Hamano, Shoko (1998) The Sound Symbolic System of Japanese,Studies in Japanese
Page 4 of 25
References
Linguistics, Stanford and Tôkyô: CSLI Publications and Kurosio.
—— (2000) ‘Voicing of obstruents in Old Japanese: evidence from the sound symbolic
stratum’, Journal of East Asian Linguistics 9–3: 207–225.
Han, Mieko (1994) ‘Acoustic manifestations of mora timing in Japanese’, Journal of the
Acoustical Society of America 96: 73–82.
Haraguchi, Shôsuke (1977) The Tone Pattern of Japanese: an Autosegmental Theory of

Tonology, Tôkyô: Kaitakusha.
—— (1991) A Theory of Stress and Accent, Dordrecht: Foris Publications.
—— (1999) ‘Accent’, in The Handbook of Japanese Linguistics, Tsujimura, N. (ed.),

Oxford: Blackwell, 1–30.
—— (2003) ‘The phonology–phonetics interface and syllable theory’, in van de Weijer, J. et

al. (eds.), The Phonological Spectrum, vol. 2, Amsterdam: John Benjamins, 31–58.
Hashimoto, Mantarô (1977) ‘On’in no taikei to kôzô [Phonemic structure and system]’, in
Iwanami Kôza Nihongo 5, Tôkyô: Iwanami Shoten, 1–29.
Hashimoto, Shinkichi (1928) ‘Ha-gyô shiin no hensen ni tsuite [About the change of the h-
sound]’, reprinted in Hashimoto, S., 1950, 29–45.
—— (1950) Kokugo on’in no kenkyû [Research in the phonology of Japanese] Tôkyô:

Iwanami Shoten.
Hattori, Shirô (1951) ‘Genshi nihongo no akusento [The accent of prehistorical Japanese]’,
in Terakawa, K. (ed.), Kokugo akusento ronsô, Tôkyô: Hôsei Daigaku Shuppan, 45–65.
—— (1954) ‘On’inron kara mita kokugo no akusento [The Japanese accent from the point
of view of phonology]’, Kokugo Kenkyû 2–3, reprinted in Sibata et al., 1980, 364–403.
—— (1955) ‘On’inron [Phonology]’, Kokugogaku 22, reprinted in Sibata et al., 1980, 157–
185.
—— (1960) Gengogaku no hôhô [Methods of linguistics], Tôkyô: Iwanami Shoten.
—— (1961) ‘Prosodeme, syllable structure and laryngeal phonemes’, Bulletin of the

Summer Institute of Linguistics 1, reprinted in Sibata et al. (1980), 186–218.
—— (1976) ‘Jôdai nihongo no boin taikei to boin chôwa [The vowel system of Archaic
Japanese and vowel harmony]’, Gengo 6: 2–14.
Hayashi, Chikafumi (1992) ‘Ha-gyô ten’on ha naze Heian jidai ni okotta ka [Why did the
change of the h-row occur in the Heian period]’, Kokugo to Kokubungaku 69–11: 110–
119.
Page 5 of 25
References
Hayashi, Ooki (ed.) (1982) Zusetsu Nihongo [Japanese through graphs], Tôkyô: Kadokawa
Shoten.
Hayata, Teruhiro (1980) ‘Nihongo no onseimen de no shôryaku [Phonetic elision in

Japanese]’, Gengo Seikatsu 339: 48–54.
—— (1999) Onchô no taiporoji [A prosodic typology], Tôkyô: Taishukan.
Hayes, Bruce (1989) ‘Compensatory lengthening in moraic phonology’, Linguistic Inquiry

20: 253–306.
—— (1995) Metrical Stress Theory – Principles and Case Studies, Chicago: University of
Chicago Press.
—— (2009) Introductory Phonology, Madden, MA: Wiley-Blackwell.
Hibiya, Junko (1999) ‘Variationist Sociolinguistics’, in Tsujimura, N. (ed.), The Handbook

of Japanese Linguistics, Oxford: Blackwell, 101–120.
Higurashi, Yoshiko (1983) The Accent of Extended Word Structures in Tokyo Japanese,
Tôkyô: Educa Inc.
Hirano, Takanori (1974) ‘Rendaku no kisokusei to kigen [The origin and regularity of
rendaku]’, Bungaku Kenkyû 71–3: 21–43.
Hirayama, Teruo et al. (eds.) (1993) Gendai Nihongo Hôgen Daijiten [Great Dictionary of
Modern Japanese Dialects], 9 vols., Tôkyô: Meiji Shoin.
Hoffmann, J. J. (1868) A Japanese Grammar, Leiden: A. W. Sythoff.
Homma, Yayoi (1981) ‘Durational relationships between Japanese stops and vowels’,
Journal of Phonetics 9: 273–281.
Howland, Douglas R. (2002) Translating the West: Language and Political Reason in
Nineteenth-Century Japan, Honolulu: University of Hawai’i Press.
Hyman, Larry (2001) ‘Tone Systems’, in Haspelmath, M. et al. (eds.), Language Typology
and Language Universals, vol. 2, Berlin: Walter de Gruyter, 1367–1380.
—— (2003)[1985], A Theory of Phonological Weight, Dordrecht: Foris.
—— (2009) ‘How (not) to do phonological typology: the case of pitch accent’, Language
Sciences 31–2/3: 213–238.
Idemaru, Kaori (2005) An Acoustic and Perceptual Investigation of the Geminate and
Singleton Stop Contrast in Japanese, PhD thesis, University of Oregon.
Imada, Shigeko (1981) Hatsuon [Pronunciation],Kyôshiyô nihongo kyôiku handobukku 6,
Page 6 of 25
References
Tôkyô: Kokusai Kôryû Kikin.
Imae, Kuniharu (1960) ‘Nihongo no digram no sôtai hindo to sono tokusei [Frequency
correlations and characteristics of Japanese digrams]’, Shinrigaku Hyôron 4–1.
Inoue, Fumio (1971) ‘Ga-gyô shiin no bunpu to rekishi [The history and distribution of
/g/]’ Kokugogaku 86: 26–41.
—— (1993) ‘Shiin no hatsuon no henka [Changes in the pronunciation of consonants]’, in

Miyaji, S. et al. (eds.), Kôza nihongo to nihongo kyôiku, 2 Nihongo no onsei – on’in (1),
Tôkyô: Meiji Shoin, 109–134.
—— (2002) ‘Seiyôgo no hatsuon no eikyô [The influence of Western words on

pronunciation]’, in Hida, Y. and Satô, T. (eds.), Gendai Nihongo Kôza 3, Tôkyô: Meiji Shoin,
59–84.
Irwin, Mark (2009) ‘Prosodic size and rendaku immunity’, Journal of East Asian
Linguistics 18–3: 179–196.
Itô, Junko (1988) Syllable Theory in Prosodic Phonology, New York: Garland Publishing.
—— (1990) ‘Prosodic minimality in Japanese’, Chicago Linguistic Society 26, vol. 2, Deaton,
K. et al. (eds.), Parasession on the Syllable in Phonetics and Phonology, Chicago: Chicago
Linguistic Society, 213–239.
—— and Mester, Armin (1986) ‘The phonology of voicing in Japanese: theoretical

consequences for morphological accessability’, Linguistic Inquiry 17: 49–73.
—— —— (1995a) ‘Japanese phonology’, in Goldsmith, J. (ed.), The Handbook of

Phonological Theory, Cambridge and Oxford: Blackwell, 817–838.
—— —— (1995b) ‘The core-periphery structure of the lexicon and constraints on

reranking’, in Beckman, J. N., Walsh Dickey, L., and Urbanczyk, S. (eds.), Papers in
Optimality Theory, GLSA, University of Massachusetts, Amherst, 181–209.
—— —— (1996) ‘Stem and word in Sino-Japanese’, in Otake, T. and Cutler, A. (eds.),
Phonological Structure and Language Processing, Cross-linguistic Studies, Berlin/New
York: Mouton de Gruyter, 13–44.
—— —— (1997) ‘Correspondance and compositionality: the ga-gyô variation in Japanese

phonology’, in Roca, I. (ed.), Derivations and Constraints in Phonology, Oxford: Oxford
University Press, 419–462.
—— —— (1999) ‘The phonological lexicon’, in Tsujimura, N. (ed.), The Handbook of

Japanese Linguistics, Cambridge, MA: Blackwell, 62–100.
—— —— (2003) Japanese Morphophonemics: Markedness and Word Structure,

Cambridge: MIT Press.
Page 7 of 25
References
—— —— (2008) ‘Rhythmic and interface categories in prosody’, paper presented at The
18th Japanese/Korean Linguistics Conference, November 13–15, CUNY (available at ).
—— —— and Padgett, Jaye (1995) ‘Licensing and underspecification in Optimality

Theory’, Linguistic Inquiry 17: 49–73.
Itô, Junko and Mester, Armin (1999) ‘Lexical classes in Japanese: a reply to Rice’,
Phonology at Santa Cruz 6: 39–46.
Jidai Betsu Kokugo Daijiten, Jôdai-hen [Great Dictionary of Archaic Japanese] (1967),
Tôkyô: Sanseidô.
Jinbô, Kaku (1928) ‘Kokugo no onseijô no tokushitsu [Phonetic particularities of

Japanese]’, Kokugo to Kokubungaku 4-4, reprinted in Sibata et al., 1980, 5–15.
Jôo, Hakutarô (1977) ‘Gendai nihongo no on’in [The phonemes of contemporary

Japanese]’, in Ôno, S. and Sibata, T. (eds.), Iwanami Kôza Nihongo 5 On’in, Tôkyô:
Iwanami shoten, 107–145.
Kager, René (1997) ‘Rhythmic vowel deletion in Optimality Theory’, in Roca, I. (ed.),
Derivations and Constraints in Phonology, Oxford: Clarendon Press, 463–499.
—— (1999) Optimality Theory, Cambridge Textbooks in Linguistics, Cambridge:

Cambridge University Press.
Kageyama, Tarô (1976/1977) ‘Incorporation in Sino-Japanese verbs’, Papers in Japanese

Linguistics 5: 117–156.
Kaji, Shigeki (ed.) (1999) Cross-linguistic Studies of Tonal Phenomena: Tonogenesis,

Typology and Related Topics (Proceedings of the Symposium), Institute for the Study of
Languages and Cultures of Asia and Africa, Tôkyô University of Foreign Studies, Tôkyô.
—— (ed.) (2001) Cross-linguistic Studies of Tonal Phenomena: Tonogenesis, Japanese

Accentology and Other Topics (Proceedings of the Symposium), Institute for the Study of
Languages and Cultures of Asia and Africa, Tôkyô University of Foreign Studies, Tôkyô.
Kamei, Takashi (1956) ‘Ga-gyô no kana [The g-row kana]’, Kokugo to Kokubungaku,
reprinted in Kamei, 1984, 1–25.
—— (1970) ‘Suzume siu siu [‘Siu siu’ says the sparrow]’, reprinted in Kamei, 1984, 447–
464.
—— (1984) Nihongo no sugata to kokoro 1 [The figure and heart (= form and meaning)
of the Japanese language], Tôkyô: Yoshikawa Kôbunkan.
—— (1997) ‘Nihongo no rekishi [History of the Japanese language]’, in Kamei, T., Kôno,
R., and Chino, E. (eds.), Nihon rettô no kotoba, Tôkyô: Sanseidô, 107–153.
Page 8 of 25
References
Kano, Yoshimitsu (1998) Kanji no naritachi jiten [Dictionary of kanji etymology], Tôkyô:
Tôkyôdô.
Karlgren, Bernhard (1957) Grammata serica recensa (reprinted, SMC, Taiwan, 1996).
Katada, Fusa (1990) ‘On the representation of moras: evidence from a language game’,
Linguistic Inquiry 21: 641–646.
Katayama, Motoko (1995) ‘Loanword accent and minimal reranking in Japanese’,

Phonology at Santa Cruz, vol. 4, 1995, 1–12.
Kawagoe, Itsue and Arai, Masako (2002) ‘Shakuyôgo ni okeru sokuon [Consonant
gemination in loanwords]’, Onsei Kenkyû (Journal of the Phonetic Society of Japan) 6–1:
53–66.
Kawahara, Shigeto (2006) ‘A faithfulness ranking projected from a perceptability scale: the
case of [+voice] in Japanese’, Language 82–3: 536–574.
—— (2007) ‘Half rhymes in Japanese rap lyrics and knowledge of similarity’, Journal of
East Asian Linguistics 16: 113–144.
—— Nishimura, Kohei, and Ono, Hajime (2003) ‘Unveiling the unmarkedness of Sino-
Japanese’, in McClure, William (ed.), Japanese Korean Linguistics vol. 12, CSLI
Publications, Stanford, California, 140–151.
Kawakami, Shin (1957) ‘Jun akusento ni tsuite [About the derived accent]’, Kokugo
Kenkyû 7: 44–60, reproduced in Kawakami, 1995, 92–113.
—— (1961) ‘Kotoba no kireme to onchô [The end of words and intonation]’, Kokugakuin
Zasshi 62–5, 67–75, reproduced in Kawakami, 1995, 130–142.
—— (1977) Nihongo onsei gaisetsu [Overview of Japanese phonetics], Tôkyô: Ôfûsha.
—— (1995) Nihongo Akusento Ronshû [Papers on the Japanese Accent], Tôkyô: Kyûko
Shoin.
Kelly, Michael H. (2004) ‘Word onset patterns and lexical stress in English’, Journal of
Memory and Language 50: 231–244.
Kenstowicz, Michael (1994a) Phonology in Generative Grammar, Oxford: Blackwell.
—— (1994b) ‘Sonority-driven stress’, ms., Rutgers Optimality Archives.
Kess, Joseph F. and Miyamoto Tadao (1999) The Japanese Mental Lexicon,
Psycholinguistic Studies of Kana and Kanji Processing, Amsterdam: John Benjamins.
Kindaichi, Haruhiko (1942) ‘Gagyô bion-ron [The velar nasal]’, reprinted in Kindaichi,
1972a, 168–197.
Page 9 of 25
References
—— (1950) ‘Satooya to satôya [The words satooya ‘foster parent’ and satôya ‘sugar
shop’]’, reprinted in Kindaichi, 1972a, 133–153.
—— (1960) ‘Nihongo no akusento no tokushitsu [The particularity of the Japanese

accent]’, reprinted in Kindaichi, 1972a, 198–230.
—— (1972a) Nihongo on’in no kenkyû [Research on Japanese phonology], Tôkyô:

Tôkyôdô Shuppan.
—— (1972b) ‘Onsetsu, môra oyobi haku [Syllable, mora, and beat]’, reprinted in
Kindaichi, 1972a, 58–77.
—— (1974) Kokugo akusento no shiteki kenkyû [Historical research on Japanese

accent], Tôkyô: Hanawa Shobô.
—— (1976) ‘Rendaku no kai [Remarks on rendaku]’, Sophia Linguistica 2: 1–22.
—— (1977) ‘Akusento no bunpu to henkô [Accentual change and distribution]’, in

Iwanami Kôza Nihongo 11 Hôgen, Ôno, S. and Shibata, T. (eds.), Tôkyô: Iwanami shoten,
129–180.
—— and Akinaga, Kazue (eds.) (2001) Shinmeikai Nihongo Akusento Jiten [Shinmeikai
dictionary of Japanese accent], Tôkyô: Shinmeikai.
—— and Maës, Hubert (1978) Phonologie du japonais standard [Phonology of standard

Japanese],Travaux de linguistique japonaise, vol. V, Paris: Université Paris 7.
Kishida, Takeo (1984) Kokugo on’in henka no kenkyû [Research on phonological changes
in Japanese], Tôkyô: Musashino Shoin.
KKK, see Kokuritsu Kokugo Kenkyûjo (National Institute for Japanese Language and
Linguistics).
Kobayashi, Akemi (1981) ‘Ennin no kijutsu suru sansukuritto onsetsu ca no onka [The
phonetic value of the Sanscrit ca syllable as observed by Ennin]’, Ôsaka gaikokugo
Daigaku Gakuhô 52: 63–80.
Kokugo Gakkai (The Society for Japanese Linguistics) (ed.) (1992) Kokugo-shi shiryô-shû
[A Collection of Materials for the History of Japanese], Tôkyô: Musashino Shoin.
Kokuritsu Kokugo Kenkyûjo (1964) Gendai zasshi 90-shu no yôgo yôji daisan bunsatsu
[The Vocabulary and Chinese Characters in Ninety Contemporary Magazines, vol. 3],
Kokuritsu Kokugo Kenkyûjo hôkoku 25, Tôkyô: Hideo Shuppan.
—— (1966) Nihon Gengo Chizu / Linguistic Atlas of Japan, Kokuritsu Kokugo Kenkyûjo,
Tôkyô. Maps available at the following address: .
—— (1984) Goi no kenkyû to kyôiku (1) [Research and Teaching in Lexicology] (Nihongo
Page 10 of 25
References
Kyôiku Shidô Sankôsho 12), Tôkyô: Kokuritsu Kokugo Kenkyûjo.
—— (2004) Nihongo hanashi kotoba corpus [The corpus of spontaneous Japanese],
Komatsu, Hideo (1981) Nihongo no on’in [Japanese Phonology] (Nihongo no Sekai 7),
Tôkyô: Chûô Kôronsha.
Kondo, Mariko (1997) Mechanisms of Vowel Devoicing in Japanese, PhD thesis,

University of Edinburgh.
—— (2000) ‘Vowel devoicing and syllable structure in Japanese’, in Nakayama, Mineharu

and Quinn, Charles J. Jr, Japanese Korean Linguistics vol. 9, Stanford, California: CSLI
Publications, 125–138.
—— (2003) ‘Speech rhythm and consonant sequence production in Japanese’, in

Proceedings of the 6th International Seminar on Speech Production, Sydney, 138–143.
Koori, Shirô (1997) ‘Nihongo no intonêshon, kata to kinô’ [Patterns and function of
Japanese intonation], in Nihongo no onsei 2. Akusento, intonêshon, rizumu to pôzu,
Sugito, M. (ed.), Tôkyô: Sanseidô, 169–202.
—— (2003) ‘Intonêshon [Intonation]’, in Kitahara, Yasuo and Uwano, Zendô (eds.),

Asakura Nihongo Kôza 3, Tôkyô: Asakura Shoten, 109–131.
Kubozono, Haruo (1985) ‘Speech errors and syllable structure’, Linguistics and
Philology 6: 220–243.
—— (1987) The Organisation of Japanese Prosody, PhD thesis, University of Edinburgh.
—— (1989) ‘The mora and syllable structure in Japanese: evidence from speech errors’,
Language and Speech 32–3: 249–278.
—— (1993a) ‘Kodomo no shiritori to môra no shûtoku [Children’s shiritori and the

acquisition of the mora]’ in Nihongo no môra to onsetsu kôzô ni kansuru sôgôteki kenkyû
(2), Research Report, University of Tsukuba, 130–137.
—— (1993b) The Organization of Japanese Prosody, Tôkyô: Kurosio.
—— (1994) ‘Nihongo no onsetsuryô ni tsuite [On syllable weight in Japanese]’,

Kokugogaku 178: 7–17.
—— (1995a) Gokeisei to on’in kôzô [Word formation and phonological structure], Tôkyô:
Kurosio.
—— (1995b) ‘Constraint interaction in Japanese phonology: evidence from compound

accent’, Phonology at Santa Cruz 4: 21–38.
—— (1995c) ‘Perceptual evidence for the mora in Japanese’, in Cornell, B. and Arvaniti, A.
Page 11 of 25
References
(eds.), Papers in Laboratory Phonology IV, Cambridge: Cambridge University Press,

141–156.
—— (1996) ‘Syllable and accent in Japanese: evidence from loanword accentuation’, Onsei
Kaihô 211: 71–82.
—— (1997) ‘Lexical markedness and variation: a nonderivational account of Japanese

compound accent’, in Proceedings of the 15th West Coast Conference on Formal
Linguistics, Agbayani, B. and Tang, S-W. (eds.), Stanford, California: CSLI publications,
273–287.
—— (1998a) ‘Akusento kenkyû no topikkusu [Topics in accentual research]’, Nihongogaku

17–3: 4–13.
—— (1998b) ‘Môra to onsetsu no huhensei [On the universality of mora and syllable]’,
Onsei Kenkyû 2–1: 5–15.
—— (1999a) ‘Mora and Syllable’, in The Handbook of Japanese Linguistics, Tsujimura, N.

(ed.), Oxford: Blackwell, 31–61.
—— (1999b) Nihongo no Onsei [Japanese Phonetics], (Gendai Gengogaku Nyûmon),

Tôkyô: Iwanami.
—— (2000) ‘Nihongo no goi to onsetsu kôzô – jo(u)o(u) ha joô ka jôô ka [The Japanese
lexicon and syllable structure – should jo(u)ou (‘queen’) be pronounced joô or jôô ?’]’,
Nihongo Kenkyû 20: 1–18.
—— (2001a) ‘Comments on “Privatives Tones” in Bantu’, in Proceedings of the

Symposium Cross-Linguistic Studies of Tonal Phenomena: Tonogenesis, Japanese
Accentology, and Other Topics, Kaji, S. (ed.), 2001, Tôkyô: Tôkyô University of Foreign
Studies, 259–265.
—— (2001b) ‘Epenthetic vowels and accent in Japanese: facts and paradoxes’, in Issues in
Japanese Phonology and Morphology, van de Weijer, J. and Nishihara, T. (eds.), Berlin:
Mouton de Gruyter, 111–140.
—— (2003) ‘The syllable as a unit of prosodic organisation in Japanese’, in The Syllable in

Optimality Theory, Féry, C. and van de Vijver, R. (eds.), Cambridge: Cambridge
University Press, 99–122.
—— (2005) ‘Rendaku: its domain and linguistic conditions’, in van de Weijer, J., Nanjo, K.
and Nishihara, T. (eds.), Voicing in Japanese (Studies in Generative Grammar 84),
Berlin/New York: Mouton de Gruyter, 5–24.
—— (2006a) ‘Phonetic and phonological organization of speech in Japanese’, in Nakayama,

Mineharu, Mazuka, Reiko, and Shirai, Yasuhiro (eds.), The Handbook of East Asian
Psycholinguistics vol. II: Japanese, Cambridge: Cambridge University Press, 191–207.
Page 12 of 25
References
—— (2006b) ‘Where does loanword prosody come from? A case study of Japanese
loanword accent’, Lingua 116–7: 1140–1170.
—— (2008) ‘Japanese accent’, in Miyagawa, S. and Saitô, M. (eds.) The Oxford Handbook
of Japanese Linguistics, Oxford: Oxford University Press, 165–192.
—— and Fujiura, Yayoi (2004) ‘Morpheme-dependent nature of compound accent in

Japanese: an analysis of “short” compounds’, in On’in Kenkyû 7, The Phonological Society
of Japan (ed.), Tôkyô: Kaitakusha, 9–16.
—— Itô, Junko, and Mester, Armin (1997) ‘On’in kôzô kara mita go to ku no gyôkai:
hukugô meishi akusento no bunseki’ [The phonological boundary between word and
phrase: an analysis of Japanese compound noun accentuation], in Sugitô, M. (ed.), Bunpô
to onsei 1, Tôkyô: Kurosio, 147–166.
Kuginuki, Tooru (1982) ‘Jôdai Nihongo ra-gyôon-kô [Remarks on r in Archaic Japanese]’,

Toyama Daigaku Jinbungakubu Kiyô 6: 192–206.
Kurematsu, Akira (1997) ‘Inritsu jôhô no jôhô kagaku he no tekiyô [Application of

intonational information to information science]’, in Sugitô, Miyoko (ed.), Nihongo Onsei
(2), akusento, intonêshon, rizumu to pôzu, Tôkyô: Sanseidô, 303–318.
Kurisu, Kazutaka (2000) ‘Richness of the base and root fusion in Sino-Japanese’, Journal
of East Asian Linguistics 9–2: 147–185.
—— (2001) The Phonology of Morpheme Realization. Doctoral dissertation of the

University of California, Santa Cruz.
Kuroda, Shigeyuki (1967) ‘Sokuon oyobi hatsuon ni tsuite [On gemination and nasal
mora]’, Gengo Kenkyû 51: 85–99.
Labrune, Laurence (1993) ‘À propos d’un trait typologique du japonais: l’absence de r à

l’initiale des mots indépendants de Yamato kotoba [About a typlogical feature of Japanese:
the absence of r at the beginning of Yamato words]’, Ebisu-Études Japonaises 2: 7–21.
—— (1996) ‘La formation des désinences verbales en japonais ancien [The formation of
verbal ending in Ancient Japanese]’, Cipango Cahiers d’Études Japonaises 5: 141–171.
—— (1998a) ‘Les consonnes voisées (dakuon) dans la phonologie du japonais et l’histoire

de leur représentation graphique [Voiced consonants (dakuon) in Japanese phonology
and the history of their representation in writing]’, Japon pluriel 2, Arles: Philippe
Picquier, 295–304.
—— (1998b) ‘De l’iconicité en japonais: suffixes en r + voyelle et réduplication [On

iconicity in Japanese: r + vowel suffixes and reduplication]’,Cahiers de Linguistique Asie
Orientale 27–1: 79–118.
Page 13 of 25
References
—— (1999) ‘Variation intra- et inter-langue: morpho-phonologie du rendaku en japonais et

du sai-sios en coréen’ [Intra- and inter-language variation: the morphophonemics of
Japanese rendaku and Korean sai-sios], Cahiers de Grammaire 24: 117–152.
—— (2002) ‘The prosodic structure of simple abbreviated loanwords in Japanese: a

constraint-based account’, Onsei Kenkyû, Journal of the Phonetic Society of Japan 6–1:
98–120.
—— (2006) La phonologie du japonais [Japanese phonology], Société de Linguistique de

Paris (collection Linguistique n° 90), Leuven/Paris: Peeters.
—— (2007) ‘Morphophonologie des emprunts abrégés en japonais: aspects prosodiques

et mélodiques’ [The prosodic and melodic aspects in the morphophonology of truncated
loanwords in Japanese], in Des sons et des sens, données et modèles en phonologie et en
morphologie, Delais-Roussarie, E. and Labrune, L. (eds.), Paris: Hermès, 173–198.
—— (2012) ‘Questioning the universality of the syllable: evidence from Japanese’,

Phonology 29.1.
—— (forthcoming) ‘The phonology of Japanese /r/: a panchronic account’.
Ladefoged, Peter (1982) A Course in Phonetics, San Diego: Harcourt Brace Jovanovich.
—— (1990) ‘Some Proposals Concerning Glottal Consonants’, Journal of the International

Phonetic Association vol. 20, no. 2, 24–26.
Lawrence, Wayne P. (1999) ‘Ha-gyôon no mae no sokuon – gendaigo ni okeru /Qh/

[Geminate h in modern Japanese]’, Kokugogaku 199: 16–27.
Loveday, Leo J. (1996) Language Contact in Japan, a Sociolinguistic History, Oxford:

Clarendon Press.
Lowenstamm, Jean (1996) ‘CV as the only syllable type’, in Current Trends in Phonology,
vol. 2, Laks, B. and Durand, J. (eds.), Paris: CNRS / ESRI / University Paris X, 419–441.
Lyman, Benjamin (1894) ‘Change from surd to sonant in Japanese compounds’, Oriental
Studies, Oriental Club of Philadelphia, 1–17.
Mabuchi, Kazuo (1959) ‘Jôdai – chûko ni okeru sa-gyô tôon no onka [The phonetic value
of s in tôon in Archaic and Ancient Japanese]’, Kokugo to Kokubungaku 36–1: 60–67.
—— (1971) Kokugo On’inron [Japanese Phonology], Tôkyô: Kasama Shoin.
Maddieson, Ian (1978) ‘Universals of tone’, in Universals of Human Language vol. 2

Phonology, J. Greenberg (ed.), Stanford: Stanford University Press, 335–365.
—— (1984) Patterns of Sounds, Cambridge: Cambridge University Press.
Page 14 of 25
References
—— (1985) ‘Phonetic cues to syllabification’, in Fromkin, Victoria (ed.), Phonetic

Linguistics, Essays in Honor of Peter Ladefoged, Orlando, CA: Academic Press, 203–221.
Maekawa, Kikuo (1993) ‘Boin no museika [Vowel devoicing]’, in Miyaji, S. et al. (eds.),
Kôza nihongo to nihongo kyôiku, 2 Nihongo no onsei – on’in (1), Tôkyô: Meiji Shoin, 135–
153.
Martin, Samuel (1952) Morphophonemics of Standard Colloquial Japanese (Language

Dissertation 47), Baltimore: The Linguistic Society of America.
—— (1966) ‘Lexical evidence relating Korean to Japanese’, Language 12–2: 185–251.
—— (1987) The Japanese Language through Time, New Haven and London: Yale
University Press.
Mase, Yoshio and Satô, Hirokazu (1993) ‘Tôkyôgo akusento no tayôsei [Variety of Tôkyô
accent]’, in Miyaji, S. et al. (eds.), Kôza nihongo to nihongo kyôiku, 2 Nihongo no onsei –
on’in (1), Tôkyô: Meiji Shoin, 206–232.
Matsumori, Akiko (2003) ‘Akusento kenkyû no dôkô to tenbô 2 (Gendaigo chûshin)

[Trends and perspective in accentual research 2 (Modern Japanese)]’, in Kitahara, Yasuo
(ed.), Asakura Nihongo Kôza 3, Tôkyô: Asakura Shoten, 261–277.
Matsumoto, Katsumi (1975) ‘Kodai nihongo boin soshiki kô [Notes on the vowel system of
archaic Japanese]’, Kanazawa daigaku hôbungaku-bu ronbun (bungaku-hen) 22: 83–152.
Matsuno, Kazuhiko (1971) ‘Some allophones of Japanese /r/ and their stylistic value’,
Onseigakkai Kaihô 136: 5–8.
McCarthy, John and Alan, Prince (1994) ‘The emergence of the unmarked: optimality in
prosodic morphology’, in Gonzalez, M. (ed.), Proceedings of the North East Linguistic
Society 24: 333–379.
—— —— (1995) ‘Faithfulness and reduplicative identity’, in Beckman, J.N., Dickey, L. W.

and Urbanczyk, S. (eds.), Papers in Optimality Theory (University of Massachusetts
Occasional Papers 18), Amherst: GLSA. 249–384.
McCawley, James (1968) The Phonological Component of a Grammar of Japanese, The

Hague: Mouton.
—— (1977) ‘Accent in Japanese’, in Hyman, L. (ed.), Studies in Stress and Accent, Los
Angeles, University of Southern California: SCOPIL, 261–302.
Mehler, Jacques, Dommergues, Jean-Yves, Frauenfelder, Ulrich, and Segui, Juan (1981)
‘The syllable’s role in speech segmentation’, Journal of Verbal Learning and Verbal
Behavior 20: 298–305.
Mester, Armin and Itô, Junko (1989) ‘Feature predictability and underspecification:
Page 15 of 25
References
palatal prosody in Japanese mimetics’, Language 65: 258–293.
Miyake, Takeo (1943) Kokumin gakkô akusento kaisetsu daiichigakunen yô [Comments

on accent for first year level in public schools], Tôkyô: Kokugo Bunka Kenkyûjo.
Miyata, Kôichi (1927) ‘Atarashii akusento-kan to akusento hyôkihô [New perspectives on

accent and accent transcription]’, Onsei no Kenkyû 1, reprinted in Sibata, T. et al., 1980,
327–332.
Morais, José (1994) L’Art de lire [The Art of Reading], Paris: Odile Jacob.
Morishige, Satoshi (1975) ‘Jôdai tokushu kanadukai to ha nani ka [The archaic “special
kana” writing system]’, Man’yô 89: 1–47.
Morita, Takeru (1977) ‘Nippo Jisho ni mieru goon renketsujô no ichikeikô [About a
tendency of sound combination in the Japanese–Portuguese Dictionary]’, Kokugogaku
108–3: 20–32.
Nakada, Norio and Hayashi, Chikahumi (1982) Nihon no Kanji [Sinograms in Japan],
(Nihongo no Sekai 4), Tôkyô: Chûô Kôronsha (re-edited 2000, Chûkô bunkô).
Nakai, Yukihiko (2001) ‘The History of the Kyôto Accent’, in Cross-linguistic Studies of
Tonal Phenomena: Tonogenesis, Japanese Accentology and Other Topics (Proceedings of
the Symposium), Kaji, S. (ed.), Tôkyô: Institute for the Study of Languages and Cultures
of Asia and Africa, Tôkyô University of Foreign Studies, 129–143.
—— (2002) Keihankei Akusento Jiten [Accent Dictionary of the Kyôto–Osaka Dialects],

Tôkyô: Benseisha.
—— (2003) ‘Akusento no Henkô [Accent change]’, in Asakura Nihongo Kôza 3, Kitahara,

Yasuo and Uwano, Zendô (eds.), 2001, Tôkyô: Asakura Shoten, 85–108.
Nasu, Akio (1996) ‘Niji kango ni okeru sokuonka genshô – saitekisei riron ni yoru bunseki
[Gemination in Sino-Japanese binomials, an OT analysis]’, Onsei Gakkai Kaihô 213: 27–39.
Nespor, Marina and Vogel, Irene (1986) Prosodic Phonology, Dordrecht: Foris.
NHK (Nippon Hôsô Kyôkai) (1985) Onsei gaido [A Phonetic Guide], Tôkyô: Nippon Hôsô
Shuppan Kyôkai.
—— (1998) Nihongo hatsuon akusento jiten [Dictionary of Japanese Pronunciation and

Accent], NHK Hôsô Bunka Kenkyûjo (ed.), Tôkyô: Nihon Hôsô Shuppan Kyôkai.
Odden, David (1999) ‘Typological issues in tone and stress in Bantu’, in Proceedings of
the Symposium. Cross-Linguistic Studies of Tonal Phenomena: Tonogenesis, Typology,
and Related Topics, Kaji, S. (ed.), 1999, Tôkyô: Tôkyô University of Foreign Studies, 187–
215.
Page 16 of 25
References
Ogasawara, Naomi and Warner, Natasha (2009) ‘Processing missing vowels: Allophonic
processing in Japanese’, Language and Cognitive Processes 24: 376–411.
Ogawa, Shinji (2003) ‘Nihongo (Tôkyô hôgen) no 3 môra meishi akusento [The accent of 3-
mora nouns in Tôkyô Japanese]’, paper presented at the 28th Workshop of the Kansai
Linguistic Society, Kôbe.
—— (2004) ‘Sino-Japanese word accent and syllable structure’, in The Phonological

Society of Japan (ed.), On’in Kenkyû / Phonological Studies 7, Tôkyô: Kaitakusha, 41–48.
—— (2008) ‘Accentual asymmetry in trimoraic Sino-Japanese words’, in Kubozono, H.

(ed.) Asymmetries in Phonology, Tôkyô: Kurosio, 183–191.
Ohnishi, Masayuki (1987) ‘Dôteki jinkô kôgai ni yoru nihongo shiin no kôsatsu [A study of
Japanese consonants by dynamic palatography ]’, Onsei Gakkai Kaihô 186: 19–24.
Ohno, Kazutoshi (2000) ‘The lexical nature of rendaku in Japanese’, in Nakayama,

Mineharu and Quinn, Charles J., Jr, Japanese Korean Linguistics vol. 9, Stanford,
California: CSLI Publications, 151–164.
—— (2005) ‘Sei-daku: diachronic developments in the writing system’, in van de Weijer, J.

Nanjo, K. and Nishihara, T. (eds.), Voicing in Japanese (Studies in Generative Grammar
84), Berlin/New York: Mouton de Gruyter, 47–69.
Okumura, Mitsuo (1972) ‘Kodai no on’in [The phonemes of Archaic Japanese]’, in Nakada,
Norio (ed.), Kôza kokugoshi 2, on’in – moji, Tôkyô: Taishûkan.
—— (1980) ‘Rendaku’, in Kokugo Gakkai (1980), 925–926.
Ôno, Susumu (1980) Nihongo no seiritsu [The development of Japanese] (Nihongo no

Sekai 1), Tôkyô: Chûô Kôronsha.
Ota, Mitsuhiko (2003) The development of prosodic structure in early words (Language
Acquisition and Language Disorders 34), Amsterdam/Philadelphia: John Benjamins.
—— (2004) ‘The learnability of a stratified lexicon’, Journal of Japanese Linguistics 20:
19–40.
Otaka, Hiromi (2009) Phonetics and Phonology of Moras, Feet and Geminate Consonants
in Japanese, Lanham, MD: University Press of America.
Otake, Takashi and Cutler, Anne (1996) Phonological Structure and Language
Processing, Cross Linguistic Studies, Berlin/New York: Mouton de Gruyter.
—— Hatano, Giyô, Cutler, Anne, and Mehler, Jacques (1993) ‘Mora or syllable: speech
segmentation in Japanese’, Journal of Memory and Language 32: 358–378.
Otsu, Yukio (1980) ‘Some aspects of rendaku in Japanese and related problems’,
Page 17 of 25
References
Theoretical Issues in Japanese Linguistics, MIT Working Papers in Linguistics 2, 207–

227.
Pierrehumbert, Janet and Beckman, Mary (1988) Japanese Tone Structure, Cambridge,
MA and London: MIT Press.
Piggott, Glyne (1992) ‘Variability in feature dependency: the case of nasality’, Natural
Language and Linguistic Theory 10: 33–78.
Poser, William (1984) The Phonetics and Phonology of Tone and Intonation in Japanese,
PhD thesis, MIT.
—— (1990) ‘Evidence for foot structure in Japanese’, Language 66–1: 78–105.
Prince, Alan and Smolensky, Paul (1993) Optimality Theory: Constraint Interaction in
Generative Grammar, ms., Rutgers University Center for Cognitive Science.
Ramsey, S. Robert (1979) ‘The Old Kyoto dialect and the historical development of
Japanese accent’, Harvard Journal of Asiatic Studies 39–1: 157–175.
—— (1982) ‘Language change in Japan and the odyssey of a teisetsu’, Journal of Japanese
Studies 8–1: 97–131.
Rice, Keren (1997) ‘Japanese NC clusters and the redundancy of postnasal voicing’,
Linguistic Inquiry 28–2: 541–551.
—— (2005) ‘Sequential voicing, postnasal voicing, and Lyman’s Law revisited’, in van de
Weijer, J. Nanjo, K. and Nishihara, T. (eds.), Voicing in Japanese (Studies in Generative
Grammar 84), Berlin/New York: Mouton de Gruyter, 25–45.
Rodriguez, João (1604–1608) Arte da lingoa de Iapam, Nagasaki. Modern reedition with
Japanese translation by Doi Tadao, Nihon Daibunten, Tôkyô: Sanseidô, (1955).
Rosen, Eric (2003) ‘Systematic irregularity in Japanese Rendaku: how the grammar
mediates patterned lexical exceptions’, Canadian Journal of Linguistics 48: 1–37.
Sagisaka, Yoshinori (1999) ‘Nihongo on’in no jikanchô seigyo to chikaku [Perception and
control of temporal duration of Japanese phonemes]’, Gekkan Gengo 28–9: 51–56.
—— and Tohkura, Yoh’ichi (1984) ‘Phoneme duration control for speech synthesis by
rule’, Transactions of the IECE J67-A, 629–636.
Saitô, Yoshio (1997) Nihongo onseigaku nyûmon [Introduction to Japanese Phonetics],

Tôkyô: Sanseidô.
—— (2003) ‘Gendai nihongo no onsei, bunsetsuon to onseikigô [Phonetics of Modern

Japanese, phones and phonetic symbols]’, in Kitahara, Yasuo (ed.), Asakura Nihongo Kôza
3, Tôkyô: Asakura Shoten, 11–21.
Page 18 of 25
References
Sakuma, Kanae (1929) Nihon Onseigaku [Japanese Phonetics], Tôkyô: Hokuseidô,

(reedited by Kasama in 1963).
Satô, Hirokazu (1989) ‘Hukugôgo ni okeru akusento kisoku to rendaku kisoku [Accent
rules and rendaku rules in compound words]’, in Miyaji, S. et al. (eds.), Kôza nihongo to
nihongo kyôiku, 2 Nihongo no onsei – on’in (1), Tôkyô: Meiji Shoin, 233–265.
Satô, Ryôichi (2002) ‘Gendai nihongo no hatsuon bunpu [Geographical distribution of

phones in modern Japanese]’, in Hida, Y. and Satô, T. (eds.), Gendai Nihongo Kôza 3,
Tôkyô: Meiji Shoin, 20–39.
Scheer, Tobias (2004) A Lateral Theory of Phonology: What is CVCV, and why should it
be?, Berlin: Mouton de Gruyter.
Schourup, Lawrence (2000) ‘Japanese number mnemonics’, Journal of the Association of

Teachers of Japanese 34: 131–158.
—— and Tamori, Ikuhiro (1992) ‘Japanese palatalization in relation to theories of

restricted underspecification’, Gengo Kenkyû 101: 107–145.
Seeley, Christopher (1991) A History of Writing in Japan, Leiden: E. J. Brill.
Selkirk, Elisabeth and Tateishi, Koichi (1988) ‘Constraints on Minor Phrase Formation in
Japanese’, in Papers from the 24th Annual Regional Meeting of the CLSChicago: Chicago
Linguistic Society, 316–336.
Shibatani, Masayoshi (1990) The Languages of Japan, Cambridge: Cambridge University

Press.
Shimizu Han, Mieko (1962) Japanese Phonology: An Analysis Based upon Sound
Spectograms, Tôkyô: Kenkyûsha.
Shinohara, Shigeko (2000) ‘Default accentuation and foot structure in Japanese: evidence
from Japanese adaptations of French words’, Journal of East Asian Linguistics 9: 55–96.
—— (2002) ‘Metrical constraints and word identity in Japanese compound nouns’, MIT
Working Papers in Linguistics 42: 311–328.
Shirose, Ayako, Kubozono, Haruo, and Kiritani, Shigeru (1997) ‘The acquisition of
compound accent in Japanese’, Proceedings of the 16th International Congress of
Linguists (Cd Rom), Pergamon, Oxford, article n° 0186.
Sibata, Takeshi (1955) ‘Nihongo no akusento taikei [The accentual system of Japanese]’,
Kokugogaku 21, reprinted in Shibata et al., 1980, 404–442.
—— (1994) ‘Gairaigo ni okeru akusentokaku no ichi [The location of the accent nucleus in
Western words]’, in Satô, Kiyoji (ed.), Kokugo Ronkyû 4 Gendaigo – Hôgen no kenkyû,
Tôkyô: Meiji Shoin, 418–388 (1–31).
Page 19 of 25
References
—— Kitamura, Hajime, and Kindaichi, Haruhiko (eds.) (1980) On’in [Phonology], (Nihon
no Gengogaku, vol. 2), Tôkyô: Taishûkan.
—— and Shibata, Ritei (1990) ‘Akusento ha dôongo wo dono teido benbetsu shiuru ka [To
what extent can accent distinguish homophones?]’, Keiryô Kokugogaku 17–7: 311–323.
Smith, Jennifer L. (1998) ‘Noun faithfulness: evidence from accent in Japanese dialects’, in
Akatsuka, N. et al. (eds.), Japanese-Korean Linguistics, vol. 7, Stanford, California:
Stanford CSLI, 611–627.
—— (2002) ‘Toward a Compositional treatment of positional constraints: the case of

positional augmentation’, in Carpenter, Angela et al., Papers in Optimality Theory II,
GLSA, University of Massachusetts, Amherst, 337–370.
Sugitô, Miyoko (1965) ‘Shibata-san to Imada-san: tango no chôkakuteki benbetsu ni tsuite

no ichi kosatsu [Mr Shibata and Mr Imada: a note on word distinction at the auditory
level], Gengo Seikatsu 165: 64–72.
—— (1982) Nihongo akusento no kenkyû [Research on Japanese Accentuation], Tôkyô:

Sanseidô.
Sukegawa, Yasuhiko, Maekawa, Kikuo, and Uehara, Satoshi (1999) ‘Nihongo chôboin no
tanboinka genshô wo meguru shoyôin no jikken onseigakuteki kenkyû to onsei kyôiku he
no shisa [Research in experimental phonetics on the factors of long vowel reduction in
Japanese – some hints for phonetic teaching]’, in Alam Sasaki, Yukiko (ed.), Gengogaku to
Nihongo Kyôiku, Tôkyô: Kurosio, 81–94.
Suzuki, Takao (1977) ‘Writing is not language, or is it?’, Journal of Pragmatics 1–4: 407–
420.
Takayama, Michiaki (1992) ‘Seidaku shokô [Remarks on voicing oppositions]’, in Tajima,

Ikudo and Niwa, Kazuya (eds.) Nihongo Ronkyû 2. Koten Nihongo to jisho, Tôkyô: Izumi
Shoin, 17–56.
Takayama, Tomoaki (1995) ‘Sokuon ni yoru hukugô no takuritsu [The prosodic role of
gemination and compound construction]’, Kokugogaku 182: 15–27.
—— (2003) ‘Gendai nihongo no on’in to sono kinô [The phonemes of contemporary

Japanese and their function]’, in Kitahara, Yasuo (ed.), Asakura Nihongo Kôza 3, Tôkyô:
Asakura Shoten, 22–42.
—— (2005) ‘A survey of rendaku in loanwords’, in van de Weijer, J. Nanjo, K. and

Nishihara, T. (eds.), Voicing in Japanese (Studies in Generative Grammar 84), Berlin/New
York: Mouton de Gruyter, 177–190.
Takeuchi, Lone (1999) The Structure and History of Japanese, from Yamatokotoba to
Nihongo, London: Longman.
Page 20 of 25
References
Tanaka, Shin‐ichi ( ) (2001) ‘The emergence of the “unaccented”: possible patterns and
variations in Japanese compound accentuation’, in van de Weijer, J. and Nishihara, T.
(eds.), Issues in Japanese Phonology and Morphology, Berlin: Mouton de Gruyter, 159–
192.
—— and Yamane Noriko (2000) ‘Paradokkusu ka joretsu-gae ka: kango ni okeru akusento
to sônyû boin [Paradox or ranking modification: accent and vowel insertion in Sino-
Japanese words]’, in Proceedings of the 24th Annual Meeting of the Kansai Linguistic
Society, 66–76.
Tanaka, Shin’ichi ( ) (1998) ‘Hutto-nai ni okeru boin no kikoe to hukugôgo akusento

[Vowel sonority in foot and compound accentuation]’, Onsei Kenkyû 2–1: 50–62.
—— (2008) Rizumu-akusento no ‘yure’ to on’in-keitai kôzô [Accent/rhythm variation and

the phonological and morphological structure of words], Tôkyô: Kurosio.
—— and Kubozono, Haruo (1999) Nihongo no hatsuon kyôshitsu [Introduction to

Japanese pronunciation], Tôkyô: Kurosio.
Tanomura, Tadaharu (1999) ‘Gairaigo akusento ni okeru gengo no hatsuon no kan’yo ni

tsuite – yon-môra ika no go wo chûshin ni [The influence of the source language
pronunciation in gairaigo accentuation: with special attention to words shorter than four
moras]’, Nihongo Kagaku 5: 67–88.
Tateishi, Kôichi (1990) ‘Phonology of Sino-Japanese morphemes’, Amherst: University of

Massachusetts Occasional Papers in Linguistics 13: 339–362.
Terao, Yasushi (1992) ‘Bunsanshutsu kenkyû kara mita onsetsu – môra [Mora and
syllable from the point of view of sentence production research]’, Nihongo no môra to
onsetsu kôzô ni kansuru sôgôteki kenkyû (1), Research Report, University of Tsukuba,
62–75.
—— (2002) Iimachigai ha dôshite okoru? [Why do speech errors happen?], Tôkyô:

Iwanami Shoten.
Tôdô, Akiyasu (ed.) (1996) Gakken Kanwa Daijiten [Gakken Great Dictionary of Sino-
Japanese] Tôkyô: Gakken.
Tokugawa, Munemasa (1972) ‘Towards a family tree for accents in Japanese dialects’,
Papers in Japanese Linguistics 1–2: 301–320.
Topintzi, Nina (2010) Onsets, Suprasegmental and Prosodic Behaviour, Cambridge:

Cambridge University Press.
Trubetzkoy, N. S. (1939) Grundzüge der Phonologie [Principles of Phonology], Travaux

du cercle linguistique de Prague 7, Prague.
Page 21 of 25
References
Tsuchida, Ayako (2001) ‘Japanese vowel devoicing: cases of consecutive devoicing

environments’, Journal of East Asian Linguistics, 10: 225–245.
Tsuchiya, Shin’ichi (1965) ‘Hanashikotoba no naka no kango [Sino-Japanese words in the

spoken language]’, Gengo Seikatsu 169: 52–55.
Tsujimura, Natsuko (ed.) (1999) The Handbook of Japanese Linguistics, Cambridge, MA

and Oxford: Blackwell.
Tsukishima, Hiroshi (1977) Kokugo no rekishi [The History of Japanese], Tôkyô: Tôkyô
Daigaku Shuppan.
Tsuzuki, Masaki and Lee, Hyun Bok (1992) ‘A phonetic study of the Korean and Japanese
lateral, flap and nasal’, in the Linguistic Society of Korea (ed.), Proceedings of the 1992
Seoul International Conference on Linguistics, Seoul: Koryo University, 761–780.
Ueda, Kazutoshi (1898) ‘P on-kô [Remarks on P]’, Teikoku Bungaku 4–1: 41–46.
Ujihira, Akira and Kubozono, Haruo (1994) ‘A phonetic and phonological analysis of
stuttering in Japanese’, Proceedings of the Third International Conference on Spoken
Language Processing, vol. 3, Yokohama, 1195–1199.
Uwano, Zendô (1977) ‘Nihongo no akusento [The accent of Japanese]’, in Iwanami Kôza
Nihongo 5 On’in, Ôno, S. and Sibata, T. (eds.), Tôkyô: Iwanami shoten, 281–321.
—— (1997) ‘Hukugô meishi kara mita nihongo shohôgen no akusento [Accent in Japanese
dialects from the point of view of compound nouns]’, in Sugitô, M. (ed.), Akusento,
intonêshon, rizumu to pôzu, Tôkyô: Sanseidô, 231–270.
—— (1999) ‘Classification of Japanese accent systems’, in Kaji, S. (ed.), 1999, Cross-

linguistic Studies of Tonal Phenomena: Tonogenesis, Typology, and Related Studies
(Proceedings of the Symposium), Institute for the Study of Languages and Cultures of
Asia and Africa, Tôkyô University of Foreign Studies, Tôkyô, 151–178.
—— (2003) ‘Akusento no taikei to shikumi [System and mechanisms of accent]’, in

Kitahara, Yasuo and Uwano, Zendô (eds.), Asakura Nihongo Kôza 3, Tôkyô: Asakura
Shoten, 61–84.
Vance, Timothy (1982) ‘On the origin of voicing alteration in Japanese consonants’, Journal
of the American Oriental Society 102–2: 333–341.
—— (1987) An Introduction to Japanese Phonology, Albany: State University of New York

Press.
—— (1992) ‘Lexical Phonology and Japanese vowel devoicing’, in Brentari, D. et al. (eds.),
The Joy of Grammar, Amsterdam: John Benjamins.
—— (1995) ‘Final accent vs. no accent: utterance-final neutralization in Tokyo Japanese’,
Page 22 of 25
References
Journal of Phonetics 23: 487–499.
—— (2005) ‘Rendaku in inflected words’, in van de Weijer, J. Nanjo, K. and Nishihara, T.

(eds.), Voicing in Japanese (Studies in Generative Grammar 84), Berlin/New York:
Mouton de Gruyter, 89–103.
—— (2008) The Sounds of Japanese, Cambridge: Cambridge University Press.
Venditti, Jennifer J., Maekawa, Kikuo, and Beckman, Mary (2008) ‘Prominence marking in
the Japanese intonation system’, in Miyakawa, Shigeru and Saitô, Mamoru (eds.), The
Oxford Handbook of Japanese Linguistics, Oxford: Oxford University Press, 456–512.
Vocabulario da lingoa de Iapam, the Society of Jesus, year MDCIII, Paris copy,
reproduced Benseisha, Tôkyô (1976).
Wada, Minoru (1966) ‘Daiichiji akusento no hakken: ibuki-jima [Discovery of primary

accent: Ibuki-jima]’, Kokugo Kenkyû 22: 24–28.
Warner, Natasha (1997) ‘Japanese final-accented and unaccented phrases’, Journal of

Phonetics 25: 43–60.
—— and Arai, Takayuki (2001) ‘Japanese mora-timing: a review’, Phonetica 58–1: 1–25.
—— Otake, Takashi, and Arai, Takayuki (2010) ‘Intonational structure as a word-

boundary cue in Tokyo Japanese’, Language and Speech 53: 107–131.
Wenck, Günther (1954–1959) Japanische Phonetik [Japanese Phonetics], Vol. I–IV,

Wiesbaden: Otto Harrassowitz.
—— (1966) The Phonemics of Japanese, Questions and Attempts, Wiesbaden: Otto

Harrassowitz.
—— (1987) ‘An early corollary of initial voicing as a means of Japanese word formation’
Pratum Japanisticum, 110–117.
Yamada, Bimyô (1892) ‘Nihon onchô-ron [A theory of Japanese prosody]’, in Nihon

Daijiten, reprinted in Sibata et al., (1980), 272–310.
Yamaguchi, Yoshinori (1988a) Kodai nihongo bunpô no seiritsu no kenkyû [Research on

the development of Archaic Japanese grammar], Tôkyô: Yûseidô.
—— (1988b) ‘Kodaigo no hukugôgo ni kansuru hito-kosatsu – rendaku wo megutte [A

remark on compounds in Archaic Japanese: about rendaku]’, Nihongogaku 7–5: 4–12.
—— (1997) ‘Nihongo no rekishi: on’in [The history of Japanese: phonology]’, in Kamei, T.,
Kôno, R., and Chino, E. (eds.), Nihon rettô no kotoba, Tôkyô: Sanseidô, 153–166.
Yamaguchi, Yukihiro (1998) Nihongo hôgen ikkei akusento no kenkyû [Research on One-
Page 23 of 25
References
pattern Accent Dialects], Tôkyô: Hitsuji Shobo.
Yamane-Tanaka, Noriko (2005) ‘The implicational distribution of prenasalized stops in

Japanese’, in van de Weijer, J. Nanjo, K. and Nishihara, T. (eds.), Voicing in Japanese
(Studies in Generative Grammar 84), Berlin/New York: Mouton de Gruyter, 123–156.
Yanagida, Seiji (1985) Muromachi jidai no kokugo [The Japanese Language at the
Muromachi Period], Tôkyô: Tôkyôtô Shuppan.
—— (1991) ‘Dô no seiritsu [The development of dô]’, Muromachi jidaigo shiryô ni yoru
kihon goshi no kenkyû, Tôkyô: Musashino Shoin, 1–28.
Yasuda, Naomichi (1982), ‘Jôdaigo no boin ha ikutsu atta ka [How many vowels were
there in Archaic Japanese?]’, Kokubungaku 27–16: 68–72.
Yip, Moira (2002) Tone, Cambridge Textbooks in Linguistics, Cambridge: Cambridge

University Press.
Yokotani, Teruo (1997) ‘Hutto gyôkai wo koeru akusento idô: Tôkyô hôgen hukugô meishi
kara no shôko [Accent shift beyond the foot boundary: evidence from Tôkyô Japanese
compound nouns]’, Onsei Kenkyû 1–1: 54–62.
Yoshida, Kanehiko (1976) Nihongo Gogengaku no Hôhô [Methods in Japanese

Etymology], Tôkyô: Taishûkan.
Yoshida, Natsuya (2002) ‘Onsei kankyô ga boin no museika ni ataeru eikyô ni tsuite [The
influence of the phonetic environment on vowel devoicing]’, Kokugogaku 53–3: 34–47.
Yoshida, Shohei (1990) ‘A government-based analysis of the “mora” in Japanese’,

Phonology 7: 331–351.
—— (1991) Some Aspects of Governing Relations in Japanese Phonology, PhD thesis,

University of London.
Yuzawa, Tadayuki and Matsuzaki, Hiroshi (2004) Onsei on’in tankyûhô [Research
Methods in Phonetics and Phonology] (Series Nihongo tankyûhô 3), Tôkyô: Asakura
Shoten.
Zoll, Cheryl (1998) Positional Asymmetries and Licensing, ms, MIT.
Page 24 of 25
Index

Laurence Labrune

Print ISBN-13: 9780199545834
DOI: 10.1093/acprof:oso/9780199545834.001.0001
(p.287) Index
*F[m constraint 173–4, 204, 206–13
/a/ 25–7, 34, 48, 76, 88, 90, 91, 151; frequency 57–8
prosodic shortening 47–8
accent 29, 31, 33, 34, 37–8, 50 (n. 7), 51, 54, 55, 102, 113 (n. 4), 159–60; see also initial
dissimilation; NADM principle; deficient mora; atonic
default accent 155, 157, 174, 181, 201, 202, 203, 204, 205, 207
and devoicing 172, 189
dominant 193–6, 197
on double vowels 47
and foot 172, 173, 205, 206, 224, 228
and heavy syllable 146, 156–157, 204
final 31, 179, 182–4, 186, 196, 199, 200, 201, 202, 217, 218, 219, 220–1, 221, 222,
224, 227–8, 231, 240, 242, 243, 244, 254, 258
initial 179, 180, 184, 186, 191, 201, 204, 205, 208, 210, 212, 213, 214, 217, 218,
219, 228, 230, 231, 237, 238, 239, 240, 251, 254, 255, 258
of loanwords 157–158, 171, 172, 174, 176, 181, 201–14
of no 158, 201
notation 179, 190 (n 4)
Page 1 of 22
Index
and prosodic word 174

recessive 157, 194, 196
of reduplicated words 119
in relation to sonority 169–70, 221–2
secondary 155
and stress 156, 258–260
and voicing 104, 130, 131
and special segments 138–40, 154, 155, 156–7, 159, 164, 169, 170, 173, 188, 200,
228
accent of compounds 171, 176, 188, 254, 257, 266
accent deletion 158, 194, 200–201, 236, 250
accent inversion 240
accent loss 185
accent nucleus 174, 179, 188, 197, 205, 216, 235–6, 249, 253, 259, 265
accent shift 155, 164, 168, 172, 188–189, 200, 203, 221, 239, 240, 242, 258
accentual phrase, see minor phonological phrase
accentless dialects 251–257, 261
adverb, adverbial 10, 49, 50, 51, 52, 70 (n 4), 94, 96, 104, 193, 199, 214–15, 237, 241,
243, 248
affricates 39, 59, 60, 62, 64, 66–8, 152 (n 4), 161
Ainu 14, 22, 75
Akamatsu, Tsutomu 34, 38, 40, 81, 133
Akinaga, Kazue 37, 78, 95, 143, 155, 163, 157, 196, 215, 216, 238–9, 240, 243, 252, 265
akusento アクセント 178, 258, 263, 264; see also accent
akusento kaku アクセント 179, 265; see also accent nucleus
akusento ku アクセント 175; see also minor phonological phrase
Alderete, John 89
alternations 12, 20
between special segments 140
CV / Q 20, 28, 30, 32, 137
e / a 27, 91
h / p 20, 70–4
i / o 27
i / u 20, 28, 31
s / sh 67 (n 2), 68
voiced / voiceless 105, 129
voiced C / CC (geminate) 111, 115
y / i 89
y / zero 68, 89
Amanuma, Yasushi 34, 41, 53, 85
apposition 236
Arai, Masako 68 (n 3)
Arai, Takayuki 143
Archaic Japanese 4, 5, 26, 27, 42, 53 (n), 57–8, 62, 75, 76, 77, 88, 91, 93, 96, 99–100,
107, 108, 166, 257, 261
Page 2 of 22
Index
Arisaka, Hideyo 6, 26, 46, 143

Aston, William George 28, 76
atamadaka-gata 179, 186; see also accent (initial)
atonic 179, 181, 182–7, 193–5, 213 (n)
adjectives 196–9
adverbs 214–5, 243
compounds 160, 217, 219, 220, 222–4, 225–6, 229, 232–4, 236, 237, 239–40, 242–4
in dialects 252, 253, 257
deictics 214
four-mora words 29, 164, 202, 210 (n), 214, 217, 239
Haraguchi's treatment 189–90
no particle 158
numerals 242–4, 245
phonological phrase 248–251
Pierrehumbert and Beckman's treatment 192–3
and rendaku 125
shika particle 156–7, 195
in Sino-Japanese 200–1, 239–40, 245
spreading of atonic pattern 186–8, 257–8
to particle 140, 157
verbs 196–8, 240
(p.288) and vowel devoicing 34
Auer, Peter 176 (n 23)
Ayusawa, Takako 172
/b/ 60, 61, 66, 72, 73, 87, 98, 100–1, 107, 133, 135; frequency 99–101; see also voiced
obstruents
[β] 98
babibu language 146
baby talk; see child language
Bantu 255, 260, 261, 263, 264
Beckman, Jill 159
Beckman, Mary 33, 39, 176, 183, 190–3, 248, 251, 253 (n 19), 265
bidakuon 78; see also velar nasal
bunsetsu 176; see also utterance
Burzio, Luigi 170
Campbell, Nick 57, 58, 151
catathesis 175, 248
Chen, Matthew Y. 261
child language 49, 50, 61, 66, 68, 90, 92, 129, 146 (n 5)
Chinese; see also kango; kanji; kanbun
as a tone language 110, 178, 238 (n), 258, 259, 263, 266
Chinese materials 107
Chinese words coined in Japan (wasei kango) 20
Chongming Chinese 261
diphthongs 55
Page 3 of 22
Index
in Kojiki and Nihonshoki 109

influence of Chinese phonology 107–8, 137
labial glide 91
linguistic tradition 5, 6, 103, 262–3
loans from Chinese 5, 6, 13, 14, 15, 16, 18, 19, 20, 21, 22, 75, 89, 228
Modern Chinese 20
Old Chinese 19, 30, 42, 75, 91, 138
writing system 145 (n 3)
Chinese character; see kanji
chô akusento ku アクセント 175; see also major phonological phrase
chokuonka 28, 89
Chomsky, Noam 3, 7, 140, 163 (n 13)
chôon 132; see also /R/
Clements, George Nick 111, 260, 261 (n 21)
coalescence 27, 43–4, 47
coda 33, 95, 98, 149, 150–3, 156, 160, 165
Coda Condition 95, 140–1
Coleman, John 60
Collado, Diego 34, 263
compensatory lengthening 43–4, 47, 52
compounds 193; see also rendaku
accentuation 29, 138, 197, 200, 215–48
adjectives 118
ancient compounds 216
boundary 42, 81, 125
determinative compounds 114
dvandva 114, 124, 186, 215, 216, 236, 247, 249
extra-long compounds 222, 223, 226
gemination as compounding mark 73, 115
hybrid 14, 15, 70, 116, 214
lexicalization 33, 115, 194, 216, 240
long compounds 117, 122, 123, 124, 125, 126–7, 185, 220–1
[ᵑ] 80–2, 85
neo-classical 20
numeral 23, 71, 73–4, 119, 186, 216, 217, 241–8
of Western origin 16, 23, 99, 116
[p] and [h] in compounds 71–2, 77
personal names 120, 154, 159–60
Right Branch Condition 118–19
short compounds 117, 122, 123, 126–7, 217
Sino-Japanese 17, 30, 31, 33, 61, 70, 73, 80–1, 217, 237–40
transparent 77, 216
truncation 171, 172
verbs 93, 113, 114, 118–9, 122, 240
vowel alternation as compounding mark 27, 91
Page 4 of 22
Index
Yamato 41, 73–4, 86, 115, 118, 216, 219, 236, 239
Coyaud, Maurice 239
Creissels, Denis 260, 261 (n 21)
culminative function 130, 179, 185, 259, 260
Cutler, Anne 146, 150
CV → /Q/ alternation 20, 28, 30, 32, 137
/d/ 62–3, 65, 97, 107, 112; frequency 99–101; see also voiced obstruents
Daigenkai 263, 264
dakuon 102; see also voicing
dakuten 10, 72, 108, 109, 110–11, 131; see also voicing
Daniels, F. J. 60
Davis, Stuart 165 (n 16)
days of the week 51
days of the month in -ka 243
de Chêne, Brent 96
de-accentuation 158–9, 184, 185–6, 219, 244; see also accent deletion; accent loss
default accent 155, 157, 174, 181, 201, 202, 203, 204, 205, 207
deficient mora 162–70, 176; see also NADM principle; special segments; *F[m
accentuation 160, 188–9, 201–2, 204, 205–13, 217, 221, 253
and epenthetic vowels 29
final in Sino-Japanese and Western short morphemes 227–8
foot structure 172–4
initial dissimilation 154–5, 180 (n 3), 191
degenerate foot 126, 160, 171
deictic 49, 50, 96, 214
delabialization 92
demarcative function 130, 154–5, 180, 185, 251, 259
devoicing (of vowel) 28, 29, 32, 34–9, 104, 136, 163–4, 240, 258
dialects 4, 5, 6, 10, 40, 54, 65, 66, 67, 68, 78, 92, 95, 98, 112, 140, 259, 261, 265, 266;
see also Ryûkyû
accent systems 251–8
(p.289) Amami 27
central dialects 262
dialectal vowel systems 26–7
Eastern dialects 4, 261
Ibuki 255, 256, 257, 261
Izu 54, 142, 157, 169
Kagoshima 65, 142, 251, 252, 255, 256
Kansai 4, 52, 69, 76 (n 7), 77, 142, 263
Kantô 34, 69, 261
Kumamoto 255
Kyûshû 34, 40, 65, 88, 91, 142, 185, 251, 252
Miakejima 27
Miyako-no-jô 251, 255, 256, 257
Niigata 27
Page 5 of 22
Index
Nôbi 27
Oita 256
peripheral dialects 262
Sendai 27, 252, 255, 257, 261
Shikoku 65, 107, 256
Shizukuishi 255
Tôhôku 65, 75, 91, 107, 142, 185
Western dialects 4, 76
Yonaguni 27
diminutives 43, 90; see also hypocoristics
diphthongs 42, 44, 53–6, 161
dissimilation 119, 120, 121; see also initial dissimilation
Dommergues, Jean-Yves 147
double vowel 44–7; see also diphthong; /R/
dvandva compounds 114, 124, 186, 215, 216, 236, 247, 249
/e/ 17, 25, 27–8, 30, 34, 44, 66, 88, 89, 93; frequency 57–8
long /e/ 40, 48, 53, 132
/ei/; see /e/ (long /e/)
empty onset 46b (n 6), 162, 165, 166, 168, 169; see also onsetless vowel
empty position 163, 165, 169; see also empty onset; onsetless vowel.
English; see also wasei eigo, Japenglish; loanwords
clipping 184
diphthongs 55
palatalization of velars in loans 63–4, 89
phonic adaptation 61, 98, 138
stress 131, 178, 215, 258, 259
stuttering 167
syllable 147, 151, 155, 156, 157, 165 (n)
epenthetic consonant 94, 96
epenthetic glide 41
epenthetic vowel 28–34, 162–6, 169–70, 176, 205, 206, 239
and accent 188, 202, 204, 208–9, 210, 213 (n 10)
Everett, Daniel 165
Everett, Karen 165
expressive (connotation; words) 13–14, 55, 60–62, 67, 68, 70, 79, 80, 90, 102, 135, 137,
138; see also diminutives; hypocoristics; slang
/f/, [f] 69, 97, 98, 99
Faber, Alice 36
final accent; see accent (final)
focus 180 (n 3), 250 (n 17)
foot 170–6; see also NADM principle
and accent 188, 223, 239
degenerate 126, 160, 171
monomoraic foot 126, 160
in OT accounts 205–9, 213, 224, 227, 228, 235
Page 6 of 22
Index
and rendaku 126

structure 142, 143
and syllable / mora 150, 151, 153, 160
and vowel devoicing 37
and vowel sonority 37, 216
formal Sino-Japanese 14
Frauenfelder, Ulrich 147
Frellesvig, Bjarke 137
French 33, 55, 63, 147, 184, 261
Canadian French 30
loans 13, 20, 138, 204
Fujitani, Nariakira 262
Fujiura, Yayoi 217 (n 11), 219
Fukuda, Shinji 114
Fukuda, Suzy 114
full mora 109, 168, 172, 173, 174, 176, 180 (n 3)
/g/ 63–4, 78–87; see also [ᵑ]; voiced obstruents
frequency 99–101
gw 91, 92
palatalization 88, 89
prenasalization 107, 112
ga-gyô bion が 78; see also velar nasal
gaikokugo 14, 21, 23
gairaigo 13, 14, 15, 16, 20–2, 23, 27, 61; see also loanwords
gemination; see /Q/
German 2, 20, 98
Germanic 55
giseigo 13; see also ideophones; mimetics
gitaigo 13; see also idéophones; mimetics
glottal stop 45, 54, 136
go reading 18–19
gojûonzu 9
Goldsmith, John 3, 189, 260, 261
Golston, Chris 152
goon 18–19
gôon 26–7
goro awase 106
gôyôon 91
Grootaers, Willem A. 265
/h/ 17, 25, 28, 38, 59
allophone [F] 98
and /b/ 60, 73
frequency 99–101
and gemination 70, 68 (n), 136
and /N/ 133, 134
Page 7 of 22
Index
neutralization with /s/ 64

(p.290) and /p/ 60–2, 68 (n), 72, 92, 136
and palatalization 88
/H/ (phonological notation of vowel length) 44; see also /R/
Hagège, Claude 92
ha-gyô tenkoon ハ 75; see also /h/
haiku 17, 145
haku 132, 133, 142, 143, 144; see also mora
haku hôgen 142
Halle, Morris 3, 7, 140, 162 (n 13)
Hamada, Atsushi 1, 41, 45, 49, 50, 66, 68, 75, 105, 108, 110, 133, 137, 140
Hamano, Shoko 61, 76, 89–90, 94, 140, 157
handakuten 10, 73; see also /p/
haneru oto ねる 132; see also mora nasal
haplology 107
Haraguchi, Shôsuke 1, 146, 149, 154, 158, 180 (n 3), 183, 189–90, 191, 253 (n 19)
Hashimoto, Mantarô 54
Hashimoto, Shinkichi 1, 6, 26, 75, 133
Hatano, Giyô 150
hatsuon 132; see also /N/
Hattori, Shirô 1, 3, 6, 26, 45, 46 (n 6), 68, 132 (n), 143, 154, 180 (n 3), 256, 264–5
Haudricourt, André 92
Hayashi, Chikafumi 17, 19, 76, 238
Hayashi, Ooki 15, 101, 141, 184
Hayata, Teruhiro 52, 107, 266
Hayes, Bruce 2, 149, 170 (n)
heiban-shiki , heiban-gata 179; see also atonic
Hepburn transcription 65, 67, 88, 98, 153
hiatus 16, 45, 47
Hibiya, Junko 78, 80, 81
hierarchy
between two components 124
of constraints 82–5, 127, 168, 169, 174, 206, 207–8, 212, 224, 226, 229, 234
of moras 169–70, 173
prosodic hierarchy 143, 173, 174–175, 191, 248
Higurashi, Yoshiko 148, 157, 193, 216, 251
hihaku hôgen 142
hiku oto く 132; see also /R/
hiragana 8–15, 40, 44, 108, 109, 113
Hiraide, Takashi 114 (n), 124
Hirano, Takanori 114
Hirayama, Teruo 142, 264–5
Hoffmann, J. J. 66, 76 (n)
Homma, Yayoi 150
hondaku 110
Page 8 of 22
Index
Howland, Douglas 264

Hyman, Larry 3, 148 (n 7), 149, 156, 161, 163 (n 14), 176 (n), 260, 261, 262
hyôjungo 4
hybrid compounds 14, 15, 70, 116, 214
hypocoristics 49, 171; see also diminutives
/i/ 25, 28
alternation i / o 27
consonant allophonies induced by /i/ 59, 62 (n), 63, 65–66, 69, 76 (n), 92, 151
devoicing 34–39
in diphthongs 44
duration 151
frequency 57–58
impossibility of *yi 88
insertion and deletion 28–34
moraic /i/ 55, 132, 139, 140, 143, 144, 146, 147, 152, 153, 154, 155, 157, 200, 206
palatalization induced by /i/ 89
prosodic shortening 48
in Sino-Japanese morphemes 25, 28
Idemaru, Kaori 150
ideophones 13, 51, 136, 214; see also mimetics
Imada, Shigeko 38, 133
Imae, Kuniharu 101, 141
Indonesian 261
initial dissimilation 154–5, 180, 181, 185–6, 249, 253
Inoue, Fumio 78, 87, 97, 107
interjection 10, 49, 51, 135, 136, 138 (n)
intermediate phrase; see major phonological phrase
interrogative 49, 51, 74, 214
inversion (accent -) 240, 259
Irwin, Mark 117, 128
Italian 98, 155, 261
Itô, Junko 1, 15, 23–4, 30, 31, 33, 52, 61, 81–5, 90, 94, 95–6, 102, 121, 124, 125, 126,
127–8, 129, 140, 175, 235
Japanese; see also dialects
Ancient 5, 61, 62 (n), 99, 140
Archaic 4, 5, 26, 27, 42, 57–8, 62, 75, 76, 77, 88, 91, 93, 96, 99–100, 107, 108, 166,
257, 261
Old 4, 5, 16, 26, 42, 44, 93, 96, 107
Middle 5, 42, 62 (n), 107
Proto 27, 75, 94, 251, 256
Japenglish 22
Jesuits 66, 77 (n 9), 107; see also missionaries
jiamari り 165
Jidai Betsu Kokugo Daijiten 100
Jinbô, Kaku 264–5
Page 9 of 22
Index
jiritsu haku 133, 144; see also full mora

Jôo Hakutarô 54, 132, 140
jûboin 53; see also diphthong; double vowel
jukugo 17; see also compounds (Sino-Japanese)
/k/ 18, 30, 33–4, 63–4, 75, 81–2, 88–9, 91–2, 133;
frequency 99–101
Kager, René 3, 163 (n 14)
Kageyama Tarô 52
kaion 26–7
kakekotoba 105–6
Kamei, Takashi 1, 66, 80, 85, 130
kan'on 18–19
kan'yôon 19, 42
(p.291) kana 2, 5, 8–12, 15, 33, 39 (n), 40, 44, 47, 51, 52, 65, 66, 72, 73, 75, 78,
97–8, 101, 108; 109, 135, 136, 140, 144, 145, 146; see also hiragana; katakana
kanbun 110, 111
kango 13–16, 17–20, 21, 22, 23, 24, 27, 186, 189; see also Sino-Japanese
kanji 2, 5, 8, 9, 11–13, 15, 17, 18, 23
Kano, Yoshimitsu 42
Karlgren, Bernhard 42
Katada, Fusa 146
katakana 8–13, 15, 21, 23, 97, 108, 109, 264; see also kana
Katayama, Motoko 203
Kawagoe, Itsue 68 (n)
Kawahara, Shigeto 24, 104, 106, 136
Kawakami, Shin 53, 192, 265
Keichû 262
keihan-shiki akusento アクセント 251; see also dialects
Kelly, Michael H. 165 (n)
Kenstowicz, Michael 3, 170 (n)
Kess, Joseph 7, 9
kihuku-shiki akusento アクセント 179
Kindaichi, Haruhiko 1, 3, 6, 7, 44, 45, 46, 68, 78, 81, 85, 86–7, 121, 143, 155, 157, 215,
216, 238–9, 240, 243, 253, 256, 262, 264–5
kinship terms 49
Kiritani, Shigeru 207
Kishida, Takeo 95, 103, 137
Kitazawa, Shigeyoshi 172
kô boin 26
Kobayashi, Akemi 66
Kochetov, Alexei 89
Kojiki 108, 109, 111, 262
Kokugo Gakkai 110
kokugogaku 1, 177
Kokuritsu Kokugo Kenkyûjo (KKK) 78 (n 11), 15, 48, 100
Page 10 of 22
Index
Komatsu, Hideo 52, 76, 103, 125, 130, 137, 140, 265
Kondo, Mariko 38, 39
konshugo 14; see also hybrid compounds
Koori, Shirô 251
Korean 4, 14, 20, 21, 22, 75, 108
Korean materials 5, 107, 111
Kubozono, Haruo 1, 29, 31, 33, 50, 52, 61, 114, 120, 124, 126, 146 (n 5), 147, 149,
151, 153, 157–8, 159–60, 165, 167, 171, 175, 181, 184, 186, 197, 198 (n), 201–4, 207,
210 (n), 213 (n), 216, 217, 219–20, 222–4, 226–9, 231, 232, 235, 236, 251, 266
Kuginuki, Tooru 93, 94
Kurematsu, Akira 92
Kurisu, Kazutaka 31, 127
Kuroda, Toshiyuki 94, 140
kyôtsûgo 4
Labrune, Laurence 7, 48, 76, 93, 94, 95, 96, 110, 111, 130, 131, 154, 172, 173, 175,
177, 185, 260
Latin 20, 29, 55, 148 (n)
Latin alphabet; see rôma-ji
Lawrence, Wayne 70
Lee, Hyun Bok 92
lenition 76–7, 91, 137, 138, 174
loanwords; see also gairaigo, and relevant language entries
accent 155–8, 164, 174, 181, 184, 185, 186, 188, 201–14, 216, 217, 219, 221,
227–9, 239, 241, 242, 247–8, 257, 266
deficient moras 202
geminates 68 (n), 95, 104, 136
long vowels 41
mimetics 13 (n)
neutralization of i and u 28
new sound combinations 59, 62, 63, 69, 90, 97–9
prosodic structure 24, 172
in the overall lexicon 14–17, 20–2, 264
/p/ 61
palatalization 44, 67, 89
/r/ 95
transition glide 53
voicing 106, 1115–116, 119, 129, 136
vowel epenthesis 28–9
long vowel; see /R/
Loveday, Leo 16, 20
Lowenstamm, Jean 167
Lyman, Benjamin 114
Lyman's law 119, 120, 121, 127, 130
/m/ 77–8, 87, 92 112, 133–4;
Page 11 of 22
Index
frequency 99–101
Mabuchi, Kazuo 25, 66
Maddieson, Ian 26, 151, 259
Maekawa, Kikuo 34, 47, 48
Maës, Hubert 7
major phonological phrase 175, 242, 248, 249, 250, 251
man'yôgana 5, 8, 109–110
Man'yôshû 99, 109, 165
Martin, Samuel E. 4, 7, 33, 52, 75, 76, 113, 122, 158
Mase, Yoshio 257
Matsumori, Akiko 257
Matsumoto, Katsumi 26
Matsuzaki, Hiroshi 81
McCarthy, John 84, 94
McCawley, James 1, 7, 31, 33, 61, 71–2, 148, 156–7, 175, 197 (n), 216, 218, 251, 253
(n 19)
Mehler, Jacques 146, 147
Meikai Nihongo Akusento Jiten 143, 155, 215, 216, 238, 239, 240, 243, 265
Mester, Armin 1, 15, 23–4, 30, 31, 33, 61, 81–5, 90, 94, 95–6, 102, 121, 124, 125, 126,
127–8, 129, 175, 235
Middle Japanese 5, 42, 62 (n), 107
mimetics
/e/ 27
/h/ 61, 62, 70, 73, 75
(p.292) [ᵑ] 79
/p/ 61, 62, 70, 73, 75
/r/ 92–95
accent 119 (n), 214, 216, 237, 247, 249
as a lexical class 13–14
ending in -ri 104–105, 138, 214
foot 171
gemination 94–5, 104, 136, 140, 157
palatalization 88, 89–90, 93, 94
prosodic lenghthening 49, 51
pseudo mimetics 119 (n)
reduplicated 80, 82, 84, 119, 124, 129, 237
voicing 103–4, 114
writing of mimetics 10
minor phonological phrase 175, 191, 192, 248, 249
missionaries 26, 66, 76, 77 (n 9), 88, 90, 107, 109,148, 263
mixed compounds 14, 15, 70, 116, 214
Miyake, Takeo 158
Miyamoto, Tadao 7, 9
Miyata, Kôichi 264
Mizutani, Osamu 34, 41, 53, 85
Page 12 of 22
Index
mora 8, 9, 44, 53, 91, 95, 120, 133, 142–70, 175, 224; see also deficient mora; special
mora; full mora
mora and syllable 54, 147–170, 207, 224, 227, 259
mora nasal; see /N/
môra onsô モーラ 132; see also special segments; /N/; /Q/; /R/
Morais, José 145 (n 3)
Morishige, Satoshi 26
Morita, Takeru 130
Motoori, Norinaga 262
mu-akusento アクセント 251; see also accentless dialects
museika 34; see also devoicing (of vowel)
/n/ 78, 87; frequency 99–101
[ᵑ] 64, 78–87, 107
/N/ (mora nasal) 59, 132–135, 143, 147, 153, 171
alternation with other segments 78, 95, 138, 140
and accent 139, 154–5, 156, 169, 218, 157, 188
as a deficient mora 163, 176, 206
as a result of lenition 78, 95, 138
before [d͡ z] 64
before /h/ 70–71
before /p/ 60, 72
before /r/ 95
before voiceless consonants 17, 24, 128–30
frequency 141
in Sino-Japanese morphemes 18, 22
in the hierarchy of moras 169 (n 17), 170
insertion of /N/ 95, 104, 105
nasalization of preceding vowel 152
phonological structure 72, 140, 162–3
post nasal voicing 128–130
syllabification 135, 152–3
words ending with /N/ 29
writing 33, 135, 136, 140, 145
NADM principle 138, 155, 159, 168–9, 172, 188–9, 200, 201, 217, 219 (n), 221, 228,
239, 242
Nakada, Norio 17, 19, 238 (n)
nakadaka-gata 179
Nakai, Yukihito 253
nasality, nasalization 80, 81, 86, 105, 107, 111, 112, 133–4, 140, 152, 162; see also
prenasalization
Nasu, Akio 32
Nespor, Marina 153
neutralization 28, 59, 63, 65, 159
NHK (Nippon Hôsô Kyôkai) 4, 34, 79, 80, 85, 174 (n), 216, 218, 245, 265
Page 13 of 22
Index
nigoriten り 10, 72, 108, 109, 110–11, 131; see also voicing
Nihonshoki 108, 109, 111
nijûboin 53; see also diphthong, double vowel; /R/
Nishimura, Kôhei 24
nisshô 138
Non Finality 205, 206, 207, 224–6, 227, 228, 229, 231, 234, 235
nucleus (of syllable or mora) 47, 54, 56, 60, 149, 150 (n), 151–2, 156–7, 160–3, 167,
170, 206; see also accent nucleus
numbers 51–52, 106;
accent 241–8
numeral 23, 52 (n 9), 73–4, 80, 196, 249
arabic 8, 9, 12, 13
compounds 23, 71, 73–4, 119, 186, 216, 217, 241–8
Sino-Japanese 245–8
Yamato 243(6
nyôbô kotoba 171–2
/o/ 25–27, 48, 143
after y 88–9
after w 92
as epenthetic vowel 29
devoicing 34–5
frequency 57–8
OCP (obligatory contour principle) 224, 228
odaka-gata 179; see also final accent
Odden, David 259, 260
Ogasawara, Naomi 39
Ogawa, Shinji 239
Ohnishi, Masayuki 92
Ohno, Kazutoshi 106, 110, 114, 117
Okumura, Mitsuo 73, 122, 123
Old Japanese 4, 5, 16, 26, 42, 44, 93, 96, 107
onbin 5, 42, 137 (n)
onchô 264
Ono, Hajime 24
Ôno, Susumu 99
onomatope オノマトペ 13 (n 6); see also onomatopoeia; mimetics
onomatopoeia 13, 51, 136, 214; see also mimetics
(p.293) onset optimization 135, 152
onsetless vowel 16, 38, 53 (n 11), 55, 100, 144, 162–5, 165, 169, 170, 173, 174, 189;
see also empty onset
onsetsu 143; see also syllable
Ootsubo, Kazuo 34, 41, 53, 85
Optimality Theory 1, 3, 4, 38, 82–85, 96, 107, 127–8, 130 (n), 148, 168, 198 (n), 202–14,
222–35
orthography 7, 41, 51, 61, 99, 108, 110, 131; see also writing system
Page 14 of 22
Index
Ota, Mitsuhiko 15, 23, 129, 173

Otake, Takashi 146, 150
otsu boin 26
Otsu, Yukio 117–18
oxytonic; see final accent; odaka-gata
/p/ 9–10, 17–18, 20, 30, 60–62, 68 (n), 70–77, 75, 92, 106, 134, 135, 138; frequency
99–101
Padgett, Jaye 15, 129
palatalization 88–90, 143, 144, 171; see also /y/
affective palatalization 67 (n), 89
alternation between CyV and CV 68
and /e/ 27, 67
and /u/ 25
as phonaestheme 89
diachronic development 44, 53, 89
frequency 101
in kango 17, 18, 22
of /k/, /g/ 63–4
of /n/ 78
of /r/ 90, 92, 94–5
phonological representation 56
scarcity 17
in writing 10, 146
Peak Prominence 168, 228
periodization 4–5
phonaestheme 89
phonemization 69, 87, 97–8
phonological phrase; see major phonological phrase; minor phonological phrase
phonologization 69, 87, 97–8
phrase; see major phonological phrase; minor phonological phrase
Pierrehumbert, Janet 175–6, 183, 190–3, 248, 251, 253 (n 19), 265
Piggott, Glyne 111
place names; see toponyms
Portuguese 14, 20, 23, 29, 44, 61, 66, 80, 116
Poser, William 50, 94, 153, 171, 176, 192, 197, 216, 251
prefix 50, 94, 153, 171, 175, 180, 192, 197, 216, 251
prenasalization 65, 66, 87, 94–5, 104, 107–8, 111, 112; see also nasality
Prince, Alan 3, 84, 94, 152, 168
prosodic hierarchy 143, 173, 174–175, 191, 248
prosodic lengthening 41, 43, 47, 49–52, 53, 253
Prosodic Word 82, 126–7, 142, 174–5, 180, 193, 206, 236, 242, 248, 251
Proto-Japanese 27, 75, 94, 251, 256
/Q/ 59, 132, 135–7, 143, 147, 171
and accent 154–5, 157, 253
alternation with /N/ 140
Page 15 of 22
Index
as a deficient mora 163, 176, 206

before /h/ 70, 72, 68 (n)
before /p/ 60, 61, 62, 70, 72
before /r/ 17, 94–5
before /s/ 38, 68, 72
CV → /Q/ alternation 20, 28, 30, 32, 137
frequency 141
in gairaigo 21, 238
in the hierarchy of moras 169 (n 17), 170, 176
insertion of /Q/ 104–5
phonological structure 72, 140–1, 162–3
in Sino-Japanese morphemes 17–18, 22, 30, 32, 33, 70, 239
in super heavy syllables 150 (n)
and voicing 17, 73, 104–5, 111, 115
and vowel devoicing 38, 39
vowel length before /Q/ 150–1
writing 11, 33, 140, 145, 146 (n 4)
/r/ 93–96, 135–6; frequency 99–101
/R/ (vowel length) 44–7, 132–3, 137, 143, see also diphthong; double vowel
and accent 139, 154–5, 157, 200
alternation with other special segments 140
as a deficient mora 162, 163, 164, 166, 169
Ramsey, Robert 256, 262
reduplication 77
and accent 237
and rendaku 118, 119, 124, 129
and velar nasal 79–80, 82, 84–5
in baby talk 50, 119 (n 6)
in kinship terms 49
of verb base 51, 52, 171
regular mora; see full mora
rendaku 110, 112–30
and /h/ 60, 73
and foot 174
and gemination 73
and lexical strata 13, 23
of /k/ and /g/ 80, 81–2, 85, 86
renjô 13 (n)
rhyme (rime) 149–51
Rice, Keren 129–30
Right Branch Condition 118–19
Rodriguez, João 109, 148 (n 6), 263
rôma-ji ローマ 8, 9, 10, 12, 14, 21, 107, 108, 109
Rosen, Eric 115, 117, 118, 121–3, 124–7, 128
Russian 20, 131, 258, 259
Page 16 of 22
Index
Ryûkyû, Ryukyuan 4, 27, 75, 92

/s/ 28, 38, 64–6, 68, 69, 90; frequency 99–101
Sagisaka, Yoshinori 92, 151
(p.294) Saitô, Yoshio 54, 98, 133
Sakuma, Kanae 183, 264, 265
Sanskrit 14, 22, 111
Satô, Hirokazu 120, 121, 215, 216, 257
Satô, Ryôichi 142
Scheer, Tobias 167
Schourup, Lawrence 95, 106
secret language 52, 146, 171–2
Seeley, Christopher 8, 109
Segui, Juan 147
sei, shô, koe 262–3
Selkirk, Elisabeth 175
Serbo-croatian 261
Shibatani, Masayoshi 72, 256
Shimizu Han, Mieko 25, 40
shindaku 110, 116
Shinohara, Shigeko 173, 204–7, 216, 220
shiritori 146
Shirose, Ayako 207
Shoji, Atsuko 39
Sibata, Ritei 181
Sibata, Takeshi 181, 185, 186, 253 (n), 264–5
Sino-Japanese 17–20
/e/ 40, 41
/p/ 60, 61, 70, 71, 72, 75
accent 184, 185, 186, 199–201, 217, 227–8, 242
as a lexical stratum 13, 14, 15, 17, 21, 23, 24, 27, 41, 216, 257
compounds 17, 30, 31, 33, 61, 70, 73, 80–1, 217, 237–40
CV → /Q/ alternation 28–34, 137, 138
fixed compounds 217, 237–40
formal Sino-Japanese 14
glides 91, 98
introduction of, 5
long vowels 41 (n), 42–3, 47, 52
numbers 51, 106
numerals 241, 242, 245–8
Old Sino-Japanese 30
one-character words 200–1, 227–8
palatalization 88, 89
post nasal voicing 129
reading 12, 15, 18–19, 23
Page 17 of 22
Index
reduplicated 80, 237

rendaku 115, 116, 119, 128
sequences of two vowels 53
specifiers 244, 245, 246
velar nasal 80, 81, 86
verbs in suru 199
vulgarized Sino-Japanese 14, 115
Sino-Korean 20, 75
slang 10, 13, 52, 72, 103
slips of the tongue 147, 153
Smith, Jennifer 127, 197–198 (n)
Smolensky, Paul 3, 84, 94, 152, 168
sokuon 132; see also /Q/
sonority 101, 170, 172, 216, 221–2
Spanish 76, 98, 258
special mora 132–141, 146, 155, 156, 157, 159, 164, 202, 210; see also deficient mora;
mora; special segment
special segment 44, 104, 132–41, 180, 188, 200, 210, 217; see also /N/; /Q/; /R/;
deficient mora; special mora
specifier 73, 119, 241–8; see also numerals
speech errors 147, 153
strengthening 72, 90, 104, 137, 138
strict layer hypothesis 153
stuttering 167
Sugitô, Miyoko 120, 183
Sukegawa, Yasuhiko 47, 48
Suzuki, Takao 7
Swahili 261 (n)
Swedish 261
syllabary 5, 8, 10, 78, 113, 135, 145 (n); see also kana; writing system; orthography
syllable 142, 147–67, 189, 202–5, 207; see also mora and syllable
and accent 180 (n 3), 259
adaptation of syllabic structure in loanwords 98
in Archaic Japanese 53 (n)
closed syllable 33, 39
in Chinese 137, 138, 266
in dialects 257
and foot 171, 172, 173
Hattori Shirô 143 (n)
and rendaku 126
in Sino-Japanese words 239
Non Finality 224–6
onsetless 54–5
syllable-based dialects 142, 257
weight 55, 126, 133, 140, 146, 147–67, 172, 210 (n), 213, 226, 227–8, 230
Page 18 of 22
Index
writing 8 (n 2), 10, 145 (n 3)

/t/ 18, 30, 39, 60, 62–3, 67, 68, 88, 90, 98, 112, 134, 135, 151; frequency 99–101
Takayama Michiaki 114
Takayama Tomoaki 14, 23, 41, 53, 76, 115, 116, 159
Takeuchi, Lone 48, 124 (n 11), 125, 238
Tamori, Ikuhiro 95
Tanaka Shin’ichi ( ) 55 (n), 145, 170 (n 17, 18), 171, 172, 180 (n 3), 185, 203, 204, 213
(n), 216, 220, 221–2, 239
Tanaka Shin-ichi ( ) 31, 37
tanka 17, 165
Tanomura, Tadaharu 203, 213 (n)
Tateishi, Kôichi 30, 175
Terao, Yasushi 149, 153
Tôdô, Akiyasu 19, 42
Toki, Satoshi 172
Tokugawa, Munemasa 256
tokushu haku 132, 133, 144; see also special segments; deficient mora
tokushu onso 132; see also special segments; deficient mora
(p.295) tôkyô-shiki akusento アクセント 251; see also dialects
tone 110, 130, 179, 189–93, 248, 251, 258–66;
of Chinese 138, 178, 238 (n), 263
tone-bearing unit 4, 189, 193
Tonga 261
tôon 19
Topintzi, Nina 165 (n)
toponyms 154, 157, 174, 196, 202–5, 208–13
tôsôon 19
Trubetzkoy, Nicolas 148, 161
truncation 21, 48, 52, 153, 171–2, 184, 214
Tsuchida, Ayako 38
Tsuchiya, Shin'ichi 15
Tsukishima, Hiroshi 110
Tsuzuki, Masaki 92
tsumaru oto つまる 132; see also /Q/
/u/ 25–6, 28–39
after /h/ 75, 76
after /t/ 62 (n)
after /w/ 89
after /y/ 56
as glide 90
deaccentuation 55 (n)
epenthetic 204
frequency 57–8
length 48
neutralization of /d/ and /z/ before /u/ 65
Page 19 of 22
Index
slightly rounded 92
Ueda, Kazutoshi 75
Uehara, Satoshi 47, 48
Ujihira, Akira 167
unaccented; see atonic
utterance 175–6, 181, 191–2, 248, 251
Uwano, Zendô 54, 146, 156, 157, 158, 159, 183, 184, 215, 216, 254, 255, 258, 266
Vance, Timothy 6–7, 18, 19, 31, 32, 33, 36, 38, 77, 81, 114, 118, 120, 122, 123, 128,
135 (n), 140, 150, 157, 158, 159, 180 (n3), 183
van der Hulst, Harry 152
velar nasal; see ᵑ
Venditti, Jennifer 175, 251
voiced obstruents 49, 60, 102–31; see also post-nasal voicing; rendaku; dakuon;
dakuten
distribution 16–17, 22, 102–4
frequency 100
non gemination 18, 32, 104–5, 136
prenasalization 65, 87, 94–5, 107, 152 (n 10)
writing 10, 72, 108–11
Vogel, Irene 153
vowel devoicing; see devoicing (of vowel)
vowel harmony 17
vowel length; see /R/
vulgarized Sino-Japanese 14, 115
/w/ 90–92, 133, 135, 136
frequency 99–100
h 〉 w evolution 69, 76–7
lenition 41
phonotactics 88, 89
in recent loanwords 97
Wada, Minoru 257
wago 13–24, 239; see also Yamato stratum
Warner, Natasha 39, 143, 180, 183
wasei eigo 21; see also loanwords
wasei kango 20
wasei yôgo 21; see also loanwords
Wenck, Günther 7, 50, 103
Western borrowings; see gairaigo; loanwords
writing system 7–13; see also kanji; hiragana; katakana; orthography; syllabary
and lexical strata 15, 22
development 5
diphthongs 44
four kana merging 65, 67
loanwords 98
mora 145
Page 20 of 22
Index
of /Q/ 135
of p,b,h 73
relation with phonology 54, 145 (n 3)
voicing 72–3, 108–11, 131
vowel length 40, 51
/y/ 88–90; see also palatalization
*yi, *ye 56, 91
/N/ before /y/ 133, 135
and rendaku 112, 120
consonant allophones before /y/ 62, 63, 64, 65, 66, 67, 69, 78
frequency 99–101
in recent loanwords 97–8
non gemination 136
Yamada, Bimyô 264
Yamaguchi, Yoshinori 107, 114, 119, 125, 130
Yamaguchi, Yukihiro 256, 261
Yamane (Yamane-Tanaka), Noriko 31, 107
Yamato stratum 13–17, 22–4; see also wago
[ᵑ] 86
[tsa] 68
*[ti], *[di] 62, 63
*[sye] 67
/a:/ 41–2
/e/, /e:/, /ei/ 27, 40
/h/, /p/, /pp/ 61, 61, 68, 69, 70–2, 73, 75, 77
/r/ 93, 95
s / sh alternation 68
accent 184, 185, 186, 187, 193–6, 200, 216, 217, 219, 257
acronyms 21
compounds 41, 73–4, 86, 115, 118, 216, 219, 236, 239
frequency of consonants 99–100
frequency of vowels 57
long vowels 41–3, 47, 49, 52
numerals 51, 52, 241, 243–5, 247–8
palatalization 89
(p.296) pseudo Yamato words 19, 22, 116
sequences of two vowels 53–4
voiced obstruents 102–4, 111, 125, 128–9
vowel deletions 34, 137, 166
Yanagida, Seiji 50 (n 8), 103
Yanagita, Kunio 262
Yasuda, Naomichi 26
/ye/ 56, 67, 77, 88, 91, 97, 98
Yip, Moira 260
Yokotani, Teruo 37, 170 (n)
Page 21 of 22
Index
yôon 57, 88, 143; see also /y/; palatalization

Yoruba 258
Yosa Buson 145
Yoshida, Kanehiko 93
Yoshida, Natsuya 38
Yoshida, Shohei 150 (n)
yotsugana no kondô つの 65
Yuzawa, Tadayuki 81
/z/ 25, 28, 59, 63, 64–5, 67, 88, 104, 107, 134–4; frequency 99–101; see also voiced
obstruents
zero consonant 46 (n), 93, 100, 132 (n); see also onsetless vowel
Zoll, Cheryl 127
zûja-go ズージャ 52, 171
zûzû-ben 65; see also dialects

Fonologiajapanese

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Fonologiajapanese

Diunggah oleh

Hak Cipta:

Format Tersedia

Title Pages

University Press Scholarship Online

Oxford Scholarship Online

The Phonology of Japanese

Print publication date: 2012

The Phonology of Icelandic and Faroese

The Phonology of Danish

The Phonology of Dutch

The Phonology of Standard Chinese, second edition

The Phonology of Polish

The Phonology of English

The Phonology of Italian

The Phonology of Norwegian

The Phonology of Japanese

The Phonology of Portuguese

Maria Helena Mateus and Ernesto d’Andrade

The Phonology and Morphology of Kimatuumbi

The Lexical Phonology of Slovak

The Phonology of Hungarian

Péter Siptár and Miklós Törkenczy

The Phonology of Mongolian

Jan-Olof Svantesson, Anna Tsendina, Anastasia Karlsson, and Vivan Franzén

The Phonology of Armenian

The Phonology and Morphology of Arabic

The Phonology of Catalan

The Phonology of German

The Phonology of Tamil

The Phonology of Welsh

The Phonology of Turkish

The Phonology of Latin

The Phonology of Spanish

The Phonology of Greek

The Phonology of Swedish

The Phonology of Washo

Great Clarendon Street, Oxford OX2 6DP

Oxford University Press is a department of the University of Oxford.

Auckland Cape Town Dar es Salaam Hong Kong Karachi

Oxford is a registered trade mark of Oxford University Press

Published in the United States

© Laurence Labrune 2012

The moral rights of the author have been asserted

First published 2012

All rights reserved. No part of this publication may be reproduced,

British Library Cataloguing in Publication Data

Library of Congress Cataloging in Publication Data

Typeset by SPI Publisher Services, Pondicherry, India

University Press Scholarship Online

Oxford Scholarship Online

The Phonology of Japanese

Print publication date: 2012

I acknowledge with gratitude several scholarships from the Japanese Ministry of

University Press Scholarship Online

Oxford Scholarship Online

The Phonology of Japanese

Print publication date: 2012

(p.xi) Tables, Figures, and Maps

Table 1.1. Hiragana (basic symbols) 8

Figure 2.2. Spectrogram and oscillogram of a ki demo (no devoicing of i) 36

University Press Scholarship Online

Oxford Scholarship Online

The Phonology of Japanese

Print publication date: 2012

(p.xii) Notes on Transcription, Abbreviations, and Other Matters