(Studies in Generative Grammar 51) Jeroen Maarten Van de Weijer, Tetsuo Nishihara - Issues in Japanese Phonology and Morphology (2001, de Gruyter) PDF

Issues in Japanese Phonology and Morphology
WDE
G
Studies in Generative Grammar 51
Editors
Harry van der Hulst
Jan Köster
Henk van Riemsdijk
Mouton de Gruyter
Berlin · New York
Issues in Japanese Phonology
and Morphology
Edited by
Jeroen van de Weijer
Tetsuo Nishihara
Mouton de Gruyter
Berlin · New York 2001
Mouton de Gruyter (formerly Mouton, The Hague)
is a Division of Walter de Gruyter G m b H & Co. KG, Berlin.
The series Studies in Generative Grammar was formerly published by

Foris Publications Holland.
© Printed on acid-free paper which falls within the guidelines

of the ANSI to ensure permanence and durability.
Library of Congress Cataloging-in-Publication Data
Issues in Japanese phonology and morphology / edited be Jeroen van

de Weijer, Tetsuo Nishihara.
p. cm. - (Studies in generative grammar ; 51)
Includes bibliographical references and indexes.
ISBN 3 11 016958 4 (alk. paper)
1. Japanese language - Phonology. 2. Japanese language -
Morphology. I. Weijer, Jeroen Maarten van de, 1965- II. Nishi-
hara, Tetsuo, 1961 - III. Series.
PL528 .188 2001
495.6'15-dc21 2001044669
Die Deutsche Bibliothek — Cataloging-in-Publication Data
Issues in Japanese phonology and morphology / ed. by Jeroen van de

Weijer ; Tetsuo Nishihara. - Berlin ; New York : Mouton de Gruy-
ter, 2001
(Studies in generative grammar ; 51)
ISBN 3-11-016958-4
© Copyright 2001 by Walter de Gruyter G m b H & Co. KG, D-10785 Berlin.

All rights reserved, including those of translation into foreign languages. No part of this
book may be reproduced in any form or by any means, electronic or mechanical, including
photocopy, recording, or any information storage and retrieval system, without permission
in writing from the publisher.
Disk conversion: OLD-Satz digital, Neckarsteinach.
Printing & Binding: Hubert & Co, Göttingen.
Cover Design: Christopher Schneider, Berlin.
Printed in Germany.
Preface
We are pleased to thank a number of institutions and individuals for their

help during the progress of the present volume. This book was first con-
ceived on the occasion of the meeting of the Phonological Society of Ja-
pan in Kobe in September 1998, where the editors could contact some of
the present authors and discuss the plan for a book in person.
The Holland Institute of Generative Linguistics (HIL) at Leiden Uni-
versity afforded an excellent academic environment for developing the
project and generously provided copying and Internet facilities.
We would like to thank a large number of reviewers in The Nether-
lands, Japan and other countries for their generous input into all of the pa-
pers.
We thank the series editor, Prof. Harry van der Hulst, for his support
and interest during all stages of finalizing the present volume. We would
also like to thank Dr. Anke Beck and Dr. Ursula Kleinhenz of Mouton de
Gruyter for their enthusiasm and support.
Last but certainly not least, it has been a pleasure and an honour to
work with the authors in this volume. We are therefore very pleased to
publish this volume shortly after the year 2000, when the Netherlands and
Japan celebrated 400 years of good relations.
Leiden, The Netherlands

Sendai, Japan
Summer 2001
Contents
Preface ν
I. Studies in Japanese Phonology
Yukiko Akasaka and Koichi Tateishi

Heaviness in Interfaces 3
Shosuke Haraguchi
The Accent of Tsuruoka Japanese Reconsidered 47
Takeru Honma
How should we Represent 'g' in loge in Japanese Underlyingly? . . 67
Haruka Fukazawa and Mafuyu Kitahara
Domain-Relative Faithfulness and the OCP: Rendaku Revisited.. 85
/
Haruo Kubozono
Epenthetic Vowels and Accent in Japanese: Facts and Paradoxes 111
Hidetoshi Shiraishi
Prosodie Structure and Sandhi Phenomena in the Saru Dialect
of Ainu 141
Shin-ichi Tanaka
The Emergence of the 'Unaccented': Possible Patterns and
Variations
Shohei in Japanese Compound Accentuation
Yoshida 159
An Element-Based Analysis of Affrication in Japanese 193
Yuko Z. Yoshida and Hideki Zamma

The Accent System of the Kyoto Dialect of Japanese
A Study on Phrasal Patterns and Paradigms 215
Vili Contents
II. Studies in Japanese Morphology
Taro Kageyama
Word Plus: The Intersection of Words and Phrases 245
Takayasu Namiki
Further Evidence in Support of the Righthand Head Rule
in Japanese 277
Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo

Against Headedness in Compound Truncation:
English Compounds in Japanese 299
III. Studies in Contrastive Japanese-English Phonetics

and Phonology
Yosihiro Masuya
Two Different Kinds of Rhythm: Japanese and English 327
Noriko Yamane
sC Clusters as Complex Segments: Evidence from the Contrastive
Phonology of English and Japanese 357
Author Index 389

Language Index 394
Subject Index 396
List of Contributors 401

I. Studies in Japanese Phonology
Heaviness in Interfaces*
Yukiko Akasaka and Koichi Tateishi

Osaka University and Kyoto University of Foreign Studies
Introduction
Weight distinctions in linguistics have been considered to belong in the

realm of phonology in the past three or four decades.1 Weight here usual-
ly refers to counting of phonological units such as segments, moras, and
syllables. Recently, it has been found that there are cases of "weight"-
bound phenomena sensitive to word-counts, e.g. right-branch effects. In
syntax, we find only a few cases of "weighf'-sensitivity, such as Heavy-NP
Shift in English and other languages. However, partly because the num-
ber of cases found is limited and partly because relying on notions like
"weight" in syntax contradicts the tacitly assumed modularity assumption
in linguistic theory, the syntactic notion of "weight" has not been made
clear enough.
In this paper, however, we propose that phonological and syntactic
(and morphological) notions of "weight" have a feature in common,
namely branching. Branching at some level of linguistic representation
makes the terminal string "heavy," and all the cases of "heaviness" effects
are actually cases of alignment of "heavy" branching constituents at the
edge of some upper-level constituent. The view in no sense is novel. For
example, Hayes (1980, 1982) and Haraguchi (1991) have already noted
the relation between branching, heaviness and foot-edge alignment in
stress theory. We here attempt to generalize it to other "heaviness" ef-
fects, using examples from English and Japanese.
This, in other words, may cast doubt on the aforementioned modular-
ity assumption of components of grammar. We have in mind a picture of
grammar not as a fork-like organization as in recent minimalist program
(Chomsky 1995), but as a picture of grammar which processes phonolo-
gy and syntax in a parallel fashion, and the so-called LF derives from
some level of syntax-phonology evaluation by a set of operations per-
haps to be called "Interpret a." This result in turn supports our view of
grammar as proposed in Akasaka (1996) and Tateishi (1999), in which
4 Yukiko Akasaka and Koichi Tateishi
we argue that phonology and syntax are more or less intermingled with
each other.
1. Problem settings
What is branching, which we identify as "heaviness"? Branching is a node

in a constituent structure that dominates more than one node.
(1) Branching: a node A branches iff it immediately dominates Β and C

and neither of Β and C dominates the other
Β C
a. Branching in phonology 2
Word
Foot Foot
Syl Syl Syl

I I I
a nec dote
b. Branching in morphology
Word
Stem Suffix
Stem Suffix
I I
Christ ian ity
c. Branching in syntax
Tense
John Tense
Tense ν
John ν
ν loves
loves ν
loves Mary
The phonological and morphological trees above are mostly left-branch-

ing, and the syntactic one is right-branching. The notion of branching
plays a role in phonology in two ways. First, the morphological and stress
feet basically consist of two moras or syllables, so that they mostly branch
into moras/syllables. Because branching is only found in an output of
structure building, we can call such cases branching in Structural Change.
Second, in domain construction processes like foot formation, whether
the syllable branches or not plays a significant role, so so-called weight-
sensitivity effects are observed (Hayes 1980). In such cases, branching
functions as the trigger of a grammatical process, so it can be called
branching in Structural Description. In traditional generative grammati-
cal terms, branching in phonology plays a role both in Structural Descrip-
tion and in Structural Change.
In syntax, however, branching only plays a role in Structural Change.
The basic structure construction process in syntax these days (Chomsky
1995) is that a pair of words/phrases are put into a set and behave as a
new phrase. The process is called Merge. For example, in the above syn-
tactic tree, the first process of syntactic structure construction is to
choose the words loves and Mary and put them into a new set (= node in
traditional syntactic terms) which is labeled loves again. As structure
building always creates a branching structure, we can say that the Struc-
tural Change of Merge always branches. On the other hand, branching
hardly plays a role as a conditioning factor of another kind of syntactic
operation: Move. For example, we have never encountered a language in
which processes like Wh-Movement or Passive are triggered only if the
noun phrase branches: It is very unnatural for a language that the noun
phrase Dogs can passivize while the dog cannot or the other way around.
Also, we cannot imagine a language in which who can Wh-move while
who the hell cannot. This shows that branching functions as an output
condition for almost all structure building processes in all components of
grammar related to forms (we will come back to morphology later be-
cause morphology behaves like phonology in some respects and like syn-
tax in others) as opposed to meanings.
However, there is a group of movement rules that seems to be triggered
by some kind of branching/heaviness. English Heavy NP Shift and Japa-
nese Scrambling are canonical cases of this. The common feature of the
two processes is that they both are optional and they both do not conform
to the "Last Resort" view of movement rules which is assumed as stan-
dard in current syntactic theories like the minimalist program (Chomsky
1995). In the current theory of syntax, movement must be triggered by
some purely (morpho-)syntactic reasons. For example, a movement pro-
cess which is generally called Passive must be triggered by the Case-as-
signing/checking properties of verbs and nouns, and Wh-Movement is
also triggered by the necessity of attraction of the feature [+Wh] by
COMP. Because there is a well-established trigger, it is currently assumed
that all movement processes must be obligatory. However, a process like
Heavy NP Shift is optional, and, moreover, it does not have a good can-
didate of its trigger. Heavy NP Shift occurs only because the NP is
"heavy" in some sense, as illustrated in (2).
(2) Heavy NP Shift

a. I found the word pneumonoultramicroscopicsilicovolcanoconiosis
in the dictionary.
b. I found in the dictionary the word pneumonoultramicroscopicsili-
covolcanoconiosis.
*I found in the dictionary the word tsetse.
We will come back to such examples later in section 4 and conclude

that such cases should be accounted for in terms of prosodie branching
(i.e. this long single word consists of two or more prosodie words). Of
course, the notion of prosodie word is phonological. Does this mean
that syntax utilizes the phonological terms, casting doubt on the mod-
ularity issue of components of grammar? We will answer this question
by "Yes and no." Yes, phonological processes like prosodie category
formation can affect syntax, and vice versa (Selkirk and Tateishi 1988,
1991). No, this does not show that syntax is not a separate module. It is
the nature of the component called syntax that has been wrongly un-
derstood. Tateishi (1999) points out that the semantics-oriented view of
syntax as is currently assumed in the minimalist program cannot stand
as is, because such a view cannot properly "spell out" phonological
strings out of syntactic trees and cannot properly describe various phe-
nomena related to phonology-syntax interface, such as word order and
the distinction between function and content words. Akasaka (1996),
on the other hand, argues that the semantics/LF-oriented view of syn-
tax cannot properly explain a simple syntactic fact of Binding. Based
on these studies, we both conclude that syntax must be more "phonol-
ogy/form-oriented." Contrary to the view of syntax in Chomsky (1995)
that syntax builds up a structure which can be properly interpreted in
semantics and that phonology "strips off" a phonological matrix out of
such a structure (SPELL OUT), we say that syntax builds up a struc-
ture which is pronounceable, and semantics "strips off" semantic infor-
mation out of it (Interpret a). This, we believe, is an appropriate form
of components of grammar which can ultimately explain form/meaning
interfaces.
In what follows, we will first introduce heaviness effects in phonology
and morphology (sections 2 and 3), and then point out problems related
to syntactic heaviness effects: Scrambling and Heavy NP Shift (section
4). Section 5 summarizes our view of grammatical components and their
interactions, and concludes the paper.
2. "Heaviness" in phonology:
Foot, accent, and weight-sensitivity
As phonology is basically concerned with linear strings of sounds, the rel-

ative length of strings plays an important role. For example, many stress
systems of the world's languages have so-called "weight-sensitive" sys-
tems. In this section, we would like to look at cases of weight-sensitivity
in English and Japanese.
English and Japanese are weight/quantity sensitive in apparently dif-
ferent ways. In English, the weight-sensitivity effects are attested in the
stress system. Let us first consider a very simple and well-known case of
English. 3
(3) English Main Stress (Nouns 4 )

a. The final syllable is ignored even though it is heavy
ínseeí, dúmpfeg, jéoparëy
b. The main stress is placed on the penultimate syllable if it is heavy,
banana, Massachusetts, excursus, poíáto
c. Otherwise, the main stress falls on the antepenultimate syllable.
América, fóllaby, policy, agony
If we follow Hayes (1995) and analyze the English case as right-to-left

noniterative moraic trochee construction with final syllable extrametri-
cality, the patterns are schematized as follows:
(4) a. The final syllable is ignored.

[Ft(sylMi)]^Sy)SeC0
b. The penultimate syllable gets stress if heavy.
(syibaJlnisyiSDlV»*
c. The antepenultimate gets it otherwise
(sylA)[Ft(sylmé)(sylri)](sytea)
(where crossed-out syllables are extrametrical and underlined ones
are heavy)
(5) Moraic Trochee in English

[word-[FtH(eavy)]S¥L]
[ W o r d ...[ F t L(ight) L ( i g h t ) ] S ¥ L ]
(where underlined syllables are stressed)
Except for (4a), where, with the final syllable extrametricality, only one
candidate is left to place accent, this pattern of stress assignment can be
accounted for in terms of branching into moras. The main stress feet in
(4b) and (4c) have the following structure:
(6) a. (4a) ba[ F t na]na b. (4b) A[ F t men]ca

Foot Foot
I
Syllable Syllable Syllable
η a a
We can generalize such cases in the following way:
(7) The main stress foot must be branching into moras.
This is a case of branching condition on Structural Change (and Structur-

al Description, at the same time), i.e. an output. Here, we have our first
case of branchingness effect, in phonology.
In Japanese, the heaviness effect is a little different from English. Jap-
anese is a typical pitch accent language in that accent is expressed by an
abrupt fall of pitch on a designated syllable. The heaviness effect in Japa-
nese is observed when an accent is placed on a heavy syllable. In Japanese,
the vocabulary is classified into four classes, and, as Itô and Mester (1995)
argue, these classes adopt different ranges of constraints. For example,
while we cannot predict the place of accent with Yamato (= Native Japa-
nese) nouns, as McCawley (1968) correctly points out, Foreign nouns usu-
ally have mostly predictable accent patterns, i.e., the accent is on the an-
tepenultimate mora.5
(8) Yamato nouns

a. miyako 'city' (no accent)
b. kábuto 'helmet' (initial accent)
c. kokóro 'heart' (penultimate accent)
d. atamá 'head' (final accent)
(9) Foreign nouns6·7

a. rikuésuto 'request'
b. intaanétto 'internet'
c. súpai 'spy'8
d. depáato 'department store'
However, if the antepenultimate mora happens to be the second mora of

a heavy syllable, the accent shifts to the preantepenultimate mora.
(10) The preantepenultimate patterns with heavy syllables

a. karéndaa 'calendar'
b. rifcéfchia 'rickettsia'
c. nyuuyóokaa 'New Yorker'
d. sooraökaa 'solar car'
This means that, whatever the exact content is, we have constraints of the
following kind in an OT-type constraint form.
( 1 1 ) ANTEPENULT 9
Place an accent on the antepenultimate mora.
( 1 2 ) ALIGN(ACCENT, LEFT, *BR(SYL)) 1 0 :

Accent on the branching syllable
(= heavy syllable) must be on the left position.
(12) is higher-ranked than (11), so that the antepenultimate regularity

does not block the shifting of an accent to the initial position of a heavy syl-
lable. We see here another mode of reference to branching in phonology.
3. "Heaviness" in morphology
Heaviness effects are observed in morphology, too. In this section, we will

introduce four kinds of examples of branching effects in the realm of mor-
phology. The four patterns constrain different kinds of branching, show-
ing that reference to branching is widespread inside grammar. We also
note that the morphological reference to branching/heaviness ranges
from micro-level constituents such as segments and feet to macro-level
constituents like words. This in turn casts doubt on the status of morphol-
ogy as an independent component of grammar.
3.1. Minimal word in English
In English, a person's name can be contracted to form a hypocoristic form

(= nickname).
(13) English hypocoristics

Elisabeth > Betty
Catherine > Cathy
Robert > Bob
Mary > Mar [mae:r]
Beatrice > Bee
Margaret > Peggy
etc.
These hypocoristics all consist of two moras, as pointed out by McCarthy

and Prince ( 1 9 8 6 ) .
(13')[b[ Mora e]t[ Mora i]]n

[^[μοΓ30Ε]Θ[ΜΟΓ31]]
[Minora0] [\lorab]]
[m[Moraae][Moraaer]] (where a long vowel is represented as two consec-
utive identical vowels, only conventionally)
[b[Morai][Morai]]
[P[Morae]g[Mora1]]
A nickname with one mora cannot exist in English:
(14) Betty > *Be [be]

etc.
nor can a lexical word with one mora exist.
(15) s p a : ΜΜΟ^ΗΜΟ^Ι]
tea:
Minorai] [Moral]]
CUe: [kj[MoraU][MoraU]]
The derived words such as hypocoristics are minimally bimoraic and the
minimal size of content words is also bimoraic. Based on this and other
facts from other languages, McCarthy and Prince (1986) conclude that the
minimal size of the word universally is a bimoraic foot.
(16) Minimal Word: [Foot Mora Mora]
In other words, we can say the following:
(17) A word must be minimally complex/heavy (i.e. branching). 12
Here, we see that branching into moras is an output condition on word

formation.
3.2. Japanese hypocoristics
Japanese also has hypocoristic word formation patterns (Poser 1990,

Tateishi 1989) just like English. It is possible to form various kinds of
nicknames from names and other words.
(18) name nickname

Saburo Saa-chan, Sabu-chan, Buu-chan, Sai-chan
Hiroshi Hii-san, Hiro-san
Yoneko Voí-chan, Yone-san, Yoo-chan, Neko-san
Haruka Ηαα-chan, Haru-chan, //ai-chan
etc.
(where -san, -chan are vocative suffixes)
Like English, the size of hypocoristic forms (if we ignore suffixes) is all bi-
moraic. This is the second argument for the minimal complexity require-
ment (17).
3.3. Word contraction in Japanese
Itô (1990) observes interesting patterns of word contraction forms in Jap-

anese. Japanese has a fairly productive process of word contraction (see
also Nishihara, Van de Weijer and Nanjo, this volume). The contracted
forms can be used in place of full forms in most cases.
(19) Japanese contracted words
a. Bimoraic Patterns
sutoraiki > suto 'strike'
demonsutoreeshon > demo 'demonstration'
negachibu > nega 'negative'
hisuterii > hisu 'hysteria'
modan booi > mobo 'modern boy'
etc.
b. 4 mora patterns
paasonaru konpyuutaa > paso-kon 'personal computer'
purinto kurabu > puri-kura 'Print Club (a kind of instant camera
booth popular among youths)'
Hirakata Paaku > Hira-Paa 'Hirakata Park (the name of an
amusement facility in Osaka)'
dezitaru kamera > dezi-kame 'digital camera'
poritikaru saiensu > pori-sai 'political science'
Kimura Takuya > Kimu-Taku (the name of a popular singer/actor)
As is obvious, examples in (19a) are all bimoraic and the examples in

(19b) are all compounding of two bimoraic forms. It appears that the mo-
rale branching output condition also accounts for these cases.
However, as Itô correctly points out, when the word begins with a
heavy syllable, we cannot have bimoraic contraction. We have trimoraic-
ity in such cases.
(20) maikurofon > maiku, *mai 'microphone'

shinpashii > shinpa, *shin 'sympathy'
konbineeshon > konbi, *kon 'combination'
kooporeeshon > koopu, *koo 'cooperation' 13
Note that the bimoraic patterns like nega above all branch on the root
node, while ill-formed examples like *kon for konbineeshon do not.
( 2 1 ) [word[syln[Morae]][sylg[Moraa]]]] [wordtsyl^ Llora0][fdora11]]]
That is, when the words are analyzed into syllables, we see bisyllabic
branching in nega and other bimoraic cases and no branching in kon and
other ill-formed bimoraic cases. In compounded 4 mora cases, we see
branching into words on the root node.
(22) [Word [Word [Sy|p [ Mora a] ] [Sy,s [ Mora o] ] ] [ Word [ Syl k [MoraO] [ M o r a n ] ] ] ]
Also, the well-formed trimoraic forms like konbi show branching into syl-
lables on the root node.
(23) [word[syl^[Mora 0 ][tvlora 1 1 ]][syl^[[vlora']!]
Thus, we can generalize the pattern as follows:
(24) (Unlike nicknames) contraction forms require branching on the root

node.
We have no idea why hypocoristics and contraction forms require differ-

ent kinds of branching. However, we can at least say that the notion of
branching can generalize all these word-formation patterns. 14
3.4. Compounding and right-branch effects
In this section, we will introduce a completely different kind of branching

effect in morphology and phonology, which pertains to the domain of ac-
centuation. First, English has a rule of stress assignment to compounds
which assigns the main stress on the first word of the two component
words of compounds.
(25) [ N bláckbird]
[ N tówel rack]
[ N Énglish teacher] 'English teacher (not a math teacher)'
This leftward stress assignment is kept intact in compounds with three or

more words.15
(26) [ N [ N towel rack] factory]

[ N [ N Énglish teacher] union]
[ N [ N video game] cartridge]
However, this generalization has a group of important exceptions. All

complex examples in (26) are left-branching structures. In the right-
branching structure, the main stress of the whole compound falls on the
main-stress of the right-branch (cf. Chomsky and Halle 1968, Liberman
and Prince 1977).
(27) [ N bathroom [ N tówel rack]]

[NJapan [ N [ N Énglish teacher] union]]
[ N nintendo [ N video game]]
If we call the domain on which the main stress of the compound falls Ac-
centual Phrase (Acc) 16 , we see the following schematic patterns.17
(28) [ n A B ] : [ A c c A ' B ]
U N / A C C A B ] C]:[AccA'BC]
[nA [N/ACCB C]]: [AC<A][accB' C]
The compound stress rule assigns the main stress on the leftmost word in
the Accentual Phrase, and, if two or more Accentual Phrases exist in one
compound word, the rightmost Accentual Phrase bears the main stress,
following the Nuclear Stress Rule (Chomsky and Halle 1968) or some
regularities which correspond to it, which assigns the main stress on the
rightmost element. To establish this Accentual phrasing, we need the fol-
lowing constraint:
(29) ALIGN(BR(WORD),LEFT,ACC-PHRASE): Align the left edge of the

branching words with the left edge of an Accentual Phrase.
We see here our first case of branching in Structural Description in mor-

phology.
The same constraint seems to be at work in Japanese, too. As is dis-
cussed in Kubozono (1988), Japanese compound accentuation is also sub-
ject to the right-branching effects. For the purpose of exposition, we will
only discuss compounding of "long" 18 words in Tokyo Japanese, but the
other classes of words have basically the same branching effects. In this
word class, accent in compound nouns always falls on the final word, and
this accent will be the only accent that is found in the word (see McCaw-
ley (1968) and others).
(30) Japanese compound nouns

yamá 'mountain' + nobori 'climbing' > yamanóbori 'mountain climbing'
udé 'arm' + tokee 'clock/watch' > udedókee 'wrist watch'
sentaku 'washing' + hasami 'scissors' > sentakubásami 'pinch'
ishi 'stone' + atamá 'head' > ishiátama 'stubborn (person)'
However, the pattern does not apply to more complex compound nouns.
With the left-branching structure, the above compound pattern applies.
(31) Left-branching compounds

huyú 'winter' + yamá 'mountain' + nobori 'climbing' > [ N [ N huyuya-
ma] nóbori]
'climbing a mountain of winter'
(c.f. huyuyama 'mountain in winter')
udé 'arm' + tokee 'clock/watch' + koojóo 'factory' > [ N [ N udedokee]
kóojoo]
'wrist watch factory'
kasétto 'cassette' + téepu 'tape' + rekoodaa 'recorder'
> [ N [ N kasettoteepu] rekóodaa] 'cassette tape recorder'
keetai 'pocket-size' + denwa 'telephone' + bangóo 'number'
> [ N [ N keetaidenwa] bángoo] 'number of a cellular phone'
In all the examples in (31) and other left-branching compounds, accent

falls on the final word, following the pattern in (30). With the right-
branching structure, on the other hand, the pattern breaks down.
(32) Right-Branching Compounds

tánki 'short period' + káigai 'overseas' + ryuugaku 'studying
abroad'
> [ N tánki [ N kaigairyúugaku]] 'short-term study abroad'
(Kubozono 1988 p. 48)
Kyóoto 'Kyoto' + gaikokugo 'foreign language' + daigaku 'universi-
ty'
> [ N Kyóoto [ N Gaikokugo Dáigaku]] 'Kyoto University of Foreign
Studies'
rika 'natural science' + shuuchuu 'intensive' + kooza 'lecture'
> [ N rika [ N shuuchuukóoza]] 'intensive lecture on natural science'
With the three-word right-branching structure, accent on the first word is

kept intact, unlike the other structures, and the branching on the right
also bears accent. This means that we have two Accentual Phrases in the
right-branching structure, and we have only one in the left-branching and
simple compounds. The pattern is exactly the same as (28) above for En-
glish compound stress. Thus, we have the constraint (29) for Japanese,
too.19 Branching works as the trigger of phonological phrasing in both
languages.20
3.5. Why branching? and Why not branching?
Thus far, we have introduced phonological and morphological phenome-

na in which branchingness of some kind plays a crucial role. Whether a
constituent branches or not seems to determine the constituent's
"weight" in many ways. In the English Main Stress Rule, the weight of a
syllable determines the location of a foot. In the Japanese Accentuation,
the weight of a syllable attracts an accent. All the minimal-word con-
straints show that the word must have some weight to be a word. Finally,
Accentual Phrasing sees the "heavy" constituent in the middle and avoids
from becoming too heavy as a prosodie category. There simply are many
branching/heaviness-sensitive regularities in phonology and morphology.
Why is that so? We would say that branching/heaviness is prominent
in phonology and morphology because we have contrasts between light
(non-branching) and heavy (branching) in phonology and morphology.

Contrasting with bimoraic long/heavy syllables, there always are mono-
moraic light syllables. Stress systems of languages have some sort of pa-
rameter which determines whether the system allows for non-branch-
ing degenerate feet. Due to the nature of structure building in
phonology and morphology, we have to have an option between light
and heavy.21 Thus, we can "see" heaviness as something that contrasts
with lightness.
Then, why can't syntax work this way? Why can't there be a language
in which a wh-phrase moves only if the phrase branches, and so on? We
think that this is due to the nature of structure-building in syntax per se.
In the current theory of syntax like the minimalist program (Chomsky
1995), phrasal projections and concatenations are reduced to an ex-
treme. For example, a noun phrase like dogs, which we have analyzed as
in (33a) until quite recently, is analyzed as in (33b) in the minimalist
program.
(33) a. [ NP [ N .[ N dogs]]]
b. dogs
That is, dogs is dogs, and we do not need any unnecessary projection of it
unless dogs concatenates with something else and forms a new phrase.
Thus, we can say:
(34) There is no unary phrasal projection in syntax.
Of course, nouns like dogs concatenate with some other words, like a
transitive verb likes. Likes also does not project into unnecessary projec-
tions, so the concatenation of likes and dogs (called Merge in recent
terms) derives the following structure:
d
(35) [likeslikeS °gs]
All other syntactic projections and concatenations function in this way, so

we can say:
(36) In syntax, structure-building processes are always binary.

Because syntax is binary, binarity in syntax is not something that we can

mark as prominent. We think that this is why there is no branchingness
condition on syntactic processes.
With all these considerations, we can hypothesize the following:
(37) If a grammatical process/derivation is conditioned/constrained by

branching/ heaviness, it is not syntactic.
With this in mind, we will look at two syntactic phenomena, Japanese

Scrambling and English Heavy NP Shift. We will argue that branching-
ness effects exist for these two processes and that at least a portion of the
two processes is phonological.
4. "Heaviness" in syntax-phonology interfaces
4.1. Scrambling and self-embedding
Scrambling in Japanese has been a matter of controversy throughout the

history of Japanese linguistics. Japanese is known for its relatively free
word order except for the fact that the verb and the tense always come at
the final position. This word-order permutation is called Scrambling.
(38) Japanese Scrambling

a. Taroo-ga Hanako-o ai-si-teiru.
Taro-NOM Hanako-ACC love-do-PROG
'Taro loves Hanako.'
b. Hanako-o Taroo-ga ai-si-teiru.
(where a phrase in boldface is Scrambled out)
This has triggered a controversy over whether Japanese, unlike English,

has a VP which dominates the verb and its object so that the two constit-
uents are located in fixed positions (Ross 1986, Hale 1980, Saito 1985,
among others). Even if the existence of VP in Japanese is persuasively ar-
gued for by Saito and Hoji (1983), Miyagawa (1989) and others, Scram-
bling is in the center of syntactic controversies. The central issue about
Scrambling relates to its optionality. In (38) above, the a. and b. sentences
are both well-formed. We can see the same kind of optionality in more
complex Scrambling cases.
(39) a. Taroo-ga Hanako-ni Hiroshi-o shookai-shi-ta.

Taro-NOM Hanako-DAT Hiroshi-ACC introduce-do-PAST
'Taro introduced Hiroshi to Hanako.'
b. Taroo-ga Hiroshi-o Hanako-ni shookai-shi-ta.
c. Hanako-ni Taroo-ga Hiroshi-o shookai-shi-ta.
d. Hanako-ni Hiroshi-o Taroo-ga shookai-shi-ta.
e. Hiroshi-o Taroo-ga Hanako-ni shookai-shi-ta.
f. Hiroshi-o Hanako-ni Taroo-ga shookai-shi-ta.
In a current syntactic theory like the minimalist program, all movement

rules move constituents for a reason. For example, the classical Passive
rule is now a subcase of movement of a DP to get its Case feature checked
or attraction of a Tense (the head of TP) to discharge its EPP (Extended
Projection Principle; the Tense must have the subject (or, SPEC)) feature.
(40) [TenseTense [ v ... V ... DP-Object]] -> [ Tense DP-Objcct, LTenseTenSe [v ...
V-tJ]]
By being in a SPEC-head relation with the head Tense, the Nomina-

tive Case feature of the DP is checked and the EPP feature discharged.
A typical A-bar-movement like Wh-Movement is also triggered by
some feature checking necessity of COMP. There is no movement pro-
cess in syntax which has no reason. Hence, it follows that all movement
processes are obligatory. The optionality of Scrambling directly chal-
lenges this.
Scholars are divided into two groups on this issue. Saito, Fukui, Ta-
kano and others (Fukui 1993, Saito 1992, Saito and Fukui 1998, Fukui
and Takano 1998, Takano 1998) argue that Scrambling is a really option-
al movement process in syntax and Universal Grammar must have some
device to allow for such optional movement with no trigger. On the oth-
er hand, Miyagawa (1997) argues that Scrambling indeed has a driving
force, such as Case-checking and focus, that is, Scrambling must move ei-
ther the focus leftward or must be a subcase of Α-Movement for Case-
checking/agreement. In this paper, we will take Saito, Fukui and Takano
to be correct and we assume that Scrambling is an optional movement
process. We basically follow Takano's (1998) argument for this position.
For example, if we follow Miyagawa (1997) and say that the short-Scram-
bling to VP is a case of focus movement to the left, we need to conclude
that nothing can move and adjoin to VP unless it is a focus. This simply
is not the case.
(41) Q: Taroo-wa [ VP ehon-o itu ni-satu kat-ta-no]?

Taro-TOP picture book-ACC when two-books buy-PAST-Q
'When did Taro buy two picture books?'
A: Taroo-wa [VPehon-o kinoo ni-satu kat-ta]./
Taroo-wa [VPkinoo ehon-o ni-satu kat-ta],
'Taro bought two picture books yesterday.'
(where italicized elements are focused and boldfaced elements are
Scrambled)
A Scrambled element need not be focused. Also, if we follow Miyagawa

(1997) and if medial-distance Scrambling to IP is a case of A-Movement
triggered by Case-checking/ agreement, we get into trouble when a PP is
Scrambled to IP, as PP need not be Case-checked.
(42) a. Taroo r ga jibun¡-no-hurusato-kara tayori-o yokoshi-ta.

Taro-NOM self-GEN-birthplace-from letter-ACC send-PAST
'Taro sent a letter from his country.'
b. Jibuiij-no-hurusato-kara Taroo r ga tayori-o yokoshi-ta.
Thus, we will assume that Scrambling is optional. We will discuss later

Boskovic and Takahashi (1998), the third type of approach to optionality
of Scrambling.
However, we would like to say that Scrambling is partly triggered by
phonology. The so-called self-embedding construction (Kuno 1973) is a
case in point. 22
(43) a. ?*Hanako-ga [ IP Jiro-ga [ IP Taro-ga [IPMichiyo-ga byooki-da]-to it-

ta]-to hookoku-shi-ta]-to omot-ta.
Hanako-NOM Jiro-NOM Taro-NOM Michiyo-NOM sick-PRES-
that say-PAST-that report-do-PAST-that think-PAST
'Hanako thought that Jiro reported that Taro said that Michiyo
was sick.'
b. liplipfrpfrpMichiyo-ga byooki-daj-to fIPTaro-ga t it-ta]j-to [ [P Jiro-
—
ga t hookoku-shi-tall-to [, P Hanako-garomot-tal1.
(where boldfaced, italicized and underlined constituents are moved

out of corresponding trace positions)
According to Kuno, self-embedding is defined in the following way:

(44) Self-Embedding
A non-terminal node of category X constructs a self-embedding
structure iff
[χΑΥΒ]
where A and Β are not empty and Y either is or contains some ele-
ment of category X.
Note that the self-embedding effect is not caused by the processing bur-
den of the stack of DPs with the same case-marker. If this is the case, the
so-called multiple-subject construction must all be unparsable. This sim-
ply is not the case.
(45) a. [jpSakana-ga tai-ga Akashi-ga ii]-to Hanako-wa it-ta.

fish-NOM red snapper-NOM Akashi-NOM good-that Hanako-
TOP say-PAST
'Hanako said that it is red snappers in Akashi among fishes which
are good.'
b. [ipKyoto-ga otera-ga tatemono-ga hurui]-no-wa yuumee-da.
Kyoto-NOM temple-NOM building-NOM old-that-TOP famous-
PRES
'It is famous that, in Kyoto, buildings of temples are old.'
Also, the A strings of (44) in the self-embedding construction need not be

of the same category. In (46), the A elements in the embedded IP's are all
different, but the self-embedding effect shows up.
(46) ?*Hanako-ga [ IP kinoo [1Pkanojo-ni [jpMichiyo-ga byooki-da]-to Jiro-

ga it-ta]-to Taro-ga hookoku-shi-ta]-to omot-ta.
Hanako-NOM yesterday her-DAT Michiyo-NOM sick-PRES-that
Jiro-NOM say-PAST-that Taro-NOM report-do-PAST-that think-
PAST
'Hanako thought that Taro reported yesterday that Jiro said to her
that Michiyo was sick.'
Also note that the self-embedding effect is not caused by structural ambi-
guity. For example, the following case of multiple across-the-board 0-as-
signment causes no processing disorder.
(47) Taro-to Hanako-to Michio-to Yuji-ga kaimono-shi, ryoori-shi, tabe,

katazuke-ta.
Taro-and Hanako-and Michio-and Yuji-NOM shopping-do cooking-

do eat clean-PAST
'Taro, Hanako, Michio and Yuji did shopping, cooking, eating and
cleaning the dishes together.' or,
'Taro shopped, Hanako cooked, Michio ate, and Yuji cleaned the
dishes.'
Even though (47) can have two possible 0-marking patterns, we can inter-
pret the sentence with the two patterns with no difficulty.
As the self-embedding constructions are all formed by legitimate syn-
tactic structure-building, the only option left to us is only to explore the
possibility of phonological explanation. In this respect, note that the In-
tonational Phrase (i.e. the domain where the intonation pitch contours
are reset to its higher position) building in Japanese always places the
boundary at the clause initial position.
(48) Taro-ga [ IP Hanako-ga tensai-da]-to it-ta.

[ IntP Taro-ga][ IntP Hanako-ga tensai-da-to it-ta] 23
Assuming that the left boundary of the clause corresponds to an Intona-

tional Phrase and that the left boundary of the maximal projection like
VP corresponds to Phonological Phrase basically following Selkirk and
Tateishi (1991), we will get the following structure for (43a).
(49) ?*[ IP Hanako-ga [ VP [ IP Jiro-ga [ VP [ IP Taro-ga [ VP [ IP Michiyo-ga [ VP byoo-

ki-da]]-to it-ta]]-to hookoku-shi-ta]]-to omot-ta]].
[utt[mtp[phpHanako-ga]][IntP[PhPJiro-ga]][IntP[PhPTaro-ga]][IntP[PhP.
Michiyo-ga][ PhP byooki-da-to it-ta-to hookoku-shi-ta-to omot-ta.]]]
(where Utt is Utterance)
In (49), we can immediately note that there are three Intonational Phras-
es with a single Phonological Phrase in a row at the utterance-initial posi-
tion. In terms of intonational rhythm, this is very unstable, as prosodie
categories as we have discussed above prefer binary rhythm patterns. On
the other hand, (43b) with Scrambling of the embedded IPs has binary in-
tonational phrases if we assume that Scrambling in this sentence is IP-ad-
junction.
(50) [ T p\\vi¡pfifMichiyo-κα byooki-daj-to íIPTaro-ga t it-tall-to [1PJiro-ga

t hookolai-shi-tal]-to [ IP Hanako-ga t omot-ta]].
—>
Ut[intp[phpMichiyo-ga] [ PhP byooki-da-to]] [ IntP [ PhP Taro-ga] [ PhP it-ta-

tQ
]] [intptphpJiro-ga] [ PhP hookoku-shi-ta-to]] [ lntP [ PhP Hanako-ga][ PhP o-
mot-ta]]]
International Phrases are the domain where a single intonational contour

is uttered, such as a clause. Thus, as an Intonational Phrase is another type
of prosodie category, preferably it should contain two Phonological
Phrases to keep binary rhythmic patterns. Scrambling of IP's in (43) does
just that. The example in (46) can also be made better by placing two el-
ements in the A string of (44). (51) shows one example.
(51) a. ?*[ IP Hanako-ga [ IP kinoo [ IP kanojo-ni [ IP Michiyo-ga byooki-da]-to

[ IP Jiro-ga it-ta]]-to [ IP Taro-ga hookoku-shi-ta]]-to omot-ta],
[utt[intp[phpHanako-ga]][ IntP [ PhP kinoo]][ IntP [ PhP kanojo-ni]][ IntP [ PhP _

Michiyo-ga][ PhP byooki-da-to]][ IntP [p hP Jiro-ga][ PhP it-ta-to]][ IntP [ PhP Ta-
ro-ga][ PhP hookoku-shi-ta-to omot-ta]]].
(three non-branching Intonational Phrases in a row at the Utter-
ance-initial position)
b. [ IP [ IP Taro-ga kinoo [ IP Jiro-ga kanojo-ni [ IP Michiyo-ga byooki-da]-
to it-ta]-to hookoku-shi-ta]-to [ IP Hanako-ga omot-ta]].
[ u t t [ i n t p [ p h p T a r o - g a ] [ PhP kinoo]] [ IntP [ PhP Jiro-ga] [ PhP kanojo-ni]] [ IntP [ PhP .

Michiyo-ga][ PhP byooki-da-to it-ta-to hookoku-shi-ta-to]][ ImP [ PhP Ha-
nako-ga][ PhP omot-ta]]].
(Intonational Phrases are all binary)
Note that in (51b), there are more arguments whose roles are not yet de-
termined in the middle of the sentence processing than in (43a) or (46),
because, until the word byooki-da-to appears, the exact semantic role of
the five phrases preceding it cannot be determined. This shows that the
self-embedding effect cannot be a matter of sentence processing.
We would say that the following constraint takes control over the
Scrambling in the self-embedding cases at least.
(52) IntPBR: A n Intonational Phrase must branch into Phonological

Phrases. 24
Thus, we would like to say that at least some cases of Scrambling are pro-
sodically-determined. We will come back to the theoretical implications
of this in Section 5.
4.2. Heavy NP Shift
Heavy NP Shift in English is troublesome for the syntactic mechanism

when we come to the level where the word order is determined. This is
due to the nature of English movement rules. Movement is almost always
leftward in English and current syntactic theories like the minimalism
only allow leftward movements. However, Heavy NP Shift appears to be
a rightward movement. In English, the direct object immediately follows
the verb. Basically, the sentence is ungrammatical if the direct object is
not adjacent to the verb.
(53) a. He attributed the fire to a short circuit,

b. *He attributed to a short circuit the fire.
However, when the object DP 25 is 'complex' or 'heavy,' the object DP

need not be adjacent to the verb.
(54) a. He attributed to a short circuit the fire which destroyed most of my

factory.
b. He attributed the fire which destroyed most of my factory to a short
circuit.
Traditionally, this phenomenon has been analyzed as the transformation

by which the object DP moves rightward when it is "heavy." This trans-
formation has been called Complex NP Shift or Heavy NP Shift (Ross
1986, Fiengo 1974).
4.2.1. Why is Heavy NP Shift a problem?
In a syntactic theory based on principles and parameters, Heavy NP Shift

is one of the rightward movements such that the constituent of a sentence
moves rightward by Move a. In the minimalist program (Chomsky 1995,
Kayne 1994), however, this rightward characteristic makes Heavy NP
Shift notorious for syntactic theory. Let us show why.
The word-order determination in the minimalist program is the cause

of the trouble. In Kayne (1994), a syntactic element A which asymmetri-
cally c-commands another element Β precedes Β without exception. 26
Thus, in the structure in (55), the word order relation is always SVO.
(55) [ IP Subj I [ v p V [ NP Obj]]] (order information not included)

Subj asymmetrically c-commands V and Obj, so that Subj precedes
V and Obj.
V asymmetrically c-commands Obj, so that V precedes Obj.
The syntactic tree itself does not include any ordering information.
For Kayne, the very basic word order for the world's language thus is
SVO, and the SOV order as seen in Japanese is derived by application
of Move a.
Fukui and Takano (1998), on the other hand, agree with Kayne in as-
suming orderless trees but differ in their mechanism of word-order deter-
mination. Putting aside trivial details, they claim that the word-order de-
termination is done with a universal syntax-phonology mapping rule
which says, "The projection of the head comes later." As the head of a
phrase always come at the end in their mechanism, the default word order
for them is SOV, and an English-style SVO order is "derived."
For both Kayne (1994) and Fukui and Takano (1998), the problem
which Heavy NP Shift poses is the same: Heavy NP Shift appears to be
a rightward movement. As we all know, Heavy NP Shift places a
"Heavy" DP in the final position of the sentence. If we assume that the
position of the Heavy DP is a VP-adjoined position, we have the follow-
ing structure: 27
(56)
VP
VP "Heavy" DP
Because the Heavy DP asymmetrically c-commands all elements domi-

nated by VP, for Kayne, the linear ordering of the Heavy DP must be al-
ways to the left, contrary to the fact. For Fukui and Takano, because the
Heavy DP is not a lower projection of the root node, it must precede the
rest of the VP. In either theory, the Heavy DP cannot be placed on the
right, which it must be.
The solutions to this problem for them look similar. Let us first intro-
duce Kayne's account of Heavy NP Shift. Kayne (1994) proposes that
Heavy NP Shift is a kind of Scrambling, and claims that Heavy NP Shift
involves leftward movement, not rightward movement, following his gen-
eralization about asymmetric c-command. Thus, an example like (57):
(57) John gave to Bill all his old linguistics books.
must have a structure as in (58) before Spell-Out and application of his

word-order determination.
(58)... [all his old linguistics books [gave to Bill]].
The constituent to Bill moves leftward independently of Verb-Raising,

which raises the verb gave to an upper position. As a result, the structure
in (59) is derived.
(59) John gave [[to Bill]i [Xo [[all his old linguistic books] [Y° [e]¡...
This leftward movement is regarded as a kind of Scrambling which is op-

tional. Because it is optional, (60) without Scrambling can be generated
(60) John gave all his old linguistic books to Bill.
Although Fukui and Takano (1998) deal with Heavy NP Shift only in
passing, they say that Heavy NP Shift is a process in the phonological
component, as Chomsky (1995) suggests. Although we cannot say any-
thing conclusive about what they think of when they say that Heavy NP
Shift is phonological, we can at least say that their spirit is similar to
Kayne's in that they think that Heavy NP Shift is optional and is not trig-
gered by anything syntactic.
The two theories of word order determination have one claim in com-
mon. Heavy NP Shift does not have any syntactic trigger. They do not
specify what the trigger of Heavy NP Shift is, but, from our point of view,
syntactic processes without syntactic trigger are likely to have a phono-
logical trigger. With Heavy NP Shift, we may expect that the notion of
heaviness may be phonological in nature. Thus, in the next section, we will
review the theories of triggers of Heavy NP Shift and conclude that the
trigger of Heavy NP Shift is in fact phonological.
4.2.2. Triggers of Heavy ΝΡ Shift
As far as we know, the only potential trigger of Heavy NP Shift is focus

other than something purely phonological. For example, Rochemont and
Culicover (1990) and Takano (1998) both argue that Heavy NP Shift is
triggered by some kind of (presentational) focus. It is only natural that
Heavy NP Shift is somehow related to focus, as Heavy NP Shift places
some heavy element in the sentence final position and the sentence final
position in the information flow tends to convey new information.
(61) To whom did you give the book your mother bought you for your
birthday?
- a. I gave the book my mother bought me for my birthday to my son.
b. *?I gave to my son the book my mother bought me for my birth-
day.
As in (61), it is at least odd to place the focus in the sentence-medial po-

sition on purpose by Heavy NP Shift. However, consider (62):
(62) What did you give to your son?

- a. I gave the book my mother bought me for my birthday to my son.
b. I gave to my son the book my mother bought me for my birthday.
If Heavy NP Shift is triggered by something like focus, why is it optional?

Also, if focus can always Heavy NP Shift, why can't "light" NP's shift?
(63) What did you give to your son?

- a. I gave books to my son.
b. *I gave to my son books.28
This shows us that some kind of "Heaviness" is necessary for Heavy NP

Shift.
Thus, we must say that a reason other than focus is needed for Heavy
NP Shift though Heavy NP Shift is associated with focus more or less, per-
haps indirectly.
Then, what is "heaviness" for Heavy NP Shift? Ross (1986) has already
had an answer to this problem. Ross defines "complex" NP (the term for
Heavy NP by Ross) as follows:
(64) A noun phrase is complex if it dominates the node S. ([NP...[S···]])
(64) may well be paraphrased as follows, because the only case of a noun
phrase dominating S in English is an adnominal relative clause construc-
tion which thus necessarily is right-branching:
(65) A DP is heavy if it is right-branching.
By (65), examples like (66) nicely fit in, as the shifted DP is right-branch-
ing.
(66) John invited to the party [ DP [ D phis][ N p[ A pClosest]friend]].
However, the definition in (65) does not necessarily fit all Heavy NP Shift
constructions. Consider the following example with Heavy NP Shift of a
left-branching DP:
(67) John invited to the party [Dp[DpLnp[Dph's][Npmother's]][NP

friend's]] [ NP father]]
Thus, we need (68) for the definition of Heaviness:
(68) A DP is heavy if it is branching.
As we had conceptually argued for in (37), repeated below, some gram-

matical process that refers to branching is likely to be non-syntactic.
(37) If a grammatical process/derivation is conditioned/constrained by

branching/ heaviness, it is not syntactic.
Moreover, there are two kinds of examples which show that the notion
"branching" here is indeed phonological. First, even though it is branch-
ing in a strict sense, examples like (69) generally do not undergo Heavy
NP Shift.
(69) *?I introduced to Bill [DP[DPmy][NPfiancée]].

This is perhaps due to the fact that function words such as pronouns are
phonologically weak and they do not constitute a prosodie category of
their own. Second, even though it is not branching syntactically, exces-
sively long words can undergo Heavy NP shift.
(70) a. *I looked up in the dictionary tsetse.

b. I looked up in the dictionary pneumonoultramicroscopicsilicovol-
canoconiosis.
(71) a. *I found in the dictionary the [ NP [ N word][ N math]].

b. I found in the dictionary the [ NP [ N word][ N flaucinaucinibilipilifica-
tion]].
In (71), even though the two words are branching at the word level, they
are not at the phrasal level like all the examples above. This shows that
something other than syntactic branching is at work.
As Heavy NP Shift appears to be purely optional as the studies above
say, we cannot immediately make it follow from some phonological con-
straint. However, we can at least say that Heavy NP Shift moves some-
thing "heavy" phonologically or stays in situ (if other elements indeed
move, as Takano (1998) argues). We basically follow Zee and Inkelas
(1990) and propose that "Heaviness" can be defined as branching at the
Phonological Phrase level.29 In the relative clause (complex NP) case, the
clause always makes an independent Phonological Phrase (and an inde-
pendent Intonational Phrase, as we have argued for above), so that the
complex NP always branches into two phonological phrases at least, the
head noun and the relative clause. With (66) and (67), if we assume with
Selkirk (1986) that, in English, the boundary of a Phonological Phrase is
marked at the right boundary of X max , we can have the following prosodie
structure with multiple "heavy" stacks of Phonological Phrases.
(72) John invited to the party [DP[DPhis][Np[ApClosest]friend]].

-> [PhPhis][PhPclosest][PhPfriend] or
[ PhP his closest] [ PhP friend]
(if we assume that function words are weak and cannot be indepen-
dent prosodically)
(73) John invited to the party [Dp[Dp[Dp[Dphis][Npm°ther's]][NP

friend's]] [ NP father]]
->• [phphis][phpmother's][phpfriend's][phpfather] or
[ PhP his mother's] [ PhP friend's] [ PhP father]
Examples like my fiancée do not usually undergo the shift because pro-
nouns are phonologically weak. The excessively long words can undergo
the shift because they are too long to pronounce in a single phonological
phrase. For example, pneumonoultramicroscopicsilicovolcanoconiosis
would be pronounced with several Phonological Phrases as in what fol-
lows:
(74) [ PhP pneumonoultra] [ PhP microscopic] [PhPsilicovolcano] [ PhP coniosis]
Thus, we can safely say that Heavy NP Shift is prosodically controlled.

However, this means that, as we will discuss in the following section,
Heavy NP Shift is a syntactic rule that is sensitive to phonology. This
leads us to doubt the generally-assumed organization of components of
grammar. 30
4.2.3. Apparent Problems
We have seen in the previous section that an optional syntactic process

like Scrambling often functions as a rhythmic adjustment. Also, we have
seen that Heavy NP Shift in English directly targets a prosodically-
branching DP and linearizes it to the right. What does the fact that Heavy
NP Shift can be prosodically controlled mean theoretically? This means
that we do not have any purely syntactic processes that refer to branching
in the phrase structure. This has led us to conclude that Heavy NP Shift is
phonological in nature. However, do we, or must we, conclude that Heavy
NP Shift is purely phonological? It appears that this is not the case. Al-
though it is triggered by "heaviness" in a prosodie sense, Heavy NP Shift
is subject to various syntactic conditions.31
(75) The subject cannot shift even if it is heavy.

*She/*It/*i impressed John very much [the woman who saved a kid's
life hit by a car].
(76) A prepositional object and an indirect object cannot shift.

a. *I talked to t about Jonnie's problems [all of the teachers],
b. *John gave t a book about roses [the man in the garden].
(77) A DP cannot shift over an embedded clause.

*I told t that we were in trouble [a man who had a kind face].
All these data show that, even though the Structural Description of
Heavy NP Shift involves phonological notions, Heavy NP Shift is sensi-
tive to syntactic distance, such as subject/object asymmetry and syntactic
locality. D o we have to assume such a mixture of levels in a single move-
ment process?
We believe that these facts can be explained once we think seriously of
how the rightward shift can be established in the aforementioned order-
less tree theories like Kayne (1994) and Fukui and Takano (1998). In
Kayne's and Fukui and Takano's theories, the word order information is
mapped based on the hierarchical information in tree structures. Due to
the nature of their mapping procedures, any moved element must move
to the left. In Kayne's theory, adjunction of an element X to, say, an IP
creates the following structure:
With this structure, X necessarily precedes all terminals dominated by IP,

because it asymmetrically c-commands all nodes under IP. Thus, all move-
ment rules for Kayne is leftward. For Fukui and Takano, X is demerged
and concatenated to the left at an IP level, because it is the first non-pro-
jection of the head under IP. In either theory, movement in syntax always
goes to left. There is no room for the rightward shift.
In such a situation, we have only two interpretations of the rightward
shift like Heavy NP Shift: The rightward shift occurs in phonology, or the
"shifted" element actually is left unmoved. We cannot take the first op-
tion, because the aforementioned mapping theories map structures into
strings and there is no constituent structure left after mapping. We cannot
attribute the shift to phonological constituents like phonological phrase,
because what we need here for Heavy NP Shift is two phonological phras-
es, a non-constituent string. A non-constituent string cannot behave as a

unit in a linguistic operation. Thus, we need to conclude that "Heavy NP"
is left unmoved. Takano (1998) proposes that the DP-PP pattern of the di-
transitive construction, which most likely triggers Heavy NP Shift, actu-
ally is derived from the PP-DP construction, as (79) shows. Takano bases
his conclusion on some syntactic facts like binding. If we follow Takano
and assume that the DP-PP pattern of the ditransitive construction in-
volves movement of the DP over PP, we can say that, in the Heavy NP
Shift case, the movement is deferred due to the "heaviness" of the DP.32
(79) Ditransitive Construction for Takano (1998)

John [gives to Mary books] -» John [gives books¡ to Mary t¡]
(80) Heavy NP Shift

John [gives to Mary [ DP ( PhP his diamond)( PhP rings)]] (no move-
ment due to heaviness)
Heavy NP Shift here is the deferring of syntactic Move a due to a syntactic

constituent's branching into phonological phrases. With this in mind, we
can readily account for all pseudo-syntactic characteristics of Heavy NP
Shift. First, the fact in (75) relates to the base-generable position of the
subject. Even if we generate the subject vP-internally as is generally as-
sumed, we have the following structure:
(81) It [ vP [the woman who saved a kid's life hit by a car] ν [ VP impressed
John] very much].
The verb impress raises to the projection of v, the light verb, and adjoins
to it.33
(82) It [vPimpressed¡[vP[the woman who saved a kid's life hit by a car] ν

[VPt¡ John] very much]].
If we defer the NP-movement of the subject due to heaviness of the sub-

ject, the derivation ends here. The word order we would have is *It im-
pressed the woman who saved a kid's life hit by a car John very much,
which is clearly different from the Heavy NP Shift case.34
(76) can also be explained easily. Let us first consider the prepositional
object case, which would have the following structure before the "Shift."
(83) *I talked [ PP to [all of the teachers]] [ PP about Jonnie's problems].
Now that rightward movement is not available, the only choice for us to
have (76a) would be to move the PP about Jonnie's problems between to
and all of the teachers. Such an operation is not allowed in syntax, because
it is a lowering and because it is an adjunction to an intermediate projec-
tion of a preposition. About the indirect object case (76b), we already
have an explanation. Suppose we follow Takano, which we do, and as-
sume that the natural (hierarchical) order of elements in a ditransitive
construction is Goal-Theme, as we do in the DP-PP construction derived
from the PP-DP construction. If so, we have, for example, John gave [the
man in the garden] a book about roses as the base. Now, this construction
have already all its Case checking relations satisfied as is (of course, by
later NP-movement and other checking mechanisms) and is in fact li-
censed as a grammatical sentence in English. With the Goal-Theme or-
der, there is no need for the shift of either object. As there is no option of
rightward movement in grammar, this is the only ordering relation we
have for this sentence and there is no Heavy NP Shift. (77) can also be ex-
plained if we assume that Goal comes first hierarchically. If Goal comes
first among internal arguments, we only have I told [a man who had a kind
face] (Goal) that we were in trouble. Again, without rightward movement,
we cannot have Heavy NP Shift. Heavy NP Shift (or, more precisely,
Heavy NP Non-Shift) is triggered by phonological branching, and its ap-
parent purely syntactic behavior is a by-product of the facts about English
sentence structures.
About Scrambling, we have a very interesting alternative. Boskovic
and Takahashi (1998) propose that Scrambling is not a matter of move-
ment. Rather, they say that the Scrambling construction is base-generat-
ed. According to them, later operations "lower" Scrambled elements to
its 0-marked positions.
(84) a. Pronounced structure: scrambled

Hanako-o Taroo-ga yon-da.
Hanako-ACC Taro-NOM call-PAST
'Taro called Hanako.'
b. Interpreted Structure (LF?): Reconstructed
Taroo-ga Hanako-o yon-da.
Boskovic and Takahashi's motivation for their analysis is to give a reason

for a seemingly optional operation like Scrambling: Reconstruction for Θ-
assignment. Given that Scrambling is partly controlled by prosody, their

proposal, which in a sense denies a syntactic analysis of Scrambling, is
worth serious consideration.
However, they have two conceptual problems. First, it is not clear if Θ-
roles can make derivations uninterpretable so that they trigger lowering
in this case. We do not have much to say about this, as this is a matter of
core syntax. The second problem with Boskovic and Takahashi is why a
structure like (84a) can be base-generated in the first place. Of course, we
can always say, "Because principles of the phrase structure construction
allow for it." However, why is it possible to have one additional lowering
derivation while we have always an option of direct base-generation in a
0-marked position? The only answer we can have here is that (84a) is pos-
sible for word-order determination, i.e. for phonology. Thus, even if
Boskovic and Takahashi are correct in their analysis of word-order, we
still have to explore into the "phonology" of Scrambling.
In sum, we have seen in this section that the two seemingly syntactic op-
erations, Heavy NP Shift in English and Scrambling in Japanese, are part-
ly controlled by phonology. They both refer to prosodie branching, as we
have seen in previous sections with phonological and morphological phe-
nomena. However, they are also syntactic in that they necessarily refer to
syntactic constituency and hierarchies. This may suggest that phonology
and syntax must be processed in a more parallel way.
5. Conclusion and related issues
Recall the morphological right-branch effects in English and Japanese.

The schematic patterns are repeated below.
(85)[NAB]:[ACCA'B]
UN/ACCA B] C] : [ Acc A' Β C]
[ n A [ n / A c c B C]]: [ Acc A][ Acc B· C]
Left-branching compound words (including simple binary ones) consti-

tute a single Accentual Phrase, while right-branching ones constitute two.
When can Accentual Phrase be formed? It cannot be in phonology or in
a special Morphology component, as Chomsky (1995) appears to assume.
In the theory of phrase structure proposed in the minimalist program, we
find two departures from the traditional X-bar-based theory of phrase
structure. First, as introduced in the previous section, there is no word or-

der specification in the syntactic tree. Second, there is no X-bar theory per
se. In the minimalist program, a phrase like big white dog is represented
as follows.
(86)
dog
dog
As there is no X-bar theoretic category in a phrase structure, the tradi-

tional categorial notions like Ν and NP play no role in syntax. The nodes
in a tree are only "labeled" with the name of the head. Consider the fol-
lowing compound noun:
(87)
statistics
statistics
We cannot say anything decisive about whether compound noun forma-

tion is an adjunction or substitution. Perhaps it is the former. But we can
at least say that we cannot distinguish between phrases and compounds
just by looking at the trees. The existence of an independent component
for word-formation presupposes the existence of distinctiveness of the
notion word, but this is just not available in current syntactic theory. As
the pruning of the X-bar tree notation is a welcome move for minimizing
idiosyncratic specifications in syntax, we cannot get rid of this assump-
tion. If so, the only thing we can do is to refer to the tree's interpretational
aspects directly or indirectly. If so, we can only construct prosodie catego-
ries at the same time the syntactic trees are constructed. Spell-Out or
whatever mechanisms which strips off phonological strings out of trees

cannot distinguish between words and phrases.
Also, the fact that the Nuclear Stress Rule (Chomsky and Halle 1968),
the stress-assignment mechanism for phrases, is sensitive to the selection-
al property of a verb and semantic predictability, as pointed out by Bol-
inger (1972) and re-stated by Zubizarreta (1998), may show that phono-
logical structures are already there deep inside syntax.
(88) a. I have a point to émphasize.

b. I have a point to make.
(Zubizarreta 1998, p. 70)
In (88b), for the verb make to come in the final position is highly predict-
able by the use of the word point, and thus the Nuclear Stress Rule choos-
es point as the target of the main stress.
If the above arguments are all correct, the organization of grammar will
be as follows:
(89)
Semantics
^ / ^ ^ I n t e r p r e t Alpha
Lexicon, Interpretable
Numeration Tree
Merge, Move and \^Speel-Out?
Prosodie Category
Formation (Stress?)
Phonetic
Representation
Syntax, phonology and perhaps semantics are all communicating with

each other in a single interface level ("Interpretable Tree"), and it is at
this level where the aforementioned prosodically-oriented syntactic oper-
ations occur. This is reminiscent of Zubizarreta's (1998) idea about the or-
ganization of grammar and Selkirk's (1984) conception of the S-Structure
with focal and intonational information. We would go further and say that
fully phonological constituents, such as phonological phrases, is involved
in syntax.35
What is the nature of an Interpretable Tree? In the minimalist syntax

as proposed in Chomsky (1995), the nodes of the tree are all set-theoretic
object of the following form:
(90) {α,{α,β}}
This reads as "a merger of a and β with the head a," that is, a's projec-
tion.36 The righthand member of the set expresses the syntactic and se-
mantic relations of elements, and the lefthand member of the set express-
es the headedness of the node. However, we immediately find that the
lefthand member tells us a redundant information, because, once we get
to know what a and β are, we immediately know which of the two is likely
to be the head of the phrase. This set-theoretic notation is also mysterious
in that, even though Chomsky considers syntax as communicating with
Semantics and Phonology, syntactic objects in Chomsky's tree do not con-
tain any phonological information although it does contain semantic in-
formation in some form. We conjecture that it is the lefthand member of
the node set which should contain phonological information like prosodie
categorization. The Interpretable Tree for us is a tree which contains both
relational (semantic/ syntactic) and prosodie information bits.37
In the minimalist program, derivations are intended to make a struc-
ture both semantically and phonologically interpretable. In the current
syntactic discussions, we see arguments about how we can make a struc-
ture semantically interpretable, such as issues on quantifier scope, Wh-
movement and reconstruction and 0-marking, and operations for that
purpose are syntactic derivations. In this sense, syntax and semantics in
the current syntactic theory are parallel. In the discussions with these in
mind, one basic assumption has been considered something inviolable:
Inclusiveness. Inclusiveness rules all syntactic derivations and bans some-
thing other than information from lexical items inserted in a tree. This
perhaps follows the tradition of compositional formal semantics. Howev-
er, for a tree to become phonologically interpretable, it is unavoidable to
insert something not in the lexicon. For example, it is simply impossible
to have a complete stress grid structure up to the Utterance level in the
lexicon. Phonology is just not made this way. However, as we have seen
above, rules like the Nuclear Stress Rule, which adds a stress grid, must
be applicable in syntax. We would say that the Inclusiveness theme must
be loosened to allow for pure insertion of grids and prosodie categories
for the phonological purpose. Unless we do so, we cannot have syntactic
derivations which are both semantically and phonologically interpret-
able, as we have argued for above. Strictly LF-oriented inclusive syntax

fails. As syntactic derivations construct both syntactic and phonological
structures, it is only natural that operations like Heavy NP Shift refer to
both syntax and phonology.
Tateishi (1999) has pointed out that the theory of word-order determi-
nation of Kayne (1994) or Fukui and Takano (1998) cannot distinguish
between functional categories and lexical categories, and they all are
Spelt-out as independent words. This is troublesome when we look for an
account of reduction of function words such as Auxiliary Contraction.
Without ample amount of categorial information and Prosodie Phrase
formation, we cannot distinguish between something strong and nonre-
duced and something weak and reduced. Moreover, it is well known that
there are strong syntactic conditions to Contraction, which suggests we
need two kinds of structures for the account.
Also, Akasaka (1996) points out that the well-known Binding Theoret-
ic patterns like (91) cannot be accounted for solely at LF.
(91) a. MaXj criticized himself¡.

b. *Max¡ criticized him¡.
Note that both anaphors and pronominals, except for reciprocals which
necessarily involve some quantificational operation, are variables in se-
mantics, at least truth-functionally.
(92) a. Max¡ criticized himself;.:

Ax(criticize(x)(x))(Max)
b. Max¡ thinks that Mary criticizes him¡.:
Ax(think(A(Az(Ay(criticize(y)(z))(x))(Mary))))(Max)
If the two sentences in (91) are interpreted in semantics of this kind, we

have (92a) for both, and we cannot distinguish one from the other. Thus,
the Binding Theory cannot be semantic, Akasaka argues. With the orga-
nization of grammar we have proposed for various "heaviness" phenom-
ena, we can say that Binding is a form-determination process (i.e. a pho-
nological process) which refers to indexing or whatever mechanisms that
marks an antecedent and anaphoric expressions. Binding has been
pushed out of the realm of syntax partly because Binding has been con-
sidered an S-Structure phenomenon, and the S-Structure vanished from
syntax due to the ban on non-interface intermediate levels in syntax in the
minimalist program. In our view in this paper, S-Structure is revived in
some way, as a pre-Spell-Out structure with full syntactico-semantic and

phonological characterizations. Spell-Out splits off prosody out of this
level.
All prosodie structures are already there in the pre-Spell-Out structure.
Along with syntactic structure-building such as Merge and Move, the pro-
sodie structure-building goes on. Only with this assumption, we can find
a proper account of "mixed-level" phenomena such as Heavy NP Shift,
Scrambling, Nuclear Stress Rule, and so on. This is in a sense a revival of
the notion of S-Structure, but more in a sense of Zubizarreta (1998) and
Selkirk (1984). However, this does not mean that we need a fixed level
called S-Structure. Spell-Out "splits off" only a prosodie structure at the
"Utterance" level on the phonological side of a coin, and structure-build-
ing continues until we reach that level, so we do not need to specify "S-
Structure" as something special. This makes a once-abandoned S-Struc-
ture explanation of phenomena like Binding possible again, but this time
as a theory of form determination.
Finally, we have the following three major problems to solve if what we
argue for in this paper is correct:
(93) a. Because Spell-Out now splits off all phonological materials out of
a tree, all that is left after Spell-Out should be phonetic implemen-
tations. This in a sense is well fit into the framework of grammar like
Chomsky's (1995), because there is only a real interface with mo-
tors and perceptions after a "fork" of organization of grammar. So,
we need to re-examine and see whether anything "phonological" is
left after Spell-Out.
b. Also, we need to have an account of word-order (and prosodic-or-

der) determination, as syntax with prosody can now incorporate
word-order.
c. Finally, we need to have a theory of pronunciation of empty cate-

gories of syntax. Especially, in a theory like Chomsky (1995) where
traces are copies of an antecedent, we need to have a principled ac-
count of which element in a syntactic chain is "prosodically signif-
icant."
However, these should be left for future investigations.

Notes
* The order of the authors is alphabetical and the two authors are equally responsible for
the content of this paper. The authors thank the editors of this book for the chance to
contribute this paper to the present volume. We have benefited much from comments
from an anonymous reviewer of this volume, Francis Michinao Matsui, and Elisabeth
Selkirk. We would also like to thank our colleagues at Osaka University and Kyoto Uni-
versity of Foreign Studies for their encouragement and comments. All mistakes and er-
rors which are left in this paper are, of course, solely our own.
1. However, since SPE's (Chomsky and Halle 1968) unfortunate neglect of the notion of
syllable and the use of [± tense] for length distinctions, the weight distinction in phonol-
ogy had also been neglected until the (re)birth of the metrical theory as presented by
Liberman and Prince (1977).
2. This is an abbreviated tree. There are nodes like mora under Syl(lable), for example.
3. Of course, there are many exceptions to the regularities that we introduce here, as is
mentioned in studies like SPE and others. In this paper, we will ignore all these excep-
tional cases and only discuss the core cases.
4. We will ignore the so-called "Ambisyllabicity" (Kahn 1976). However, the existence of
ambisyllabicity itself may show that the sensitivity to branching is real in a sense.
5. Yamada (1990) argues that Yamato nouns basically have an antepenultimate accent
pattern and that Japanese accent is predictable in all vocabulary classes. Yamada bases
his argument on the fact that this pattern is dominantly attested in derived words such
as compounds and newly coined words like foreign borrowings. However, we think that
these facts do not save us from specifying the place of accents in all those words with
unpredictable accent patterns.
6. There are some identifiable groups of exceptions to this regularity. Most typical of these
are those words whose accent patterns are taken from accent patterns of original lan-
guages, e.g. Chómusukii 'Chomsky,' ákusento 'accent.'
7. Japanese syllable structure is basically (C)V(V)(C), where the coda C is either a nasal
or the first half of a geminate. All segments except onsets are counted as moras. There
are some exceptional trimoraic syllables (Tateishi 1997,1998b).
8. Some speakers shift the accent to the right, due to devoicing of the vowel [u], which is
not substantially relevant to the point here.
9. Probably, this derives from the final two moras constituting a foot and from the foot in
question being extrametrical (Tateishi 1985,1992), together with an End Rule of some
kind.
10. In (12), *BR(SEG) designates a heavy syllable that is accented. We need this kind of des-
ignation to capture the fact that accent never falls on the second mora of a heavy sylla-
ble in other vocabulary classes either. We avoid formalizing this constraint as
ALIGN(ACCENT, LEFT, SYL), because this formulation is partly vacuous and redundant
as accent necessarily falls on the left (and right) edge of a light syllable anyway. Note
that the constraint in question is not undominated in an Optimality term as an anony-
mous reviewer claims, because we occasionally find non-head accentuation even in the
Tokyo dialect of Japanese (Tateishi 1997).
11. In McCarthy and Prince's theory of prosody, onsets are outside of the moraic structure.
We follow their way of representing moras.
12. As onsets and extra codas are adjuncts to syllables in McCarthy and Prince (1986), we
can say that these are not counted as branches. We avoid using the term foot explicitly,
because the status of the notion foot in morphology is not as clear as that in the stress
theory, and because using the term foot may make it impossible to distinguish between
a non-branching degenerate foot and a binary foot. What we need here is a binary
branching foot.
Function words are naturally exceptions to this generalization, as they often behave
as a syntactic affix.
13. Examples like sinkuro 'synchronization (sinkuronaizeeshon), synchronized swimming
(sinkuronaizudosuimingu)' perhaps fit into the bimoraic compounding pattern.
14. Perhaps, in the case of nicknames, the word must be only minimally complex because
hypocoristics generally are used with vocative suffixes like -san, -chan, -ko and so on.
Combined with these suffixes, the word becomes truly complex. On the other hand,
without the help from such suffixes, contraction forms must be truly complex/heavy.
Cases like retii < remontii 'lemon tea' are probably formed following such complexity
requirement.
15. We only discuss the location of the main stress.
16. We call Accentual Phrase what McCawley (1968) calls minor phrase. It comes between
Phonological Word and Phonological Phrase and bears at most one accent.
17. Compounds are of course not only formed out of nouns. Non-nominal compounds are
often subject to idiosyncracies.
18. "Length" of words in Japanese for this purpose is measured by the mora count. Roughly
speaking, if a word has three or more moras, it is long (McCawley 1968).
19. Incidentally, this Accentual Phrasing works as the domain of another Japanese phono-
logical rule called rendaku (Otsu 1980, Ito and Mester 1986, Fukazawa and Kitahara,
this volume) and the rule of rendaku also has a right-branching effect.
20. As Elisabeth Selkirk (personal communication) has correctly pointed out, in Japanese,
a right-branching Adjective-Adjective-Noun sequence, e.g. [ao_i [uma_i [ame]]] 'blue
tasty candy', is prosodically parsed into [ PPh ao_i uma_i ame]. This means that adjectives
are regarded as non-maximal in syntax-phonology mapping, if we follow Selkirk and
Tateishi (1991) and assume that it is the left edge of a syntactic maximal projection that
coincides with a boundary of a Phonological Phrase. More elaborated investigations on
the syntax of attributive adjectives are in order.
21. For example, in a language with bisyllabic stress alternation, of course a word can have
an odd number of syllables. In such a case, Universal Grammar should give us an option
of having a degenerate monosyllabic foot and of not parsing a left-over syllable into a
foot.
22. Kuno (1973) actually uses relative clause constructions. The point of self-embeddedness
is kept intact, however.
23. There are cases where (48) is pronounced with one Intonational Phrase: [ IntP Taro-ga
Hanako-ga tensai-da-to it-ta], with no resetting of the pitch contour and continuous
lowering of pitch. In such cases, our intuitive interpretation is either that Taro-ga is fo-
calized or that Hanako-ga tensai-da-to it-ta is defocalized. We only discuss the default
out-of-the-blue cases in this paper.
24. Perhaps, this constraint must be interpreted as a relative one. Other factors like focus
and topic may make a unary-branching Intonational Phrase sound better.
25. The terms NP and D P are here used interchangeably.
26. A asymmetrically c-commands Β if and only if A c-commands Β but not vice versa.
27. The point still goes through even if Heavy NP Shift is IP-adjunction.
28. Note that this sentence may be fine only if we put emphatic stress on books. Stress
marks the boundary of some kind of prosodie category, and this strengthens our posi-
tion that Heavy NP Shift is triggered by something phonological.
29. See also Verhijde and Van de Weijer (1992) for a similar conclusion independently
reached.
30. Why is a heavy NP "shifted"? We conjecture that this would be due to the fact that pro-
sodie heaviness (non-unarity) obligatorily creates binary rhythm. With two words we
can compose a hierarchical structure, but we cannot with only one.
(i) a. Unary Phrase

*
A
b. Binary Phrase
*
* *
[A B]
With something binary and with a rhythmic head at the end, we can easily adjust the
rhythm of the whole utterance. Thanks go to an anonymous reviewer for this point.
We need an account of truly focal "light" NP Shift.
(ii) She bequeathed to the library her PAPERS and to the radio station her CD's.
(Elisabeth Selkirk (personal communication))
In such a case, we assume that a focus forces prosodie bracketing.
(ill) *
*
[PAPERS]
Thus, heaviness here corresponds to some prosodie hierarchy which is either forced by
binarity or focus.
31. There is another interesting syntactic fact pertaining to Heavy NP Shift. A D P can shift
over the object-oriented secondary predicate, while it cannot shift over the subject-ori-
ented secondary predicate.
(i) a. Jews never eat t raw [fish over two days old].
b. *John left t angry [the reception for the ambassador from Ulan Bator],
However, examples like (ia) can "shift" the D P even if it is light.
(ii) Jews never eat t raw some fish.

Thus, we must say that this DP-predicate permutation needs another explanation. We
are grateful to Makoto Kurata for reminding us of this fact.
32. Takano's (1998) D P Shift is actually a complex of two movement operations: one for
reaching the domain of Case-assigner and the other for Case checking. Moreover, Ta-
kano differs from us in that he interprets Heavy N P Shift as a PF interpretation (= pro-
nunciation) of a trace instead of a target. H e also attributes "heaviness" to focus, which
we reject as shown in discussions in previous sections.
We do not adopt Kayne's (1994) extra-Scrambling analysis which we introduce above
simply because Kayne's theory involves some complexity in his analysis of SOV lan-
guages.
33. Here we simply follow Fukui and Takano's (1998) assumption that the syntactic trees
must keep "expanding" throughout the course of derivations, which they attribute to
Chomsky (1995). Thus, there is no head movement per se and the so-called head move-
ment is also a case of phrasal adjunction.
34. The reason of ungrammaticality of (82) perhaps is attributed to an independent matter
of Case-checking of the subject. Even if we shift the object and the adverb later by some
sort of Scrambling, they should come to the left of the verb and we would have * It John,
very much, impressed the woman who saved a kid's life hit by a car.
35. Perhaps, we are leaving the minimalist spirit of strict modularity among grammatical
components. According to our view of grammar, syntactic trees can contains anything
phonological and semantic, including something like phonemes. Being invisible and be-
ing irrelevant are separate matters, we believe.
36. Adjoined nodes may have a different form, which is not substantially relevant to the dis-
cussion in this paper.
37. See Akasaka (2000) and Akasaka and Tateishi (2000) for details.
References
Akasaka, Yukiko
1996 Binding as the Realization of Overt Variables. Master's thesis, Kyoto Univer-
sity of Foreign Studies.
2000 Extended Prosodie Syntax: Case Studies in English. Master's thesis, Osaka
University.
Akasaka, Yukiko and Koichi Tateishi
2000 "Metrical Branching and Phonology-Syntax Interface," Phonological Studies
3, 61-42
Bolinger, Dwight
1972 "Accent is Predictable (If You are a Mind Reader)." Language 48, 633-644.
Boskovic, Zeljko and Daiko Takahashi
1998 "Scrambling and Last Resort." Linguistic Inquiry 29, 347-366.
Chomsky, Noam
1972 "Remarks on Nominalization." Studies on Semantics in Generative Grammar,
7-61, Mouton, The Hague.
1986 Barriers. The MIT Press, Cambridge, MA.
1995 The Minimalist Program. The MIT Press, Cambridge, MA.
Chomsky, Noam and Morris Halle
1968 The Sound Pattern of English. Harper and Row, New York.
Fiengo, Robert
1974 Semantic Conditions on Surface Structure. Doctoral dissertation, MIT.
Fukui, Naoki
1993 "Parameters and Optionality." Linguistic Inquiry 24, 399-420.
Fukui, Naoki and Yuji Takano
1998 "Symmetry in Syntax: Merge and Demerge." Journal of East Asian Linguistics
7,27-86.
Hale, Kenneth
1980 "Remarks on Japanese Phrase Structure: Comments on the Papers on Japa-
nese Syntax." MIT Working Papers in Linguistics 2,185-203.
Haraguchi, Shosuke
1991 A Theory of Stress and Accent. Foris, Dordrecht.
Hayes, Bruce
1980 A Metrical Theory of Stress Rules. Doctoral dissertation, MIT.
1982 "Extrametricality and English Stress." Linguistic Inquiry 13, 227-76.
1995 Metrical Stress Theory; Principles and Case Studies. The University of Chicago
Press, Chicago.
Itô, Junko
1990 "Prosodie Minimality in Japanese." CLS 26,213-239.
Itô, Junko and R. Armin Mester
1986 "The Phonology of Voicing in Japanese: Theoretical Consequences of Mor-
phological Accessibility." Linguistic Inquiry,17, 49-73.
1995 "Japanese Phonology." The Handbook of Phonological Theory, ed. by John
A. Goldsmith, 817-838, Blackwell, Oxford.
Jackendoff, Ray
1977 X' Syntax: A Study of Phrase Structure. The MIT Press, Cambridge, MA.
Kahn, Daniel
1976 Syllable-Based Generalizations in English Phonology. Doctoral dissertation,
MIT.
Kayne, Richard S.
1994 The Antisymmetry of Syntax. The MIT Press, Cambridge, MA.
Kubozono, Haruo
1988 The Organization of Japanese Prosody. Kurosio Publishers, Tokyo.
Kuno, Susumu
1973 Nihon Bunpo Kenkyu [Studies in Japanese Grammar], Taishukan, Tokyo.
Liberman, Mark and Alan Prince
1977 "On Stress and Linguistic Rhythm." Linguistic Inquiry 8, 249-336.
McCarthy, John and Alan Prince
1986 "Prosodie Morphology." ms., University of Massachusetts, Amherst and
Brandeis University, Waltham.
McCawley, James D.
1968 The Phonological Component of a Grammar of Japanese. Mouton, The
Hague.
Miyagawa, Shigeru
1989 Syntax and Semantics 22: Structure and Case Marking in Japanese. Academic
Press, San Diego.
1997 "Against Optional Scrambling." Linguistic Inquiry 28,1-25.
Otsu, Yukio
1980 "Some Aspects of Rendaku in Japanese and Related Problems." MIT Work-
ing Papers in Linguistics 2, 207-227.
Poser, William J.
1990 "Evidence for Foot Structure in Japanese." Language 66, 78-105.
Prince, Alan and Paul Smolensky
1993 Optimality Theory: Constraint Interaction in Generative Grammar. Technical
Report 2, The Rutgers Center for Cognitive Science, New Brunswick.
Rochemont, Michael S. and Peter W. Culicover
1990 English Focus Constructions and the Theory of Grammar. Cambridge Univer-
sity Press, Cambridge.
Ross, John Robert
1986 Infinite Syntax! Ablex Publishing Corporation, Norwood, NJ.
Saito, Mamoru
1985 Some Asymmetries in Japanese and Their Theoretical Implications. Doctoral
dissertation, MIT.
1992 "Long Distance Scrambling in Japanese." Journal of East Asian Linguistics 1,
69-118.
Saito, Mamoru and Naoki Fukui
1998 "Order in Phrase Structure and Movement." Linguistic Inquiry 29, 439-474.
Saito, Mamoru and Hajime Hoji
1983 "Weak Crossover and Move-α in Japanese." Natural Language and Linguistic
Theory 1, 245-259.
Selkirk, Elisabeth
1984 Phonology and Syntax: The Relation between Sound and Structure. The MIT
Press, Cambridge, MA.
1986 "On Derived Domains in Sentence Phonology." Phonology Yearbook 3, 371-
405.
Selkirk, Elisabeth and Koichi Tateishi
1988 "Constrains on Minor Phrase Formation in Japanese." CLS 24, 316-336.
1991 "Syntax and Downstep in Japanese." Interdisciplinary Approaches to Lan-
guage: Essays in Honor ofS.-Y. Kuroda, ed. by Carol Georgopoulos and Rob-
erta Ishihara, 519-543, Kluwer Academic Publishers, Dordrecht.
Takano, Yuji
1998 "Object Shift and Scrambling." Natural Language and Linguistic Theory 16,
817-889.
Tateishi, Koichi
1985 Accent of Compound Nouns in the Standard Japanese: A Nonliner Approach.
Master's thesis, International Christian University.
1989 "Theoretical Implications of the Japanese Musician's Language." Proceedings
ofWCCFL 8, 384-398.
1992 "Maximal Non-Exhaustive Invisibility." Kansai Linguistic Society 12,134-144.
1997 "Nihon-go Tokyo-Hogen Washa-ni Okeru Akusento-no Yure-to Onsetsu-
Kozo [Variability of Judgement on Accentuation among Tokyo Japanese
Speakers and the Syllable Structure]." a paper presented at the 115th meeting
of the Linguistic Society of Japan, Kyoto University.
1998a "Foot or Syllable?: Binarity Effects in Japanese and English Phonology." ICU
English Studies Vol. 7, Society of ICU English Studies, 1-22.
1998b "Non-Repair Strategies in Japanese Syllables." Phonological Studies 1, 123-
130.
1999 "Are Interface Studies Possible?" Journal of the Phonetic Society of Japan 3,
4-19.
Verhijde, Mark and Jeroen van de Weijer
1992 "On Deriving Minimal NPs." Proceedings of the Leiden Conference for Junior
Linguists 3, ed. by L.C.J. Barbiers, M. den Dikken and C.C. Levelt, 293-306.
Leiden University Press, Leiden.
Webelhuth, Gert
1989 Syntactic Saturation Phenomena and the Modern Germanic Languages. Doc-
toral dissertation, University of Massachusetts, Amherst.
Yamada, Eiji
1990 "Stress Assignment in Tokyo Japanese." ms., Fukuoka University, Fukuoka.
Zee, Draga and Sharon Inkelas

1990 "Prosodically Constrained Syntax." The Phonology-Syntax Connection, ed.
by Sharon Inkelas and Draga Zee, 365-378, The University of Chicago Press,
Chicago.
Zubizarreta, Maria Luiza
1998 Prosody, Focus, and Word Order. The MIT Press, Cambridge, MA.
The Accent of Tsuruoka Japanese Reconsidered
Shosuke Haraguchi
Meikai University
Introduction
Japanese is generally regarded as having a pitch accent system. However,

it has a number of dialects, which are classified as unaccentual systems
(e.g., Kagoshima Japanese), as well as accentual systems (e.g., Tokyo-type
Japanese and Kansai-type Japanese). For detailed discussion, see Mc-
Cawley (1968), Haraguchi (1977,1991), among others.
Accentual systems of Japanese dialects have a number of distinct char-
acteristics. For example, in almost all dialects, verbs and adjectives are di-
vided into two classes: either accented and unaccented like Tokyo Japa-
nese, or two different melodies HL & LHL like Osaka Japanese, LHL &
LH like Kagoshima Japanese, etc. The former normally has accent on the
penultimate mora of the Present forms of verbs and adjectives, which is
equivalent to the stem-final mora/syllable in (1).
(1) (C)]..«n, Γ (r)u] -I verb

ι i] L,
On the other hand, nouns tend to have more accentual varieties than
those of verbs and adjectives, depending on surface accentual/tonal real-
ization. Consider the following table, which lists some of the tonal pat-
terns frequently observed in Japanese dialects.
(2) No. of surface analyses

tonal patterns
a. 1 one-melody unaccentual system
b. 2 two-melody unaccentual system
c. 3 accentual system
d. η accented words only
e. n+1 accented and unaccented words
f. n+2 accented and unaccented words + Accent
Shift (or Tone Shift)
g- 2n+2 two-melody system
h.
3n+3 three-melody system
48 Shosuke Haraguchi
If a dialect has one or two tonal patterns, the system is interpreted as a

one-melody unaccentual system or a two-melody unaccentual system. Miy-
akonojo Japanese, Izumi Japanese, and Nakamura Japanese are examples
of (2a). Kagoshima Japanese in southern Kyushu is a celebrated example of
(2b). Both patterns are different from all the remaining ones in that the
former are interpreted as unaccentual systems and the latter as accentual
systems. As far as I know, Kumi Japanese is a good case of (2c) (see Hara-
guchi 1977, 1998, among others). Fukuoka Japanese is the only attested
case of (2d); Tokyo Japanese is a typical example of (2e); Osaka Japanese
and other Kansai-type dialects are examples of (2g); and Old Kyoto Japa-
nese is the only attested case of (2h) (see Haraguchi 1997, among others).
Note that (2f), of which Tsuruoka Japanese is an example, is a rather
strange possibility, because one-melody dialects can have just one unac-
centual word, not two, for η-mora words.
The purpose of this paper is two-fold: (i) To discuss this seemingly-
strange tonal property of Tsuruoka Japanese in detail, reexamining the
major tonal facts of nouns, verbs, and adjectives, and reviewing my previ-
ous analysis incorporating an accent shift rule; (ii) To propose an alterna-
tive analysis with a phonetic process which moves the Η-tone to the right
and to discuss its theoretical implications.
1. Tsuruoka Japanese as an n+1 underlying accentual pattern
As noted above, the existence of this type of dialect would be rather sur-
prising. This is because if lexical items have η moras (or syllables), we
would expect that the largest number of their accentual pattern be n+1,
taking the unaccented lexical items into account. This n+1 pattern is ob-
served in (3).
(3) n+l\ Tokyo Japanese
noun-Nom. gloss analysis

a. hasi(-ga) edge unaccented
LH Η
b. hasí(-ga) bridge final-accented
LH L
c. hási(-ga) chopstick initial-accented
HL L
Here and below, the location of (surface) accent is represented by the

acute accent on a vowel.
Japanese is unique in that it has a dialect with the n+2 surface accentual
patterns: Tsuruoka Japanese (as well as a number of dialects of the Tohoku
District). This dialect has long attracted the attention of a large number of
Japanese linguists, who have tried to propose a variety of analyses, account-
ing for the seemingly-strange accentual facts. See Shibata (1957), Kindaichi
(1953), Haraguchi (1979,1997,1998), Nitta (1994) among others.
To see the tonal properties of Tsuruoka Japanese, let us begin by exam-
ining the facts of nouns and noun phrases. This dialect has two accentual
classes for one-mora nouns; four accentual classes for two-mora nouns;
five accentual classes for three-mora nouns; and six accentual classes for
four-mora nouns.
First, let us examine the following cases of two-mora and three-mora nouns
in (4). All the examples discussed in this paper are cited from Nitta (1994), ba-
sically following his notation with minor notational changes. Thus, the nota-
tions /'í, 'á, 'ú, 'ó, 'á/ stand for [i, e, u, o, a] respectively; Β, for prenasalized labial
[b]; g for velar nasal [η]; and the acute accent, for accent on the mora.
(4)
No. of Gloss In isolation ~ + s a ~+do ~+no ~-i-gara
Mora 'Acc' 'and' 'Gen' 'from'
2a. nose hana hanasa hanado hanano hanagara
L L L LL L L L L L L LLLL
2b. leg 'ásí 'ásisá 'ásidó 'àsino 'ásigára
LH LLH LLH LL L LLHL
2c. shoulder kadá kadása kadádo kadáno kadágara
LH LH L LH L LH L LH LL
2d. autumn 'àgi 'ágisa 'ágido 'ágino 'ágigara
HL H LL HLL HLL HLLL
3a. fish sagana saganasa saganado saganano saganagara
LL L L L LL L L L L L L L L L L L LL
3b. head 'ádamá 'ádamasá 'ádamadó 'ádamano 'ádamagára
L L H L L LH L L L H L L L L L L LHL
3c. heart kogoró kogorósa kogoródo kogoróno kogorógara
L LH L LH L L LH L L LH L L LH L L
3d. rabbit 'úsági 'úságisa 'úságido 'úságino 'úságigara
LH L LH LL LH L L LH L L LHLLL
3e. helmet káBudo káBudosa káBudodo káBudono káBudogara
H L L H L L L, H L L L H L L L H LLLL
All the L-toned nouns of the 2a and 3a types are analyzed as unaccented
by all linguists, while the other nouns with an Η-tone are analyzed as ac-
cented, with the accent on the Η-toned mora. When the accusative
marker -sa, the Genitive marker -no, -do 'and' corresponding to -to in
Tokyo Japanese, and -gara 'from' corresponding to -kara of Tokyo Japa-
nese, are attached to the nouns, the unaccented nouns remain L-toned.
However, in the accented type 2b and 3b nouns, only L-tone appear
when the Genitive marker -no is attached to a noun, which suggests that
the word-final accent is deleted when it is adjacent to the Genitive mark-
er. Also, before the other particles, the Η-tone appears one-mora to the
right, which is interpreted by most Japanese linguists as a shift of accent
to the right. In other cases, the accent seems to surface in its underlying
position.
Haraguchi (1979), paying attention to the fact that accent shifts to the
right only when the vowel of the post-accented mora is [e, o, or a], while
the shift is blocked when the vowel is [i or u], proposes that the seemingly
peculiar properties of Tsuruoka Japanese can be accounted for if we as-
sume the following accent shift rule contingent on the height of the post
accented vowel.
(5) Accent Shift:

Shift accent one-mora to the right if the vowel of the post-accented
mora is [-high] (i.e., [e, o, a]).
This rule is formally represented as in (5') in Haraguchi (1979), in which

the star (*) stands for the accent and V, for a vowel or a mora:
(51) Accent Shift:

*
V
V V ' V
-high -high
Assuming this accent shift rule, we can analyze the nouns of the 2c and 3c
types as having an underlying penultimate accent, which is moved one
mora to the right after the application of the accent deletion rule before
no. Haraguchi (1979) also proposes that, assuming this accent shift rule,
this accentual system is nothing but a variant of Tokyo-type pattern: the
n+1 accentual pattern.
Let us now examine four-mora nouns in (6).
(6)
No. of Gloss In isolation ~+sa ~+do +no ~+gara
4a. chicken niwadori niwadorisa niwadorido niwadorino niwadorigara
LLLLLLLLLLLLLL LLLLL LLLLLL
4b. wild pig 'ínosisí 'ínosisisá "ínosisidó 'ínosisino 'ínosisigára
L L LH L L L L H L L L L H LLLLL LLLLHL
4c. sound of feet 'asi'odó 'ási'odósa 'ási'odódo 'ási'odóno 'ási'odógara
LLLH LLLHL LLLHL LLLHL LLLHLL
4d. onion tamanégi tamanégisa tamanégido tamanégino tamanégigara
L L H L L L H LL L L H LL LLHLL LLHLLL
4e. morning glory 'ásága'o 'ásága'osa 'ásága'odo 'ásága'ono 'ásága'ogara
LHLL LHLLL LHLLL LHLLL LHLLLL
4f. mushroom mázidage mázidagesa mázidagedo mázidageno mázidagegara
HLLL HLLLL HLLLL HLLLL H LLLLL
Parallel to what we have seen in (4), the L-toned nouns are unaccented
and the final-accented nouns of the 4b type lose accent before the Geni-
tive marker no. The nouns of the 4c type have an underlying accent on the
penultimate mora, shifting it to the right by the accent shift rule in (5).
Those of the 4d type have the same surface and underlying accent on the
penultimate mora, and those of the 4e and 4f types have an underlying ac-
cent on the initial mora, with the subsequent accent shift to the second
mora in the case of the 4e type.
Consider now the cases of one-mora nouns, which have just two types,
unaccented and accented.
(V)
No. of Gloss In isolation ~+sa ~+do ~+no ~+gara
la. mosquito ka kasa kado kano kagara
L LL LL LL LLL
lb. picture 'é 'esá 'édo 'eno 'egára
H LH HL LL LHL
In isolation, unaccented nouns are L-toned just like other unaccented

cases of longer nouns, while accented nouns are Η-tones as in 'é. When
the Accusative marker sa is attached to them, the unaccented nouns re-
main L-toned, but the accented nouns surface as LH-toned, which is in-
terpreted as a shift of accent to the right. When the particle do 'and', cor-
responding to in Tokyo Japanese, is attached, the accent does not shift to
the right for some reason unclear to me at present. When the Genitive
marker no is attached, the word-final accent is deleted, while it is shifted
to the right in the case of the particle gara 'from', corresponding to kara
in Tokyo Japanese.
I assume that the accent shift rule in (5) applies to lexical items in the
lexicon and to what McCawley (1968) calls the 'minor phrase'. Recall that
this rule moves the accent to the right if the height of the following vowel
is non-high. Once we assume this rule, the data in (4, 6, and 7) can be an-
alyzed as follows:
(8) No. of Mora Gloss Sample Underlying Accent

la mosquito ka o unaccented
lb picture 'é ó initial-accented
(or final-accented)
2a nose hana 0 o unaccented
2b leg 'ásí o ó final-accented
2c shoulder kadá ó o initial-accented
2d autumn 'àgi ó o initial-accented
3a fish sagana oo 0 unaccented
3b head 'ádamá oo ó final-accented
3c heart kogoró 0ó o penultimate-accented
3d rabbit 'úsági oó o penultimate-accented
3e helmet káBudo ó 0 o initial-accented
4a chicken niwadori o o 0 o unaccented
4b wild pig 'ínosisí OO0 ó final-accented
4c sound of feet 'ási'odó O 0 ó o penultimate-accented
4d onion tamanégi o ó o o antepenultimate-accented
(or penultimate-accented)
4e morning glory 'ásága'o Ó 0 o o initial-accented
4f mushroom mázidage Ó o 0 o initial-accented
The existence of one more additional surface accentual pattern in Tsu-

ruoka Japanese is accounted for by the Accent Shift rule in (5). This rule
plays an important role in accounting for a variety of phenomena found
in verbal inflection and adjectival inflection of this dialect, to which I will
return in the next section.
2. The tonal system of Tsuruoka Japanese
As discussed in section 2, Tsuruoka Japanese is peculiar in that it appears

to have an n+2 surface tonal distinctions. This seemingly strange charac-
teristic of Tsuruoka Japanese is shown in the section above to be due to
the existence of an accent shift rule dependent on the height of vowels, re-
peated here for ease of reference:
(5) Accent Shift:

Shift accent one-mora to the right if the vowel of the post-accented
mora is [-high] (i.e., [e, o, a]).
In this section, I will discuss a number of tonal properties of this dialect to

shed light on some of its peculiar properties.
2.1. The tonal system of Tsuruoka nouns
Let us first discuss the tonal system of nouns in this dialect, and determine
what the best way is to analyze its melodic properties. The basic tonal
facts of nouns are illustrated in (4, 6,7) in section 2.
I will begin with a discussion of the accentual system of this dialect. A
careful examination of the above facts shows that accent seems to be
shifted to the right when the following mora contains a non-high vowel [e,
o, or a]. One case which does not conform to this accent shift is 'é 'picture'
in (lb) followed by do 'and'. For some unexplained reason the phrase
'édo does not undergo accent shift. Thus, if this is a fact, we must regard
this phrase as an exception to the rule.
The all-L tone in the no-phrases of the lb, 2b, 3b, and 4b types compels
us to set up an accent deletion rule, which deletes accent immediately in
front of the Genitive marker no:
(9) Accent Deletion before no:

* -»· 0 / -no
Notice that unless we assume that this rule applies before Accent Shift in
(5), we would not account for the difference between the 2b, 3b, and 4b
types and the 2c, 3c, and 4c types. This shows that we need derivational
information to distinguish the b classes and the c classes.
An examination of the list immediately shows that only one H tone ap-
pears in a word or a phrase and everything else is L. Let us first discuss
how this fact is analyzed within the framework of Haraguchi (1977) and
(1979). The basic tone melody of this dialect is LHL. Assuming that the
H tone is associated with the accented mora first, as in numerous other
Japanese dialects, all other moras are associated with the L tone of the
LHL melody. Notice also that the absence of a contour tone (i.e., HL fall-
ing tone or LH rising tone) indicates that this dialect prohibits association
of two or more tones to a single tone-bearing unit, which leaves the L tone
unassociated with the peripheral mora of the initial-accented word or of
the final-accented word.
The remaining thing for us to do is to account for the all-L tone of un-
accented words. One possible solution is to assume that Tsuruoka Japa-
nese has no system to associate the basic tone melody LHL with unac-
cented words, which leaves the unaccented words toneless. The toneless
words later receive the default L tone by a default tone association con-
vention. The second possible solution is that, just like Tokyo Japanese, the
H tone of the LHL melody is associated with the final mora of the unac-
cented words, which brings about the following schematic melodic shape:
(10) . . . X X X X
\ l / I
L H L
The H tone associated with the unaccented mora is then deleted and the
floating L tone is associated with the toneless final mora. The third possi-
ble solution is to assume that the H tone of the LHL melody is associated
with the initial mora of a word. Thus, the derivation is the mirror image
of the second solution, as indicated in (11):
(11) X X X X
I I / /
L H L
Which solution is the best? It is rather difficult to argue for one over the
others. One might claim that the best guess would be to choose the last
one, taking the following observation by Nitta (1994: 85) into consider-
ation.
(12) Classes la, 2a, 3a, etc. are phonologically represented as having, LL,
LLL, LLLL, etc. but these all L melodies, as Kindaichi has indepen-
dently pointed out, sound like ML, MML/MLL, MMLL/MLLL, etc.

respectively. It seems the latter melodies are normally the case.
To derive the melodies ML, MLL, MLLL, etc., what would be needed is
to simply assume that the initial H tone associated with the unaccented
mora is phonetically realized as the M tone. The M tone which appears
optionally on the second mora will be handled by assuming an optional
phonetic spreading of the M tone one mora to the right. Assuming that
this solution is correct, Tsuruoka Japanese would be just the opposite of
Tokyo Japanese in the association of the H tone with an unaccented word.
Admitting the initial plausibility of this analysis, I will elect to choose
the second solution here. This solution makes it necessary to introduce an
H Deletion rule, which erases the H tone on the unaccented mora, induc-
ing the subsequent spreading (or association) of the L tone. One of the
reasons that I choose this option is to make this dialect parallel to other
Tokyo-type Japanese in handling the association of the H tone with the
final mora of unaccented words. This analysis is supported by features of
verbs in this dialect, which are discussed in the next subsection.
Overall, we can say that tonal facts of Tsuruoka Japanese are more
complicated than those of Tokyo Japanese in that the former requires the
accent shift rule dependent on the height of vowels, the H Deletion rule
for nouns, and a couple of phonetic rules.
Tonal facts of Tsuruoka nouns discussed in section 1 can be summa-
rized as follows:
(13) a. All L-toned nouns are unaccented.

b. Accented words have just one Η-toned mora.
c. Accent Shift, which moves accent one mora to the right contingent
on the height of the post-accented vowel, makes it possible to an-
alyze a number of nouns as having underlying accent one-mora to
the left of the surface accent, as illustrated in (8).
d. An accent deletion rule applying before the Genitive no makes the
tone of the final-accented nouns all L-toned.
I will not go into the details of exceptions to the analysis presented above.
For further discussion of ramifications on marked irregularities and a
number of exceptions, see Haraguchi (1997) and (1998).
2.2. The tonal system of verbs in Tsuruoka Japanese
Let us now turn to tonal phenomena of verbs in Tsuruoka Japanese. An

examination of the tonal system of verbs gives further supports to the
above analysis incorporating Accent Shift (5).
Verbs are, just like other Tokyo-type Japanese, divided into two cate-
gories: accented and unaccented. The following list will make this clear:
(14) a. Unaccented Class

Present Melody — sage Negative ~ sage Neg-Imperative Conditional
kiru LH kiruságe' kine' kine'ságe' kiNna kideba
neru LH neruságe' nene' nene'ságe' neNna nedeba
'igu LH 'íguságe' 'ígane' 'ígane'ságe 'iguna 'ígeba
ka'u LH ka'uságe' kawane' kawane'ságe 'ka'una kae'ba
mageru LLH mageruságe' magene' magene'ságe' mageNna magedeba
kariru LLH kariruságe' karine' kariruságe' kariNna karideba
'oguru LLH 'óguruságe' 'ógurane' 'óguraneságe' 'óguNna 'ógureba
cuka'u LLH cuka'uságe' cukawane' cukawane'ságe' cuka'uNna cuka'eba
kasaneru LLLH kasaneruságe' kasanene' kasanene'ságe' kasaneNna kasanereba
hadaragu LLLH hadaraguságe' hadaragane' hadaragane'ságe' hadaraguna hadarageba
b. Accented Class
déru HL dérusage' dené' dené'sage' déNna dedéba
míru HL mírusage' miné' miné'sage' míNna midéba
kúru HL kúrusage' koné' koné'sage' kúNna kudéba
kú H kúsage' kuwané' kuwané'sage' kúna kéba
tóru HL tórusage' torané' torané'sage' tóNna todéba
kágu HL kágusage' kagané' kagané'sage' káguna kagéba
tadéru LHL tadérusage' tadené' tadené'sage' tadéNna tadedéba
toZíru LHL toZírusage' toZiné' toZiné'sage' toZíNna toZidéba
cugúru LHL cugúrusage' cugurané' cugurané'sage' cugúNna tuguréba
hará'u LHL hará'usage' harawané' harawané'sage' hará'una hara'éba
hé'ru LHL hé'rusage' he'rané' he'rané'sage' hé'Nna heréba
kaZo'éru LLHL kaZo'érusage' kaZo'ené' kaZo'ené'sage' kaZo'éNna kaZo'edéba
'ugogásu LLHL 'ugogásusage' 'ugogasané' 'ugogásauné'sage' 'ugogasuna 'ugogaséba
where e' stands for open [ε], Ζ for prenasalized [z], and Ν for moraic
nasal.)
Notice that the unaccented verbs have the Η tone on the final mora of the
Present. This means that the Η tone of the LHL basic tone melody is as-
sociated with the final mora of verbs in the autosegmental analysis. Recall
that for unaccented nouns we have chosen the second solution in which
the H tone is associated with the final mora, with the subsequent erasure
of the H tone. Verbs are different from nouns in that this H deletion does
not apply to verbs. However, the association of the H tone with the final
mora of unaccented cases is common to both nouns and verbs.
Accented class is assumed to be assigned accent by the following rule,
which is identical to that of Tokyo Japanese:
(15) Assign accent to the stem-final mora.
This accent assignment applies before the accent shift rule in (5). The
stem-final or penultimate accent does not undergo accent shift because
the Present morpheme ru has a high vowel. The implication of this as-
sumption will be clear immediately below.
First consider the morpheme sage'. An examination of both the unac-
cented class and the accented one shows that this morpheme is accented.
Within the present framework assuming an accent shift rule, we must fur-
ther assume that this morpheme belongs to the so-called 'Pre-accenting'
class, which is represented schematically as follows:
(16) *
* *
sa ge'
This makes it possible for the accent shift rule to move accent to the right,
as illustrated in (16):
(17) * *
* * * * * *
ku ru- sa ge' ku ru -sa ge'
Notice that this accent normally does not surface when the morpheme is
attached to accented verbs, with a few exceptions with which we are not
concerned here.
Turn next to the Negative ne\ which is derived from the underlying
form ai by Coalescence. This bound morpheme should be classified as un-
accented, because there is no indication of the existence of accent when
it is attached to unaccented verbs. The fact that it has accent when it is at-
tached to accented verb stems can be easily handled by the accent shift
rule (5) introduced independently for nouns.
Notice that the Negative Imperative na is also unaccented. What is in-
teresting is that when it is attached to the Present form of verbs, the ru is
turned into the geminate Ν immediately before η of the Negative ne'. This
segmental process seems to be common to most dialects of the Tohoku
district and the northern Kanto district (e.g., various Ibaragi dialects). Ir-
respective of the relative order of this segmental rule and the accent shift
rule, we can account for the reason why accent shift is blocked in cases
where the accent is followed by the geminate N. This will be clear from
the discussion below of the accent shift of the Past form.
Consider now the Conditional form (d)eba, which corresponds to
(r)eba of Standard or Common Japanese. Thus, when the Conditional is
attached to the consonant-ending (or consonantal) verbs, eba is selected
as illustrated in:
(18) a. 'igeba 'ig + eba 'go' + Conditional

b. kae'ba kaw + eba 'buy' + Conditional
c. 'ógureba 'ógur + eba 'send' + Conditional
d. cuka'eba cukaw + eba 'use' + Conditional
e. kasanereba kasaner + eba 'pile' + Conditional
f. hadarageba hadarag + eba 'work' + Conditional
g· kéba kúw +eba 'eat' + Conditional
h. kagéba kág+ eba 'write' + Conditional
i. tuguréba tugúr + eba 'make' + Conditional
j· hara'éba haráw + eba 'pay' + Conditional
k. heré-ba hér + eba 'enter' + Conditional
1. 'úgogasé'ba 'úgogás +eba 'move' + Conditional
Otherwise (i.e., when it: is attached to the vowel-ending verbs), deba is se

lected, as illustrated in:
(19) a. kideba ki + deba 'wear' + Conditional

b. nedeba ne + deba 'sleep' + Conditional
c. magedeba mage + deba 'bend, twist' + Conditional
d. karideba kari + deba 'borrow' + Conditional
e. dedéba dé + deba 'go out' + Conditional
f. midéba mí + deba 'see' + Conditional
g· kudéba kú + deba 'come' + Conditional
h. todéba tó + deba 'take' + Conditional
i. tadedéba tadé + deba 'build' + Conditional
The distribution of (d)eba is identical with that of (r)eba of Standard Jap-

anese and other Tokyo-type Japanese. The alternation of r and d is also
found in a number of Japanese dialects. The Conditional (d)eba should be

unaccented by nature, but it gets accent when it is attached to the accent-
ed verbs. This accent is due to the accent shift rule (5) discussed above.
Let us now turn to the Past morpheme ta, which is realized as da by a
phonetic voicing rule which applies in the following environment:
(20) (Cvo.ced)r V , V
IN J
Otherwise ta remains as it is. Just like other forms, the Past ta is unaccent-
ed by nature. Thus, when it is attached to unaccented verbs, the Past forms
have no accent at all. However, when it is attached to accented verbs, it
sometimes gets accent as illustrated in the examples of the column I:
(21) I. Accent on the Past II. Accent on verb stem

a. dedá 'went out' káida 'wrote'
b. midá 'saw' dásita 'put out: sent'
c. kitá 'came' yóNda 'read'
d. 'áttá 'met' tadáida 'struck'
e. 'úttá 'hit' 'éráNda 'chose'
f. tottá 'took' 'árúida 'walked'
g· ku(t)tá 'ate' hé'tta 'entered'
h. cukuttá 'made' cigazú(i)da 'came near'
i. harattá 'paid' 'úgogásita 'moved'
j· ayamattá 'apologized' 'ósikóNda 'pushed in'
These facts are analyzed as follows: The Past morpheme is extratonal or

invisible, and accent is assigned to the final syllable of the verbal stem.
The accent shift in question then applies to the examples of the left col-
umn and the accent is moved onto the Past morpheme. Here the geminate
consonant is ignored since it cannot bear accent. In contrast, all the exam-
ples of the right column do not display accent shift, because the high vow-
els and the moraic nasal Ν block it. Notice that the moraic nasal Ν is in-
terpreted as equivalent to the high vowels in a number of cases. The only
example which requires comments is hé'tta 'entered', which is an appar-
ent exception to the accent shift in question. Notice however that the
word hé'tta is a surface form of /háir-ta/, which is turned into [háit-ta] by
assimilation and then turned into [hé't-ta] by Coalescence of /ai/ and the
subsequent (or simultaneous) shortening into je']. Thus we can correctly
account for the reason why accent shift is blocked in this case. The under-
lying high vowel, which exists when the rule in question applies, prevents
accent shift from applying to this case.
Notice that in the discussion above, I omitted a couple of seemingly re-
calcitrant cases, which will be briefly examined at this point:
(22) a. tadéda 'built'

b. toZida 'closed'
Recall that the Past morpheme ta is invisible and accent is assigned to the
final syllable of the verbal stem. This means that accent is assigned to the
initial mora of these cases. The accent shift rule under consideration
should move the accent of (22a and b), while it predicts, contrary to fact,
that it should be blocked from applying to derive (22b).
The problem is how to account for the bocking of accent shift in (22b).
Since the relevant examples are scarce, I cannot say anything conclusive at
present. However, the best guess is that accent shift is permitted to apply
under certain conditions even if the following mora contains a high vowel.
I suspect that the voiced consonant has something to do with this. Howev-
er, I have to leave the problem of the blocking of accent shift open here.
The core cases discussed in this section clearly show the necessity of ac-
cent shift and of derivations.
2.3. The tonal system of adjectives in Tsuruoka Japanese
Before concluding the discussion of the tonal system of Tsuruoka Japa-

nese, I will briefly discuss that of adjectives. Since the factual investigation
into the data of adjectives are limited and the quality of the data does not
seem to be so reliable, the following analysis of the tonal system of adjec-
tives is tentative at best.
According to Nitta (1994), adjectives in Tsuruoka Japanese are also di-
vided into two classes: accented and unaccented. Notice that this is also
parallel to most of the Tokyo-type dialects. Consider first the following
cases:
I. Accented II. Unaccented

2a. kói 'thick,dense' suppe' < sui 'sour'
b. ée,íi 'good' (gap?)
né' 'not'
3a. samúi 'cold' 'áme' 'sweet'
b. sirói, siré 'white' kure' 'dark'

c. 'á'ói 'blue' 'âge' 'red'
4a. 'ókkí 'big' kurusi 'hard'
b. miziké' 'short' kuyasi 'upset; frustrating'
c. 'óisí 'tasty' 'áBune' 'dangerous'
The accented class receives accent on the penultimate mora (or equiva-
lently on the stem-final mora), which is common to almost all of the To-
kyo-type dialects.
The reader might wonder how to handle the fact that some adjectives
like siré in 2b, 'okki in 4a, and 'oisi in 4c appear to have final accent. This
fact will pose no problem if we assume that the underlying forms of these
adjectives are respectively as follows:
(24) surface form underlying form

a. siré < sirói 'white'
b. 'ókkí < okkíi 'big'
c. 'óisí < 'óisíi 'tasty'
Assuming that accent is assigned to the underlying forms and then the fol-
lowing rules apply to derive the surface forms, we can easily handle these
properties of accentual behavior:
(25) a. Coalescence: o+i ->• ee

b. Shortening: {e, i} ->• 0 / {e, i} +
Notice that the word-final open e' is also the result of Coalescence, which
is universally found in a variety of dialects and languages, and subsequent
shortening in (25b):
(26) Coalescence: ai ->• ee'
Looking at the samples in (23), the astute reader might have noticed that
the adjective né' is classified as belonging to the accented class. However,
we have seen in the section above that the bound form ne' is unaccented.
How should we account for this apparent incompatibility?
A careful examination of facts concerning adjectives shows beyond
reasonable doubt that the adjective né' 'not' is accented. To see this, con-
sider the following cases:
(27) Accented class:

adjective gloss . . . + thing non-existent . . . + not Past Negative
a. né' 'not' né-mono né'gu-naru
b. kói 'thick' kói-mono kógu-naru kógu-ne' kógu-né'kke
c. siré 'white' siré-mono sirégu-naru sirégu-ne' sirégu-né'kke
(28) Unaccented class:

adjective gloss . . . + thing non-existent . . . + not Past Negative
'agge' 'red' Vagge'-mono'agge'gu-naru 'aggegu-ne' 'aggegu-né'kke
As is clear from the comparison of (27) and (28), the Negative ne' be-
haves parallel to the other accented adjectives. Thus, I conclude that the
adjective should be classified as accented.
Returning to the bound form ne', I assume that this bound form is a
kind of compound adjective and suggest that the form should be inter-
preted in a parallel way that Halle and Vergnaud (1987) treated some cas-
es of compounds of Dakota and English. They propose that the noun man
in English is demoted to an affix in some cases. To see this, consider the
following examples cited from Halle and Vergnaud (1987: 91):
(29) a. [ae]: garbage man front man Kennedy man garage man
b. [a]: infantryman salesman outdoorman postman doorman
They then note as follows:
(30) The forms in [(29a)] are ordinary compounds with the falling stress
contour typical of such words in English. The forms in [(29b)], on the
other hand, are compounds whose second constituent is treated not
as a full word but as an affix. This is shown by the fact that man sur-
faces without stress.
- H a l l e and Vergnaud (1987: 92)
In line with their proposal, I interpret the bound adjective ne' as being de-
moted to an affix. As a result of this demotion, the accented adjective los-
es accent, and becomes an accentless affix. This resolves the apparent in-
compatibility under consideration.
3. An alternative analysis of Tsuruoka Japanese
In the previous sections, I have assumed that Tsuruoka Japanese has the
accent shift rule in (5). We are now in a position to propose an alternative
analysis which does not incorporate accent shift. Instead of shifting ac-
cent, we can account for the tonal facts of Tsuruoka Japanese if we as-
sume a phonetic rule which moves the Η-tone one mora to the right, if the
mora contains non-high vowel [e, o, or a]:
(31) Η-tone Shift:

Move the Η-tone to the right, if the following mora contains non-
high vowel.
This is formally represented as in (32).

* *
(32) V V V ' V
-high -high
H I
H
Note that this rule can account for the tonal facts of Tsuruoka Japanese,
which were treated by the accent shift rule, without introducing addition-
al mechanisms. Since this new system has no shift of accent, we have no
need to worry about the ordering of the rule of accent deletion before no
and Η-tone Shift. The former, which is a phonological rule whose func-
tion is to readjust the accentual property of lexical items, should apply be-
fore the latter, which is a phonetic rule. I assume that phonetic rules are
ordered after phonological rules, based on Haraguchi (1998b, 1999).
The existence of Η-tone Shift is quite natural, taking into consideration
the fact that even in Common Japanese there is a tendency to delay the
lowering of Η-tone associated with the accented syllable. The reason that
this phonetic rule is dependent on the height of the vowel of the following
mora in Tsuruoka Japanese is that the low sonority of the high vowels
prevents the Η-tone from moving onto them in this dialect.
Though I will skip the discussion of how this phonetic rule can account
for the tonal facts handled by the accent shift rule (5), it should be clear
that this alternative analysis makes it easier to account for the tonal prop-
erties of this dialect. We can conclude that all the examples which are con-
sidered to be evidence for Accent Shift can now be interpreted as evi-
dence for the Η-tone Shift rule.
4. Summary of Tsuruoka Japanese
In this paper, we have examined the tonal system of Tsuruoka Japanese

in detail. First, the tonal system of nouns was discussed and it was deter-
mined that the basic tone melody is LHL and that if we introduce an ac-
cent shift rule dependent on the height of vowels, we can account for
these tonal properties. Then the tonal system of verbs was closely exam-
ined and it has been shown that the accent shift analysis of Haraguchi
(1979), (1997) and (1998) is well supported. Finally, the tonal system of
adjectives was briefly discussed and it was shown that the accent shift rule
works well here.
Then, an alternative system to account for the tonal properties of Tsu-
ruoka Japanese was proposed, which incorporates a phonetic rule: H-
tone shift. This system is natural and can account for the Tsuruoka facts
without producing any vicious effects. It contains no theoretically difficult
problems, either. The seemingly complicated tonal system of Tsuruoka
Japanese has now become simple and easier to deal with.
Note
* This paper is in part supported by The Special Research Project of the Typological Inves-
tigation into Languages and Cultures of the East and West, The COE Project of Kanda
University of Foreign Affairs, Monbusho's Grant-in-Aid for Scientific Research (A), and
Grant-in-Aid for International Scientific Research (Joint Research).
I am grateful to two anonymous reviewers of this paper, Wayne P. Lawrence, Roger
Martin, John Shillaw, Kevin Varden, and Jeroen van de Weijer for many helpful com-
ments and suggestions on an earlier draft of this article. The remaining errors and insuf-
ficiencies are of course mine.
References
Halle, Morris and Jean-Roger Vergnaud
1987 An Essay on Stress. Cambridge, MA: MIT Press.
Haraguchi, Shosuke
1977 The Tone Pattern of Japanese: An Autosegmental Theory of Tonology. Tokyo:
Kaitakusha.
1979 "Nihongo Onchou no Shosou (Aspects of Japanese tonal systems)," Gengo no
Kagaku (Science of Language) 7, 21-69.
1991 A Theory of Stress and Accent. Dordrecht: Foris.
1997 "Tsuruoka Hougen no Akusento Saikou (Accent of Tsuruoka Japanese Re-
considered)," Report (1): Researching and Verifying an Advanced Theory of
Human Language: Explanation of the Human Faculty for Constructing and

Computing Sentences on the Basis of Lexical Conceptual Features. 231-95.
Kanda University of Foreign Affairs.
1998a "Aspects of Japanese Accent," Report (2): Researching and Verifying an Ad-
vanced Theory of Human Language: Explanation of the Human Faculty for
Constructing and Computing Sentences on the Basis of Lexical Conceptual
Features. 231-95. Kanda University of Foreign Affairs.
1998b "A Theory of the Syllable," paper presented at LP '98, Columbus, Ohio, Sep-
tember 15,1998.
1999 The Phonology-Phonetics Interface and Syllabic Theory," paper presented at
HILP 4, Leiden University, January 29,1999.
Kindaichi, Haruhiko
1953 "Akusento (Accent)," Chiiki Shakai no Gengo Seikatsu: Tsuruoka ni Kansuru
Jittai Chousa (Language Life in a Regional Community: Investigation into the
Tsuruoka Dialect) 163-83. Tokyo: Shuei Shuppan.
McCawley, James D.
1968 The Phonological Component of a Grammar of Japanese. The Hague: Mou-
ton.
Nitta, Tetsuo
1994 "III Shou: Tsuruoka Hougen no Akusento (Chapter III: Accent in the Tsu-
ruoka Dialect)." Tsuruoka Hougen no Kijututeki Kenkyuu: Dai-3-ji Tsuruoka
Chousa Houkoku 1 (A Descriptive Study of the Tsuruoka Dialect: The Third
Language Survey in Tsuruoka City, The First Report) 81-140.
Shibata, Takeshi
1957 "Akusentoron no Tame ni (Remarks on Accentual Theories)," Kokugogaku
29, 67-76.
How should we Represent 'g' in toge in Japanese
Underlyingly?*
Takeru Honma
Tokyo Metropolitan University
1. Introduction
This paper will discuss the underlying representation of 'g' 1 in Japanese

within the framework of Optimality Theory (Prince and Smolensky 1993;
McCarthy and Prince 1995). In Japanese, [g] and [g] are in complementa-
ry distribution, [g] appears in the word-initial position, while [g] appears
elsewhere. As the two phones, namely [g] and [g], are in complementary
distribution, traditionally phonologists 2 have supported that they are al-
lophones of the same phoneme, say, /g/.
A version of Optimality Theory which adopts Lexicon Optimization
(LO) (Prince and Smolensky 1993: 192) makes a rather different claim.
LO leads us to posit /toge/ as the legitimate input for the surface word
[toge], since LO requires that the input forms of underived words to the
EVAL be identical with the output forms of the words. The word-initial
[g] and the word-medial [g] may have different input forms.
This situation may pose some problems to the sympathy-based analysis
of the interaction between Rendaku (or Sequential Voicing) and [g]/[g]-
allophony given by Itô and Mester (Itô and Mester 1996, 1997a,b; Itô
1999; Mester 1999), as their analysis crucially depends on the assumption
that toge is represented as /toge/ (with no nasalization on the velar stop).
In order to solve the problems, we extend Sympathy Theory beyond
the original proposal by McCarthy (1997, 1998,1999). Whereas McCar-
thy (1997, 1998, 1999) stipulates that ^-selecting constraints must be
Faithfulness constraints, we allow Markedness constraints to be ^-selec-
tors. This extension will have far-reaching consequences. We will not en-
ter into the issue further in this paper, but we will point out the possibility
of distinguishing the cases with Faithfulness constraints as ^-selectors
from the cases with Markedness constraints as ^-selectors.
This paper is organized as follows. In section 2, we will review the basic
facts about Sequential Voicing and Voiced Velar Nasalization and the
analysis of their interaction given by Itô and Mester (Itô and Mester 1996,
68 Takeru Honma
1997a, b; Itô 1999; Mester 1999). It will be shown that a standard version
of Optimality Theory cannot account for the opaque interaction between
the constraints in Japanese phonology. In section 3, we will introduce
Sympathy Theory originally proposed by McCarthy (1997, 1998, 1999)
and review the analysis of the opaque interaction in Japanese phonology
in question. In section 4, we point out that the analysis provided by Itô
and Mester (Itô and Mester 1996,1997a, b; Itô 1999; Mester 1999) is not
in consonant with the assumption of Richness-of-the-Base and Lexicon
Optimization. In section 5, we will explore the possibility of extending the
notion of ^-selector from the set of Faithfulness constraints to the set of
Faithfulness and Markedness constraints. Concluding Remarks will be
given in section 6.
2. Basic facts
In this section, we will review the basic facts about Sequential Voicing and
Voiced Velar Nasalization and the analysis of their interaction given by
Itô and Mester (Itô and Mester 1996,1997a, b; Itô 1999; Mester 1999). It
will be shown that the standard version of Optimality Theory cannot ac-
count for the opaque interaction between the constraints.
2.1. Sequential Voicing
Consider the examples in (1) which show some relevant factors of

Rendaku or Sequential Voicing. Sequential Voicing is a junctual process
in which the initial obstruent of the second member of a compound is
voiced. For example, in (la) when [tama] 'ball' is compounded with [tep-
poo] 'gun' and becomes the second member of a compound [teppoo-
dama], literally, 'gun ball', the [t] in [tama] is voiced and turns into [d].
However, Sequential Voicing is blocked, when the second member of a
compound already contains a voiced obstruent. For example, in (lb) the
voicing of the initial obstruent [t] in [taba] is blocked, because this stem
has [b] (a voiced obstruent) in it. As a result, we have [satsu-taba], in-
stead of [satsu-daba]. This blocking effect is known as "Lyman's Law." I
do not commit myself to discussing the exact nature of this constraint (see
Itô and Mester 1996,1997a, 1998: for details).
How should we Represent 'g' in toge in Japanese Underlyingly? 69
(1) a. Sequential Voicing in compounds:

tama 'ball' teppoo + dama 'bullet'
tana 'shelf' garasu + dana 'glass shelf'
b. Blocking of Sequential Voicing in stems containing voiced ob-
struents ("Lyman's Law")
taba 'bundle' satsu + taba 'wad of bills' *satsu-daba
sode 'sleeve' furi + sode 'long-sleeved kimono' *furi-zode
Itô and Mester (Itô and Mester 1996, 1997a, b; Itô 1999; Mester 1999)
propose the constraint ranking in (2) in order to account for Sequential
Voicing.
(2) Lyman's Law OCP on [+voi, -son] in a stem
SEQVOI Initials of second compound members should be voiced

I
IDENT-IO(voì) Input-Output correspondents are identically specified
for [voice]
In this analysis, Lyman's Law is considered to be a constraint on double

occurrences (OCP) of the feature combination [+voi, -son] within a
stem3.
I D E N T - I O ( v o ì ) is dominated by SEQVOI 4 . This ranking ensures the voicing
of the initial obstruents of the second members in compounds, even if this
voicing incurs the violation of IDENT-IO(voì). Tableau (3) shows that the
candidate (a), which satisfies SEQVOI, wins.
(3) SEQVOI » IDENT-IO(VOÌ)
/hana-sono/ Lyman's Law SEQVOI LDENT-IO(voi)

*
a. hana-zono
b. hana-sono *!
The figure in (4) is the schematic representation of the effect of Sequen-

tial Voicing.
70 Takeru Honma
(4) word
R(endaku) ( = [+voi, — son])
The tableau ( 5 ) shows that Lyman's Law outranks SEQVOI. This ranking
correctly chooses the candidate (b) as a winner. Lyman's Law blocks the
voicing if the input of the second member in a compound contains a
voiced obstruent.
(5) Lyman's Law > SEQVOI
/satsu-taba/ Lyman's Law SEQVOI LDENT-IO(voi)

a. satsu-daba *! *
b. satsu-taba *
The figure in (6) shows the configuration where Sequential Voicing is

blocked by Lyman's Law.
(6) word
*R J
+voi
—son
2.2. Voiced Velar Nasalization
In a conservative dialect of Tokyo Japanese, the two voiced velar seg-

ments [g] and [rj] are in complementary distribution. As is illustrated by
the data in (7), [g] and [rj] occur in the initial position of a Prosodie Word
(PrWd) and in the PrWd-medial position, respectively:
(7) a. Initial [g]

[g ] to ] gloss
geta *geta 'clogs'
giri *rjiri 'duty'
guchi *guchi 'complaint'
go *go 'game of G o '
garasu *garasu 'glass'
b. Initial [g]
[...g...] [...g...]
*kagi kagi 'key'
*kago kago 'basket'
*kaggae kaggae 'thought'
*tokage tokage 'lizard'
*igirisu igirisu 'England'
Basically, this alternation can be analyzed as a case of allophonic varia-

tion. Allophonic patterns emerge from the interaction of faithfulness con-
straints, context-free markedness constraints and contextual markedness
constraints (see, for example, Kager 1999: 30).
(8) Contextual markedness > Context-free markedness > Faithfulness
Itô and Mester (Itô and Mester 1996,1997a, b; Itô 1999; Mester 1999) pro-
pose the constraint ranking in (9).
(9) * P r W d [g [g] is prohibited PrWd-initially

I
*g Voiced velar obstruents are prohibited
I
lDENT-IO(nas) Input-Output correspondents are identically
specified for [nasal]
In this case, iDENT-IO(nas) (the faithfulness constraint) is dominated by a

segmental context-free markedness constraint (here, *g that prohibits
voiced velar obstruent). This markedness constraint is in turn dominated
by a contextual markedness constraint (here, * P r W d [g that prohibits a ve-
lar nasal η PrWd-initially).
(10) * PrW dfo > *9 » lDENT-IO(nas)

Contextual markedness Context-free markedness Faithfulness
72 Takeru Honma
*PrWd[o m u s t outrank *g. This ranking accounts for the realization of the
voiced velar stop PrWd-initially in output for the corresponding voiced
velar segment in input, as is illustrated in the tableau (11):
(li) W o » * g
/geta/ "VrWdfO *g lDENT-IO (nas)

a. g e t a *
b. geta *! *
The tableau (12) illustrates the ranking between *g and lDENT-IO(nas). In

order to satisfy the higher-ranked markedness constraint (*g), the faith-
fulness constraint (lDENT-IO(nas)) is violated.
(12) *g » lDENT-IO(nas)
/kagi/ *PrWd[0 *g lDENT-IO (nas)

a. kagi *!
®· b. kagi *
2.3. Interaction between Sequential Voicing and Voiced

Velar Nasalization
Itô and Mester (1997b) report that we can find both transparent and
opaque interactions between the constraints that are relevant to Sequen-
tial Voicing and Voiced Velar Nasalization (VVN).
In (13), the stem-initial /k/ in the input form surfaces as [Q]. S E Q V O I is re-
sponsible for the voicing of the surface correspondent of the input /k/. Sim-
ple voicing of the input /k/ into [g] will create the violation of *g. Since *g
bans voiced velar obstruents from surfacing in the output, the input /k/
will surface as a velar nasal [rj]. This interaction between two constraints
S E Q V O I and *G is transparent, in the sense that the effects of the constraints,
namely, voicing and nasalization, can be observed in the output.
(13) Transparent interaction: k ->· g -»• η

kuni 'country' yuki + rjuni 'snow country'
kami 'paper' ori + gami 'origami paper'
kaki 'writing' yoko + rjaki 'horizontal writing'
kusuri 'medicine' nuri + gusuri 'medical ointment/cream'

kirai 'dislike' onna + girai 'woman-hater'
On the other hand, the examples in (14) show that a stem-internal [g]
(14a), which is the correspondent of an input /g/, blocks the effect of
SEQVOI, even if it is not a voiced obstruent phonetically. In other words,
in spite of its sonorant nature in the output, a stem-internal [g] does not
behave like a sonorant such as in (14c), but behaves as if it were a voiced
obstruent such as in (14b) in terms of Lyman's Law and then blocks Se-
quential Voicing. This interaction between the stem-internal voiced velar
nasalization effect and the stem-initial Lyman's Law effect (i.e. the lack of
Sequential Voicing) is opaque, in the sense that Lyman's Law can take an
effect on a form with a stem-internal [g], even if the form has no voiced
obstruent triggering Lyman's Law. In other words, a stem-internal [g] be-
haves as if it were a voiced obstruent and triggers Lyman's Law and
blocks the Sequential Voicing effect.
(14) Opaque interaction: Surface [g] behaves as if it were [g]

a. togi 'sharpen' hamono + torji 'knife grinder' *hamono + dogi
toge 'thorn' saka + torje 'reverse thorn' *saka + doge
b. taba 'bundle' satsu + taba 'wad of bills' *satsu + daba
sode 'sleeves' furi + sode 'long-sleeved kimono' *furi + zode
c. tama 'ball' teppoo + dama 'bullet'
sono 'garden' hana + zono 'flower garden'
Itô and Mester (1997b) propose the constraint hierarchy (15) by connect-
ing the two ranked hierarchies (2) and (9). In this hierarchy, SEQVOI dom-
inates *g and LDENT-IO(nas).
^ ^ Lyman's Law
I
SEQVOI * [η
iDENT-IO(voi) *g
I
iDENT-IO(nas)
While the ranking in (15) can select the correct winner in the transparent
case (16), it will designate the wrong candidate as the winner in the
opaque case (17).
74 Takeru Honma
(16) Analysis of a transparent case with the ranking (15)—correct result:

/ori-kami/ Lyman's Law SEQVOI *g LDENT-IO(voi) LDENT-IO(nas)
a. ori-kami *!
b. ori-gami *! *
c. ori-Qami * *
(17) Analysis of an opaque case with the ranking (15)—wrong result:

/saka-toge/ Lyman's Law SEQVOI *g LDENT-IO(voi) LDENT-IO,„as)
a. saka-toge *! *
b. saka-doge * *
desired winner *! *
c. saka-toge
wrong winner * *
d. saka-dor)e
Itô and Mester (1997b) conclude that it is not possible to capture this kind
of opaque interaction in standard O T and they present an alternative ac-
count based on the Sympathy Theory proposed by McCarthy (1997,1998,
1999). In the next section, the sympathetic analysis presented by Itô and
Mester (1997b) will be discussed in some detail and some crucial prob-
lems will be pointed out.
3. Sympathy Theory and Sympathetic analysis
In this section, we will review Sympathy Theory originally proposed by

McCarthy (1997, 1998, 1999) and the Sympathetic analysis of the prob-
lematic case discussed above.
3.1. ^-Selecting and ^-to-O correspondence
McCarthy (1997) originally proposes Sympathy Theory in order to ac-

count for the opaque interactions among constraints in the languages
such as Tiberian Hebrew, Catalan and others. In Sympathy Theory, a new
kind of correspondence, namely the correspondence between two co-can-
didates in a single candidate set, is considered to play an important role.
One candidate is designated as the ^-candidate (or the sympathetic can-
didate) and another candidate will be selected as the optimal output in
terms of the correspondence relation between the ^-candidate and the

output. If a candidate is faithful to, or similar enough to, the ^-candidate
in one way or another, then it is selected as the optimal output. The $b-
candidate is the candidate that is optimal with respect to some designated
constraint, or ^-selecting-constraint C*. To put it more precisely, the
candidate is the best among the candidates that do not violate C*. In (18),
the more formal expression of the notions is cited from Itô and Mester
(1997b).
(18) Given a constraint hierarchy iK and a designated constraint C* e 'H,
a. C* partitions the candidate set into two subsets (as do all con-
straints):
(i) those that do not violate C*, and
(ii) those that violate C* (to whatever degree — gradience of vio-
lation is irrelevant as long as there is some candidate that does not
violate C* at all).
b. The designated sympathy candidate (= ^-candidate) is that ele-
ment of subset (i) that best-satisfies Ή - C* (the rest of the con-
straint system), in the standard optimality-theoretic sense (Prince
and Smolensky 1993).
Note that, in Sympathy Theory, two types of constraints (^-selecting con-

straints and Sympathy constraints) play very important roles. While a de-
selecting constraint (C*) determines the ^-candidate among the candi-
dates, a Sympathy constraint governs the ^-to-Output correspondence
and contributes the selection of the real output. In the next subsection, we
will examine the crucial role of a Sympthy constraint.
3.2. A Sympathetic analysis of the Japanese case
Let us return to the Japanese case. Itô and Mester (1997b) point out that
nasalization that occurs in the output obscures (or 'opacifies') the effect
of the interaction between Sequential Voicing and Lyman's Law, and
therefore Sympathy must be oriented towards a non-nasalizing candi-
date. In other words, a candidate within the set of non-nasalizing co-can-
didates must be referred to as the ^-candidate. In order to select the suit-
able ^-candidate, Itô and Mester (1997b) propose that ^-selector C*
should be lDENT-IO(nas). lDENT-IO(nas)® divides the output into two sub-
76 Takeru Honma
sets: the one is the set of nasalizing candidates which violate this con-
straint, and the other is the set of non-nasalizing candidates which ob-
serve the constraint. One of the candidates within the set of non-
nasalizing candidates will be selected as the ^-candidate.
(19)
/saka-toge/ Lyman's Law SEQVOI *g LDENT-IO(nas)* LDENT-IO(voi)
* *
Φ a. saka-toge
b. saka-doge *! * *
c. saka-toge * *
d. saka-dor)e * *
In (19), the candidates can be divided with respect to iDENT-IO(nas)® (=

C*). The candidates (19c, d) violate the C*, and then they cannot take
part in the ^-competition. The darker shading in the tableau represents
this situation. The other candidates (19a, b) do not violate the C® and they
are relevant for the ^-competition. Between (19a) and (19b), (19a) is op-
timal with respect to the remainder of the constraints and is hence desig-
nated as the (sympathy) ^-candidate.
The ^-candidate influences the selection of the actual output through
the (#-to-Output) Sympathy constraint. Itô and Mester (1997b) propose
that, in the grammar of Tokyo Japanese, the [voice]-Sympathy constraint,
namely IDENT-#0 (voí) , behaves as the Sympathy constraint. The defini-
tion of IDENT-^0 (voí) is given in (20) and its ranking and the interaction
among the constraints is shown in (21).
(20) IDENT-&0(voí): Corresponding segments in the output and the ^-can-

didate are identical in the voicing feature specification.
(21 )
/saka-toge/ Lyman's Law : IDENT- SEQVOI *g IDENT- IDENT-
1 *O(v0i) IO(nas)* IO(voi)

a. saka-toge * *
b. saka-doge *! i * *! *
' r
c. saka-toge 1 * *
d. saka-doge 1 *; * *
I
The tableau (22) makes it sure that the constraints and the ranking that
are introduced into the Sympathetic analysis can also account for the
transparent case (16).
(22)
/iro-kami/ Lyman's Law IDENT- SEQVOI *g IDENT- IDENT-
Ô(VOI) « W IO(voi)
'Ss a. iro-gami *! *
b. iro-kami *! *
"" c. iro-gami * * *
The analysis so far seems to be reasonable, but as we will see below, it em-
bodies problems related to the Richness of the Base assumption and the
principle of Lexicon Optimization.
4. Richness of the Base and Lexicon Optimization
4.1. The possibility of velar nasals in input
The analysis of the opaque cases with velar nasals in the stem-medial po-
sition (i.e. [saka-toge] case) crucially depends on the assumption that ve-
lar nasals in the output correspond to voiced (but non-nasal) velar stops
in the input. This can be a problem, because this is not compatible with
the assumption of Richness of the Base. In Optimality Theory, grammat-
ical generalizations are expressed as the results of interactions of
Markedness/Structural constraints at the level of the output or con-
straints that state correspondences between the output and the other lev-
els (the level of input, ^-candidate, etc.) but no specific property can be
stated referring to the level of input alone:
(23) Richness of the Base: No constraints hold at the level of underlying

forms, (see, for example Kager 1999:19)
Itô and Mester (1997b) take up this issue in the footnote (see Itô and
Mester 1997b: note 4). According to the footnote, Kazutaka Kurisu and
Philip Spaelti pointed out this issue to them. Under the richness-of-the-
base assumption, we cannot prevent an alternative form such as /saka-
toqe/ from being a possible input. As the result of the effect of faithful-
ness to input nasality (namely, lDENT-IO(nas)), this input, however, will
wrongly correspond to the output [saka-doge].
The tableau (24) shows that, for an alternative input /saka-torje/, the
constraint system that we have developed so far would select [saka-doqe]
78 Takeru Honma
as the Sympathy ^-candidate, and, ultimately, would designate this form

as the optimal output, which is not the case.
(24)
/saka-toge/ Lyman's Law IDENT- SEQ Voi *9 IDENT- IDENT-
Ô(VOI) IO (nas) * IO(VOI)
Φ a. saka-toge *! * * *
b. saka-doge *! * * *
desired winner *! *
c. saka-torje
Wrong winner *
d. saka-dorje
Below we will find that, whereas Lexicon Optimization provides a desired

result for the stem-initial [g], it fails to choose an appropriate input for the
stem-internal [g].
4.2. Limiting input: Lexicon Optimization
One seemingly obvious way to circumvent the problem is to provide some

principle to prevent velar nasals from appearing in the input. Lexicon Op-
timization (LO) (25) proposed by Prince and Smolensky (1993) might be
the candidate for such a principle.
(25) Lexicon Optimization (Prince and Smolensky 1993:192)

Suppose that several different inputs I2, ..., I n when parsed by a
grammar G lead to corresponding outputs O l 5 0 2 , . . . , O n , all of which
are realized as the same phonetic form Φ — these inputs are all pho-
netically equivalent with respect to G. Now one of these outputs must
be the most harmonic, by virtue of incurring the least significant vio-
lation marks: suppose this optimal one is labelled O k . Then the learn-
er should choose, as the underlying form for Φ, the input I k .
LO provides a desired result for the stem-initial [g]. The combined tab-
leaux in (26) show that two possible inputs, /geta/ and /η et a/, can corre-
spond to a desired output, [geta]. The input /geta/ will be chosen as the ac-
tual input for [geta], because the output corresponding to the input /geta/
incurs only one constraint violation and the violation is not fatal in the up-
per tableau (26), while the output corresponding to the input /geta/ incurs
two violations in the lower tableau in (26).
(26)
/geta/ *PrWdW *g lDENT-IO(nas)
*
a. geta
*
b. geta *!
/geta/ *PrWd[0 *g lDENT-IO(nas)
* *
c. geta
d. geta *!
The Lexicon Optimization Tableau5 (27) provides a comprehensible rep-

resentation of the effect of Lexicon Optimization. In the tableau, the in-
put (a) wins, because its mapping to the surface form incurs less serious
violations than that of the candidate (b).
(27)
LO input output *PrWd[0 *g lDENT-IO(nas)
<*· a. *
/geta/ [geta]
b. /geta/ [geta] *
*!
Lexicon Optimization can select a desired input for a word with word-ini-
tial [g]. This result is "desired," in the sense that the input provided by L O
is identical with the input required by the Sympathetic analysis presented
in Itô and Mester (1997b).
On the other hand, for the stem-medial case, the input provided by L O
is different from the one required by the Sympathetic analysis. For exam-
ple, in order to get the output [toge], we can assume both /toge/ and /toge/
as shown in (28) and (29), respectively:
(28) /toge/ as an input

/toge/ *PrWdh) *g lDENT-IO(nas)
a. toge *!
*
b. toge
(29) /torje/ as an input

/toge/ *PrWd[0 *g lDENT-IO(nas)
a. toge *!
b. toge
Lexicon Optimization will choose /torje/ as the input for [toge]. The L O
tableau (30) illustrates the situation.
80 Takeru Honma
(30)
LO input output *PrWdfr) *g lDENT-IO(nas)
a. /toge/ [torje] *!
b. /toge/ [toge]
As we see in the tableaux (21) and (24) above, the Sympathetic analysis
requires that the second member of the compound [saka-toge] should be
/-toge/ but not /-torje/ at the level of input, /torje/ is predicted in the iso-
lated form but /-toge/ is required for the compound.
5. The nature of ^"-selecting constraints
Itô and Mester (1997b) and Mester (1999) suggest that the crucial refer-
ence to the input made by the ^-selecting constraint C®, (namely, I D E N T -
IO(nas) in the case at hand) could be the source of the problem. They men-
tion a possible alternative that utilizes a markedness constraint against the
occurrence of [η] as the ^-selector 6 instead of an ΙΟ-Faithfulness con-
straint (namely, LDENT-IO(nas)), but they do not develop this alternative ful-
ly in their papers. It is worth while fleshing it out. Based on their sugges-
tion, if we designate a Markedness constraint as the ^-selecting constraint,
we can get the desired output [saka-toge] either from /saka-toge/ (31) or
from /saka-toge/ (32). In other words, the analysis presented here is com-
patible with the Richness-of-the-Base assumption.
(31)
/saka-toge/ LL IDENT- SEQVOI *g IDENT- IDENT-
Ô(VOi) 'O(nas) IO(voi)
& a. saka-toge * *!
* * *
b. saka-doge *!
c. saka-toge * * *
d. saka-doge *t * * *
(32)
/saka-toge/ LL IDENT- SEQVOI *g IDENT- IDENT- *g*
IO(nas) IO(voi)
Φ a. saka-toge *
*! *
b. saka-doge *L * * * *
/saka-toge/ LL IDENT- SEQVOI *g IDENT- IDENT-
IO(NAS) IO(VOI)
c. saka-torje * *
d. saka-dorje *! * *
The analysis presented here which incorporates as a ^-seletor is a

straightforward development based on the suggestion made by Itô and
Mester (1997b) and Mester (1999), and is not my original proposal. How-
ever, I want to point out the theoretical advantage of this analysis that
they fail to mention. This analysis is not only able to incorporate the Rich-
ness-of-the-Base assumption but also is in consonant with the principle of
Lexicon Optimization. The Lexicon Optimization tableau in (33) shows
that the input form in (b) (/saka-torje/) will be selected as the actual input,
because its mapping to the output [saka-toge] incurs less serious viola-
tions than the other possible input in (a) does. The form /toge/ is legiti-
mately considered as the input for the isolated word ([toge]) as we have
seen in (30), and, at the same time, the one for the second member of a
compound /saka-toge/ as in (33).
(33).
LO LL IDENT- SEQVOI *g IDENT- IDENT-
IO(nas) IO(voi)
a. /saka-toge/ *
*! *
[saka-toge]
<*" b. /saka-torje/ * *
[saka-toge]
Now we are in a position to answer the question posed in the title of this
paper: how should we represent 'g' in toge in Japanese underlyingly? The
answer is: we should represent 'g' in [toge] as /g/ underlyingly.
6. Concluding remarks
The analysis so far provides the desired theoretical consequences. We do

not have to create any unreasonable limits on the input (both /saka-toge/
and /saga-toge/ are considered to be reasonable inputs) and we can let the
principle of Lexicon Optimization select the actual input for the simplex
word [toge] and the one for the compound word [saka-toge] without any
undesired discrepancy of the input representations for the stem /-toge/.
82 Takeru Honma
Despite the advantages above, the analysis requires an extension of

Sympathy Theory beyond the original proposal of McCarthy (1997,1998,
1999). In the original proposal, McCarthy stipulates that ^-selecting con-
straints must be faithfulness constraints. The analysis above incorporates
*r) as a ^-selector and this constraint is not a faithfulness constraint but a
markedness constraint. This extension can bring about undesired results
of unlimitedness into the theory. I do not enter into this issue further, but
I want to point out a major difference between the cases discussed by
McCarthy (1997,1998, 1999) and the Japanese case discussed in this pa-
per in which Sequential Voicing and Voiced Velar Nasalization are in-
volved. The cases discussed by McCarthy involve prosodie reorganization
(that is, deletion and/or epenthesis of segments), but the case at hand
does not. This distinction can be a crucial factor to distinguish the cases
where Faithfulness constraints behave as ^-selector from the cases where
Markedness constraints behave as such.
Furthermore, I want to mention that the Japanese case is not a sole case
which can be analyzed in a straightforword way if we assume Markedness
constraints can behave as ^-selectors. Mester (1999) examines a case in
German where R/B allophony and ç/x allophony interact with each other.
He discusses that a sympathetic analysis of this interaction, in which a
Faithfulness constraint (namely, I D E N T - I O (cons)) behaves as a ^-selector,
can have the similar problem of input dependence. Honma (2000) argues
that this problem can be solved if we assume that a Markedness con-
straint against ç (low glide) behaves as a ^-selector. In addition, McCar-
thy (1999a: 14-15) examines the cases in literature in which Markedness
constraints can be considered to behave as ^-selectors, although he de-
nies the possibility in order to eliminate illegitimate Duke-of-York deri-
vations. Further empirical investigations on this issue have to be done, in
order to settle the theoretical disputes.
Notes
* This research was partially supported by the Ministry of Education, Science, Sports and
Culture, Grant-in-Aid for Encouragement of Young Scientists, No. 11710265,1999-2001.
I wish to express my gratitude to the editors of this book for offering me a chance to con-
tribute this paper to the present volume and for their patience with my delay in turning
in my manuscript.
1. I use single quotation marks (") to suggest that the representation is "orthographic."
2. Based on this allophony, Trubetzkoy (1939), for example, mentions that the opposition
between [g] and [rj] signals boundaries. "Im Japanischen besteht zwischen g und rj ein
kombinatorisches Variantenverhältnis, indem g nur im Wortanlaute und η nur zwischen

Vokalen vorkommt; hier kann also der Gegensatz g:rj keine Wortpaare unterscheiden,
wohl dient aber dieser Gegensatz zur Wortabgrenzung, indem g immer den Anfang eines
Wortes angibt." Trubetzkoy (1962: 244)
3. The O C P can be regarded as self-conjunction of markedness constraints. See also Itô and
Mester (1996, 1998); Alderete (1997); Fukazawa (1999) for details. Fukazawa and Ki-
tahara (this volume) deal with the issue of specification of domain on the self-conjunction
and propose that such specification should be eliminated.
4 . One might point out that the constraint S E Q V O I is excessively phemomenon-specific or,
at best, nothing but a restatement of a derivational rule that changes a stem-initial voice-
less obstruent of the second member of a compound into a voiced counterpart. Itô and
Mester (1998) propose another way of looking at the issue. They divide SEQVOI into a
phenomenon-specific part (a linking morpheme [νοί] ρ ) and a universal part (REAL-
I Z E M O R P H E M E ) . Rendaku voicing is encoded as a linking morpheme [voi] e and R E A L -
I Z E M O R P H E M E (a constraint that is violated when a certain morpheme has no output re-
alization) requires the morpheme [voi] e be parsed, if this parse does not incur any
violation of Lyman's Law (see Itô and Mester 1998:27). Fukazawa and Kitahara (this vol-
ume) present their analysis of Sequential Voicing based on this proposal of R E A L I Z E M O R -
PHEME.
5. See Inkelas (1995: 290) for details
6. Mester (1999) suggests * a [g as the ^-selector.
References
Alderete, John
1997 Dissimilation as Local Conjunction. Proceedings of North East Linguistics So-
ciety 27,17-32.
Fukazawa, Haruka
1999 Theoretical Implications of OCP Effects on Features in Optimality Theory. Ph.
D. thesis, University of Maryland.
Honma, Takeru
2000 Sympathetic Analysis of German R-Vocalization. Ms., Downloadable at http:/
/member.nifty.ne.jp/honmat/papers.html.
Inkelas, Sharon
1995 The Consequences of Optimization for Underspecification. NELS 25, 287-
302.
Itô, Junko
1999 Optimality Theory I: The Foundations. Handout of the lecture at Tsukuba
University on 21st, July, 1999.
Itô, Junko and Armin Mester
1996 Rendaku I: Constraint Conjunction and the OCP. Ms. [ROA-144], Download-
able at http://www.ruccs.rutgers.edu/roa.html.
1997a Correspondence and Compositionality: The Gagyo Variation in Japanese
Phonology. In Iggy Roca (ed.), Derivations and Constraints in Phonology, 419-
462. Oxford: Clarendon Press.
1997b Featural Sympathy: Feeding and Counterfeeding Interactions in Japanese. In
Rachel Walker, Motoko Katayama, and Daniel Karvonen (eds.), Phonology
84 Takeru Honma
at Santa Cruz, Volume 5, 29-36. Santa Cruz: The Linguistics Reseach Center,
University of California.
1998 Markedness and Word Structure: OCP Effects in Japanese. Ms. [ROA-255],
Downloadable at http://www.ruccs.rutgers.edu/roa.html.
Kager, René
1999 Optimality Theory. Cambridge Textbooks in Linguistics. Cambridge: Cam-
bridge University Press.
McCarthy, John
1997 Sympathy and Phonological Opacity. Handout at Hopkins Optimality Theory
Workshop/Maryland Mayfest 1997, May 1997.
1998 Sympathy and Phonological Opacity. Ms. [ROA-252], Downloadable at http:/
/www.ruccs. rutgers.edu/roa.html.
1999 Sympathy and Phonological Opacity. Phonology 16, 331-399.
1995 Faithfulness and Reduplicative Identity. In Jill N. Beckman, Laura Walsh
Dickey, and Suzanne Urbanczyk (eds.), University of Massachusetts Occasion-
al Papers 18, 249-384. Graduate Linguistic Student Association.
Mester, Armin
1999 OT: Some Current Developments. Handout of the lecture at Tsukuba Univer-
sity on 21st, July, 1999.
1993 Optimality Theory: Constraint Interaction in Generative Grammar. RuCCs
Technical Report No. 2. Piscataway, New Jersey: Rutgers University Center
for Cognitive Science.
Trubetzkoy, Nikolaj Sergeevich
1939 Grundzüge der Phonologie (1st ed.). Prague: Cercle Linguistique de Prague.
Principles of Phonology. Trans, by Christiane A. M. Baltaxe. Berkeley: Uni-
versity of California Press., 1969.
1962 Grundzüge der Phonologie (3rd ed.). Göttingen: Vandenhoeck & Ruprecht.
Domain-Relative Faithfulness and the OCP:
Rendaku Revisited*
Haruka Fukazawa and Mafuyu Kitahara
Kyushu Institute of Technology and Indiana University
1. Introduction
The Obligatory Contour Principle (henceforth the OCP) was first pro-
posed at the advent of autosegmental phonology (Leben 1973; Goldsmith
1976) for a restrictive representation of autosegments. It was modified
along with the development of phonological theories in the 1980's and has
been adapted to the analyses of a wide variety of languages and phenom-
ena (see McCarthy 1986 for a concise review and Yip 1988 for a typolog-
ical consideration with respect to the OCP). The frequently cited defini-
tion of the OCP is as follows (McCarthy 1986):
(1) At the melodic level, adjacent identical elements are prohibited.
This definition crucially relies on the notion of adjacency which can be

specified at the non-skeletal level. However, the definition also implicitly
relies on the notion of locality where the domain in which the adjacency
is computed is limited to a certain linguistically natural unit, such as syl-
lable, root, word, and so on. Otherwise the scope of adjacency could be
infinitely wide and any two identical elements in a discourse might be
banned. However, the domain specification of the OCP has been treated
in a rather ad-hoc manner in the pre-OT literature.
In the early stages of OT, the OCP was treated as a monolithic con-
straint without major changes from the definition in (1). The domain
specification of the OCP was still an open question. In the later develop-
ment of OT, Itô and Mester (1996) and Alderete (1997) independently
propose the idea that the OCP is a self-conjunction of a markedness con-
straint in some local domain, which follows the formulation of Local Con-
junction (Smolensky 1993; 1995; 1997).
Local Conjunction is a convention where two lower-ranked constraints
can play the role of one higher-ranked constraint when they are conjoined
in a certain phonologically local domain such as "within a morpheme",
86 Haruka Fukazawa and Mafuyu Kitahara
"syllable boundary", and so on. 1 By definition, a domain specification is

explicitly mentioned in every set of conjoined constraints and thus the
proposal that the OCP is a self-conjunction inherits this explicit domain-
specificity. This is taken as an advantage over the blurred status of domain
specification in the monolithic OCP.
However, the domain specification for self-conj oined constraints is still
unrestricted and stipulative. Though the domain is chosen from a rather
limited set of familiar categories in phonology and morphology, such as
segment, syllable, stem, word, and so on, the selection of the domain is de-
termined in an ad-hoc manner for each case in question. This stipulative
nature of the domain issue in the OCP/self-conj unction may have serious
repercussions in the process of language acquisition. For example, the
OCP/self-conjunction is at work in a syllable-sized unit in Alderete
(1997). Is such domain specification universal in all languages? We should
say clearly not because that domain is irrelevant for the OCP, for exam-
ple, in Rendaku in Japanese. Giving up the universality of the domain
leaves us to stipulate that it is language-specific and somehow learned in
the process of language acquisition. The OCP-related phenomena are
even diverse within a single language. In other words, it is not only lan-
guage-specific but also process/phenomenon-specific. For example, a
pitch accent can occur only once in an accentual phrase in Tokyo Japa-
nese. An accentual phrase can contain three or more morphemes (Kubo-
zono 1993). If we apply the OCP/self-conjunction analysis to this phe-
nomenon, we need a domain of that size. On the contrary, Lyman's Law
in Rendaku has traditionally been analyzed as a clear case of the OCP
within a morpheme, which we will see in more detail in the later section.
The two cases have very different size of domains suggesting that domain
specification is not only language-specific but also process/phonomenon-
specific. Children need to know the domain for the OCP/self-conjunction
for each process in the language, which is an undesirable situation. There
is no guarantee that a child can learn such an arbitrary specification of do-
main for each OCP-sensitive case. By eliminating the stipulation of the
domain in the OCP/self-conjunction apparatus, we can construct a sim-
pler and more plausible grammar.
In the present paper, we take Rendaku (Sequential Voicing), which has
been frequently discussed in the context of the OCP (Itô and Mester 1986;
1996; 1998 among others), as a sample case and attempt to show that we
can eliminate the stipulation of domain from the OCP when we relativize
faithfulness constraints with respect to the domain. This is along the line of
the recent developments in Correspondence Theory in which the possibil-
Domain-Relative Faithfulness and the OCP: Rendaku Revisited 87
ity of relativizing faithfulness with respect to a variety of categories or do-

mains has been pursued. For example, a number of studies report that
faithfulness can be relativized with respect to certain sub-groupings in the
lexicon (Benua 1995; 1997; Urbanczyk 1995; 1996; Lubowicz 1997; Fuka-
zawa 1999). Positional faithfulness is another area in which the idea of rel-
ativizing faithfulness constraints has been applied successfully (Beckman
1995). She claims that special positions in the phonological structure, such
as the onset of a syllable can be more sensitive to faithfulness relations
than other positions. Based on these theoretical considerations for relativ-
izing faithfulness constraints, handing over the burden of domain specifi-
cation from the OCP to faithfulness constraints seems a plausible solution.
Moreover, we propose a common overall schema of rankings in which a
markedness constraint intervenes between the two sets of faithfulness
constraints relativized from a single faithfulness constraint.
The structure of this paper is as follows: the basics of Rendaku and
analyses within O T are summarized in the next section. Section 3 reviews
the idea of relativizing faithfulness constraints and its application to var-
ious data. Our new analysis of Rendaku is outlined in section 4. Section 5
points out the common schema of rankings involving relativized faithful-
ness constraints reviewed in section 3 and our analysis of Rendaku in sec-
tion 4, which concludes the paper.
2. Rendaku and Lyman's Law
Let us take a look at the basic facts of Rendaku. 2 The examples in (2)
show typical compound words which have voicing at the juncture of the
two elements.
(2)
a./koi/ 'love' + /fumi/ 'letter' [koibumi] 'love letter'
b./ama/ 'rain' + /tare/ 'drop' —> [amadare] 'raindrop'
c./tsuri/ 'hang' + /kane/ 'bell' —> [tsurigane] 'hanging bell'
d./kaimono/ 'shopping' + /fukuro/ 'bag' —> [kaimonobukuro] 'shopping bag'
When the second element of the compound contains any voiced ob-
struent, Rendaku does not take place. This is known as Lyman's Law. The
examples in (3) show sample cases with hypothetical ungrammatical
forms where two voiced obstruents cannot cooccur within a word.
(3)
a. /take/ 'bamboo' + /higo/ 'stick' —> [takehigo] 'bamboo stick'
*[takebigo]
b. /shira/ 'white' + /sagi/ 'heron' [shirasagi] 'white heron'
*[shirazagi]
c. /kita/ 'north' + /kaze/ 'wind' —> [kitakaze] 'north wind'
*[kitagaze]
d. /mukashi/ 'old days' + /katagi/ —> [mukashikatagi] 'old-fashioned spirit'
'character' *[mukashigatagi]
Voiced obstruents in the first element of the compound do not affect Ly-
man's Law at all. The examples in (4) show that two voiced obstruents can
cooccur within a word in such cases.
(4)
a./tabi/ 'travel' + /hito/ 'person' —> [tabibito] 'traveller'
b./sabi/ 'rust' + /tome/ 'stopper' —> [sabidome] 'anti-corrosive'
c./shibu/ 'bitter' + /kaki/ 'persimmon' —> [shibugaki] 'bitter persimmon'
d./ichigo/ 'strawberry' + /hatake/ 'field' —> [ichigobatake] 'strawberry field'
Ito and Mester (1986; 1998) propose that voicing in Rendaku can be con-
sidered as a morpheme which attaches to the second member of the com-
pound as an abstract prefix ρ. This prefix is assumed to bear the feature
[voice] underlyingly.
(5)
Stem 1 Stem 2
ama ρ tare
[voice]
In order for the prefix to be fully parsed at the surface, the [voice] feature
of the prefix must be realized as a part of the initial obstruent of the sec-
ond member. Ito and Mester propose a constraint REALIZE-MORPHEME to
force the feature to surface in the output. This constraint can only see
whether the input morpheme is realized in the output.
(6)
/ama ρ tare/
1 REALIZE-MORPHEME lDENT[voice]
[voice]
a. [ama + tare] *!
[voice] e
b. [ama + dare] *
/
[voice] e
As tableau (6) shows, REALIZE-MORPHEME penalizes candidate (a) since

the input morpheme [voice]e is not linked to any segment and thus cannot
surface. REALIZE-MORPHEME outranks i D E N T f v o i c e ] , which makes candi-
date (b) with Rendaku win.
However, when the second member of the compound already includes
a [voice] feature, Rendaku does not take place because of Lyman's Law.
/kita+Q + kaze/ [kitakaze]
[voice] e [voice] [voice] e [voice]
OCP violation
Lyman's Law has been analyzed as a violation of the OCP on the [voice]
feature (Itô and Mester 1986; 1998). As a result of the OCP violation, the
[voice]Q feature from the ρ morpheme cannot be linked to any segment
and thus does not surface. Let us tentatively regard the OCP as a mono-
lithic constraint and incorporate it in the previous constraint ranking. In
order for Lyman's Law to block Rendaku, the OCP on [voice] has to be
ranked higher than REALIZE-MORPHEME. This ranking correctly selects
the desired output as shown in tableau (8).3
(8)
/ k i t a + ρ + k a ζ e/
REALIZE- IDENT
1 1 OCPfvoice]
MORPHEME [voice]
[voice] [voice]
a. [k i t a + k a ζ e] *
I
[voice]e [voice]
b. [k i t a + g a ζ e] *! *
/ I
[voice]e [voice]
When the first member of the compound contains the [voice] feature,
Rendaku is not blocked even though there are two [voice] features seem-
ingly adjacent to each other. To account for this asymmetry, the domain of
the OCP constraint is specified as a morpheme in the previous analyses.
(9)
/ t a b i + ρ + h i t o/ OCP[voice]
REALIZE- IDENT
1 1 within a
MORPHEME [voice]
[voice] [voice] morpheme
a. [t a b i + h i t o] *!
I
[voice] [voice]Q
b. [t a b i + b i t o] Not violated *
I \
[voice] [voice]Q
As mentioned in the introduction, we attempt to eliminate this domain

specification from the OCP by relativizing faithfulness constraints. Be-
fore going into the analysis of Rendaku with relativized faithfulness, let
us review the major theoretical developments in Correspondence Theory
which gave rise to the idea of relativized faithfulness.
3. Relativization of faithfulness constraints
In this section, we will make clear that, unlike markedness constraints,

faithfulness organizes a set of constraints such as M A X , DEP, IDENT[F],
etc., and the set has a possibility of multiplying for any linguistic relations
such as Input-Output, Base-Reduplicant, and so on. By assuming that the
grammar generates such a various set of faithfulness constraints, previous
research has shown that the interaction of the faithfulness constraints
from different sets accounts for the cases in which a phonological alterna-
tion is observed in some domain, and not observed in another within a
grammar of a language. We will briefly review cases of the Emergence of
the Unmarked (TETU) (McCarthy and Prince 1994), positional faithful-
ness constraints (Beckman 1995), relativized faithfulness constraints for
distinct morphemes (Urbanczyk 1995; 1996; Benua 1995; 1997; Fukazawa
1998; Fukazawa, Kitahara, and Ota 1998; Itô and Mester 1999).
3.1. Faithfulness and Correspondence Theory
Correspondence Theory (McCarthy and Prince 1995) revises the original

approach to input-output faithfulness in Optimality Theory (Prince and
Smolensky 1993):
(10) Definition of Correspondence

Given two strings SI and S2, correspondence is a relation from the
elements of Sj to those of S2. Elements a e Si and β e S 2 are referred
to as correspondents of one another when α 9Î β.
All correspondence relations are generalized under this definition. At-

tention is given to correspondence between representations, and faithful-
ness constraints are itemized from the segmental, featural, or structural
viewpoint: {MAX, D E P , I D E N T [ F ] , CONTIGUITY, LINEARITY, INTEGRITY, U N I -
FORMITY, A N C H O R , ALIGN}.
Moreover, Correspondence Theory recognizes identity between dis-
tinct types of the representations such as Input-Output (IO), Output-
Output (OO), Base-Reduplicant (BR), Tone-Tone-bearer (TT), etc. The
definition of correspondence in (10) subsumes all types of linguistic rela-
tions, and every relation generates a full set of faithfulness constraints:
IO:{MAX-IO, DEP-IO, IDENT[F]-IO, INTEGRITY-IO, ...}; 00:{MAX-00,
...}; etc. All the sets
D E P - O O , I D E N T [ F ] - 0 0 , ...}; B R : { M A X - B R , D E P - B R ,
of faithfulness constraints which are generated in a grammar attend the
ranking of the grammar to evaluate the correct output.
The simultaneous attendance of all those faithfulness constraints can
account for the phenomena in which some alternation is observed only in
some specific domain. First, McCarthy and Prince (1994) claim that even
within a language in which a marked structure is generally allowed, it is
often the case that the same marked structure is banned in a special do-
main. In such a domain, an unmarked structure emerges. This is called
"The Emergence of the Unmarked (TETU)". TETU is derived from the

interaction of a markedness constraint(s) for the phenomenon and two or
more kinds of faithfulness constraints.
For example, in Dakota coronal dissimilation takes place due to the ef-
fect of the OCP on [coronal] in reduplicated verb morphemes. On the oth-
er hand, coronal dissimilation is not observed in other morphemes. Thus,
Fukazawa (1999) concludes that the two adjacent [coronal] features are al-
lowed in general in Dakota, while the sequence is banned in the redupli-
cated verb forms because of a TETU ranking, MAx[cor]-IO » OCP[cor] >
MAx[cor]-BR.
Tableau (11) shows that OCP[cor] can be violated in the lexical com-
pounds. Although candidate (11a) violates OCP[cor], it still becomes op-
timal. Candidate ( l i b ) is penalized by its violation of the higher ranked
constraint MAx[cor]-IO. The coronal feature of the first segment in the in-
put does not have a correspondent in the output in candidate (b).
(11) No coronal dissimilation in general domains

(lexical compounds) in Dakota
/sdot + chi-ya/ Ί know you.'
I I ' MAx[cor]-IO O CP [cor] MAX [cor]-BR
[cor] [cor]
cr
a. sdod + chi-ya *
1 1
[cor] [cor]
b. sdog + chi-ya *!
1
[cor]
In contranst to lexical compounds in tableau (11), the OCP cannot be vi-

olated in the reduplicative verb morphemes. In tableau (12), the violation
of OCPfcor] penalizes candidate (a), because there is a better candidate
(b) which violates only the lower-ranked MAx[cor]-BR.
(12) Coronal dissimilation in a specific domain

(i.e. Reduplicative verb morphemes) in Dakota
/RED + suta/ 'strong'
1 MAx[cor]-IO OCP[cor] MAx[cor]-BR
[cor]
a. sut sut *!
I 1
[cor] [cor]
*
·»· b. suk sut
1
[cor]
The constraint interaction with the two kinds of MAxfcor] constraints,

namely, for the input-output, and for the base and reduplicant, accounts
for why coronal dissimilation takes place in a reduplicative verb form,
while it is not observed in other morphemes.
Second, Beckman (1995) suggests that faithfulness constraints are rel-
ativized to positions on the basis of the observation that prosodie head
positions tend to retain distinctions that are lost in other positions. As an
example, in Shona, the mid vowels [e] and [o] are contrastive only in root-
initial syllables. They appear in subsequent positions only when preceded
by a mid vowel in root-initial syllables. To account for this asymmetry, she
proposes two kinds of i D E N T f h i g h ] constraints; one is for a root-initial syl-
lable, LDENT ol [high] which outranks two markedness constraints *MID
and *HIGH, and the other is for general positions, iDENTfhigh] which is
ranked lower than those two markedness constraints. The ranking she
proposes is as follows: lDENT o l [high] > * M I D , * H I G H » ÏDENT[high].
Building on Beckman's idea, Lombardi (1995) proposes the unmarked
ranking, iDENTÜNsflar] » lDENT[lar], in which the input-output faithful-
ness constraint for the laryngeal feature in the onset position is ranked
higher than that in general positions. She states that this ranking is un-
marked because preserving the distinctions in the onset is of most impor-
tance. For instance, this ranking can account for Dutch voicing assimila-
tion. In Dutch, the [voice] feature is retained in the onset position /krab
+ en/ -»• [kra.ben] 'to scratch', while it is lost in other positions /krab/ ->•
[krap] 'scratch', /krab + sel/ —» [krap.sel] 'scratchings'.
As in Shona, a markedness constraint *LAR4 intervenes between the
two faithfulness constraints. This ranking, iDENTÜNsflar] > * L A R »
LDENT[lar] correctly accounts for the asymmetry between the voicing as-
similation in the onset and in the other positions in Dutch. In the onset
position, the [voice] feature surfaces because the faithfulness constraint

for the laryngeal feature in the onset position outranks the markedness
constraint for the laryngeal feature.
(13)
/krab + en/ lDENTONs[lar] *LAR lDENT[lar]
*
a. kra.ben
*
b. kra.pen *!
As in tableau (13), candidate (a) with the voice feature in the onset position
becomes optimal due to the proposed ranking. On the contrary, in other po-
sitions, the markedness constraint * L A R penalizes the candidate which con-
tains voiced obstruents. As shown in tableau (14), candidate (b) wins, be-
cause the violation of *Lar is fatal due to the fact that lDENTONs[lar] is
satisfied by both candidates.
(14)
/krab/ lDENTONs[lar] *LAR iDENTflar]
a. krab *!
*
b. krap
Thus, on the basis of those previous studies, we conclude that relativiza-

tion of faithfulness constraints can deal with the asymmetry of phenome-
na between specific and general domains within a language.
3.2. Relativization of faithfulness constraints

for each morpheme
Further development of the idea of relativizing faithfulness has been car-

ried out in papers such as Urbanczyk (1995; 1996), Benua (1995; 1997),
Fukazawa (1998), Fukazawa, Kitahara, and Ota (1998) and Itô and
Mester (1999). They suggest that faithfulness constraints are relativized
not only for the general/specific domains but also for each morphological
unit within a language.
For example, Urbanczyk (1995; 1996) notices that there are two pat-
terns of reduplication in Lushootseed depending on the reduplicative
morpheme: Diminutive or Distributive. The distributive morpheme redu-
plicates the first Q V ^ from the base, while the diminutive morpheme
copies only the first CjVj. For example, the distributive of [badá?]
(C1V1C2V2C3) 'child, offspring' is not *[b9-b9dá?] (C1V1-C1V1C2V2C3)
but [bad-badá?] (C1V1C2-C1V1C2V2C3) 'children'. On the other hand, the
diminutive form for [cabs] 'hand' is [ca-cabs] (C1V1-C1V1C2V2C3) 'little
hand', and a C1V1C2-C1V1C2V2C3 form, *[cal-cabs], is incorrect.
Urbanczyk analyzes this as the avoidance of a coda which results in a
CV-shape for the diminutive. In contrast, codas are possible in the distrib-
utive morpheme, creating a CVC-shape. Therefore, the markedness con-
straint prohibiting codas, NOCODA, is respected in the diminutive redupli-
cation at the expense of a violation of the faithfulness constraint against
deleting a segment, MAX (NOCODA > MAX). However NOCODA is violated
to satisfy MAX in the distributive (MAX > NOCODA). TO resolve this con-
flict between the two rankings, she claims that each of the reduplicative
morphemes has its own correspondence relation to the base; hence, there
are two different Base-Reduplicant (BR) relations in Lushootseed.
Consequently, two full sets of BR faithfulness constraints are generated
in the grammar of Lushootseed: BR-Diminutive (Dim): {MAX-BR-Dim,
DEP-BR-Dim, LDENT[F]-BR-Dim, ...} and BR-Distributive (Dis): {MAX-
BR-Dis, D E P - B R - D Í S , IDENT[F]-DÍS, ...}. These faithfulness constraints
are placed in a single ranking with the markedness constraints. Both the
diminutive CV-shape and the distributive CVC-shape result from ranking
M A X - B R - D Í S » NOCODA » M A X - B R - D i m .
Let us take a look at how the constraint ranking works with the actual
analysis. The given ranking accounts for the CV-shape of the reduplica-
tive forms of the diminutive.
(15) The Diminutive CV-shape:

/RED + calas/ MAX-BR-DÍS NOCODA MAX-BR-Dim
<*" a. ca-calas *
b. cal-calss *!
The same ranking provides the correct analysis for the CVC-shape of dis-
tributive reduplication.
(16) The Distributive CVC-shape:

/RED + bad á?/ MAX-BR-DÍS NOCODA MAX-BR-Dim
a. ba-badá? *!
*
b. bad-badá?
Thus, the two full sets of faithfulness constraints for both distributive and
diminutive reduplicative morphemes are instantiated in the grammar of
Lushootseed.
Benua (1995, 1997) shows that there are two patterns of affixation in
English. For example, certain coda clusters are simplified in both root
morphemes: [kAndem], *[kAndemn] ('condemn') 5 , and class 2 affixation
such as when -ing is attached to the root: [kAndemirj], *[kAndemmr)]
('condemning'), but not in class 1 affixation: [kAndemneysAn], *[kAnde-
meysAn] ('condemnation').
Benua notes that class 1 and class 2 affixal morphemes each display a
different correspondence relation to the output of the root morphemes;
hence, there are two kinds of Output-Output ( O O ) faithfulness relations
in English. Thus, two full sets of faithfulness constraints, namely, OO-class
1 affix: {MAx-OO-class 1 affix, DEP-OO-class 1 affix, IDENT[F] OO-class 1
affix,...} and OO-class 2 affix:{MAX-00-class 2 affix, DEP-OO-class 2 af-
fix, LDENT[F]00-class 2 a f f i x , . . . } are found in the grammar of English.
First, in an unaffixed word, the ranking between *mn] a » MAX-ΙΟ ex-
plains why the output is not [kAndemn] but [kAndem].
(17) Unaffixed word 6

/kAndemn/ *mn] 0 MAX-IO
e a. kAndem *
b. kAndemn *!
The clusters are simplified due to the ranking in (17). Now, from this out-
put form, two kinds of affixation are possible: class 1 and class 2.
In class 1 affixation, the clusters are not simplified, because D E P - O O -
class 1 affix is lower ranked than the phonological constraint *mn] 0 and
MAX-IO.
(18) Class 1 affixation 7

Tab » /kAndemn + eysAn/ DEP-OO- DEP-OO-
*mn] 0 MAX-IO
class 2 affix class 1 affix
a. kAndem.neysAn * *
b. kAndemeysAn **1
On the other hand, the clusters are simplified in class 2 affixation, because
DEP-OO-class 2 affix and *mn] 0 are higher ranked than M A X - I O .
Domain-Relativ e Faithfulness and the OCP: Rendaku Revisited 97
(19) Class 2 affixation

Tab > /kAndemn + ιη/ DEP-OO- : * , DEP-OO-
mn J MAX-IO
class 2 affix ! ° class 1 affix
¡
*! * *
a. kAndem.nir)
¡
b. kAndemir) 1 **
I
Thus, a single ranking explains the occurrence of cluster simplification de-

pending on the class of each affix in English.
Both Urbanczyk's and Benua's research suggest that phonological pat-
terns can vary depending on the difference between morphological cate-
gories within a language: a pattern observed in one category may not
occur in another. Each morphological group gives rise to its own corre-
spondence relation; therefore, it is possible for each of the basic pairs (IO,
OO, BR, TT, etc.) to bear multiple full sets of faithfulness constraints for
each morphological class within a language. We conclude from this that
the full set of faithfulness constraints in Universal Grammar (UG) has the
potential of propagation for any correspondence relation in a language.
Fukazawa (1998) and Itô and Mester (1999) independently propose
that there are five kinds of IO faithfulness constraints depending on sub-
lexica in Japanese: IO-Yamato, IO-Sino-Japanese, IO-Mimetics, IO-For-
eign, and IO-Alien. In Japanese, there are some phonological phenomena
which are observed only in a certain lexical stratum (or strata). The inter-
action of all of those faithfulness constraints with other constraints ac-
count for such lexical stratification in Japanese.
Let us take an example from Japanese. Post Nasal Voicing (PNV) is a dis-
tributional fact that all obstruents become voiced after a nasal in Yamato
and Mimetics (Itô and Mester 1995, Itô, Mester and Padgett 1995, Fukaza-
wa, Kitahara, and Ota 1998). This does not apply in lexical strata other than
Yamato and Mimetics and thus voicing in obstruents is contrastive after a
nasal in, for example, Sino-Japanese. To account for this asymmetry, there
must be two kinds of MAx[voice] constraints for Yamato and Sino-Japanese
respectively. The ranking with these relativized faitufulness constraints is:
MAx[voice]-IO-Sino-Japanese » PNV > MAx[voice]-IO-Yamato.
(20) PNV in Yamato

MAx[voice]-IO- MAx[voice]-IO-
/kan-ta/ 'bite (past)' PNV
Sino-Japanese Yamato
a. kanda *
b. kanta *!
(21) No PNV in Sino-Japanese

MAx[voice]-IO- MAx[voice]-IO-
/han-tai/ 'opposite' PNV
Sino-Japanese Yamato
a. handai *!
b. hantai *
As shown in tableaux (20) and (21), a single ranking can account for the
occurrence of PNV in Yamato and its non-occurrence in Sino-Japanese.
Thus, the simultaneous attendance of the relativized faithfulness con-
straints account for the asymmetric phonological phenomena among stra-
ta in Japanese.
3.3. Possibility of relativizing markedness constraints
As we have seen in the previous section, the relativization of faithfulness

constraints has broadened the scope of Correspondence Theory and has
given us a number of fruitful insights. However, why only faithfulness
constraints can be relativized is rarely asked in the above mentioned lit-
erature. If any markedness constraints can also be relativized with respect
to a certain group of items, we might not need to relativize the faithfulness
constraints to obtain the desired outcome.
Itô and Mester (1999) attempt to give a formal answer to this question.
The main point of their argument is that faithfulness constraints are dif-
ferent from structural markedness constraints in that, as a function, the
former always take two arguments while the latter take only one. In other
words, faithfulness constraints always compare two representations and
return the number of differences between them as violations, while struc-
tural markedness constraints only check one representation and give the
number of violations. In (22a), for example, i and o are arguments of the
function [F( )]( ) which produces the list of violations as return values.
(22)
a. Faithfulness constraint: [F(i)](o) = list-of-violations
b. Structural constraint: S (o) = list-of-violations
Itô and Mester call [F(¿)] part of the faithfulness constraint an "instanti-
ated constraint" which is derived from the two-argument schema of the
function. They further argue that because this instantiated constraint is
derived for each input, there is a possibility that different instantiated
constraints can be ranked in different positions. We adopt their formal

reasoning why only faithfulness constraints can be relativized and assume
that markedness constraints never split.8
3.4. Interim summary and proposal
In this section, we have discussed the nature of faithfulness constraints

with respect to its relativization for various kinds of linguistic relations.
Also, we have confirmed that some asymmetric phenomena depending
on domains within a language are explained by the interaction of those
relativized faithfulness constraints.
As an extension of the relativization of faithfulness constraints, we pro-
pose that among the set of faithfulness constraints proposed in Corre-
spondence Theory, UNIFORMITY can also be relativized depending on each
phonological domain.
For the case of Rendaku, UNiFORMiTY[voice]: the [voice] feature in the
output cannot have multiple correspondents in the input (prohibition of fu-
sion of the two [voice] features) is the relevant constraint to be relativized.
We will claim that UNiFORMiTYfvoice] for "within a domain" is higher
ranked than that for other general domains as in a ranking " U N I F O R M I -
TY[voice]-within a morpheme > UNIFORMITY [voice]-general". This proposal
would imply two points. First, there is a possibility that this ranking schema
which requires being more faithful in a smaller domain than in a larger (or
more general) domain is universal. Second, this ranking leaves a possibility
of relativization of UNiFORMiTY[voice] not only for within a morpheme but
also for within other distinct larger domains. We will leave those two kinds
of implications for future research, and will concentrate on discussing how
the proposed ranking above accounts for the asymmetric phenomenon that
Rendaku is observed in one case, and is blocked in another.
4. Analysis of Rendaku
4.1. Typology of the OCP on features: Three kinds of repair

strategies for the OCP on features
In the framework of autosegmental phonology, following McCarthy's

(1986) formalization of the OCP, Yip (1988) proposes that the OCP some-
times plays the role of a rule trigger. She also indicates that there are four
kinds of repair strategies for the OCP violations, namely degemination,
dissimilation, assimilation, and epenthesis. Therefore, it seems that there
are logically five types of languages from the perspective of the OCP ef-
fect. Fukazawa (1999) reexamined Yip's categorization of languages spe-
cifically for the OCP on features. She eliminates some of the possibilities
from Yip's classification such as epenthesis of a segment, and reorganizes
some of the categories such as dissimilation, assimilation, and deletion. 9
She concludes that there are four types of languages regarding the OCP
on features:
(23) Typology of the OCP effects on features:

Type 1 language: O C P violation is observed.
Type 2 language: O C P violation is not allowed, and featural fusion takes place.
Type 3 language: O C P violation is not allowed, and feature deletion
and feature insertion both occur.
Type 4 language: O C P violation is not allowed, and feature deletion,
or feature deletion leads to segmental deletion.
Thus, we can consider three kinds of repair strategies in the languages in

which the OCP violation is not allowed.
Considering Japanese compounds, the OCP on [voice] ranks high
enough to be satisfied in Japanese as we have already observed in section 2.
(24) OCP on [voice] in Japanese repeated from (8)

/kita + Q + kaze/
REALIZE- IDENT
1 1 OCP[voice]
MORPHEME [voice]
[voice] [voice]
a. [k i t a + k a ζ e] *
I
[voice]e [voice]
b. [k i t a + g a ζ e] *! *
/ I
[voice]e [voice]
In tableau (24), it seems that the first [voice] feature deletes to satisfy
OCP[voice] at the expense of violation of R E A L I Z E - M O R P H E M E as in can-
didate (a). However, in addition to this featural deletion, there are other
possible repair strategies to satisfy OCP[voice] according to Fukazawa's
claim, such as featural fusion, featural change, and segmental deletion.
Among them, featural change and segmental deletion are irrelevant since
they are not observed. Therefore, we examine the case where featural fu-
sion takes place.
(25)
/kita + Q + kaze/
REALIZE- IDENT
1 1 OCPfvoice]
MORPHEME [voice]
[voice] [voice]
a. [kita + kaze] *!
1
[voice]
<desired output>
»! *
b. [kita + gaze]
/ 1
[voice] [voice]
*
*®"c. [kita + gaze]
\ /
[voice]e
<wrong winner>
Candidate (c) in tableau (25) has a featural fusion for repairing the OCP
violation, but it is not the desired output. Another constraint which
penalizes candidate (c) must be introduced. UNIFORMITY[voice], which
prohibits fusion of two [voice] features, is the appropriate constraint to
specifically penalize the fused structure in candidate (25c). U N I F O R M I -
TY[voice] outranks REALIZE-MORPHEME, which leads candidate (a) to be
the output.
(26)
/kita + Q + kaze/
UNIFORMITY REALIZE- IDENT
1 1 OCPfvoice]
[voice] MORPHEME [voice]
[voice] [voice]
*
a. [ k i t a + k a z e ]
1
[voice]
*
b. [ k i t a + g a z e ] *!
/ 1
[voice]e [voice]
*
c. [ k i t a + g a z e ] *!
\ /
[voice]e
Now, let us examine another example in which the first member of the
compound contains the [voice] feature under the same constraint ranking.
(27) The first member contains [voice]
It a b i + ρ + h i t o/
UNIFORMITY REALIZE- IDENT
1 1 OCP[voice]
[voice] MORPHEME [voice]
[voice] [voice]
*«" a. [ t a b i + h i t o ] *
ι
I
[voice]
<wrong winner>
b. [ t a b i + b i t o ] *! *
Ι I
1 1
[voice] [voice] e
c. [ t a b i + b i t o ] *! *
\ /
[voice] e
<desired output>
The ranking given in (26) incorrectly lets candidate (a) win in tableau
(27). The correct output should be determined as either candidate (b) or
candidate (c).
Here, we propose that the faithfulness constraint UNIFORMITY[voice] is
relativized depending on the domain as we have already mentioned in
section 2. Similar to the ranking schema for positional faithfulness, the
general UNIFORMITY constraint UNIFORMITY[voice]-G must be ranked low-
er than the specific UNIFORMITY constraint for within a morpheme, UNI-
FORMITY[voice]-M. The overall ranking including relativized UNIFORMITY
is shown in (28).
(28) Ranking with relativized Uniformity

UNIF0RMITY[v0ice]-M, UNIFORMITY[vOÌCe]-G,
REALIZE-MORPHEME
OCP[voice] iDENTfvoice]
Let us examine this ranking by reanalyzing tableaux (26) and (27). First,
when the second member of the compound contains a [voice] feature,
candidate (a) in tableau (29) wins as in tableau (26). Thus, relativizing
UNiFORMiTYfvoice] does not complicate the picture.
(29) reanalyzed version of (26)

/kita + Q + kaze/
UNIFORMITY REALIZE- UNIFORMITY
1 1 OCP[voice]
[voice]-M MORPHEME [voice]-G
[voice] [voice]
ar
a. [k i t a + k a ζ e] *
[voice]
b. [k i t a + g a ζ e] *!
/ I
[voice] [voice]
c. [k i t a + g a ζ e] *! *
\ /
[voice]e
On the other hand, when the first member of the compound contains a
[voice] feature, candidates in which featural fusion takes place win, which
is the desired result. In tableau (30), candidate (c) does not violate UNI-
FORMITY[voice]-M because the fusion takes place across two morphemes.
Hence, not only the OCP[voice] violation but also the violation of R E A L -
I Z E - M O R P H E M E becomes fatal in the candidate set.
(30) reanalyzed version of (27)

/tabi + Q + hito/
UNIFORMITY REALIZE- UNIFORMITY
1 1 OCP[voice]
[voice]-M MORPHEME [voice]-G
[voice] [voice]
a. [t a b i + h i t o] *!
ι
I
[voice]
b. [ t a b i + b i t o ] *!
I 1
1 1
[voice] [voice]e
c. [ t a b i + b i t o ] *
\ /
[voice]e
As we have seen so far, by introducing the relativized UNIFORMITY con-

straint, the data for Rendaku and Lyman's Law have been accounted for
without specifying the domain for the OCP[voice].
5. Conclusion
As we have pointed out in the previous section, the ranking schema we

have seen in our analysis of Rendaku has one thing in common with other
analyses in Correspondence Theory, such as the case of TETU, positional
faithfulness, and lexical strata: a markedness constraint intervenes be-
tween the two faithfulness constraints relativized from a single source.
(31) Ranking schema in various analyses

Exemplar Relativized Markedness Relativized
language Faithfulness-A Faithfulness-B
•TETU Dakota MAx[cor]-IO OCP[cor] MAx[cor]-BR
• Positional Dutch lDENTONs[lar] *LAR lDENT[lar]
Faithfulness
• Lexical strata Japanese MAx[voice]-IO-SJ PNV MAx[voice]-IO-
(PNV as a sample) Yamato
• Rendaku and Japanese UNIFORMITY REALIZE- UNIFORMITY
Lyman's Law [voice]-M MORPHEME [voice]-G
As summarized in table (31), relativized faithfulness-Α and -B are always

propagated from a single faithfulness constraint for different domains,
positions, or categories.10 The interaction of the intervening markedness
constraint with surrounding relativized faithfulness constraints accounts
for asymmetric phenomena which may seem difficult to treat within a sin-
gle grammar.
As the summary above shows, building on the previous research based
on Correspondence Theory, we have developed the notion of relativiza-
tion of faithfulness constraints for distinct phonological domains. We
have observed a case of Rendaku and Lyman's Law in Japanese where
the violation of UNIFORMITY on [voice] is more serious in a specific pho-
nological domain than in general domains. By introducing the notion of
relativized faithfulness depending on phonological domains, we have suc-
ceeded in eliminating the domain specification of the OCP.
Also, we have introduced a new interpretation of Rendaku and Ly-
man's Law: Rendaku is blocked when the second member of a compound
contains [voice], because it would result in an OCP violation on [voice].
Since UNIFORMITY [voice]-within a domain is ranked higher than REALIZE-
MORPHEME, featural fusion does not take place in this case. On the other
hand, when the first member of the compound contains [voice], Rendaku
does occur. We have analyzed that featural fusion takes place in this case
to satisfy OCP[voice] and Realize-morpheme at the expense of the viola-

tion of UNiFORMiTYfvoice] for general domains.
Thus, we have concluded that whether Rendaku is observed in Japa-
nese compounds depends on whether featural fusion is possible or impos-
sible in the compounds. This asymmetry has been accounted for on the
basis of the relativized faithfulness constraints for phonological domains.
Notes
* We would like to thank the editors of this volume, Jeroen van de Weijer and Tetsuo Nish-
ihara, for offering us a chance to contribute this paper. We also would like to thank Stuart
Davis and Laura W. McGarrity for their helpful comments.
1. Following Smolensky (1993, 1995, 1997), we assume that a local conjunction of con-
straints A and Β (represented as A&B) also plays as another constraint. In other words,
once two (or even more) constraints are conjoined into a local conjunction, it should be
treated as an independent constraint. Therefore, no quantitative factor in the evaluation
procedure is introduced there.
If Local Conjunction is a type of constraint, it must be in Universal Grammar (UG).
However, if it is in UG, it must be cross-linguistically valid. A question now arises: Are
all possible local conjunctions truly in U G ? If so, U G grows extremely large. Following
Fukazawa and Miglio (1998), we consider that the possibility of the conjunction is in UG,
in other words, the " & " operator for conjunction is in UG. However, the choice of which
two constraints to be conjoined is language specific. Furthermore, there must be specific
restrictions on which two (or more) constraints can be conjoined. For the detailed discus-
sion on the restrictions of Local Conjunction, see Fukazawa & Miglio (1998).
2. There are numerous exceptions to Rendaku and much literature is devoted to account
for those exceptions. First, though Rendaku has been considered to take place mostly in
the Yamato (Native) stratum in Japanese lexicon, some Sino-Japanese or Mimetic vocab-
ularies show Rendaku, while some Yamato compounds do not (Itô and Mester 1999,
Vance 1996). These observations are based on etymological analyses of lexical stratifica-
tion. Our position to these exceptions is that a purely phonological lexical stratification
can also be considered as a synchronic grammar in our cognitive system. For a more gen-
eral discussion on lexical stratification, see Fukazawa, Kitahara, and Ota (1998) and Itô
and Mester (1999). Second, compounds with more than two elements have systematic ex-
ceptions to Rendaku due to the internal structure of the compound (Otsu 1980). Third,
pitch accent assignment and Rendaku are argued to be related in some cases (Kibe 1978).
Sato (1989) reviews these and other exceptional cases to Rendaku. Finally, Itô, Mester
and Padgett (1995), Rice (1997), Itô and Mester (1998) and Honma (2000, this volume)
give a detailed discussion of the relationship between Rendaku and Post Nasal Voicing
and voicing specification of sonorants which apparently seems a problem for an OCP-
based analysis of Lyman's Law.
Although we are aware of those exceptions and theoretical complications around
Rendaku, our main focus in this paper is not to explore them in detail but to give a new the-
oretical direction within a general framework of Correspondence Theory and Local Con-
junction. Further research is of course necessary for a more thorough treatment of Rendaku.
3. In tableau (8), another candidate, where the stray voice is from the underlying /z/, can be
considered. A similar situation arises when the second member of the compound has a
voiced initial obstruent. A possible candidate is that the input [voice] is delinked from the
second member and the [voice]Q is re-linked there. For example, /hana/ 'flower' + /gara/
'pattern' is realized as [hanagara] with [voice]e as the voicing feature for [g]. These cases
can be taken care of by additional faithfulness constraints for association lines such as
MAxfassoc] and DEpfassoc]. As shown in tableau (8)', candidate (a) is more faithful to
the association line configuration in the input than (c) though they are phonetically
equivalent. Both the [voice] e feature from the prefix ρ and the [voice] feature from input
[z] are delinked in (8c)', which causes two violations of MAxfassoc]. Moreover, the
delinked [voice] e is re-linked to [z], which causes a violation of DEp[assoc].
(8)'
/kita + Q + kaze/
1 1 MAx[assoc] DEp[assoc]
[voice] [voice]
*
τ a. [k i t a + k a ζ e]
I
[voice]p [voice]
*
c. [k i t a + k a ζ e]
[voice]e [voice]
Though MAx[assoc] incurs violation(s) in other cases with Rendaku, we do not include it
in the following tableaux since the point of discussion here is not to layout the entire
ranking but to review the gist of Itô and Mester's analysis on Rendaku.
4. Lombardi assumes that all features are privative. Therefore, [b] bears [voice], and [p]
does not. The definition of the constraint * L A R is "Do not have the [laryngeal] feature."
The feature [voice] is considered to depend on [laryngeal]. Thus, [b] violates * L A R be-
cause it has [laryngeal], while [p] does not.
5. It could be pointed out that the vowels in the words "condemn", "condemning", and
"condemnation" should be replaced by the vowels [θ] or [a]. We cite the data exactly
from what Benua (1997) mentions.
6. For the discussion of the constraint *mn] 0 see Benua (1997).
7. Tableaux (18) and (19) only show some parts of the recursive tableaux which are utilized
in Benua (1997). For the whole analyses, see Benua (1997).
8. It has been a controversial issue whether the markedness constraints can be split like
faithfulness constraints. Although we adopt Itô and Mester's reasoning in this paper, we
recognize that the claim will have to be supported by valid empirical evidence and further
investigation.
9. The OCP on features differs from the OCP on segments in various ways. For instance, the
repair strategies are different. OCP on segments can be repaired by assimilation, dissim-
ilation, epenthesis, metathesis, and so on, as Yip (1988) points out. On the contrary, some
of these phonological processes cannot repair violations of the OCP on features. For ex-
ample, epenthesis cannot fix a violation of the OCP on features. Consider a case where a
segment C is epenthesized between two adjacent segments A and Β which both have an
identical feature F. After epenthesis, segments A and Β are not adjacent anymore, but the
feature F are still adjacent unless the epenthetic segment C bears the oppositte value of
F. Fukazawa (1999) follows the recent proposal in phonology in which all the features are
privative. When all features are privative, the segment C never bears the oppositte value
of F. Thus, epenthesis of a segment never rescues the OCP on features. This is why Fuka-
zawa (1999) claims that the typology of the OCP on features has to be treated differently
from that on segments. For a more detailed discussion, see Fukazawa (1999).
10. It is a matter of controversy whether multiplied faithfulness for Input-Output, Base-Re-

duplicant, Base-Truncated etc., are a genuine case of relativization for the case of
TETU. Relations such as IO, BR, BT may reside in all the grammars and thus may be
universal. But, we still clearly see the similarity between the ranking schemata of the
cases in table (31).
References
Alderete, John
1997 "Dissimilation as Local Conjunction," Proceedings of North East Linguistics
Society 27:17-32.
Beckman, Jill
1995 "Shona Height Harmony: Markedness and Positional Identity," In Jill Beck-
man, Laura Walsh Dickey, and Suzanne Urbanczyk (eds.), University of Mas-
sachusetts Occasional Papers: Papers in Optimality Theory 18: 53-76, GLSA,
Amherst, Mass.
Benua, Laura
1995 "Identity Effects in Morphological Truncation," In Jill Beckman, Laura Walsh
al Papers: Papers in Optimality Theory 18: 77-136. GLSA, Amherst, Mass.
Benua, Laura
1997 Transderivational Identity: Phonological Relations between Words. Doctoral
Dissertation. University of Mass. Amherst.
Fukazawa, Haruka
1998 "Multiple Input-Output Faithfulness Relations in Japanese." ms., University
of Maryland, College Park. [ROA-260-0598]
1999 Theoretical Implications of O C P Effects on Features in Optimality Theory.
Doctoral dissertation, University of Maryland, College Park.
Fukazawa, Haruka, Mafuyu Kitahara, and Mitsuhiko Ota
1998 "Lexical Stratification and Ranking Invariance in Constraint-Based Gram-
mars," in M. Catherine Gruber, Derrick Higgins, Kenneth S. Olson, and Tam-
ra Wysocki (eds.), CLS 34: The Panels, 47-62.
Fukazawa, Haruka, and Miglio Viola
1998 "Restricting Conjunction to Constraint Families," in Proceedings of Western
Conference on Linguistics (WECOL 96) vol 9:102-117.
Goldsmith, John A.
1976 Autosegmental Phonology. Doctoral dissertation, MIT, Cambridge.
Honma, Takeru
2000 "How should we Represent 'g' in toge in Japanese Underlyingly?" in this vol-
ume.
Itô, Junko, and Armin Mester
1986 "The phonology of voicing in Japanese: theoretical consequences for morpho-
logical accessibility," Linguistic Inquiry 17: 49-73.
1995 "Japanese Phonology," in John Goldsmth (ed.), The Handbook of Phonolog-
ical Theory: 817-838. Basil Blackwell, Cambridge, Mass.
1996 "Structural Economy and O C P Interactions in Local Domains," Paper pre-
sented at Western Conference on Linguistics ( W E C O L ) 1996, University of
California, Santa Cruz.
1998 "Markedness and Word Structure: OCP Effects in Japanese," ms., University
of California, Santa Cruz. [ROA-255-0498].
1999 "The Phonological Lexicon," In Natsuko Tsujimura (ed.), The Handbook of
Japanese Linguistics. Blackwell Publishers, Oxford.
Itô, Junko, Armin Mester, and Jaye Padgett
1995 "Licensing and Underspecification in Optimality Theory," Linguistic Inquiry
26: 571-613.
Kibe, Nobuko
1978 "Keetai akusentoronteki ichi koosatsu: fukugoogo akusento to gokoosee-
rendaku o megutte [A morphoaccentual analysis on compound accent, word
structure and sequential voicing]," Gobun Kenkyuu (Kyushu University) 46.
Kubozono, Haruo
1993 The Organization of Japanese Prosody, Tokyo: Kuroshio.
Leben, William
1973 Suprasegmental Phonology. Doctoral dissertation, MIT, Cambridge.
Lombardi, Linda
1995 "Why Place and Voice are different: Constraint Interactions and Featural
Faithfulness in Optimality Theory." ms., University of Maryland, College Park.
Lubowicz, Anna
1997 "Derived Environment Effects in OT." ms., University of Massachusetts, Am-
herst. [ROA-239-0198]
McCarthy, John
1986 "OCP effects: Gemination and Antigemination," Linguistic Inquiry 17: 207-
63.
1988 "Feature Geometry and Dependency: A Review," Phonetica 43: 84-108.
McCarthy, John, and Alan Prince
1994 "The Emergence of the Unmarked." ms., University of Massachusetts, Am-
herst, and Rutgers University.
1995 "Faithfulness and Reduplicative Identity," In Jill Beckman, Laura Walsh
al Papers: Papers in Optimality Theory 18: 249-384. GLSA, Amherst, Mass.
Otsu, Yukio
1980 "Some Aspects of Rendaku in Japanese and Related Problems," in Y. Otsu
and A. Farmer, eds. MIT Working Papers in Linguistics 2, Theoretical Issues
in Japanese Linguistics: 207-228, MIT Press, Cambridge, MA.
Prince, Alan, and Paul Smolensky
1993 "Optimality Theory: Constraint Interaction in Generative Grammar." ms.,
Rutgers University, New Brunswick, NJ; University of Colorado, Boulder.
Rice, Keren
1997 "Japanese NC Clusters and the Redundancy of Postnasal Voicing," Linguistic
Inquiry 28: 541-551.
Sato, Hirokazu
1989 "Fukugoogo ni okeru Akusento Kisoku to Rendaku Kisoku [Accent Rules
and Sequential Voicing Rules in Compounds]," in Miyoko Sugito ed., Kooza
Nihongo to Nihongo Kyooiku [Lectures on Japanese and Japanese Education]
vol. 2: 233-265.
Smolensky, Paul
1993 "Harmony, Markedness, and Phonological Activity." Paper presented at Rut-
gers Optimality Workshop-1, Rutgers University, October 23,1993.
1995 "On the Internal Structure of the Constraint Component Con of UG." Collo-
quium presented at U C L A , April 7,1995.
1997 "Constraint Interaction in Generative Grammar II: Local Conjunction." Pa-
per presented at the Hopkins Optimality Theory Workshop/University of
Maryland Mayfest 1997, May 8-12,1997.
Urbanczyk, Suzanne
1995 "Double Reduplications in Parallel," In Jill Beckman, Laura Walsh Dickey,
and Suzanne Urbanczyk (eds.), University of Massachusetts Occasional Pa-
pers: Papers in Optimality Theory 18: 499-531. GLSA, Amherst, Mass.
1996 Patterns of Reduplication in Lushootseed. Doctoral dissertation, University
of Massachusetts, Amherst.
Vance, Timothy J.
1996 "Sequential Voicing in Sino-Japanese." Journal of the Association of Teachers
of Japanese 30: 22-43.
Yip, Moira
1988 "The Obligatory Contour Principle and Phonological Rules: A Loss of Iden-
tity," Linguistic Inquiry 19: 65-100.
Epenthetic Vowels and Accent in Japanese:
Facts and Paradoxes
Haruo Kubozono
Kobe University
1. Introduction
Many languages exhibit a phonological phenomenon known as vowel

epenthesis in an attempt to conform to syllable structure constraints such
as N O C O D A , which prohibits coda consonants, and *COMPLEX, which pro-
hibits consonant clusters (Prince and Smolensky 1993). This phenomenon
is especially prevalent in loanword phonology, where foreign words are
adapted to fit the phonological structure of the borrowing language. The
loanword phonology of Japanese is no exception in this respect. Since
Japanese favors the basic CV structure as its canonical syllable form, it
has a number of loanwords—Sino-Japanese (SJ) words (i.e. loanwords
from Chinese) and foreign loanwords (i.e. loanwords from a language
other than Chinese)—which have undergone vowel epenthesis as part of
the borrowing process1. The historical process of vowel epenthesis itself
is interesting enough, but it is more interesting when considered in con-
junction with other phonological processes from a synchronic perspec-
tive. One such process is that of accent assignment, which we will discuss
in this paper.
A standard idea about the interaction between epenthetic vowels and
synchronic accent rules will be that words are subject to accent rules only
after epenthetic vowels have been inserted, namely, that epenthetic vow-
els are fully visible to synchronic accent rules and behave just like under-
lying, i.e. non-epenthetic, vowels. However, a careful analysis of a wide
range of accentual phenomena in modern Japanese suggests that at least
some epenthetic vowels are invisible to some accent rules under certain
circumstances, i.e. that some epenthetic vowels behave as if they did not
exist at all2. Phonological invisibility of epenthetic vowels is documented
in other languages and dialects, too 3 , but the case of modern Tokyo Japa-
nese seems much more complicated.
Equally interesting is the behaviour of consonants which have been
elided, i.e. deleted, in the course of the history. Many native words in Jap-
112 Haruo Kubozono
anese have undergone consonant elision which has turned CVCV into
CVV: e.g. kofi koi 'carp'. In terms of syllable structure, this process has
had the effect of producing a heavy (bimoraic) syllable out of a sequence
of two light (monomoraic) syllables. Here, again, a conventional theory
would predict that elided consonants should be no longer visible to pho-
nological rules in the contemporary grammar of the language, and so pho-
nological rules apply to the output, not to the input of the segmental
change. While this orthodox account seems to hold in most cases, some
accent rules in Japanese exhibit cases where CVV syllables derived from
CVCV sequences pattern with underlying CVCV sequences, i.e. elided
consonants sometimes behave as if they were still present in the syn-
chronic grammar.
The purpose of this paper is to describe and generalize the peculiar be-
haviour of epenthetic vowels and elided consonants in modern (Tokyo)
Japanese in relation to a variety of accentual phenomena including loan-
word and compound accentuation. In order to achieve this goal, we will
first briefly look at the history of epenthetic vowels and elided consonants
in Japanese (section 2). We will then consider how these segments behave
in accent rules said to be sensitive to the syllable structure of words (sec-
tion 3). The final section (section 4) summarizes the basic facts and para-
doxes behind the peculiar behaviour of epenthetic vowels and elided con-
sonants, and demonstrates the difficulties they might imply for theoretical
analyses in the future.
The syllable-sensitive accent rules to be discussed in this paper are list-
ed below together with a summary of the metrical (in)visibility of epen-
thetic vowels and elided consonants for each rule. All of these rules are
more or less productive in the synchronic grammar of Tokyo Japanese;
they readily apply to new and nonsense words as well as existing words.
(1) a. the loanword accent rule known as the 'antepenultimate rule' (section 3.1.)
- Some epenthetic vowels seem invisible, while others are visible.
b. accent rules triggering 'unaccentedness' in loanwords (section 3.2.)
- Some epenthetic vowels are invisible, while others are visible.
c. the ordinary noun-noun compound accent rule (section 3.3.)
- Epenthetic vowels in SJ morphemes are generally invisible, while those in
foreign morphemes are visible. Elided consonants in native Japanese
morphemes are no longer visible.
d. the compound accent rule for personal names involving taroo (section 3.4.)
- All epenthetic vowels are visible, but elided consonants are somehow also
visible.
e. the compound accent rule for personal names with the SJ morpheme -iti
(section 3.5.)
- All epenthetic vowels are visible, but elided consonants are somehow also
visible.
f. the accent rule for verbs and adjectives (section 3.6.)
- Elided consonants in native morphemes are no longer visible.
g. the so-called 'pre-no deaccenting rule' (section 3.7.)
- Epenthetic vowels in SJ morphemes are fully visible.
As can be grasped from the summary above, epenthetic vowels are gen-
erally visible and elided consonants are generally no longer visible to
most of the syllable-sensitive accent rules. However, there are many ex-
ceptions. For example, some epenthetic vowels in foreign loanwords are
invisible to the loanword accent rule and the accent rules which yield un-
accented loanwords, although other epenthetic vowels in the same type of
words are fully visible to the same accent rules. Moreover, epenthetic
vowels in SJ morphemes behave as if they were invisible to the ordinary
noun-noun compound accent rule, while they are fully visible to the other
accent rules. Furthermore, elided consonants in native Japanese mor-
phemes behave as if they were still present in the X-taroo and X-iti com-
pounds, but not in other types of words. These puzzling facts not only
undermine the simple and transparent distinction between excrescent
(phonology-invisible) and epenthetic (phonology-visible) vowels (Levin
1987), but also seem to challenge theoretical analyses, derivational and
non-derivational alike.
2. Vowel epenthesis and consonant elision
Japanese is a typical CV language in which codas and consonant clusters

are prohibited in principle. Old Japanese permitted only CV structures, i.e.
onsets were obligatory and codas were strictly prohibited. Although these
constraints operate less rigidly in modern Japanese, they have constrained
loanword phonology throughout the history of the language, often trigger-
ing vowel epenthesis in SJ and foreign loanwords. SJ morphemes were
originally monosyllabic (as are most morphemes in tone languages), and
became bisyllabic through vowel insertion in order to conform to the basic
CV structure of Japanese syllables. As can be seen from the examples in
(2a, b), the epenthetic vowels (placed in < >) in SJ morphemes are either
114 Haruo Kubozono
Iii or lui, the two high vowels which are prone to vowel devoicing in (mod-
ern) Japanese (dots indicate syllable boundaries) 4 .
(2) a. ek -* e.k 'railway station'

sek -»• se.k 'seat'
b. rak ra.k 'ease, comfort'
zyuk ->· zyu.k 'a cram school'
zyut ->• zyu.t 'art, technique'
ket ->· ke.t 'decision'
Foreign words have undergone a similar epenthesis process as illustrated

in (3). The epenthetic vowel in these words is lu/, loi, or liI, as exemplified
in (3a-c), respectively, of which the first represents the default vowel, loi
and HI are chosen in specific phonological contexts: loi is inserted after the
two dental stops [t] and [d], while Iii is epenthesized after the palato-alve-
olar affricates [tj] and [d3] and also after [k] in some words. It is not clear
why loi was inserted after [t] in foreign loanwords, but not in SJ words (cf.
(2b)).
(3) a. ba.s 'bus'

k.ri.s.ma.s 'Christmas'
in.k 'ink'
s.rii 'three'
b. bat.t<o> 'bat'
s.t<o>.re.s 'stress'
c. bii.tf 'beach'
kee.k 'cake'
Vowel epenthesis in foreign loanwords can actually be divided into two

kinds depending on the specific syllable structure constraint that is re-
sponsible for triggering it. In one kind, vowels are inserted to avoid a coda
consonant, just as in the SJ morphemes illustrated in (2): e.g. /ba.s/
'bus'. Vowel epenthesis in this case has the effect of converting the origi-
nal coda consonant into the onset of a newly-created syllable. In the other
kind, vowels are inserted to avoid a consonant cluster, a structure which
is strongly disfavored in Japanese: e.g. /s.rii/ 'three'. Both of these two
syllable structure constraints, namely, N O C O D A and *COMPLEX, force the
creation of CV syllables rather than other syllable structures.
Vowel epenthesis triggered by the NOCODA constraint illustrated in (2)

and (3) has the effect of converting a heavy closed syllable into a se-
quence of two light syllables. This effect is illustrated in (4).
(4) CVC CV.C<V>
While loanwords have undergone the process of vowel epenthesis in con-

formity to the basic CV structure of Japanese, some native words under-
went segment deletion processes which very often added CVC and other
syllable types to the inventory of syllable structure.
The process of onbin, which started as early as the Heian Period (8th-
12th century), for example, elided either a consonant or a vowel when two
morphemes were combined to form a compound or an inflected form of
a verb (or adjective). This segmental change had the effect of producing
a single heavy syllable by deleting a consonant or vowel. This process is
illustrated in (5) below: i-onbin and u-onbin yielded a CVV syllable out
of a sequence of two light syllables by deleting a consonant (very often
Ik/), whereas hatu-onbin and soku-onbin gave rise to a CVC syllable out
of the same type of syllable sequence.
(5) a. i-onbin
tu.ki + ta.ti -> tui.ta.ti 'moon, beginning; first day of the month'
b. u-onbin
ha.ya + ku ->• ha.yau (->• hayoo) 'early (continuous form)'
c. hatu-onbin
yo.mi + te -»• yom.te -> yon.de 'to read (continuous form)'
d. soku-onbin
mo.ti + te -*• mot.te 'to take (continuous form)'
Elision phenomena occurred in later periods, too, including those in (6).

Note that these operations as well as those in (5) did not affect the moraic
structure of the words involved. Eliding an onset consonant does not gen-
erally change the moraic length of a word, as illustrated in (6a). Eliding a
vowel, too, exerts no effect on the moraic structure of a word as long as it
occurs in a non-initial position. In this latter case, the onset consonant—
now without its nuclear vowel—becomes moraic, i.e. takes on monomo-
raic length by becoming a moraic nasal or a moraic obstruent, as shown
in (6b).
116 Haruo Kubozono
(6) a. ka.fi kai 'shellfish'

ta.fi -• tai 'sea bream'
ko.fi -> koi 'carp'
ha.fi hai 'ash'
to.fo.ru -* too.ru 'to pass by'
ko.fo.ri koo.ri 'ice'
b. wo.mi.na -*• on.aa 'woman'
wo.fi.to -> ot.to 'husband'
In terms of syllable structure, the segmental changes shown in (5) and (6)
had the effect of creating a heavy monosyllable (either CVV or CVC) out
of a sequence of two light syllables. This is schematically shown in (7).
(7) a. CV.CV ->• CVV

b. CV.CV -> CVC
3. Accent rules and epenthetic vowels
As is well known, Tokyo Japanese is a typical 'mora language' as opposed

to a 'syllable language' (Trubetzkoy 1969, Kubozono 1989). Accent rules
are no exception in this regard, and can be formulated by invoking the
mora as a basic descriptive unit. The application of many of these rules,
however, requires information on syllable structure as well as mora struc-
ture. In this section, we will focus on these syllable-sensitive accent rules
and examine the ways in which they interact with words involving an ep-
enthetic vowel or an elided consonant. We will especially be concerned
with those cases where epenthetic vowels are invisible to accent rules and
where elided consonants are, in contrast, visible.
3.1. Loanword accent rule
The first accent rule we will consider is the loanword accent rule conven-
tionally known as the 'antepenultimate rule'. This rule places an accent
on the third mora from the end of the word in loanwords. If this antepen-
ultimate mora is the second mora of a heavy syllable, then the accent au-
tomatically shifts one mora leftwards, i.e. onto the nucleus of the syllable.
That is, an accent is placed on the syllable containing the antepenultimate
mora (McCawley 1968)5.
As can be seen from the examples in (8), this accent rule applies to
structures in which epenthetic vowels have already been inserted. That is,
it counts all the moras in the surface structure, irrespective of whether
they contain an epenthetic vowel or not. In the examples below, the ac-
cent mark (') is placed immediately after the accented mora.
(8) a. p'.ra.s 'plus'

b. pu'.rin 'pudding'
c. s.t<o>'.re.s 'stress'
d. k.ri.s'.ma.s 'Christmas'
e. wa.si'n.ton 'Washington'
While this rule accounts for the accentuation of a majority of loanwords

in Japanese, it admits a certain number of exceptions in certain phonolog-
ical contexts. Let us consider trimoraic words here. Most trimoraic words,
including those in (9a, b), are accented on their antepenultimate mora as
predicted by the loanword accent rule. However, this rule admits a certain
number of exceptions which are accented on their penultimate mora. This
second pattern is exemplified in (9c)6.
(9) a. pu'.rin 'pudding', u'.ran 'Uranium', se'.dan 'sedan', i'.ran 'Iran',

to'.naa 'toner', sa'.tan 'Satan', ki'.raa 'killer', ba'.ree 'volleyball',
ri'.ree 'relay', ha'.wai 'Hawaii'
b. p'.ra.s 'plus', k'.ra.b 'club', g'.ra.s 'glass',
t<o>'.ri.o 'trio', d<o>'.re.s 'dress'
c. p.re'e 'play', b.ru'u 'blue', s.ri'i 'three',
g c u x r e ' e 'gray', d<o>.ra'i 'dry', t.i'n 'twin'
A careful analysis reveals that the words showing the exceptional accent
pattern in (9c) almost invariably have the two features in (10) in common
(Kubozono 1996,1999b, Kubozono and Ohta 1997)7·8.
(10) a. They consist of two syllables, a light (L) syllable followed by a

heavy (H) syllable,
b. They involve an epenthetic vowel in their initial light syllable.
Note that the accent pattern in (9c) arises when the two conditions in (10)
are both met: it surfaces neither in words which have the same syllable
structure but not the same segmental structure, as illustrated in (9a), nor
in words which satisfy the segmental condition but not the condition on
118 Haruo Kubozono
syllable structure, as illustrated in (9b)9. The accent pattern in (9a) corre-

sponds to the trochaic stress pattern characteristic of Latin words of the
same syllable structure, e.g. amo /a'.moo/ Ί love'. On the other hand,
words showing the accent pattern in (9c) are equivalent to the police-type
bisyllabic words in English, where main stress is placed on the final heavy
syllable in avoidance of the preceding light syllable (Prince and Smolen-
sky 1993, Hayes 1995).
The analysis described above has been statistically borne out in Tana-
ka's (1996) examination of the accent patterns of all the loanwords listed
in the NHK Accent Dictionary (1985). According to this study, there are
424 trimoraic loanwords showing either a Light-Heavy bisyllabic struc-
ture (146 words) or a Light-Light-Light trisyllabic structure (278 words).
Of these 424 words, only 35 instances (8%) exhibit the irregular accent
pattern in (9c): LH bisyllables show this pattern in 23 instances (16%),
whereas LLL trisyllables allow the same irregular pattern in only 12 in-
stances (4%). However, if we confine our analysis to those words satisfy-
ing both the syllable and the segmental structure conditions mentioned in
(10) above, we see a totally different tendency. There are 22 instances sat-
isfying the two conditions in (10), of which 19 instances (86%) show the
exceptional accent pattern in question. In contrast, those words which
satisfy only the syllable structure condition in (10a) show the irregular ac-
cent pattern in only four out of 124 instances (3%). These statistics are
summarized in Table 1: /LLL/ means a sequence of three light syllables
(with or without an epenthetic vowel), whereas /1H/ and /LH/ represent,
respectively, light-heavy bisyllables with and without an epenthetic vowel
in the initial light syllable.
Table 1. Percentage of the accent pattern in (9c) shown by each syllable structure
LH/1H LH IH LLL Total

23/146 4/124 19/22 12/278 35/424
(16%) (3%) (86%) (4%) (8%)
In sum, Tanaka's statistical study confirms the validity of the two condi-
tions in (10). This allows us to conclude that as far as trimoraic loanwords
are concerned, word-final heavy syllables are metrically stronger than
penultimate light syllables if the light syllables contain an epenthetic vow-
el; in contrast, they are metrically weaker than non-final light syllables
containing a non-epenthetic vowel. This reveals an interesting interaction
between syllable structure and segmental structure. Although the precise
formulation of this interaction is beyond the scope of this paper, the case
in question clearly demonstrates that the nature of epenthetic vowels
must be taken into consideration when generalizing the two accent pat-
terns shown by the trimoraic loanwords in (9), i.e. (9a, b) vs. (9c).
This result is especially interesting if compared with the analysis men-
tioned at the beginning of this section. As shown in (8), the antepenultimate
accent rule for loanwords does not rely on the distinction between epen-
thetic and non-epenthetic vowels; the accent pattern of p'.ra.s
'plus', for example, indicates that epenthetic vowels are fully visible to the
accent rule. Indeed, the very basic idea of 'antepenultimacy' is based on the
assumption that epenthetic vowels are fully visible to the rule just like non-
epenthetic vowels. However, the data given in (9) and Table 1 clearly show
that the same loanword accent rule does distinguish between the two kinds
of vowels under certain circumstances: word-initial epenthetic vowels be-
have as if they did not exist in a particular type of trimoraic words. To be
more precise, moras containing an epenthetic vowel are generally counted
by the mora-counting accent rule, but they somehow avoid bearing an ac-
cent in and only in a certain type of trimoraic words.
3.2. Unaccented loanwords
The situation is further complicated and our dilemma is further deepened

if we look at the factors responsible for unaccented words. Unaccented or
accentless words are those which do not involve an abrupt pitch drop any-
where in the phonetic output. Japanese has many words of this type. Ac-
cording to statistics previously reported, a majority of the vocabulary list-
ed in ordinary Japanese dictionaries belong to this accent type (Sibata
1994). This accent type is especially popular with four-mora words, ac-
counting for 66 % of native Japanese and S J words of this phonological
length: e.g. hi.ro.si.ma 'Hiroshima', yo.ko.ha.ma 'Yokohama', too.kyoo
'Tokyo', oo.sa.ka 'Osaka'. Apart from this factor, the distinction between
accented and unaccented words is largely lexical, i.e. it cannot be readily
predicted by the linguistic structures of indivisual words.
While unaccented words are very common in Japanese as a whole, they
are not very common in loanwords. Sibata (1994) estimates that they only
account for ten percent of loanwords in modern Tokyo Japanese, more
than five times less than the general average. This low percentage, how-
ever, contrasts with an extremely high percentage with which certain
types of loanwords exhibit the accent type in question. This suggests that
120 Haruo Kubozono
unaccentedness in loanwords is created by certain linguistic structures.

We will look here at two such structures, both of which involve an epen-
thetic vowel in one way or another.
3.2.1. #HLL# and #LLLL# four-mora words
The first type that shows a marked tendency for unaccentedness involves
four-mora words which satisfy the two conditions in (11) (Kubozono 1996).
(11) a. The word ends with a sequence of two light syllables,

b. The vowel in the final syllable is NOT an epenthetic lui.
The validity of the generalization in (11) has been confirmed by three inde-
pendent statistical studies: Kubozono (1996) examined the accent patterns
of all of the 206 foreign placenames listed in the appendix to the NHK Ac-
cent Dictionary (1985); Tanaka (1996) looked at all the loanwords and for-
eign placenames in the same dictionary, of which 931 words have four mo-
ras; and Fukui (1998) examined all the foreign loanwords in the new edition
of the NHK Dictionary (1998) in which she found a total of 792 four-mora
loanwords. All these studies confirmed that the distinction between accent-
ed and unaccented words is heavily dependent on the syllable structure in-
volved. To be specific, unaccentedness is dominant in four-mora words sat-
isfying the condition in (11a), but not in four-mora words with different
syllable structures: e.g. a.ri.zo.na 'Arizona', ai.o.wa 'Iowa' vs. a'.ma.zon
'Amazon', ro'n.don 'London'. Table 2 summarizes the statistics reported by
these studies: it shows the percentage of unaccented words as opposed to
accented words exhibited by each of the syllable structures.
Table 2. Correlation between unaccentedness and syllable structure in four-mora loanwords
Syllable type HLL LHL LLH HH Total

LLLL
Study
Kubozono (1996) 60% 5% 12% 0% 34% (= 71/206)
Tanaka (1996) 45% 9% 15% 2% 22% (= 201/931)
Fukui (1998) 50% 25% 19% 8% 29% (= 233/792)
The percentage of unaccented words gets even higher if the factor in

( l i b ) is further taken into consideration. According to Kubozono (1996),
the percentage of unaccented words goes up to 89%. Tanaka (1996) and
Fukui (1998) conclude that 82% and 83 %, respectively, of the words sat-
isfying the two conditions in (11) are unaccented. These figures are no
doubt extremely high in light of the fact that the same accent pattern ac-
counts for only ten percent of loanwords as a whole.
Stated conversely, four-mora words ending in an epenthetic vowel /u/
are much less likely to become unaccented even if they fulfill the syllable
structure condition in (11a). According to Fukui's data, unaccentedness
only accounts for 30 % of #HLL# and #LLLL# four-mora loanwords end-
ing in this particular type of vowel, which is considerably lower than the
ratio given in Table 2 for #HLL# and #LLLL# four-mora loanwords in
general. Instead of becoming unaccented, these words that involve an ep-
enthetic vowel word-finally tend to be accented according to the tradi-
tional antepenultimate rule. Some examples are given in (12).
(12) a'n.de.s 'Andes', we'e.r.z 'Wales',

ka'p.p.r 'couple', si'n.ba.r 'cymbals';
p.ro'.se.s 'process', s.t<o>'.re.s 'stress',
ba.ri'.u.m 'Barium'
Four-mora loanwords ending in an epenthetic vowel /u/ thus behave as if

they ended in a heavy syllable rather than a sequence of light syllables.
Namely, the word-final epenthetic vowel /u/ behaves as if it were invisible
to the rule that triggers unaccentedness.
Note that this peculiar behaviour is shown only by an epenthetic /u/,
and not by other types of vowels. For example, four-mora loanwords end-
ing in a non-epenthetic /u/ exhibit the unaccented pattern, and not the ac-
cent pattern in (12): e.g. ma.ka.ro.ni 'macaroni', ri.ha.bi.ri 'rehabilitation',
ho.no.ru.ru 'Honolulu'. Likewise, the other high vowel HI generally gives
rise to the unaccented pattern whether it is underlying (e.g. a.r.ka.ri
'alkali', e.s.p.ri 'esprit') or epenthetic (e.g. bi.h.te.k 'beef
steak', bo.r.s.t 'borshch'). These observations preclude the
possibility that the accent pattern in (12) is due to the devoicing or low so-
nority of the word-final vowel. Moreover, words ending in an epenthetic
loi pattern with those ending in HI or other vowels, and not with those
ending in an epenthetic lui: e.g. faa.s.t<o> 'first', o.ka.r.t<o> 'oc-
cult'. Fukui's data actually reveal that the unaccented pattern accounts
for 50 % of #HLL# and #LLLL# four-mora loanwords ending in an epen-
thetic loi, which is just the average of #HLL# and #LLLL# four-mora
loanwords in general (see Table 2).
These statistical data corroborate the two conditions given in (11) for un-
accented loanwords and, hence, the idea that the word-final epenthetic lui
122 Haruo Kubozono
among many kinds of vowels is largely invisible to the accent rule responsi-
ble for the unaccented pattern. This fact is puzzling in two ways. First, it is
strange to find that epenthetic /u/ is invisible to the accent rule for unaccent-
ed words, whereas epenthetic HI and loi are visible to the same accent rule.
A second paradox is that while epenthetic /u/ is invisible to this rule, the
same vowel seems visible to the ordinary antepenultimate accent rule. In-
deed, the accent pattern in (12) can only be accounted for by counting all
the moras in the surface structure, irrespective of whether they contain an
epenthetic vowel or not: e.g. s.t<o>'.re.s 'stress', ba.ri'.u.m
'Barium'. One solution to this second paradox may be to assume that the
consonant immediately preceding the word-final epenthetic /u/ bears a
mora on its own while the vowel itself is phonologically invisible. Even this
solution, however, fails to explain why this peculiar behaviour is shown by
epenthetic lui, and not by the other types of epenthetic vowels.
3.2.2. #LHL# four-mora words
A second source of words showing a marked tendency for unaccented-

ness are four-mora words which satisfy the two conditions in (13) (Kubo-
zono 1996).
(13) a. The word involves a heavy syllable flanked by light syllables,

i.e. LHL.
b. The final syllable contains an epenthetic vowel loi.
The condition in (13a) alone does not trigger the accent type in question.
According to Fukui (1998), only one-fourth of all LHL trisyllables are un-
accented, which is about the average of all four-mora loanwords (see Ta-
ble 2). If the second condition in (13) is added to this syllable structure
condition, however, the ratio of unaccentedness more than doubles. In
Fukui's data, 41 out of 68 words (i.e. 60%) satisfying the two conditions
in (13) are unaccented. Typical examples are given in (14a), which con-
trast with the exceptions given in (14b) 10 .
(14) a. b.ran.d<o> 'brand', f.rai.t<o> 'flight', p.ree.t<o> 'plate',

g.ree.d<o> 'grade', s.pee.d<o> 'spade'
b. pa.re'e.d<o> 'parade', po.ke't.t<o> 'pocket', e.ri'i.t<o> 'elite'
c. s.to'o.b 'stove', g.ro'o.b 'glove'
d. da.me'e.z 'damage', i.me'e.z 'image', a.re'n.z 'arrange',
s.te'e.k 'steak'
e. b.ra'n.ko 'balanco, a swing', h.ra'n.ko 'Franco'

f. sa.ba'n.na 'savanna', s.ri'p.pa 'slipper'
Interestingly, epenthetic word-final vowels other than loi do not trigger

this tendency even if the word fulfills the syllable structure condition in
(13a): Fukui reports 114 instances of lui insertion and eleven instances of
HI insertion word-finally, of which only eleven instances (10%) and two
instances (12%), respectively, are unaccented. These two types of vowel
epenthesis are exemplified in (14c, d). Moreover, non-epenthetic word-
final vowels do not generally trigger unaccentedness, as illustrated in
(14e, f): there are 28 instances in Fukui's data which involve a non-epen-
thetic vowel word-finally, but only two instances (7%) turned out to be
unaccented. Words failing to fulfill the two conditions in (13) are gener-
ally accented on their antepenultimate mora according to the antepenul-
timate rule for loanwords.
In sum, there are two types of four-mora words which display a marked
tendency towards unaccentedness. These two types can both be defined
on the basis of their syllable structure and the nature of the word-final
vowel. As for the second factor, word-final epenthetic lui prevents the
word from becoming unaccented, as illustrated in (12), whereas the epen-
thetic lo/ in the same position prompts the word to be unaccented, as il-
lustrated in (14a). It is intriguing to note that the two epenthetic vowels
thus show entirely opposite tendencies with respect to unaccentedness. A
more revealing point about the cases we have hitherto seen is that the dis-
tinction between epenthetic and non-epenthetic vowels plays a pivotal
role in the distribution of unaccented versus accented words. This is par-
ticularly interesting and, at the same time, puzzling if compared with the
fact we saw in (8), namely, that in general epenthetic vowels behave just
like non-epenthetic vowels under the antepenultimate accent rule.
3.3. Compound accent
The next accent rule to consider is the compound accent rule which is sen-
sitive to syllable structure as well as mora structure (Kubozono 1995,
1997). We will confine ourselves here to noun-noun compounds with a
'short' second member (N2), i.e. those compounds whose first member
(Nl) is more than two moras long and their N2 is only one or two moras
long. Generally speaking, the accentuation of compound nouns in Japa-
nese is determined by the phonological length and accent of N2, irrespec-
124 Haruo Kubozono
tive of the accent of NI (which is merely deleted in compounds). The type

of compound nouns in question shows the accentual regularities stated in
(15): see Kubozono (1995,1997) for a foot-based analysis.
(15) a. NONFINALITY: N 2 retains its accent as the compound accent unless

it is accented on the very final syllable,
b. Emergence of the Unmarked: If N2 is accented on the final sylla-
ble or unaccented, then a new compound accent emerges on the fi-
nal syllable of N l .
This rule is exemplified in (16).
(16) a. pe'.ru.sha + ne'.ko ->· pe.ru.sya.ne'.ko 'Persia, cat; Persian Cat'

ga.ra.su + ma'.do ->• ga.ra.su.ma'.do 'glass, window; glazed window'
b. ni'.wa.ka + yu.ki' -> ni.wa.ka'.yu.ki 'sudden, rain; sudden rain'
ka.na'.ga.wa + ke'n ->• ka.na.ga.wa'.ken
'Kanagawa, prefecture; Kanagawa Prefecture'
ne'.ku.tai + pi'n -»· ne.ku.ta'i.pin 'necktie, pin; necktie pin'
kyo'o.to + si' -»• kyoo.to'.si 'Kyoto, city; Kyoto City'
ka'.bu.to + mu.si ->· ka.bu.to'.mu.si 'helmet, bug; a beetle'
What concerns us here is the subrule in (15a), which embodies an inter-

esting interaction between two constraints: NONFINALITY (head syllable),
which prohibits a head (i.e. accented) syllable from appearing in word-fi-
nal position, and MAx-accent, which requires that the accent of the N2 in
the input be retained in the output. The regularity in (15a) is obtained by
outranking NONFINALITY (head syllable) over M A x-accent. On the other
hand, the default accent pattern described in (16b) is obtained by ranking
NONFINALITY (head foot), a constraint prohibiting the accented (bimoraic)
foot from appearing word-finally, below M A x-accent but above E D G E -
MOSTNESS. The overall constraint ranking is as follows (Kubozono 1995,
1997).
(17) Nonfin (head syllable) > Max-accent o Nonfin (head foot) > Edgemostness
The subrule in (15a) is subject to two groups of lexical exceptions; one is

idiosyncratic and the other relates to the lexical strata. Idiosyncratic ex-
ceptions consist of a subset of native Japanese morphemes like hi'me
'princess', which invariably fail to keep their non-final accent in the com-
pound: sirayuki'hime 'white snow, princess; Snow White', kaguya'hime
'Princess Kaguya'. It is difficult to explain why these morphemes show

such exceptional behaviour, while most other native Japanese mor-
phemes, including those in (16a), readily follow the rule in (15a).
The second group of exceptions consists of Sino-Japanese morphemes.
Unlike words belonging to other lexical strata, SJ morphemes almost in-
variably fail to keep their non-final accent in compound nouns. As exem-
plified in (18), these morphemes consequently take the default com-
pound accent illustrated in (16b).
(18) yo.ya.k + se'.k ->• yo.ya.k'.se.k

'reservation, seat; reserved seat'
sai.min + zyu'.t >- sai.mi'n.zyu.t 'hypnosis, art; hypnotism'
This peculiar behaviour of SJ morphemes contrasts with the behaviour of

foreign morphemes illustrated in (19) below as well as that of native Jap-
anese morphemes illustrated in (16a) above.
(19) mai.k.ro + ba'.s -> mai.k.ro.ba'.s

'micro, bus; minibus'
faa.s.to + ki'.s -> faa.s.t<o>.ki's
'first, kiss; first kiss'
mi'n.to + ga'.m ->· min.to.ga'.m
'mint, chewing gum; mint chewing gum'
It is interesting that words of foreign and SJ origins display contrastive pat-

terns although they involve very similar segmental structures. That is, for-
eign bisyllables such as bu'.s in (19) generally exhibit the regular pat-
tern described in (16a) despite the epenthetic vowel in their second syllable.
These foreign loanwords consequently pattern with bisyllabic loanwords
without an epenthetic vowel, e.g. pi'.za 'pizza', mik.k.s.pi'.za
'mixed pizza'. This pattern is crucially different from the pattern shown by
monosyllabic loanwords such as pi'n 'pin' in (16b), which do not generally
retain their accent in the compound. In contrast, SJ bisyllables behave as if
they were monosyllabic although they contain an epenthetic vowel in the
same final position. Thus, se'k and zyu't in (18) pattern with ke'n
'prefecture' and si' 'city' in (16b).
The accentual contrast between foreign and SJ bisyllables is puzzling in
two ways. First, epenthetic vowels in foreign loanwords are readily visible
to the compound accent rule, whereas those of SJ words appear to be in-
visible to the same rule. Second, epenthetic vowels in SJ words have a
126 Haruo Kubozono
much longer history in Japanese than those of foreign loanwords. Namely,

SJ morphemes underwent vowel epenthesis as described in (2) a milleni-
um ago, while foreign loanwords underwent the same phonological pro-
cess only a century or so ago. It is extremely peculiar to find that SJ mor-
phemes such as se'k and zyu't in (18) behave as if they were
monosyllabic despite the fact that they became bisyllabic via vowel epen-
thesis much earlier than foreign morphemes 11 .
The paradox can be solved, at least in part, in an optimality-theoretic
analysis. Invoking the notion of '(minimal) constraint reranking', it can be
assumed that SJ morphemes (or compound nouns containing SJ mor-
phemes as an N2 component) are subject to the constraint ranking in (20),
where NONFINALITY (head foot) as well as NONFINALITY (head syllable)
outranks MAx-accent.
(20) Nonfinality (head syllable, head foot) > MAX-accent > Edgemostness
According to this new hierarchy, N2 cannot retain its non-final accent in

the final bimoraic foot, which means that bisyllabic SJ morphemes such
as se'.k and zyu'.t in (18) cannot preserve their accent even if the
epenthetic vowels are phonologically visible. By positing two constraint
rankings in Japanese phonology, i.e. (17) and (20), it is now possible to as-
sume that epenthetic vowels are fully visible in both SJ and foreign mor-
phemes and, hence, to discard the apparent difference between the two
types of morphemes with respect to the (in)visibility of epenthetic vowels.
This is actually the analysis suggested by Kubozono (1997: 284) and pur-
sued further by Tanaka and Yamane (1999).
While the idea of (minimal) constraint reranking thus enables us to ac-
count for the apparent phonological invisibility of epenthetic vowels in SJ
morphemes, we can go one step further to ask why then SJ and foreign
morphemes are governed by different rankings (grammars) and, more-
over, why SJ morphemes obey the specific ranking in (20), and not the
one in (17). These important questions remain to be solved.
Before concluding our discussion of compound accent, let us briefly
consider the accent pattern shown by native Japanese morphemes that
became monosyllabic via the consonant elision described in (6a). These
monosyllabic words behave just like other monosyllabic words, i.e. they
abandon their lexical accent in conformity to the nonfinality constraint
mentioned in (15a). Some examples are given in (21). This suggests that
elided consonants in native morphemes are no longer visible to the com-
pound accent rule.
(21) a. sa.ku.ra + ka'i -*• sa.ku.ra'.gai 'cherry, shellfish; cherry shell'

b. ku.ro'on + ta'i ->· ku.ro.o'n.tai 'clone, sea beam; clone sea bream'
c. ni'.si.ki + ko'i --> ni.si.ki'.goi 'brocade, carp; coloured carp'
3.4. X-taroo compounds
Let us next consider a special compound accent rule applying to personal

(boys') names whose N2 is ta'roo, e.g. Kintaroo and Momo'taroo. As dem-
onstrated in Kubozono (1999a, b), this compound accent rule, too, is sen-
sitive to both syllable and mora structures. Compound nouns containing
taroo, or X-taroo compounds as I will call them for short, actually exhibit
three accent patterns depending on the phonological length of N1 (the ac-
cent of N1 being irrelevant). First, monosyllabic Nl, whether monomora-
ic or bimoraic, yields an unaccented compound, as illustrated in (22a).
Second, a compound accent emerges on the final syllable of Nl if this el-
ement is bisyllabic AND bimoraic 12 · 13 . This type is exemplified in (22b).
Finally, if Nl is three moras long or longer, the accent of N2 survives as
the compound accent, as illustrated in (22c). Note that this third accent
pattern represents the general rule for ordinary noun-noun compounds
with an initially-accented trimoraic N2: e.g. bi'iti + ba'ree bii.ti.ba'.ree
'beach volleyball'. The fact that ordinary noun-noun compounds do not
exhibit the accent patterns in (22a) and (22b) suggests that these two ac-
cent patterns are characteristic of compound names involving ta'roo.
(22) a. ki + ta'.roo ->• ki.ta.roo 'ogre, taroo; Kitaroo'

ko + ta'roo -»• ko.ta.roo 'little, taroo; Kotaro'
ki'n + ta'.roo -»• kin.ta.roo 'gold, taroo; Kintaroo'
ra'n + ta'.roo ->· ran.ta.roo 'riot, taroo; Rantaroo'
kyu'u + ta'roo kyuu.ta.roo 'Q, taroo; Kyuutaroo'
b. mo.mo + ta'.roo ->· mo.mo'.ta.roo 'peach, taroo; Momotaroo'
a.ka' + ta'.roo a.ka'.ta.roo 'dirt, taroo; Akataroo'
o.ni' + ta'.roo o.ni'.ta.roo 'ogre, taroo; Onitaroo'
c. tikara 1 + ta'roo ->· ti.ka.ra.ta'.roo 'power, taroo; Tikarataroo'
ba'.ree + ta'roo ->• ba.ree.ta'roo 'volleyball, taroo; Bareetaroo'
What is relevant to our discussion here is the syllable-based distinction

between (22a) and (22b): Bimoraic monosyllables such as ran and kyu'u
pattern with monomoraic elements like ki and ko, and not with bimoraic
bisyllables such as mo.mo and a.ka'. So this compound accent rule pro-
128 Haruo Kubozono
vides us with some reliable grounds on which to examine the syllable

structure of the bimoraic words in (2), (3) and (6a), and hence the visibil-
ity of epenthetic vowels and elided consonants.
First of all, bisyllabic SJ morphemes as given in (2) exhibit the accent
pattern in (22b). The compound accent is indeed placed on the epenthetic
vowel, as shown in (23)14. This means that the S J morphemes behave as
bisyllabic words and, hence, that their epenthetic vowels are readily visi-
ble to the accent rule in question.
(23) ra.k' + ta'.roo -»• ra.k'.ta.roo 'comfort, taroo; Rakutaroo'

e'.k + ta'.roo -• e.k'.ta.roo 'station, taroo; Ekitaroo'
zyu.t' + ta'.roo ->· zyu.t'.ta.roo 'art, taroo; Zyututaroo'
The same result is obtained for words of foreign origin. Thus, bisyllabic,
bimoraic words like ba'.s 'bus' and pi'.r 'pill' pattern with mo.mo
and a.ka' in (22b), with a compound accent on the epenthetic vowel. This
is illustrated in (24).
(24) ba'.s + ta'.roo ba.s'.ta.roo 'bus, taroo; Basutaroo'

pi'.r + ta'.roo ->• pi.r'.ta.roo 'pill, taroo; Pirutaroo'
ta'.b + ta'.roo ta.b'.ta.roo 'tub, taroo; Tabutaroo'
While epenthetic vowels seem fully visible to the X-taroo accent rule,
elided consonants display somewhat peculiar behaviour. Bimoraic words
such as those in (6a) exhibit the compound accent pattern in (22b) and
not the pattern in (22a). This is illustrated in (25)15. In other words, these
words behave as if they were bisyllabic although they have been turned
into monosyllables by the historical process of consonant elision.
(25) ko'i + ta'.roo koi'.ta.roo 'carp, taroo; Koitaroo'

ta'i + ta'.roo tai'.ta.roo 'sea bream, taroo; Taitaroo'
ka'i + ta'.roo ->• kai'.ta.roo 'shellfish, taroo; Kaitaroo'
For many speakers of Tokyo Japanese, this accent pattern contrasts with
the unaccented pattern shown by SJ morphemes which are originally
monosyllabic and bimoraic. This latter pattern is shown in (26).
(26) ka'i + ta'.roo kai.ta.roo 'sea, taroo; Kaitaroo'

ta'i + ta'.roo ->· tai.ta.roo 'team, taroo; Taitaroo'
Recall that the native Japanese bimoraic words in (25), i.e. koi, tai and kai,
behave as monosyllabic words in the general compound accent rule de-
scribed in (21), where they form the second member of noun-noun com-
pounds. It is somewhat paradoxical, therefore, that the same morphemes
exhibit the accentual behaviour characteristic of bisyllabic words in the
compound name, X-taroo.
Finally, the compound accent patterns sketched in (22) provide an in-
teresting insight into the nature of the epenthetic vowels in the three-
mora loanwords in (9c). As discussed in section 3.1., trimoraic loanwords
are exceptionally accented on the penultimate mora if this mora is in a
heavy syllable and is preceded by an epenthetic vowel. This exceptional
accent pattern suggests that the epenthetic vowel is not fully visible to the
accent rule or, at least, that the epenthetic vowel is metrically lighter than
non-epenthetic short vowels. This account does not hold, however, when
the same trimoraic words form the initial member of X-taroo compounds.
As shown in (27), these words exhibit the accent pattern in (22c) when
combined with ta'roo to form a X-taroo compound. Thus, as far as the ac-
centuation of X-taroo compounds is concerned, trimoraic loanwords such
b.ru'u 'blue' and d<o>.ra'i 'dry' pattern with ba'.ree 'volleyball',
which does not contain an epenthetic vowel.
(27) b.ru'u + ta'roo b.ruu.ta'roo 'blue, taroo; Buruutaroo'

d<o>.ra'i + ta'roo ->• dcoxrai.ta'roo 'dry, taroo; Doraitaroo'
t.i'n + ta'roo ->• t.in.ta'roo 'twin, taroo; Tuintaroo'
This puts us into another dilemma, a dilemma whereby epenthetic vowels

in the same trimoraic loanwords are fully visible to the X-taroo com-
pound rule, but not to the ordinary loanword accent rule discussed in sec-
tion 3.1.
3.5. X-iti compound
Another interesting source of data is the accentuation of boys' names

which end in the SJ morpheme iti' (pronounced as ichi), meaning 'one'.
Names with this morpheme display the two accent patterns illustrated in
(28), depending on the phonological length of the element preceding iti'
(Tanaka and Kubozono 1999). If the element is monosyllabic and bimo-
raic, the whole expression is pronounced with an unaccented pattern as
shown in (28a). Otherwise, a compound accent of some sort emerges on
130 Haruo Kubozono
the final syllable of the initial element. This second accent pattern is illus-
trated in (28b).
(28) a. koo + i.ti' ->· koo.i.ti 'happiness, iti; Koiti'

si'n + i.ti' -> sin.i.ti 'truth, iti; Shiniti'
ka'i + i.ti' -»· kai.i.ti 'sea, iti; Kaiiti'
b. ta + i.ti' -• ta'.i.ti 'fat, iti; Taiti'
hi'.ko + i.ti' -• hi.ko'.i.ti 'hiko, iti; Hikoiti'
An interesting aspect of this accentuation is that bisyllabic, bimoraic

words such as hi. ko pattern with monosyllabic monomoraic words such as
ta. Putting this aspect aside, the same accent rule offers an interesting
analysis of bisyllabic words with an epenthetic vowel and monosyllabic
words derived from bisyllables. First, the originally monosyllabic SJ mor-
phemes in (2) and foreign loanwords in (3) both display the accent pat-
tern in (28b) rather than the pattern in (28a). Indeed, they are accented
on the epenthetic vowel in X-iti names, as exemplified in (29).
(29) a. SJ morphemes
ra.k' + i.ti' -> ra.k'.i.ti 'comfort, iti; Rakuiti'
e'.k + i.ti' ->• e.k'.i.ti 'station, iti; Ekiiti'
b. Foreign loanwords
ba'.s + i.ti' ->· ba.s'.i.ti 'bus, iti; Basuiti'
pi'.r + i.ti' ->• pi.r'.i.ti 'pill, iti; Piruiti'
The accent pattern illustrated in (29) suggests that the bimoraic words in
question behave like a bisyllabic word (e.g. hi.ko in (28b)), rather than a
monosyllabic word (e.g. koo in (28a)). This, in turn, suggests that the ep-
enthetic vowels in question are visible to the accent rule.
While the accent rule under consideration applies to the output of the
vowel epenthesis rule, the same rule seems to apply to the input of the
consonant elision rule to which native Japanese morphemes were subject.
Monosyllabic, bimoraic morphemes of native origin typically exhibit the
accent pattern in (28b), as illustrated in (30).
(30) ko'i + i.ti'

ta'i + i.ti'
ka'i + i.ti'
Epenthetic Vowels and Accent in Japanese: Facts and. Paradoxes 131
For some speakers of Tokyo Japanese, the unaccented pattern is as ac-

ceptable for these names as the accent pattern in (30): e.g. koi.i.ti, tai.i.ti,
kai.i.ti. However, the accented pattern in (30) is not acceptable for bimo-
raic words which are underlyingly monosyllabic, e.g. kai.i.ti in (28a). The
gap between these underlyingly monosyllabic words and the monosyllab-
ic words derived from bisyllables seems to suggest that the consonants
historically elided some hundred years ago in native Japanese mor-
phemes are somehow visible to the contemporary synchronic accent rule
ofX-iti.
3.6. Accent of verbs and adjectives
The next accent rule we will look at is that of verbs and adjectives. As far
as end forms ('syuusikei') are concerned, verbs and adjectives in Tokyo
Japanese fall into two accentual groups: those accented on their penulti-
mate mora and those which are unaccented. These are exemplified in
(31a) and (31b), respectively. The accentedness of verbs and adjectives is
largely unpredictable and, hence, must be specified in the lexicon. Note
also that accented verbs and adjectives are accented one mora closer to
the end of the word than nouns, which we saw in section 3.1. above.
(31) a. mi'.ru 'to look', ha.na'.su 'to speak', kan.ga.e.'ru 'to think'
u.ma'i 'tasty', a.tu'i 'hot', mi.zi.ka'i 'short'
b. to.bu 'to fly, to jump', wa.ta.ru 'to cross', ku.ra.be.ru 'to compare'
a.mai 'sweet', a.tui 'thick', tu.me.tai 'cold'
While most verbs and adjectives follow either of the two accent patterns
in (31), some words are accented on their antepenultimate mora as exem-
plified in (32).
(32) to'o.ru 'to pass'

o'o.i 'many'
This exceptional accent pattern can be explained in a straightforward

manner if we allow for the fact that all these words underwent onset eli-
sion in their penultimate mora: namely, too.ru and oo.i are historically de-
rived from to.fo.ru and o.fo.i, respectively. Accent assigned to the penulti-
mate mora simply moved to the nucleus of the syllable containing this
mora: i.e. to.fo'.ru ->• too'.ru to'o.ru. Seen in this light, the accent pattern
132 Haruo Kubozono
in (32) is anything but an exception to (31a), but rather a direct conse-

quence of the historical process of consonant elision. This indicates that
the accentuation of verbs and adjectives applies to the output of conso-
nant elision, and cannot see the onset consonant that was elided in the
course of the history. This may not be very surprising by itself, but it con-
trasts very sharply with the observation in the preceding two subsections,
namely, the observation that the elided consonants in other native mor-
phemes are somehow visible to the accent rules of X-taroo and X-iti.
3.7. Pre-no deaccenting rule
Finally, let us consider the rule called 'pre-no deaccenting rule' (Akinaga
1985, Poser 1984, Kubozono 1999b). This accent rule has the effect of de-
leting an accent from the very final syllable of nouns. A s exemplified in
(33), it applies only if the noun is followed by the genitive particle no
(hence the term 'pre-no deaccenting rule'). (ga is the nominative particle).
(33) a. ha.na' + no -> ha.na.no 'of the flower'

ha.na' + ga -»• ha.na'.ga 'the flower ( N o m ) '
b. ki.no'o + no -* ki.noo.no 'of yesterday'
ki.no'o + ga ->· ki.no'o.ga 'yesteray ( N o m ) '
c. i.moo.to' + no ->• i.moo.to.no 'of my sister'
i.moo.to' + ga ->• i.moo.to'.ga 'my sister ( N o m ) '
This deaccenting rule offers a very good test for our discussion in this pa-
per since it fails to apply to monosyllabic nouns. This is illustrated in (34).
(34) a. te' + no -> te'.no 'of the hand'

b. ho'n + no -> ho'n.no 'of the book'
kyo'o + no - » kyo'o.no 'of today'
Additionally, this rule admits a certain number of lexical exceptions.

Namely, quite a few nouns somehow retain their word-final accent even
before the genitive no.
(35) koo.hi'i + no -> koo.hi'i.no 'of coffee'

hoo.ge'n + no ->• hoo.ge'n.no 'of dialects'
Despite the lexical exceptions exemplified in (35), the deaccenting rule in

question remains a good tool to test the monosyllabic vs. bisyllabic struc-
ture of some of the words we have been discussing in this paper, specifi-
cally of SJ morphemes 16 . Bisyllabic SJ morphemes fall into three catego-
ries depending on their modern pronunciation in isolation. According to
Yoshida's statistics (Yoshida 1999), the initial-accent pattern in (36a) is by
far the most common, accounting for 145 instances of the total of 241, i.e.
60 %. This is followed by the unaccented pattern in (36c), which accounts
for 64 instances, or 27 %. The final-accent pattern in (36b) is the least pop-
ular, only accounting for 32 instances (13 %).
(36) a. Initially accented

zyu'.k 'a cram school', e'.k 'station', se'.k 'seat'
b. Finally accented
zyu.t' 'art, technique', do.k' 'poison', ni.k' 'meat',
ra.k' 'comfort', yo.<ku>' 'desire', hu.k' 'clothes'
c. Unaccented
to.k 'profit', to.k 'virtue', te.k 'enemy', te.t 'iron'
Recall that all these bisyllabic forms originated from monosyllables his-
torically, with a vowel inserted in the morpheme-final position. There-
fore, the morphemes in (36b) are historically equivalent to the SJ mor-
pheme ho'η 'book' in (34b), which remains monosyllabic in modern
Japanese, but they synchronically involve the same syllable and accent
structure as ha.na' 'a flower', which is a finally-accented bisyllabic word
of native origin given in (33a).
An analysis of the accentual behaviour of the bisyllabic morphemes in
(36b) reveals that these morphemes exhibit the accent pattern in (33),
and not the one in (34), when preceding the genitive particle no. In other
words, most speakers of Tokyo Japanese readily apply the pre-no deac-
centing rule to bisyllabic SJ morphemes, as shown in (37).
(37) a. do.k' + no do.ku.no, *do.ku'.no 'of poison'

do.k' + ga ->· do.ku'.ga 'poison (Nom)'
b. ni.k' + no -• ni.ku.no, *ni.ku'.no 'of meat'
ni.k' + ga -» ni.ku'.ga 'meat (Nom)'
c. hu.k' + no ->· hu.ku.no, *hu.ku'.no 'of the clothes'
hu.k' + ga ->• hu.ku'.ga 'the clothes (Nom)'
134 Harno Kubozono
The fact that the SJ morphemes in question pattern with the bisyllabic ha-
ría' and not with the monosyllabic ho'η suggests that the epenthetic vow-
els contained in these morphemes are readily visible to the pre-no deac-
centing rule. This consequence is in full accordance with the findings
reported in (23) and (29a). However, it appears contradictory to the fact
reported in section 3.3, namely, the fact that bisyllabic SJ morphemes with
an initial accent pattern with accented monosyllabic morphemes when
they form the second member of noun-noun compounds. It is interesting
and, at the same time, quite puzzling to find that one and the same epen-
thetic vowel is visible to some accent rules but not to others.
4. Conclusion
In this paper we have seen how epenthetic vowels and elided consonants
interact with various accent rules that are more or less productive in mod-
ern Tokyo Japanese. The traditional idea about the interaction between
synchronic accent rules and diachronic segmental changes is that epen-
thetic vowels are fully visible, and elided consonants are no longer visible
to the synchronic accent rules. Namely, accent rules apply in the synchro-
nic grammar to the output, and not to the input, of historical segmental
changes. However, a careful analysis of accentual behaviour of words
with an epenthetic vowel or an elided consonant has shown that at least
some epenthetic vowels are invisible and some elided consonants are vis-
ible to some accent rules; i.e. some epenthetic vowels behave as if they did
not exist and some elided consonants behave as if they were still present.
The main facts are summarized below.
(38) a. Epenthetic vowels in SJ morphemes

Visible: X-taroo accent rule (section 3.4.),
X-iti accent rule (section 3.5.),
pre-no deaccenting rule (section 3.7.)
Invisible: ordinary compound accent rule (section 3.3.)
b. Epenthetic vowels in foreign morphemes
Visible: antepenultimate rule (section 3.1.),
loanword accent rules triggering unaccentedness
(section 3.2.),
ordinary compound accent rule (section 3.3.),
X-taroo accent rule (section 3.4.),
X-iti accent rule (section 3.5.)
Invisible: antepenultimate rule (section 3.1.),

loanword accent rules triggering unaccentedness
(section 3.2.)
c. Elided consonants in native morphemes
Visible: X-taroo accent rule (section 3.4.),
X-iti accent rule (section 3.5.)
Invisible: ordinary compound accent rule (section 3.3.),
verb/adjective accent rule (section 3.6.)
The strange behaviour of epenthetic vowels and elided consonants is

partly related to the type of word in which they appear. For example, ep-
enthetic vowels in SJ words seem totally invisible to the compound accent
rule as discussed in section 3.3., while those of foreign morphemes are vis-
ible to the same rule. This situation is further complicated by the puzzling
fact we saw in section 3.1., namely, that some epenthetic vowels in the ini-
tial syllable of foreign morphemes, e.g. b.ru'u 'blue\p.re'e 'play',
are invisible to the antepenultimate accent rule although most other ep-
enthetic vowels in foreign morphemes, e.g. p<u'>.ra.s 'plus', are fully
visible to the same accent rule. Similarly, some epenthetic vowels are vis-
ible to the accent rule for unaccented loanwords while other epenthetic
vowels are not.
Paradoxes do not stop here. It has further been pointed out that epen-
thetic vowels in a given type of word is visible to some accent rules but not
to others. For instance, some epenthetic vowels in foreign morphemes, e.g.
b.ru'u 'blue', a'n.de.s 'Andes', are invisible, at least partially, to the
loanword accent rules discussed in sections 3.1. and 3.2., although they ap-
pear to be fully visible to the X-taroo accent rule (section 3.4.). Moreover,
epenthetic vowels in SJ morphemes such as in zyu.k 'a cram
school' behave as if they were invisible to the general compound accent rule
as shown in section 3.3., while the same vowels are fully visible to other ac-
cent rules, i.e. the X-taroo compound accent rule (section 3.4.), the X-iti ac-
cent rule (section 3.5.), and the pre-no deaccenting rule (section 3.7.).
Similarly, there are cases where an elided consonant in one and the
same morpheme is visible to some rules but not to others. Thus, the elided
consonants in the native Japanese morpheme, e.g. Iii in koi (originally
ko.fi) 'carp', appear largely visible to the X-taroo and X-iti accent rules
(sections 3.4. and 3.5.), while they are no longer visible to the ordinary
compound accent rule (section 3.3.) and the accent rule for verbs and ad-
jectives (section 3.6.).
136 Haruo Kubozono
All these facts involve a paradox of one kind or another and, indeed,
pose difficulties for theoretical analyses. While a solution to one of these
paradoxes was hinted at in section 3.3., it remains an open question as to
how all the other paradoxes can be solved within an existing theoretical
framework, especially within the output-oriented framework of Optimal-
l y Theory (Prince and Smolensky 1993).
Acknowledgements
Earlier versions of this paper were read at a regular meeting of the phonology circle at the
Tokyo University of Foreign Studies (AA) and at the Spoken Language Working Group in
Nara. I am indebted to the audiences of these meetings, particularly to Zendo Uwano, Te-
ruhiro Hayata and Miyoko Sugito, for their valuable comments. I would like to thank the ed-
itors of this volume and the two anonymous reviewers for a number of useful comments and
suggestions for the improvement of this paper. I am also grateful to Frank Owens for check-
ing the original manuscript. All the errors that remain are my own. The work reported on in
this paper was partly supported by research grants from ATR Interpreting Telecommunica-
tions Research Laboratories and the Grant-in-Aid for Scientific Research from the Japanese
Ministry of Education, Science, Sports and Culture (Nos. 09610480,10044010,11111102).
Notes
1. In the rest of this paper I will refer to these two types of loanwords in Japanese as 'SJ mor-
phemes' (or 'SJ words') and 'foreign loanwords' (or simply 'loanwords'), respectively.
2. Tanaka (1999) reports that epenthetic vowels in loanwords in Japanese behave different-
ly from underlying vowels in music, too. In Japanese songs, there is usually a one-to-one
correspondence between the musical note and the mora in words, so that a musical note
is assigned to every mora, and vice versa. However, this correspondence does not neces-
sarily hold if the mora contains an epenthetic vowel; namely, epenthetic vowels often be-
have as if they did not exist in the mora-note alignment. Tanaka adds that this is a rela-
tively new tendency which is most eminently found in songs composed in the past two
decades.
3. From a theoretical viewpoint, Alderete (1995) argues for a constraint in the grammar
that dislikes epenthetic vowels in stressed positions—a kind of Head Faithfulness that
connects with a general theory of Positional Faithfulness. If this constraint is high-rank-
ing, then the prediction is that stress rules will avoid epenthetic vowels when counting syl-
lables. In descriptive terms, Michelson (1981,1988) and Potter (1994) report a similar ef-
fect of stress-epenthesis interaction in Mohawk. An interesting case is found in the Saru
dialect of Ainu, an almost extinct language of Hokkaido (northern Japan), where accent
rules apply before the rule of vowel epenthesis. Selkirk (1999) discusses a more compli-
cated case of stress-epenthesis interaction in Makassarese and Selayarese, where some
words appear to undergo stress assignment after vowel/consonant epenthesis while other
words appear to be derived in the reverse order. Similarly, Uwano (1992) gives some in-
teresting examples from Kagoshima Japanese, a syllable-based dialect spoken in south-
ern Japan (Kubozono 1999b), where accent (or tonal) rules exceptionally apply to the in-
put of a consonant deletion rule, despite the fact that they apply to the output of vowel
deletion rules.
4. See Tateishi (1989) and Itô and Mester (1996) for the linguistic factors responsible for the
choice of these two vowels. For the phonetics and phonology of vowel devoicing in Japa-
nese, see Beckman and Shoji (1984), Kubozono (1999a) and the references cited therein.
5. This antepenultimate accent rule accounts for the accentuation of most accented nouns
in Japanese, whether they are loanwords or words of other origins, i.e. native Japanese
or Sino-Japanese words (Tanaka and Kubozono 1999).
6. Note that /u/, not loi, is inserted after [t] in /tu.in/ 'twin' in (9c). This is probably because
of the semivowel /w/ that immediately follows the dental consonant in the source word.
7. Different scholars have proposed different generalizations about the accent pattern in
(9c). Akinaga (1985), for example, states that words ending in a 'tokusyuhaku' (a spe-
cial mora) such as a moraic nasal and a long vowel tend to attract an accent on the pen-
ultimate mora. While this rule is strikingly similar to the syllable-based generalization
proposed in (10a), it cannot fully capture the distribution of the accent pattern in (9c)
as opposed to the pattern in (9a). Quackenbush and Ohso (1990), in contrast, generalize
that words ending in a vowel in the source language tend to give rise to the accent pat-
tern in (9c). This analysis, too, is an overgeneralization of the fact because we recognize
a large number of words that fulfill this segmental requirement but nevertheless show
the ordinary antepenultimate accent pattern: e.g. ki'.raa 'killer' in (9a). More recently,
Tanomura (1999) suggests that the generalization in (10a) is too strong because the ac-
cent pattern in (9c) is not generally applicable to words ending in a moraic nasal: e.g.
p'.ran 'plan' (vs. t.i'n 'twin'). While this observation may be correct, it does not
argue directly against the syllable-based generalization in (10a). Because syllables with
a moraic nasal are metrically weaker than other types of heavy syllables in Japanese in
general (Kubozono 1999a, c), it should not be surprising to find that moraic nasals be-
have in the way Tanomura points out.
8. Some may argue against the generalization in (10b) and propose to attribute the pecu-
liar accent pattern in (9c) to the fact the vowels in the initial syllable/mora tend to be
devoiced or are of very low sonority. These alternative analyses can be rejected in a rea-
sonable way. First, the initial vowels in question are not always devoiced nor are they
always low in sonority. For example, unlike HI and /u/, loi does not generally get de-
voiced in Japanese but nevertheless triggers the penultimate accent: e.g. d<o>.ra'i
'dry'. In addition, lui is not prone to devoicing when it is preceded or followed by a
voiced consonant, but nevertheless gives rise to the peculiar accent pattern in (9c): e.g.
b.ru'u 'blue'. Second, underlying HI and lui do not trigger the accent pattern in (9c)
although they are, other things being equal, just as subject to devoicing as their epen-
thetic counterparts: e.g. pu'.rin 'pudding', ki'.raa 'killer' and ri'.ree in (9a).
9. There are two groups of exceptions to this generalization. First, such words as
h.ro'.a 'floor', s.to'.a 'store', and si.ka'.go 'Chicago' are accented on the penul-
timate syllable/mora although they have the same (or similar) syllable structure as the
words in (9b). Second, such words as p'.ra-n 'plan' exceptionally allow the the ac-
cent pattern in (9a) as a relatively new pronunciation along with the pattern in (9c),
p.ra'-n.
10. A comparison of the examples in (14a) with those in (14b) suggests that the unaccented
accent pattern is more likely to occur if the initial light syllable as well as the final sylla-
ble contains an epenthetic vowel.
11. A functional account may be that this peculiar behaviour of SJ morphemes has to do
with the fact that they alternate with monosyllabic forms in the synchronic grammar.
Many, if not all, SJ bisyllabic morphemes are often realized as a monosyllabic form if
they are followed by a voiceless consonant in compounds: e.g. ga'.ku + koo -* gak.koo
'learning, school; school', se'.ki + ken -· sek.ken 'seat, roll; to overwhelm'. In compari-
138 Haruo Kubozono
son, foreign morphemes do not show such a synchronic alternation; for example,
bu.s and ki.s in (19) never manifest their original monosyllabic form in the
grammar of modern Japanese. In theoretical terms, this functional account is based on
the idea of output-output correspondence.
12. Some may object to this generalization, claiming that the dictinction between the two
accent patterns in (22a, b) has to do with whether the N1 of X-taroo compounds is a free
morpheme or not, i.e. whether it is used in isolation. It is true that many of the first
members in (22a) are bound morphemes and those in (22b) tend to be free morphemes.
However, this account does not hold in many other cases. There are monosyllabic free
morphemes which exhibit the accent pattern in (22a): e.g. ran.ta.roo 'riot, taroo; Ranta-
roo', kin.ta.roo 'gold, taroo; Kintaroo'. There are also many bisyllabic bound mor-
phemes which show the accent pattern in (22b): e.g. riki'.ta.roo (or ri.ki.ta'.roo) 'power,
taroo; Rikitaroo'. These instances suggest that the morphological status of N1 is not di-
rectly relevant to the dictinction between the two accent patterns in (22a, b).
13. Note that many of the words involving this mora/syllable structure admit the accent pat-
tern in (22c) alongside the pattern in (22b). This tendency is most evident when the
vowel in the second syllable of N1 is devoiced, i.e. either li/ or /u/: e.g. su.si' + ta'.roo >
su.si'.ta.roo, su.si.ta'.roo 'susi, taroo; Susitaroo'. This accentual variation, however, does
not affect the discussion to be developed below.
14. Some speakers of Tokyo Japanese seem to prefer the accent pattern illustrated in (22c) for
these compounds and the compounds in (24) below, e.g. ra.k.ta'.roo, ba.s.ta'.roo.
This variant accent pattern can probably be attributed to the phonetic devoicing of the ep-
enthetic vowel.
15. Note that some speakers of Tokyo Japanese, especially young people, seem to permit
an unaccented pattern alongside this accent pattern: e.g. tai.ta.roo and tai'.ta.roo. Most
of these speakers, however, do not admit the accented pattern in (25) for the compound
expressions in (26).
16. The syllable structure of foreign morphemes cannot be tested by this deaccenting rule
since bisyllabic foreign morphemes, whether containing an epenthetic vowel or not, are
generally accented on their initial syllable: e.g. ba'.s 'bus', pi'.za 'pizza'.
References
Alderete, J.
1995 Faithfulness to prosodie heads, ms. University of Massachusetts, Amherst.
Akinaga, K.
1985 Kyootuugo no akusento (Accent of Standard Japanese). In NHK Pronuncia-
tion and Accent Dictionary.
Beckman, M. and A. Shoji
1984 Spectral and perceptual evidence for CV coarticulation in devoiced /si/ and
/syu/ in Japanese. Phonetica 41, 61-71.
Fukui, M.
1998 A remark on unaccented loanwords in Kansai Japanese, ms. Kobe University.
Hayes, B.
1995 Metrical Stress Theory: Principles and Case Studies. Chicago: University of
Chicago Press.
Itô, J. and A. Mester

1996 Stem and word in Sino-Japanese. In T. Otake and A. Cutler (eds.), Phonolog-
ical Structure and Language Processing: Cross-Linguistic Studies. New York:
Mouton de Gruyter.
Kubozono, H.
1989 The mora and syllable structure in Japanese: Evidence from speech errors.
Language and Speech 32-3, 249-78.
1995 Constraint interaction in Japanese phonology: Evidence from compound ac-
cent. Phonology at Santa Cruz, 4,21-38.
1996 Syllable and accent in Japanese: Evidence from loanword accentuation. The
Bulletin, (Phonetic Society of Japan) 211, 71-82.
1997 Lexical markedness and variation: A nonderivational account of Japanese
compound accent. WCCFL, 15, 273-87.
1999a Nihongo no Onset (The Sound System of Japanese). Tokyo: Iwanami Shoten.
1999b Mora and syllable. In N. Tsujimura (ed.), The Handbook of Japanese Linguis-
tics. Oxford: Blackwell. 31-61.
1999c Mora and syllable in Japanese songs, (in Japanese) In Spoken Language
Working Group (ed.), Speech and Grammar 2. Tokyo: Kurosio. 243-62.
Kubozono, H. and S. Ohta
1997 On 'in Koozoo to Akusento (Phonological Structure and Accent). Tokyo: Ken-
kyusha.
Levin, J.
1987 Between epenthetic and excrescent vowels. WCCFL 6,187-202.
McCawley, J.D.
1968 The Phonological Component of a Grammar of Japanese. The Hague: Mouton.
Michelson, K.
1981 Stress, epenthesis and syllable structure in Mohawk. In G.N. Clements (ed.),
Harvard Studies in Phonology. Vol 2, Indiana University Linguistics Club.
311-53.
1988 A Comparative Study of Lake-Iroquoian Accent. Dordrecht: Kluwer.
NHK
1985/98 Nihongo Hatuon Akusento Ziten (Dictionary of Japanese Pronunciation and
Accent). Tokyo: NHK (Nippon Hoosoo Kyookai) Syuppan.
Poser, W.
1984 The Phonetics and Phonology of Tone and Intonation in Japanese, PhD Dis-
sertation. MIT.
Potter, B.
1994 Serial optimality in Mohawk prosody. CLS 30,347-61.
Prince, A. & P. Smolensky
1993 Optimality Theory: Constraint Interaction in Generative Grammar. Technical
Report #2, Rutgers Center for Cognitive Science, Rutgers University.
Quackenbush, H. & M. Ohso
1990 Gairaigo no Keisei to Sono Kyooiku (Loanword Formation and Its Educa-
tion). Kokuritu-Kokugo Kenkyuusyo.
Selkirk, E.
1999 Some basics of Optimality Theory. A talk given at a linguistics colloquium at
Sophia University, Tokyo, November 5,1999.
140 Haruo Kubozono
Sibata, T.
1994 Gairaigo ni okeru akusentokaku no iti (On the location of accent in loan-
words). In K. Sato (ed.), Gendaigo Hougen no Kenkyuu (Studies in Modern
Japanese and Dialects). Tokyo: Meiji Shoin. 338-418.
Tanaka, Shin'ichi
1996 Nihongo no onsetu koozoo to on'in gensyoo (Syllable structure and phonolog-
ical phenomena in Japanese). M A thesis, Osaka University of Foreign Studies.
1999 Kayoo ni okeru gairaigo rizumu no hensen. (The development of loanword
rhythm in songs). A paper presented at the regular meeting of the Phonetic
Society of Japan, held at Keio University on June 12.
Tanaka, Shin'ichi and H. Kubozono
1999 Introduction to Japanese Pronunciation: Theory and Practice. Tokyo: Kurosio.
Tanaka, Shin-ichi and N. Yamane
1999 Paradox resolution and unaccentedness in foreign and Sino-Japanese com-
pounds. Paper read at the 24th meeting of Kansai Linguistic Society (KLS),
Kwansei Gakuin University, October 23,1999.
Tanomura, T.
1999 Gairaigo akusento ni okeru gengo no hatuon no kanyo ni tuite. (On the influ-
ence of the pronunciation of the source language on loanword accentuation).
Nihongo Kagaku 5, 67-88.
Tateishi, K.
1989 Phonology of Sino-Japanese morphemes. University of Massachusetts Occa-
sional Papers in Linguistics 13, 209-35.
Trubetzkoy, N.S.
1969 Principles of Phonology (trans. C.A.M. Baltaxe). Los Angeles: University of
California Press. (Original work published 1958).
Uwano, Z.
1992 Kagosima Hukiage-tyoo hoogen no hukugoomeisi no akusento (Accentua-
tion of compound nouns in Kagoshima Fukiage dialect). In T. Kunihiro (ed.),
Nihongo Intoneesyon no Zittai to Bunseki (Facts and Analyses of Japanese In-
tonation) (Grant report). 91-208.
Yoshida, Y.
1999 Kanzi on to akusento (Sino-Japanese sounds and accent). Paper presented at
the March meeting of the Phonological Association in Kansai (PAIK).
Prosodie Structure and Sandhi Phenomena
in the Saru Dialect of Ainu1
Hidetoshi Shiraishi
Sakhalin Museum of Regional Studies
1. Introduction
A well-known fact about phonological processes applying between words

is that they can be divided into two types with respect to sensitivity to syn-
tax. The sensitive ones, dubbed P-structure rules in the literature (Selkirk
1986), apply only when the relevant segments fall under certain syntactic
environments. A classic example of such a process is French Liaison, as is
illustrated in the following examples. 2
(1) a. Il y a encore deux _ après-midis 'There are still two afternoons'

b. Il y en a encore deux // après lui 'There are still two of them after him'
(Nespor and Vogel, henceforth N&V 1986: 4)
In both (la) and (lb), the structural description of the rule is met. Never-
theless, Liaison applies only to the former. This contrast is rooted in the
difference of the syntactic structure between (la) and (lb): in (la), the Li-
aison context is located within a single syntactic phrase (modifier-head)
whereas in (lb), it spans two phrases, an inappropriate domain for Liaison.
On the other hand, a number of phonological processes are blind to such
syntactic structure. Flapping rule of American English belongs to this type.
It is applicable even across the largest syntactic unit, the sentence.
(2) Have a seat. I'll be right back. > ... sea[r] I'll...
(N&V 1986: 236)
Phonological rules of this type, called Pure phonological rules (Selkirk

1986) are insensitive to any syntactic information. A similar distinction
has been proved to be valid for a number of postlexical phonological pro-
cesses cross-linguistically (cf. several contributors to Phonology Year-
book 4 and Inkelas and Zee 1990, among others), where Ainu is no excep-
tion, as will be discussed in the present work.
142 Hidetoshi Shiraishi
This paper examines the consonantal sandhi phenomena of Ainu and

points out that a similar distinction can be observed in at least two sandhi
processes centering on η and r, which we will term n-alternation and r-al-
ternation, respectively. The two processes occur in virtually every dialect
of the language and have been reported from the earliest stages of linguis-
tic research (e.g. Pilsudski 1912, Kindaichi 1931). Despite this fact, little
attention has been paid to their exact domain of application. A closer
look at the distribution of the phenomena informs us, however, that they
exhibit a significant contrast with respect to their domain. An examina-
tion of a corpus of the Saru dialect reveals that the n-alternation is sensi-
tive to a certain syntactic boundary (and hence a P-structure rule), but the
r-alternation is not (and hence is a Pure phonological rule).
Our next question, then is to what extent syntactic information is

needed to set up the appropriate domain of n-alternation. Here the Ainu
data follows the cross-linguistic tendency of prosodie phrasing, where
only impoverished syntactic information is used by phonology (Selkirk
1984, 1986, N&V 1986, Hayes 1989, etc.); prosodie phrasing is sensitive
to the phrasal rank but information such as category label (e.g. NP or
VP) is irrelevant. In section 3.1., however, we point out that syntax alone
is insufficient to predict the actual application of n-alternation. In the da-
ta, we often observe unexpected blocking of n-alternation in terms of
syntactic structure, but instead ought to be explained by reasons as fre-
quency.
Section 3.2 is devoted to the comparison of prosodie units observed

for the n-alternation and those referred to in the versification of oral lit-
erature. We will point out that the prosodie units for n-alternation and
versification show a certain degree of similarity. This further supports
the claim that one and the same prosodie phrasing is responsible for
phonological operations in general in a given language, a claim that we
repeatedly find in the literature in the area of the syntax-phonology in-
terface.
1.1. Overview of the two processes
In this section, we present the data on n-alternation and r-alternation.

Note that both are applicable between words, as well as between constit-
uents inside a word (see 3a and 4a). 3
Prosodie Structure and Sandhi Phenomena in the Saru Dialect of Ainu 143
(3) n-alternation
a. ror -un-so > roru[y] so
place of honor-at- seat
'the seat at a place of honor in a house'
b. pon yuk > po[y] yuk
little deer
'little deer'
c. yayan wakka > yaya[w] wakka
normal water
'normal water (in contrast with water from a hot spring)'
d. . . . a n w a . . . > a[m m]a
to be CON 4
'... exists and ...'
e. ... an yakka ... > a[y] yakka ~ an akka
to be CON
'...even though...'
(4) r-alternation
a. ar- rametok > a[n]rametok
real-bravery
'a real bravery'
b. yar nima > ya[n] nima
bark tray
'a tray made from bark'
c. asir cise > asi[t] cise
new house
'a new house'
d. kusur ta > kusu[t] ta
kusur LOC
'at Kusur (place name)'
There is no agreement in the literature on how to interpret these sandhi

rules. Since no attempt has been made to give a unified account to these
phenomena, it has been the norm to describe and interpret them as they
are. Under such an approach, we find, for instance, both dissimilation (4a)
and assimilation (4b-d) for r-alternation. A closer look at their phonolog-
ical nature reveals, however, that the rules centering on η and r can be giv-
en a unified account, respectively. The rules centering on η are assimila-
tory in nature; the η assimilates to the following consonant no matter of
its place of articulation. Its gliding to y or w is a special case which occurs
when either y, w, or s (the [+continuant] consonants of the language) fol-

lows. So in principle, it does not differ from cases as pon-pe > po[m] pe
'something small' in which case the η assimilates to the following stop. In
sum, the whole process (3a-e) can be understood as a regressive assimila-
tion which targets both place and continuancy.5 The 'mutual assimilation'
in (3d) is a special case, allowed only when the trigger w is part of the con-
junctive particle wa. In other function words (3e), the trigger either caus-
es gliding (a[y] yakka) or is deleted (an akka).
On the other hand, the rules centering on r assign a stricter condition
on the trigger; they are strictly restricted to the coronals, namely n, r, t and
c. As mentioned above, here the r either assimilates (4bcd) or dissimilates
(4a). On the other hand, nothing happens when the r is followed by any
other consonant, e.g. korka 'but', sermak 'behind', arpa 'go'. The various
outputs can be given a unified account if we regard this alternation as a
result of a syntagmatic constraint prohibiting r + coronal sequences. Sev-
eral sandhi patterns in other dialects exhibit that this constraint has a gen-
eral character prevailing over the phonology of the language, e.g. rs > ss
(Nothern dialects): ar-suy ->• a[s]suy 'once' karkar-se - > karka[s]se 'to
roll', and further observed in the Saru dialect are, ry > r. kor-yar -» korar
'to give' (deletion of y), a-kor yarpe a=ko[y]yarpe 'my baby' (gliding
of r). These forms indicate that any r + coronal sequence is disfavored in
the phonology of this language.
From these discussions, we conclude that the various rules above can be
grouped into two groups (namely, n-alternation and r-alternation) and
that such a classification is not an arbitrary one.
Despite their difference in their phonological nature, the two processes
share characteristics typical of "postlexical rules", e.g. they have no lexi-
cal exceptions, and are easily blocked by pause insertion. In addition, n-
alternation may create a highly marked sequence uw (in rime): a=esika-
ru[w] wenkasuno 'to long very much for - ' (Tamura 1986:12). This is pos-
sibly a violation of Structure Preservation since uw is an inadmissible se-
quence in the underlying form of the language. Furthermore, it is a well-
known fact that both rules may extend to the speaker's pronunciation of
Japanese, e.g. se[y]se instead of Japanese sense(i) 'teacher' (Kindaichi
1931: 25). This is frequently observed in a code-mixing sentence, e.g. ue-
peke[n] ni yoku yuu '[something] appears frequently in uepeker' (own
field notes, 14 Sep. 1998), where the alternation of the last r of uepeker has
a Japanese word ni as a trigger.
Apart from these properties, the two processes show significant differ-
ences in their distribution, as will be shown in what follows.
1.2. Method and source of examination
Our primary source of evidence is restricted to recordings of the Saru di-

alect, in particular Kayano (1974,1998a-c) and Tamura (1984,1985,1986,
1988a, 1989,1997). The distribution of the two sandhi processes in these
recordings was checked by the author by means of impressionistic obser-
vation.
As a linguistic corpus, the content of these sources is not well balanced,
in spite of the fact that in all 18 speakers are involved. A large part of the
material consists of recitations of oral literature performed predominant-
ly by women. Although we limited our source to those performed without
a significant musical melody, it remains an assumption that the selected
genre more or less reflects the natural speech of the language.
2. The discrepancy between r-alternation and n-alternation

in the prosodie domain
An interesting discrepancy appears when we look at the distribution of n-

alternation and r-alternation in larger syntactic units. Compare the fol-
lowing examples, where the target and the trigger of the alternation rules
span two sentences.
(5) n-alternation
a. inani un hoski arpa=an kor pirka kus hawas sekor yaynu=an //
which to first go 4 CON good PURP sounds COMP think 4
sino wen iruska poka nesi a=ki
real bad anger even very 4 do (KT6 Kayano 1998a: 110)
Ί could not decide where to go first. I was completely frustrated.'
b. e=motoho a=nukar wa an= an //
2 origin 4 look CON to be 4
yaun mosir un iwor kor kamuy a=ne wa ...
land country to field have spirit 4 be CON (KT Kayano 1998c: 58)
Ί am looking at your origin. I am the spirit governing this field in this country
c. kanna ruyno ye yan // ye yan sekor
once again say FP: IMP COMP (KKo Kayano 1998a: 74)
'Say it again. Say it again'
(6) r-alternation 7
a. ne sinrici ka e=kopuspakar _ nokan uypehe ka opitta usa muni
that root also 2 dig fine chips also all various garbage
turano e=uhuyka.
together 2 burn (KKo Kayano 1974:145)
'You dig that root. Together with the fine chips you burn all of them'
b. kotankonnispa sine matnepo kor _ nea matnepo
the village head one daughter have that daughter
ramutu uk tek hine ...
life pull out briefly CON
(HF Tamura 1985: 60)
'the village head had one daughter. [My brother] pulled out
the life of that daughter ...'
c. nea niatus heyasi a=ninpa kor _ rapokke sinki= an hine ...
that pail towards a bank 4 drag CON then to be tired 4 CON
(KT Kayano 1974: 99)
Ί dragged that pail to the bank and then became tired ...'
d. isepo kuari cironnup kuari a=eykoysanpa kor _ cironnup
rabbit trap fox trap 4 imitate CON fox
ka isepo ka a=rayke
too rabbit too 4 kill
(KKo Tamura 1988a: 12)
Ί imitated [my father's] rabbit traps and fox traps and caught rabbits
and foxes.'
e. eci= kor mosir sekor eci= haweoka ka eaykap kunihi eci=
2PL have country COMP 2PL say also impossible NOM 2PL
ramu kor _ tane anakne somo uhekote itak=an kusu ne na
think CON now TOP NEG each other talk 4 will COP FP
(HS Tamura 1986: 46)
'Keep in mind that you cannot insist that it is your country
[if lots of people are gathering] and don't talk about this any longer.'
In both (5) and (6), the structural description of the rules is met, and yet
only r-alternation occurs. T h e apparent difference with the e x a m p l e s in
(3) and (4), w h e r e n o such discrepancy was observed, is their syntactic
construction. In (5) and (6), the relevant segments b e l o n g to separate sen-
tences, i.e. the target and trigger are separated by a s e n t e n c e boundary. It
should also b e n o t e d that the t w o words in question are not separated by
a pause.
It is worth noting that the length of the sentence does not matter for the
application of n-alternation. Consider the following.
(7) oraun tumun soyo yan // soyo yan sekor ...

then dust bring outside FP:IMP COMP
'then (she said) "bring the dust outside'" (KKo Kayano 1974: 7)
Here the verb + final particle sequence is repeated, creating an n-alterna-

tion context which spans relatively short sentences. Nevertheless, n-alter-
nation does not apply. This example suggests that n-alternation is blocked
whenever there is a sentence boundary, regardless of the length of the
sentence.
All these examples inform us that the n-alternation is sensitive to sen-
tence boundaries. On the other hand, no such restriction is observed for
the r-alternation, which applies whenever the relevant segments are tem-
porarily adjacent, no matter what syntactic boundary intervenes. Unlike
the n-alternation, the r-alternation is totally blind to such syntactic infor-
mation.
From this observation, we conclude that the n-alternation should be
classified as a P-structure rule and r-alternation as a Pure phonological
rule, following the terminology of Selkirk (1986). Although both are
postlexical in nature, they behave differently with respect to sensitivity to
syntactic information and hence deserve different labeling.
3. The domain of n-alternation
Of particular interest is the distribution of the n-alternation, which

shows sensitivity to syntactic information such as sentence boundaries.
In what follows, we will examine the syntactic context of n-alternation in
more detail and show that the other blocking contexts for n-alternation
more or less involve large syntactic boundaries comparable to sentence
boundaries.
We will start with contexts in which n-alternation is constantly ob-
served. These include word internal contexts and idiomatic expressions.
(8) a. pon _ ya-un-pe

young-land-live-man
(name of a hero)
b. sasun _ sir
to have descendants-NOM
(part of an idiom)
The next context in which n-alternation is observed frequently is between

a prehead modifier (demonstratives, numerals, adjectival verbs 8 ) and its
head noun, though n-alternation in this context is subject to a certain de-
gree of optionality (see section 3.1.).
(9) a. tan _ ya ta
this land LOC
'at this land'
b. iwan _ suy
six times
'six times (a sacred number of the language denoting 'many')'
c. pon _ suma
little stone
'little stone'
The n-alternation may also apply between words belonging to different

syntactic phrases, as is illustrated in the following examples.
(10) adjunct + verb

[ppi= etoko un] _ [VPsupa kor] ...
4 0 B J before at cooking (KT Kayano 1974: 33)
'[she] has been cooking before my (return)'
(11) indirect object + verb
[NPnea pon menoko], [ PP a=hokuhu eun] _ [VPye]
that young lady 4 husband to say
(HF Tamura 1985: 20)
'that young lady said to my husband'
(12) direct object + verb
[NPtumun] _ [vpsoyo yan]
garbage putoutside FP: IMP (KKo Kayano 1974: 7)
'put the garbage outside!'
(13) subject + verb
[NP[Nkeman] _ [VPyupke] p] [VPan wa]
starvation hard NOM to be CON (HF Tamura 1984: 48)
'There was a hard starvation'
(14) adjunct + adverb

[pplskar etoko un] _ [ A D v s u y ] [ v p a r P a w a ]
upper stream to again go CON (HS Kayano 1974:214)
'[he] went again to the upper stream of Iskar river'
The overall rate of alternation in this context (10)-(14), however, is not as

high as the former two contexts (8)-(9), even within the same speaker.
The number of applications of n-alternation inside and outside a phrase
boundary is compared below for three speakers. 9 It should also be noted
that these three speakers show the most frequent application of n-alter-
nation across phrase boundaries within our primary source. 10
speaker inside X P outside X P

n-alternation number of n-alternation number of
contexts applications contexts applications
KT 5 4 9 3
HS 15 13 5 1
HF 13 10 8 4
For some speakers, n-alternation never applies across a phrasal boundary.
(15) adjunct + verb

[ PP a=uni un] // [VPsan = an]
4 house to go down 4 (KM Kayano 1998a: 38)
Ί went downwards to my house'
(16) subject + verb

[PProrunpuyar pok un][ NP tun]// [NPsike] [ V[ osura]
window beneath to two person baggage throw
(NK Tamura 1989: 38)
'The two (men) threw their baggage beneath the window'
So it seems fair to say that n-alternation across a phrase boundary (as in

(10)-(14)) is not preferred. It is thus a "marked" context, observed spo-
radically in the speech of some speakers, whereas the inside-phrase con-
text is "unmarked", observed for all speakers.
Interestingly, n-alternation was absent from the following contexts as
well.
(17) quotation + complimentizer ( s e k o r )

[s[NPi= otke humi] [VPitasasa=an] // sekor] [VPhawean kor]
4 0 B J stab feeling hurt 4 COMP say -ing
(HS Tamura 1984: 20)
'Saying „It hurts where you have stabbed me'"
(18) VP + evidential nominalizer (siri)11

[Np[ADvOpitta][vPray=an] // siri] [ne]
all die 4 NOM:EVID COP
(KT Kayano 1974: 36)
'We were (almost) all dead'
(19) relative clause + head noun 12

[NP[Pani ][VPkampinuye=an] // [Nsumi]]
with write 4 ink (Tamura 1984:74)
"The ink with which we write'
The non-application of n-alternation in these contexts is comparable to

that with the sentences in (5) in their stability. Throughout our primary
source, not a single case of n-alternation has been observed in these con-
texts.13 Here the n-alternation does not occur even though the relevant
segments are temporarily adjacent. It seems therefore fair to say that
these boundaries fall within the sentence boundary type, which consti-
tutes absolute blocking contexts for n-alternation. Within the domain de-
limited by these boundaries, n-alternation is allowed to occur for some
speakers, though with certain degree of optionality that is further sensi-
tive to phrasal boundaries.
3.1. Frequency
In this section, we will consider „optionality" which was used somewhat

informally in our discussion above. In fact, optionality should not be un-
derestimated since n-alternation shows a certain degree of variability in
its application in many contexts,14 with the exception of the inside-word
context (8). This is even true for speakers with the widest range of n-al-
ternation (i.e. speakers which exhibit n-alternation across phrasal bound-
aries). The following examples show failure of n-alternation in a modifier-
head context in the speech of such speakers (again, no pause intervenes).
(20) a. tan // Yupet un

this at (NT Kayano 1998b: 56)
'at this Yupet'
b. tapan // sisam mosir
this Japanese land (HF Tamura 1984: 32)
'this land of the Japanese'
Optionality, however, does not seem to be a simple matter of chance.

Tamura (1988b: 63) points out that n-alternation is blocked when the
word string is pronounced with „analytic consciousness".
From our own observation, it seems that words (or strings of words) of
frequent usage tend to be more subject to alternation than those of low
frequency. For instance, one speaker (KT) showed variation in her pro-
nunciation of pon sinrici 'little root' (pon _ sinrici ~ pon // sinrici), where-
as there was no variation in pon seta 'little dog' and pon su 'little pot'
(both constantly with po[y]). Although we have not yet succeeded in un-
covering all factors that play a role in this variation, it is quite possible
that several factors are at work within the possible prosodie domain of n-
alternation.
3.2. Comparison with the prosodie phrasing in oral literature
The unmarked domain of n-alternation discussed so far exhibit an inter-

esting isomorphism between constituents observed in the versification of
oral literature, in particular, with that of a genre called Yukar. This is a
desirable result if the principles of prosodie phrasing provides prosodie
constituents not only for segmental phonology (as sandhi phenomena)
but for phonological operations of the language in general, including
versification. In this section, we will compare the prosodie units dis-
cussed so far with those of Yukar. The analysis on the versification of
Yukar adopted here heavily depends on the study of Okuda (1988), who
investigated the phrasing pattern observed in Yukar of a single speaker
in the Shizunai district.15
According to Okuda (1988), the versification of Yukar reveals a certain
patterning with respect to the assignment of syntactic units (the verse) to
the rhythm pattern in its recitation. The former, called rhythmic unit by
Okuda, consists of a free morpheme followed by a bound morpheme
(note that Ainu is a postposition language). The rhythm of Yukar is cre-
ated by a hit of a stick (repni) to the edge of a fireplace, which itself con-
sists of two beats. With respect to the sequence of this metrical unit con-
sisting of two beats, the following regularity is observed.
(21) a) The left edge of a rhythmic unit should be aligned with the left
edge of a metrical unit, i.e. the first beat. 16
b)The right edge of a rhythmic unit should be placed somewhere be-
tween the second beat and the first beat of the next metrical unit.
(Okuda 1988: 40, slightly modified by the author)
Accordingly, (22a) exhibits an appropriate alignment, since the left edge

of a free morpheme (adverb) is assigned the first beat. On the other hand,
(22b) is an example of improper alignment, since the right edge of the
rhythmic unit is put on the first beat, thus violating generalization b).
(22) a II I
(poron no} 'many'
b * II I II
{po ron no}
(where || denotes the first and the second beat. The rhythmic unit is
contained in curled brackets)
(Okuda 1988: 42-43)
The next question, then, is which syntactic unit may initiate this rhythmic
unit (and which may not). As mentioned above, Okuda attributes to free
morphemes: noun, verb, conjunction, interjection, adverb, etc. On the
other hand, bound morphemes (postpositions, particles) can enter the
rhythmic unit only by cliticizing to free morphemes (1988: 37-39).
Of interest to us is the structural similarity of this rhythmic unit with the
unmarked domain of n-alternation. Similar to the latter, Okuda's rhyth-
mic unit does not require fully-fledged syntactic information (such as cat-
egory labels), as the division of free versus bound morpheme indicates.
This is exactly what we expect if the prosodie phrasing makes use only of
impoverished syntactic information. In addition, modifiers seem to form a
single rhythmic unit with its head, even though in that case the left edge of
the latter does not align with the first beat, in violation of generalization a).
(23) II I II I
{poro sin to ko} 'a big chest'
big chest
(Okuda 1988: 54)
Here the left edge of sintoko 'chest' is not aligned with the first beat even
though this can initiate a new rhythmic unit since sintoko itself is a free
morpheme (noun). This extraordinary alignment can be accounted for if
we regard the whole modifier-head phrase as forming a single rhythmic
unit. Now, recall that such a modifier-head phrase was a common n-alter-
nation domain. This 'coincidence' further supports our claim that the
same prosodie unit is provided for versification of Yukar as well as for seg-
mental rules such as n-alternation.
On the other hand, it is also true that Okuda's rhythmic unit and the
domain of n-alternation do not show a perfect match. A major discrepan-
cy between them is the phrasing of the evidential nominalizer siri (see 18).
According to Okuda, siri is usually phrased with the preceding free mor-
pheme (predicate verb), while this was never an n-alternation context in
our data. However, it seems also to be true that siri does provide an n-al-
ternation context for this speaker (Okuda p.c.), in which case this partic-
ular phrasing does not contradict our hypothesis.
Another disparity can be seen in the phrasing concerning relative
clauses. The investigation of relative clause constructions in versification
reveals that the verb in the relative clause (which is necessarily the final
constituent within the relative clause) is phrased with the following head
noun.
{a=kor totto} {kor casi or ta} 'at my mother's castle'

4 have mother have castle at
Since a free morpheme casi is left-aligned with the second beat, the rhyth-
mic unit should be interpreted as initiated by the preceding kor (other-
wise there is a violation of generalization a)). Recall, however, that the
relative clause-head noun sequence context never provides an n-alterna-
tion context (see 19). However, it should also be noted that the total num-
ber of relative clause constructions is small in our primary source of re-
search.
Although the details of the domain for n-alternation and Okuda's
rhythmic unit are not isomorphic to each other, it remains a fact that they
exhibit a certain degree of similarity, which we consider a worthwhile top-
ic for future research.
4. Conclusion
Since the domain of application for n-alternation and r-alternation are

relatively large, their different behavior with respect to syntactic informa-
tion as sentence boundary has been overlooked in the literature. In fact,
the two sandhi rules differ significantly when we consider their sensitivity
to certain syntactic boundaries. The present work has brought this differ-
ence to light and provided grounds to the current classification of sandhi
rules into two groups centering on η and r.
The fact that not a single case of n-alternation has been found across a
sentence boundary suggests that this constitutes an edge of a certain pro-
sodie domain, active in the grammar of the speakers. Whether there are
other phonological operations which share the same prosodie unit needs
to be worked out.
It is not, however, sufficient to say that syntax alone decides the phrasing
process since the actual application of sandhi processes varies a great deal,
even within the same syntactic context. In fact, the phrasing seems to be
highly dependent on (non-) linguistic information other than that provid-
ed by the syntax of the language. It still needs to be investigated what fac-
tors play a role in the actual phrasing processes, and in what priority.
Acknowledgments
A great deal of this paper constitutes a part of the author's M. A. thesis submitted to the Chi-
ba University Graduate School in September 1998.1 am grateful to the member of the thesis
committee, Professors Tohru Kaneko, Hiroshi Nakagawa and Kenji Kanno. Part of this pa-
per was presented at the Hoppogengobunka Kenkyukai (October 1997). Thanks to the par-
ticipants for their comments, especially Professor Suzuko Tamura and Hideo Kirikae. Many
thanks to Professor Osami Okuda for helpful comments. Thanks are also to my colleague at
the University of Groningen, Tjeerd de Graaf, Dirk-Bart den Ouden and especially Wouter
Jansen for hours of stimulating discussion. All the errors are my own. The research was par-
tially supported by the Netherlands Government scholarship (NUFFIC).
Notes
1. Ainu : an endangered language of Japan whose genetic affiliation with the neighboring
languages is unknown. The Saru dialect refers to the dialect once spoken in the villages
alongside the river of Saru, southwest Hokkaido.
2. _ = application, // = blocking of Liaison and of any other phonological process in this pa-
per.
3. We will use the conventional writing system of Ainu throughout this paper. The follow-
ing correspondences should be noted: y = IPA [j], r = IPA [r]. Hyphens indicate mor-
phological boundaries. Segments that have undergone alternation are bracketed.
4. Abbreviations: 2 = second person personal prefix, 4 = fourth person personal affix, in-
dicating the first person in the oral literature, hence in the examples below. These per-
sonal affixes are separated with a double hyphen (=) in the text. C O M P = complemen-
tizer, C O N = conjunctive particle, C O P = copula, E V I D = evidential, F P = final particle,
IMP = imperative, L O C = locative, N E G = negative, N O M = nominalizer, OBJ = ob-
ject, PL = plural, P U R P = purpose, T O P = topic. ~ indicates variation.
5. A feature-geometric approach following the model of Padgett (1991) is proposed by
Shiraishi (1998).
6. Hereafter we will provide the initials of the speakers.
7. The astute reader would notice that the examples (c-e) of r-alternation all involve the
conjunctive particle kor 'and, while'. This is due to the unbalanced number of r-alterna-
tion contexts, i.e., the r-n context outranks the other three (r-c, r-t, r-r), and it is difficult
to find an example without kor for the latter.
8. 'Adjectival verbs' refer to prehead intransitive verbs modifying the following head
noun. Ainu has no adjectives (morphologically speaking).
9. Of the 18 speakers, 7 show n-alternation across a phrase boundary and 5 do not. The
rest lack data of the context under discussion.
10. Note: identical expressions within a single speaker are counted as a single context. If
there is variation, it is counted as a case of application.
11. The evidential nominalizer can be regarded as a subcase of relative clause construction
as siri derives from the noun meaning 'appearance, state'.
12. Postpositional stranding as seen in this example is a common strategy for relative clause
construction in Ainu.
13. Examples of relative clause constructions containing relevant segments for n-alterna-
tion are, however, quite rare in our primary source.
14. This applies to the conjunctive particles yakne, yakka and yakun as well. We still have
no answer why these function words show unstable application of n-alternation for
some speakers, being inconsistent with the prediction of most literature on phrasal pho-
nology (e.g. Selkirk 1995).
15. The Shizunai district lies about 30km to the southeast of the Saru district and its dialect
differs slightly from that of Saru. A comprehensive description of the Shizunai dialect
has been published in 1986 by Kirsten Refsing.
16. Okuda's original principle contains an alternative: the left edge of the rhythmic unit can
also be placed somewhere within the latter half of the second beat. Since this is irrele-
vant for our discussion, it is ignored here.
References
Hayes, B.
1989 The Prosodie Hierarchy in Meter. In P. Kiparsky and Y. Gilbert (eds.), Pho-
netics and Phonology 1 Rhythm and Meter, 201-260. San Diego: Academic
Press.
Inkelas, S. and D. Zee (eds.)
1990 The Phonology-Syntax Connection. Chicago: The University of Chicago
Press.
Inkelas, S. and D. Zee

1995 Syntax phonology Interface. In J. Goldsmith (ed.), The Handbook of Phono-
logical Theory, 533-549, Cambridge: Blackwell.
Kayano, S.
1974 Uepekere Shutaisei [Collection of Uepekere] Tokyo: Arudoo.
1998a Kayano Shigeru no Ainu Shinwashusei [Shigeru Kayano's Collection of Ainu
Myths] Uepekere hen 1. Tokyo: Heibonsha.
1998b Kayano Shigeru no Ainu Shinwashusei [Shigeru Kayano's Collection of Ainu
1998c Kayano Shigeru no Ainu Shinwashusei [Shigeru Kayano's Collection of Ainu
Kindaichi, K,
1931 Ainu Jojishi Yukar no Kenkyu [Research on Ainu Yukar Epics]. Tokyo: Toy-
obunko.
Nespor, M. and I. Vogel
1986 Prosodie Phonology. Dordrecht: Foris.
Okuda, O.
1988 Shizunai chiho no yukar ni okeru rhythm to shiku no kankei ni tsuite [On the
relationship between the rhythm and verse of yukar in the Shizunai district],
M.A. thesis, Chiba University. Abridged version published in Journal of Wase-
da Institute of Language Teaching 40 (1990), 59-63.
Padgett, J.
1991 Stricture in Feature Geometry. Doctoral dissertation, University of Massachu-
setts, Amherst.
Pilsudski, B.
1912 Materials for the study of the Ainu language and folklore. Reprinted in A.
Majewicz ed. (1998). The Collected Works of Bronislaw Pilsudski vol. 2. Ber-
lin: Mouton de Gruyter.
Refsing, K.
1986 The Ainu Language: The Morphology and Syntax of the Shizunai Dialect.
Aarhus: Aarhus university press.
Selkirk, E.O.
1984 Phonology and Syntax: The Relation between Sound and Structure. Cam-
bridge, Massachusetts: MIT Press.
1986 On derived domains in sentence phonology. Phonology Yearbook 3\ 371-405.
1995 The Prosodie Structure of Function Words. In J. Beckman et al. (eds.), Univer-
sity of Massachusetts Occasional Papers 18: Papers in Optimality Theory, 439-
469.
Shiraishi, H.
1998 Ainugo no on'in kotai [The consonantal sandhi of Ainu]. M.A. thesis, Chiba
University.
Tamura, S.
Ainugo onsei shiryo [Phonetic materials of the Ainu] 1. Tokyo: Waseda Insti-
1984 tute of Language Teaching.
1988a tute of Language Teaching.
1988b Ainugo [The Language of Ainu] In. T. Kamei et al. (eds.), The Sanseido En-
cyclopaedia of Linguistics, vol. 1, 6-94. Tokyo: Sanseido.
1989 Ainugo onsei shiryo [Phonetic materials of the Ainu] 6. Tokyo: Waseda Insti-
tute of Language Teaching.
1996 Ainugo jiten [The Ainu-Japanese Dictionary Saru dialect] Tokyo: Sofukan.
1997 Ainugo onsei shiryo [Phonetic materials of the Ainu] 10. Tokyo: Waseda In-
stitute of Language Teaching.
The Emergence of the 'Unaccented':
Possible Patterns and Variations
in Japanese Compound Accentuation
Shin-ichi Tanaka
Nagoya University
1. Introduction
This paper is an in-depth examination of how Optimality Theory (hereaf-

ter, OT) can apply to the phonological system of Japanese and, converse-
ly, what theoretical consequences the intricate phonological phenomena
of Japanese imply for OT. These two issues will be examined by extensive-
ly considering the possible patterns, variations, and progressive changes
of accent in Japanese. In doing this, I will basically adopt but then extend
the set of general constraints proposed by Kubozono (1995, 1997) and
give a principled account of what patterns and variations are possible in
the accentual system of Japanese compounds and why some types of com-
pounds can be de-accented at all in this language.
The central claim is that the three possible accentuation modes of com-
pounds (full preservation of input accent, default accentuation, and de-
accentuation) fall out from a fixed hierarchy of the members of M A R K E D -
NESS constraints (i.e. the N O N - F I N A L I T Y family), in which a FAITHFULNESS
constraint (i.e. MAX) is ranked in minimally different positions. Although
the mechanism of unaccented patterns has long been a mystery for Japa-
nese phonologists, this uniform analysis makes it possible to understand
the phenomenon as a non-finality effect, in which a certain N O N - F I N A L I T Y
constraint bans accent from the final (and, crucially, single) PrWd in a
compound. I will further suggest that diachronically, MAX has been, and
will be, undergoing demotion step by step in the accentual system of Jap-
anese, which is carried out under the 'gravitation' of the members in the
FAITHFULNESS family.
The organization of this paper is as follows: we will first look at three
basic characteristics of compound accent and sketch Kubozono's (1995,
1997) treatment of them in OT (section 2); we will then consider several
problems with his analysis and elaborate a new constraint-ranking system
on the basis of various sets of data (section 3); next, I will demonstrate
160 Shin-ichi Tanaka
that my proposed system can capture the progressive change of de-accen-

tuation as well as other variations (section 4); and finally, we will reexam-
ine the new ranking system in the context of synchronic and diachronic
grammar (section 5).
2. Background and fundamental factors
2.1. Basic characteristics
First, let us review three basic properties of Japanese compounds. In gen-

eral, compound accent exhibits a clearer regularity and hence seems eas-
ier to formalize than lexical accent. Therefore, I will be concerned with
the former in the present discussion. First, as shown in ( l a ) , the word ac-
cent of a head nominal (N2) usually survives as compound accent, rather
than the accent of a lefthand non-head nominal (Nl). Moreover, as
shown in ( l b ) , compound accent appears in N2 if it is originally an unac-
cented word:
(1) N2 > NI (cf. McCawley 1977)

a. miso + síru • misosíru 'miso soup'
siifúudo + píza -> siifuudopíza 'seafood pizza'
áka + murásaki -* akamurásaki 'red purple'
yámato + támasii yamatodámasii 'Japanese soul'
b. kita + amerika -»· kitaámerika 'North America'
nisi +afurika • nisiáfurika 'West Africa'
nikú + dango ->• nikudángo 'meat ball'
yó + sakura ->• yozákura 'cherry tree at night'
Another property of compounds is that a non-final lexical accent of N2 is

usually preserved in that position (e.g. (2b)), while an accent on the final
mora or syllable is avoided and shifts leftward (e.g. (2a)):
(2) Avoidance of final position (cf. McCawley 1977, Kubozono 1995,1997)

a. nágoya + sí -• nagoyási 'Nagoya City'
róndon + hasí ->· rondónbasi 'London Bridge'
isí + atamá ->· isiátama 'hard head'
áisu + koohíi • aisukóohii 'ice coffee'
sinzyu + kái ->· sinzyúgai 'pearl shell'
súupaa + mán -> suupáaman 'superman'
Possible Patterns and Variations in Japanese Compound Accentuation 161
b. natu + míkan natumíkan 'summer orange'

satuzin + ziken ->• satuzinziken 'murder case'
furáido + póteto ->· furaidopóteto 'fried potato'
áka + murásaki akamurásaki 'red purple'
The final property is that if N2 is a short head with 1 or 2 moras, com-

pound accent falls on the syllable immediately preceding the N1-N2
boundary (e.g. (3a)); if N2 is a long head with 3 or 4 moras, however, it
falls on the syllable immediately following the boundary (e.g. (3b)):
(3) Before and after the boundary (cf. Kubozono 1995)

a. nágoya + sí -»· nagoyási 'Nagoya City'
róndon + hasí ->• rondónbasi 'London Bridge'
amerika + zin amerikázin 'Americans'
arizona + syúu - > arizonásyuu 'Arizona State'
yoyaku + séki -»• yoyakúseki 'reserved seat'
tyóu + bátu tyóubatu 'punishment infliction'
b. isí + atamá ->• isiátama 'hard head'
yámato + támasii yamatodámasii 'Japanese soul'
onná + kokóro ->• onnagókoro 'women's heart'
kita + amerika ->• kitaámerika 'North America'
From this brief review, it is evident that any formal analysis must reflect
these three properties in some way or other.
However, it is also the case that they do not always hold well as excep-
tion-free generalizations, since they often conflict with one another. For
instance, nagoyási 'Nagoya City' and rondónbasi 'London Bridge' do not
bear accent on N2, favoring the factors in (2) and (3) over the one in (1).
Conversely, such examples as bitámin + sii bitaminsíi 'vitamin C' and
biggu + bén biggubén 'Big Ben' carry over the lexical accent of their
head nomináis, preferring (1) to (2) and (3). In this sense, these three gen-
eralizations seem to hold true in some cases but to have exceptions in oth-
ers, and it matters in what case one factor is dominant over the others.
Thus, our task is to explicitly show how these characteristics can be cap-
tured by violable, hopefully general and well-motivated constraints and
how such constraints are ranked with respect to one another. Needless to
say, OT will be a promising framework for giving a principled account of
Japanese accent, which has long been considered too complicated to cap-
ture in a rule-based framework where any exceptions or violations to spe-
cific generalizations are not allowed, or constraints are inviolable.
2.2. The status of parsability among other factors
There are, of course, other factors in computing compound accent, but the
above three are particularly important and crucial, as is clear from Kubo-
zono's (1995, 1997) analysis. He proposes the constraints and associated
ranking relations in (4) and (5) to account for the accent distribution of
compounds with either long or short heads.1 The generalizations in (1), (2),
and (3) correspond roughly to (4a), (4b), and (4d), respectively:
(4) The definition of constraints in Kubozono (1995,1997)

a. PARSE (accent):
Parse the lexical accent of N2 in compound nouns. 2
b . NON-FINALITY (Μ', Σ ' ) :
The accented mora and syllable must not be final in PrWd. 3
c . NON-FINALITY ( F ' ) :
The accented foot must not be final in PrWd.
d. ALIGN-CA:
Align the accent with the boundary between N1 and N2.
e . RIGHTMOSTNESS:
A peak of prominence (i.e. accent) lies at the right edge of PrWd.
(5) Constraint ranking

N O N - F I N A L I T Y (μ' σ ' ) > P A R S E ( a c c e n t ) > N O N - F I N A L I T Y ( F ' ) , A L I G N - C A > R I G H T M O S T N E S S
Obviously, Kubozono considers the avoidance of final accent at the mo-

rale and syllabic level (i.e. (4b)) as the most dominant, the parsing of the
N2 accent (i.e. (4a)) as less dominant, and the 'around-the-boundary' ac-
cent (i.e. (4d)) as the least dominant of the three. The tableau in (6) shows
that this constraint-based system works well for the basic data presented
above:
(6) Examples
NON-FINALITY PARSE NON-FINALITY ALIGN-CA RIGHTMOST

(accent) (F)
a. /miso + síru/
miso) + (sirú) **! * * *
"" miso) + (síru) * *
misó) + (siru) *! **
miso) + (siru) *! * ****

NON-FINALITY PARSE NON-FINALITY ALIGN-CA RIGHTMOST

(Μ\Σ·) (accent) (F)
b. /isí + atamá/
isi) + (a)(tamá) **! * *
isi) + (a)(táma) * *! *! *
isi) + (á)(tama) * **
isi) + (a)(tama) * ***!
isi) + (a)(tama) * *****

*!
c. /sinzyu + kái/
sinzyu) + (gài) *
*!
sinzyú) + (gai) * *
sinzyu) + (gai) * *! **
sinzyu) + (gai) * *! ***
d. /minami + amerika/
minami) + (ame)(riká) **1 * *
minami) + (ame)(ríka) *! *! *
minami) + (amé)(rika) *! **
minami) + (áme)(rika) ***
minami) + (ame)(rika) ****!
minami) + (ame)(rika) *! *******
As shown in (6a), compound accent basically carries over the lexical ac-
cent of N2 unless it is final at the moraic or syllabic level. If N2 is final-
accented at that level as in (6b, c) or unaccented as in (6d), compound ac-
cent falls on the 'around-the-boundary' syllable within the non-final
rightmost foot. Here and below, I am assuming that foot construction pro-
ceeds from right to left (Poser 1990 and Kubozono 1995, 1997) within a
single morpheme and not across a morpheme boundary indicated here as
+ (Tateishi 1985 and Tanaka 1992). The morphology-sensitive foot pars-
ing is motivated by such examples as nagoyá)+(si), but not *nagó)(ya+si)
'Nagoya City'.
Note in the above tableau that *sinzyugái is final-accented at the syl-
labic level and thus has a single violation of N O N - F I N A L I T Y (μ\ Σ ' ) while
*misosirú is final-accented at both the moraic and syllabic levels, causing
two violations of the constraint. N O N - F I N A L I T Y (F') is motivated by the re-
traction of accent to the penultimate foot in (6b, c), which clearly shows
evidence for metrical feet from the viewpoint of accentual phonology, as
compared to oft-cited evidence from prosodie morphology (Tateishi
1989, Poser 1990, and Itô, Kitagawa, & Mester 1996, among others).
RIGHTMOSTNESS (4e) requires that, other things being equal, accent be lo-
cated as far to the right as possible, so that unaccented words have as
many violations to (4e) as their number of syllables.4 Finally, it does not
matter here whether Japanese has iambic or trochaic feet, as will be dem-
onstrated in section 3.2., because FOOT TYPE is much less dominant (hence,
is not given here) in its ranking system. In fact, what we have called 'ac-
cent' so far is not the head of a foot but the head of a prosodie word.
In more general terms, this analysis regards accentual faithfulness as
quite crucial and amounts to saying something like (7):
(7) Kubozono's generalization on compound accent

Compound accent parses (i.e. carries over) the lexical accent of N2
unless it is final at some level of analysis (i.e. mora or syllable).
The advantages of this system are as follows. First, it is no longer neces-

sary for the computation of accent to distinguish between the short and
long head classes of compounds (cf. (3)) by a judicious use of ALIGN-CA. 5
Second, this is the first convincing attempt, as far as I know, to provide in-
sight into the accentual system of Japanese compounds in OT.6
3 . NON-FINALITY of accented feet
3.1. Other data
Indeed, there is no doubt that Kubozono's analysis is insightful and pio-

neering. However, an extensive and careful survey, which I will show lat-
er, seems to suggest another generalization. For expository purposes, let
me begin this section with my findings of the survey before showing evi-
dence for them. The generalization in (8) is one of such findings:7
(8) Revised generalization

Compound accent falls on the penultimate foot, except for the full
preservation of accent in foreign heads (or in some archaic native
and Sino-Japanese heads).
(8) implies that default accentuation (i.e. penultimate-foot accent) gener-

ally overcomes full preservation of input accent in the grammar of Japa-
nese. For example, it is obvious that the words in (6b-d) bear accent on
the penultimate foot. In technical terms, I am claiming that the constraint
ranking for general lexical classes should be something like (9a), except
for the cases of accent preservation in (9b) (The symbol '=' in (9a) indi-
cates 'free ranking,' which will be discussed in section 3.2):
(9) The constraint-ranking system for compound accent

(preliminary)
a. G e n e r a l c a s e s
NON-FINALITY (μ\ σ\ F') > MAX (accent) = ALIGN-L (σ\ root) > ALIGN-R (PrWd, σ')
b. F o r e i g n h e a d s ( o r s o m e archaic n a t i v e a n d s i n o - J a p a n e s e h e a d s )
MAX (accent) > NON-FINALITY (μ\ σ', F') > ALIGN-L (σ\ root) » ALIGN-R (PrWd, σ')
c. D e f i n i t i o n s of n e w c o n s t r a i n t s
MAX (accent):
The accent of a head root has a correspondent in the compound.
ALIGN-L ( σ \ root):
The left edge of any accented syllable is aligned with the left edge of a head root.
ALIGN-R (PrWd, a'):
The right edge of any PrWd is aligned with the right edge of an accented syllable.
Note in (9) that NON-FINALITY forms a unit as a constraint family, which

can be overranked by MAX (accent) for the cases of accent preservation.
I will show in section 5 that although Japanese once had the grammar of
(9b), MAX (accent) has subsequently undergone demotion to its ranking
in the present grammar of accentuation. Another crucial point is that the
rankings of (9a) and (9b) are based on the type of lexical class (i.e. general
vs. foreign).
Here, I introduce M A X (accent), A L I G N - L ( σ \ root), and A L I G N - R
(PrWd, σ'), which have different definitions and consequences from
Kubozono's PARSE (accent), A L I G N - C A , and RIGHTMOSTNESS, respective-
ly. First, following Tanaka ( 2 0 0 1 , to appear), I adopt M A X (accent), in-
stead of PARSE (accent), because the former constraint has a distinct effect
in accent preservation between Sino-Japanese and foreign words, which I
will not be concerned with in this paper. Another reason is that the lexical
accent of Nl, not that of N2, is preserved in mimetic words (góro /
górogoro / *gorogóro 'the manner of rolling'), whose head root I assume
is Nl. 8 Second, A L I G N - C A is replaced by A L I G N - L ( Σ 1 , root) for the exact
reason of the latter's usual single-edge definition. But the crucial differ-
ence between the two lies in the fact that the latter constraint is not vio-
lated by unaccented compounds (see section 4 . 1 . ) , since an accented syl-
lable is aligned in that way if there is any. The order of the two arguments
(i.e. categories) in the statement of A L I G N is crucial here (McCarthy and
Prince 1 9 9 3 ) , and it should be contrasted with A L I G N - R (PrWd, Σ ' ) , which
is always violated by unaccented patterns. Third, the reason for adopting
the final constraint, A L I G N - R (PrWd, Σ ' ) , instead of RIGHTMOSTNESS, is re-
lated to its edge-oriented alignment of two prosodie categories, gradient
nature of violations, and general applicability to many other languages

(Walker 1995,1996).
Now let us turn to several sets of arguments in favor of the ranking sys-
tem in (9a, b). First, even the penultimate accent, which would be parsed
in accordance with (5) as was the case with misosiru in (6a), often shifts
leftward to the penultimate foot in the case of native N2 words, as Kubo-
zono (1995: 35-36,1997: 280-82) himself admits:
(10) Native: N O N - F I N A L I T Y (μ\ o\ F ' ) > M A X (accent)

(Sometimes rerankable)
ningyo + hime -• ningyóhime 'mermaid princess'
nisiki + hébi -* nisikíhebi 'reticulated python'
kansou + háda kansóuhada 'dry skin'
kansou + négi kansóunegi 'dry leek'
densyo + háto ->• densyóbato 'carrier pigeon'
kóumori + kása -> koumorígasa 'Western umbrella'
nyúudou + kúmo nyuudóugumo 'thunderhead'
níwaka + áme ->· niwakáame 'sudden rainfall'
kasure + kóe ->• kasurégoe 'husky voice'
yakata + húne yakatábune 'houseboat'
garasu + ita -> garasúita 'glass board'
onná + kokóro ->· onnagókoro 'heart of girls'
yudé + tamágo ->• yudetámago 'boiled egg'
hidari + utíwa ->· hidariútiwa 'comfortable life'
kamí + omútu kamiómutu 'paper diaper'
mata + itóko -»· mataítoko 'second cousin'
Parsed accent is also acceptable in some compounds like densyobáto,

koumorigása, and hidariútiwa but not in others like *ningyohime, *kan-
souháda, and *onnagokóro. But it is an important fact here that histori-
cally, all variations have developed from the parsed penultimate accent to
the antepenultimate one in the non-final rightmost foot. In that sense, the
parsed accent as described by McCawley (1968: 168) sounds somewhat
archaic to younger Japanese speakers of the present generation.
The second argument is that in the case of Sino-Japanese compounds
in (11a), again, even the non-final penultimate accent is not parsed but
shifts leftward, although there are some rare cases, as in (lib), where the
final accent at the syllabic level is parsed: 9
(11) Sino-Japanese: NON-FINALITY (μ\ σ\ F ' ) » MAX (accent)

(Sometimes rerankable)
a. Non-final but shifted
mán + gétu mángetu 'full moon'
tyóu + bátu -• tyóubatu 'punishment infliction'
gài + kótu -»• gáikotu 'skeletal bone'
suidou + kyóku -> suidóukyoku 'waterworks bureau'
yoyaku + séki -• yoyakúseki 'reserved seat'
syuutyaku + éki -> syuutyakúeki 'terminal station'
syouka + éki ->· syoukáeki 'digestive fluid'
ningyou + géki ->• ningyóugeki 'puppet play'
kenkyuu + situ kenkyúusitu 'study room'
kekkon + síki -> kekkónsiki 'wedding ceremony'
ténsyu + káku tensyúkaku 'castle tower'
kougeki + ryóku --> kougekiryoku 'offensive power'
b. Final but parsed
mèi + syó -> meisyó 'place of scenic beauty'
kyúu + syó -»· kyuusyó 'vital spot'
sínzitu + mí • sinzitumí 'truth'
keisatu + syó ->• keisatusyó 'police office'
nihón + zín -> nihonzín 'Japanese people'
ísi + kái -»• isikái 'society of doctors'
kasi + kin ->• kasikín 'loan'
syóu + gén -> syougén 'verbal evidence'
bán + góu bangóu 'serial number'
kán + súu kansúu 'function'
syóu + tyúu syoutyúu 'distilled spirits'
zyóu + dán -> zyoudán 'joking matter'
sén + sèi ->• senséi 'teacher'
níti + hón ->• nippón 'Japan'
Note that in ( l i b ) , nihónzin, isíkai, and kasíkin are also acceptable vari-
ants and that other words also bear accent on the non-final rightmost foot
when they undergo further compound formation as in mokugekisyóugen
'eyewitness's evidence,' denwabángou 'phone number,' sankakukánsuu
'trigonometrical function,' and so on. Thus, it can be said that (8) holds
true generally for compounds with Sino-Japanese heads.
Third, in the case of foreign compounds, however, we must admit that

they carry over their N2 lexical accent even if it is final:10
(12) Foreign: M A X (accent) > NON-FINALITY (μ\ σ\ F') (almost fixed)

a. Final but parsed
káfe + báa kafebáa 'cafe bar'
eiga + fán ->• eigafán 'movie fan'
bitámin + sii ->• bitaminsíi 'vitamin C'
sáakuru + kée • saakurukée 'circle Κ'
táfu + gài ->• tafugái 'tough guy'
bésuto + tén ->• besutotén 'best 10'
hánii + bíi ->· haniibíi 'honey bee'
wáido + byúu waidobyúu 'wide view'
bíggu + bén ->• biggubén 'Big Ben'
suizyou + sukíi -»• suizyousukíi 'water skiing'
héyaa + buróo ->• heyaaburóo 'hair blow'
kurisúmasu + turíi kurisumasuturíi 'Christmas tree'
táimu + masin -»• taimumasin 'time machine'
márizzi + burúu -• marizziburúu 'marriage blues'
súupaa + dorái -> suupaadorái 'super dry'
tákkusu + furíi ->• takkusufuríi 'tax-free'
hairáito + síin -> hairaitosíin 'highlight scene'
rozétta + sutóon ->· rozettasutóon 'Rosetta stone'
b. Non-final and parsed
róosu + hámu roosuhámu 'roast ham'
siifúudo + píza -»• siifuudopíza 'seafood pizza'
Kúrobe + dámu -+ Kurobedámu 'Kurobe dam'
tennen + gásu ->· tennengásu 'natural gas'
báioretto + fízu -> baiorettofízu 'violet fizz'
máikuro + básu -> maikurobásu 'microbus'
dorággu + suóa doraggusutóa 'drug store'
furáido + póteto -> furaidopóteto 'fried potato'
c. Final and shifted
súupaa + mán -»• suupáaman 'superman'
sáin + pén -* saínpen 'pen for signature'
furansu + pán -» furansúpan 'French bread'
rémon + tíi -> remóntii 'lemon tea'
wáido + syóu -> waidósyou 'wide show'
áisu + koohíi -> aisukóohii 'ice coffee'
The compounds given in (12a) conform to what Kubozono (1997: 280)

calls the 'snack-bar' pattern, where the original final accent is preserved.
Some cases in (12a) allow such variations as suizyousúkii, heyaabúroo,
and kurisumasutúrii, but (12a) and (12b) generally do not undergo accent
shift even when they are compounded with another word (unlike the
Sino-Japanese cases in (lib)), as in koukyuukafebáa 'high-class cafe bar,'
koukyuuroosuhámu 'high-quality roast ham,' and the like. The examples
in (12c) are nativized loans whose accent falls on the penultimate foot,
but such types are observed less frequently. It thus follows that the un-
marked pattern for foreign compounds parses lexical accent regardless of
its finality.
Next, the following three pieces of evidence for the ranking in (9a) con-
cern quadrimoraic heads whose lexical accent is originally non-final but
moves leftward when compound formation applies to them. For example,
the penultimate accent in (13a) shifts leftward obligatorily without varia-
tion (cf. (10), (lia), and (12b)). The antepenultimate accent of quadrimo-
raic heads in (13b) usually exhibits variation when they form compounds,
but in either case compound accent appears within the penultimate foot. 11
Moreover, compounds with unaccented quadrimoraic heads as in (13c)
always receive the same accent as (13a):
(13) Quadrimoraic heads: NON-FINALITY (μ', a', F ' ) » MAX (accent)

(fixed)
a. Penultimate
óo + namekúzi • oonámekuzi 'big slug'
dénki + nokogiri ->· denkinókogiri 'power saw'
dénki + kamisóri denkikámisori 'electric shaver'
tukaisute + tiritóri tukaisutetíritori 'throwaway dustpan'
siharai + matigáe ->· siharaimátigae 'payment mistake'
sóosu + yakisóba -»· soosuyákisoba 'chow mein with sauce'
siróuto + kangá siroutokángae 'naive idea'
yunyuu + houbéni yunyuuhóubeni 'imported cheek rouge'
sokuseki + misosíu sokusekimíosiru 'quickly-cooked miso soup'
b. Antepenultimate
neri + hamígaki nerihámigaki / nerihamígaki 'toothpaste'
áka + murásaki -»· akamúrasaki / akamurásaki 'red purple'
dendou + habúrasi ->· dendouháburasi / dendouhabúrasi 'electric toothbrush'
yunyuu + kudámono -> yunyuukúdamono / yunyuukudámono 'imported fruits'
irò + origami • iroórigami / iroorígami 'colored folded paper'
kínu + oirímono kinuórimono / kinuorímono 'silk fabrics'
ríron + tetúgaku rirontétugaku / rirontetúgaku 'theoretical philosophy'
síro + hatímaki sirohátimaki / sirohatímaki 'white headband'
koukuu + rikígaku -»• koukuuríkigaku / koukuurikígaku 'aerodynamics'
kanyou + syokúbutu -> kanyousyókubutu / kanyousyokúbutu 'foliage plant'

teami + kutúsita teamikútusita / teamikutúsita 'hand-made knitted socks'
síro + hukúroo - • sirohúkuroo / sirohukúroo 'snowy owl'
c. Unaccented
kita + amerika ->· kitaámerika 'North America'
sin + yokohama -»· sinyókohama 'New Yokohama'
óo + hurosiki oobúrosiki 'big cloth wrapper'
nisi + afurika ->• nisiáfurika 'West Africa
óo + kuwagata ookúwagata 'big stag beetle'
kuti + yakusoku ->· kutiyákusoku 'verbal promise'
youhuu + izakaya -»· youhuuizakaya 'Western pub'
zikasei + tukemono -> zikaseitúkemono 'hand-made pickles'
tukimi + kisimen • tukimikisimen 'flat noodles with egg'
giri + omiai giriómiai 'arranged marriage from a sense of duty'
binbou + gakusei ->• binbougákusei 'poor student'
yokoku + satuzin ->• yokokusátuzin 'announced murder'
sángyou + kakumei • sangyoukákumei 'Industrial Revolution'
roudou + kumiai ->• roudoukúmiai 'labor union'
kázan + katudou ->· kazankátudou 'volcanic activity'
gódan + katuyou ->· godankátuyou 'five-grade conjugation'
rentai + sekinin rentaisékinin 'collective responsibility'
kokuyuu + tetudou ->• kokuyuutétudou 'national railway'
kinen + satuei kinensátuei 'commemorative photographing'
In general, unaccented words seem to tell us more clearly how com-

pounds undergo unmarked (i.e. default) accentuation, since they are neu-
tral or transparent in the sense that they are not sensitive to previous ac-
centuation (i.e. irrelevant to parsability). This is also true for other
unaccented heads with fewer than four moras:
(14) Parallelism of output forms
Penultimate-Accented Unaccented
yakatá) + (bune) 'houseboat' vs. abaré) + (usi) 'spirited cow'
onna) + (gó)(koro) 'women's heart' vs. yama) + (zá)(kura) 'mountain cherry tree'
denki) + (nóko)(giri) 'power saw' vs. sin) + (yóko)(hama) 'New Yokohama'
In (14), the words in the left column are originally penultimate-accented

(i.e. húne 'boat,' kokóro 'heart,' and nokogiri 'saw,' while the ones in the
right are unaccented heads. Yet both types of compounds have their ac-
cent on the same non-final rightmost foot. Thus, unaccented heads seem
to reflect the unmarked pattern of compound accent.
Finally, let us reexamine Kubozono's (1995,1997) proposed ranking in
(5), repeated here as (15a), in terms of its reranking possibilities which
would be necessary to account for the data so far:
(15) Kubozono's reranking of constraints

a. N O N - F I N A L I T Y ( μ \ o') » PAUSE ( a c c e n t ) » NON-FINALITY ( F ' ) , A L I G N - C A > . . . f o r ( 6 )
b. PARSE promotion: PARSE (accent) » NON-FINALITY (μ\a', F1), ALIGN-CA ... for (lib) and (12a)
C . PARSE d e m o t i o n : NON-FINALITY (μ\σ\ F ' ) > PARSE (accent), ALIGN-CA ... for (10), (11a), and (13a)
D . PARSE d e m o t i o n : N O N - F I N A L I T Y ( « ' , σ ' ) > N O N - F I N A L I T Y ( F ' ) , A L I G N - C A > PARSE ( a c c e n t ) . . . f o r ( 1 3 b )
(15a, b) and (15c) would be needed for preserved accent and penultimate-
foot accent, respectively. In such parsed-accent compounds as meisyó
'place of scenic beauty,' kyuusyó 'vital spot,' and sinzitumi 'truth' in ( l i b ) ,
the reranking of N O N - F I N A L I T Y ( M ' ) » P A R S E (accent) » N O N - F I N A L I T Y (σ1,
F') (instead of (15b)) would be useless, because their accent is final at the
moraic level (note 9). In general terms, the reranking system in (15) seems
to be more complicated than the one proposed in (9a, b) in that (15a)
would have to be reranked in the three ways of (15b-d). Furthermore, the
reranking in (15d) would involve some 'non-local' process, as shown in
(16a, b), although the optimal accent in (16b) is actually a very productive
pattern. The local reranking of N O N - F I N A L I T Y ( F ' ) » P A R S E (accent),
A L I G N - C A in (16c), however, would have only the same result as (16a):
(16) Reranking for (13b)
NON-FINALITY PARSE NON-FINALITY

a. /neri + hamígaki/ ALIGN-CA RIGHTMOST
(accent) (F1)
neri) + (hami)(gakí) **! * * *

* *
neri) + (hami)(gáki) *
*! *
**
neri) + (hamí)(gaki) ***
neri) + (hámi)(gaki) *! ****
neri) + (hami)(gaki) *!
NON-FINALITY NON-FINALITY PARSE
b. /neri + hamígaki/ ALIGN-CA RIGHTMOST
(μ\σ·) (F) (accent)
neri) + (hami)(gaki) **ι * * *
*!
neri) + (hami)(gáki) * *
*!
**
neri) + (hami)(gaki) *! *
***
neri) + (hámi)(gaki) *
neri) + (hami)(gaki)
NON-FINALITY NON-FINALITY PARSE

c. /nerí + hamígaki/ ALIGN-CA RICHTMOST
(μ\ο') (F) (accent)
neri) + (hami)(gakí) **T * * *
neri) + (hami)(gáki) * * *
*!
neri) + (hamí)(gaki) * **
neri) + (hámi)(gaki) *
nerí) + (hami)(gaki) *
In contrast to (15), I will explicitly show in the next section that the new
ranking system proposed in (9), which involves only one minimal rerank-
ing, is both necessary and sufficient to account for the data in (6) and (10)-
( 1 3 ) . Furthermore, I will present some evidence for ALIGN-L ( o \ root),
which plays a vital role particularly in accounting for the accentual behav-
ior of such quadrimoraic words as those given in (13).
3.2. Analysis and Iambic/Trochaic alternation
Now let us summarize the patterns of compound accent in terms of the

number of syllables and the accent position of head nomináis:12
(17) Possible patterns and variations (accented)
syllable input accent heads compounds output accent
1 — si nagoyási 'Nagoya City' penultimate foot

— b á a (foreign) kafebáa 'cafe bar' parsed
unaccented ko tinomígo 'nursing baby' penultimate foot
2 final así sinobíasi 'stealthy steps' penultimate foot

initial áme niwakáame / niwakaáme varied
'sudden rainfall'
básu (foreign) maikurobásu 'microbus' parsed
unaccented usi abaréusi 'spirited cow' penultimate foot
3 final atamá isiátama 'hard head' penultimate foot

medial utíwa hidariútiwa / hidariutíwa varied
'comfortable life'
sutóa ( f o r e i g n ) doraggusutóa 'drug store' parsed
initial árasi yamaárasi 'mountain storm' penultimate foot
póteto (foreign) furaidopóteto 'fried potato' parsed
unaccented sakura yamazákura penultimate foot
'mountain cherry tree'
syllable input accent heads compounds output accent
4 final yorokobi nukayórokobi 'premature joy' penultimate foot

penultimate nokogiri denkinókogiri 'power saw' penultimate foot
i g u z á m u (foreign) taamuiguzámu 'term exam' parsed
antepen- hamigaki nerihámigaki / nerihamígaki penultimate foot
ultimate 'toothpaste'
toráburu (foreign) zyuumintoráburu parsed
'trouble with neighbors'
initial kámakiri ookámakiri 'big mantis' penultimate foot
bízinesu (foreign) biggubizinesu 'big business' parsed
unaccented yokohama sinyókohama 'New Yokohama' penultimate foot
In (17), all the parsed-accent cases are compounds with foreign heads and
can thus be captured by (9b), whereas varied cases with native heads are
accounted for by the free ranking of either (9a) or (9b). The varied cases
with Sino-Japanese heads whose accent can exceptionally be parsed as in
nihón + ζ in ->• nihonzín 'Japanese people' and kityou + mén -> kityoumén
'methodical nature' (i.e. ( l i b ) ) are also captured by (9b). In addition,
some nativized foreign compounds in (12c), such as súupaa + mán -*
suupáaman 'superman' and sáin + pén > saínpen 'pen for signature,' sug-
gest that they are assimilated to the usual ranking of (9a). Thus, given (9),
the analysis involves only one minimal reranking to explain the data
above, as in (18)-(20). Note that the symbol '=' in (19) indicates that the
former two constraints concerned can be ranked freely, either as in (9a)
or (9b):
(18) Foreign (parsed)
MAX (accent) NON-FINALITY (μ', o \ F ' ) ALIGN-R (PrWd, Σ')
a. /káfe + báa/
kafe) + (báa) **
kafé) + (baa) *! *
b. /máikuro + básu/
maikuro) + (básu) * *
**
maikuró) + (basu) *!
c. /dorággu + sutóa/
doraggu) + (su)(tóa) * *
**
doraggu) + (sú)(toa) *!
(19) Some archaic native and Sino-Japanese (varied)
NON-FINALITY (μ\ a', F ' ) = MAX (accent) ALIGN-R (PrWd, a')

a. /níwaka + áme/
niwaka) + (áme) * *
* **
niwaká) + (ame)
b. /hidari + utiwa/
hidari) + (u)(tiwa) * *
hidari) + (Ú)(tiwa) * **
c. /nihón + zín/
**
nihon) + (zín)
nihón) + (zin) * *
(20) General and nativized foreign (penultimate foot)
N O N - F I N A L I T Y ( μ \ o', F') MAX (accent) ALIGN-R (PrWd, σ')

a. /nágoya + sí/
nagoya) + (sí)
* *
nagoyá) + (si)
b. /abare + usi/
*
abare) + (úsi) *!
**
abaré) + (usi)
c. /yamá + árasi/
yama) + (á)(rasi) **
***
yamá) + (a)(rasi) *!
d. /súupaa + mán/
suupaa) + (mán)
suupáa) + (man) * *
It is worth pointing out that the general dominance of N O N - F I N A L I T Y (a',

O\ F') over MAX (accent) correctly excludes such impossible alternations
as in honé / *abarabóne / abarábone 'rib bones,' utá / *warabeúta /
warabéuta 'children's song,' atamá / *isiatáma / isiátama 'hard head,' and
otokól *yamaotóko / yamaótoko 'woodsman,' where the final accent in
the input never retracts to the penultimate mora and syllable in the out-
put. This is because penultimate accent violates N O N - F I N A L I T Y ( F ' ) .
In (18)-(20), I have not shown how each of the examples satisfies or vi-
olates A L I G N - L (σ1, root) because it is not crucial for these cases. Howev-
er, its importance is seen for the cases with quadrimoraic heads given in
(17) (see also (13)). It is true that their accent uniformly falls on the pen-
ultimate foot, but it still remains unclear why it doesn't falls on the ante-
penultimate but the pre-antepenultimate mora as in *nuka) + (yoró)(ko-
bi) and *denki) + (nokó)(giri), as compared to (a) (bará) + (bone) /
*a)(bára) + (bone) 'rib bones' and oya)(yubí) + (hime) / *oya)(yúbi) +
(hime) 'thumb princess.' In other words, the following two questions re-
main to be explained; that is, 1) why only quadrimoraic heads exhibit tro-
chaic pattern, while all others are iambic due to A L I G N - R (PrWd, o{) and,
moreover, 2) why only words with the original antepenultimate accent
(i.e. (13b)) exhibit an iambic/trochaic variation as in neri) + (hámi)(gaki)
/ (neri) + (hamí)(gaki). (21) clearly demonstrates that both of these are
given a convincing account by the interactive effect of MAX (accent) and
A L I G N - L ( o \ root):
(21) Quadrimoraic heads (penultimate foot)
NON-FINALITY ALIGN-R
MAX (accent) = ALIGN-L (σ\ root)
(μ\σ\Γ) (PrWd, a-)
a. /nuka + yorokobi/
nuka) + (yoró)(kobi) **
***
nuka) + (yóro)(kobi) *
b. /dénki + nokogíri/
denki) + (nokó)(giri) * *1 **
denki) + (nóko)(giri) *
c. /sin + yokohama/
sin) + (yokó)(hama) *! **
sin) + (yóko)(hama)
d. /neri + hamígaki/
"" neri) + (hamí)(gaki) * **
***
"" neri) + (hámi)(gaki) *
Note here that M A X (accent) and A L I G N - L ( o \ root) are ranked freely, as

was explicitly shown in (9a), which means that only words with the origi-
nal penultimate accent as in (21d) allow variation of either the parsed op-
tion or the left-edge option. The free ranking, of course, does not affect
the optimality of the latter (or second) candidate in each pair of (21a-c).13
This characteristic of attracting compound accent to the left edge of a
head root is also clear from the following contrast of Sino-Japanese ex-
amples:
(22) Minimal pairs
tetú) + (gaku) 'philosophy' vs. riron) + (tétu)(gaku) 'theoretical philosophy'

riki) + (gaku) 'dynamics' vs. koukuu) + (riki)(gaku) 'aerodynamics'
syokú) + (butu) 'plant' vs. kanyou) + (syóku)(butu) 'foliage plant'
gakú) + (mon) 'learning' vs. mimi) + (gáku)(mon) 'ear learning'
sakú) + (motu) 'crops' vs. nou) + (sáku)(motu) 'farm products'
The compounds in the left column originally bear accent before the
boundary due to NON-FINALITY (F'), but it shifts to the left within the pen-
ultimate foot, attracted to the new boundary, when they undergo further
compound formation as is shown in the right column (although the parsed
option is also possible as in (21d)). As a result, compound accent might
appear to be assigned either before or after the boundary as Kubozono's
A L I G N - C A defines, but this proves to come from the effect of NON-FINAL-
ITY (F ) and the usual single-edge alignment A L I G N - L (Σ 1 , root). On the
1
other hand, the surface ambiguity of either iambic or trochaic rhythm in

(21d) is an epiphenomenon of the synergistic effect and reranking of MAX
(accent) and A L I G N - L (a', root).
4. The de-accentuation phenomena
4 . 1 . NON-FINALITY of accented prosodie word
So far I have argued for the ranking system of NON-FINALITY (μ1, σ\ F') »
M A X (accent), in place of Kubozono's NON-FINALITY ( μ ' , o') » PARSE (ac-
cent) » NON-FINALITY (F'), on the basis of the varieties of possible pat-
terns and variations of accented compounds, and indeed, the data I have
presented thus far seem to favor the proposed ranking with respect to the
present state of Japanese grammar. However, I demonstrate in this sec-
tion that the most compelling and convincing evidence in support of this
ranking can be found when we attempt to account for such unaccented
compounds as those given in (23):14
(23) Unaccented compounds

a. Final-accented inputs
nihón + gó ->• nihongo 'the Japanese language'
riron + ká ->• rironka 'theorist'

éfferu + tóu ->• efferutou 'the Eiffel Tower'
hébii + kyúu hebiikyuu 'heavy weight'
kesyóu + sui kesyousui 'toilet lotion'
syokuzi + tyúu ->• syokuzityuu 'at table'
unebi + yamá -»· unebiyama 'Mt. Unebi'
gárasu + tamá garasudama 'glass ball'
b. Quadrimoraic outputs
kínu +ito kinuito 'silk thread'
dasí + siru ->• dasiziru 'soup stock'
nía + mísu -»· niamisu 'near miss'
dokú + gásu -»• dokugasu 'poisonous gas'
on + in ->· onin 'sound and rhythm, phonology'
mizu + hana -> mizubana 'running nose'
usu + azi -> usuazi 'light taste'
náka + niwa nakaniwa 'courtyard'
dán + zétu ->• danzetu 'extinction'
sóku + báku -»· sokubaku 'restraint'
sétu + zóku -»• setuzoku 'connection'
kótu + sétu ->· kossetu 'bone fracture'
c. Both conditions met (Sino-Japanese)

kóu + sài kousai 'social intercourse'
róu + dóu -> roudou 'manual labor'
kán + kóu -• kankou 'sightseeing'
kyóu + sèi kyousei 'compulsion'
hán + zái ->• hanzai 'criminal guilt'
kái+ makú ->· kaimaku 'opening screen'
sán + myakú ->• sanmyaku 'mountain range'
zéi + nikú ->· zeiniku 'superfluous flesh'
hukú + dokú hukudoku 'taking poison'
ume + gumi -»• umegumi 'Urne group'
As in (23a), final-accented heads often trigger de-accentuation in com-

pounds (cf. Kubozono 1995: 26-27,1997: 276). In addition, quadrimoraic
compounds also tend to undergo de-accentuation, as in (23b), even if
their N2 is not final-accented. (23c) shows the cases of Sino-Japanese
compounds, where both requirements are satisfied. There is a great vari-
ety of unaccented compounds in Sino-Japanese, as well as those accented
on the penultimate foot, because its typical word form (more strictly, its
minimal word) is quadrimoraic with two roots as in (23c). Note that none
of the compounds in (23a-c) have a pitch fall anywhere in their domain,
which means that they seriously violate A L I G N - R (PrWd, Σ ' ) . Thus, my
task is to explain how such an unaccented pattern is possible at all in the
grammar of Japanese accentuation by taking advantage of the constraint-
ranking system for all the patterns seen in the preceding sections.
I propose that the key to answering this question lies in the constraint
of NON-FINALITY. Specifically, it is quite natural that Japanese grammar
has not only N O N - F I N A L I T Y (JU\ A\ F ' ) but also N O N - F I N A L I T Y (PrWd') as
a member of the N O N - F I N A L I T Y family, since the inventory of prosodie cat-
egories contains not only mora, syllable, and foot, but also prosodie word.
If we add this constraint to the ranking I have argued for, it should be lo-
cated below M A X (accent), as shown in ( 2 4 ) : 1 5
(24) The constraint-ranking system for compound accent (revised)
ALIGN-L (σ\ root) NON-FINALITY (PrWd')
ALIGN-R (PrWd, σ')
Here, the dotted lines indicate the rerankable relations of MAX (accent)
to the other three constraints: the reranking of (24a) is for foreign heads
and some archaic native and Sino-Japanese heads as seen in (18) and (19);
(24b) is for quadrimoraic heads with the antepenultimate accent as (21);
and I am claiming that (24c) is exactly what is needed for the unaccented
compounds shown in (23).
Following the definition of (4b, c), we define the requirement of NON-
FINALITY (PrWd') in (25a). Note, however, that this definition logically
implies that accent must not be present in PrWd, precisely because any
compound in Japanese forms a single PrWd of its own and hence it is al-
ways final in its domain, which is contrastive with English compounds

with two PrWd's as in (25c). A comparison with other members of the
N O N - F I N A L I T Y family, schematized in (25b), will make it easier to under-
stand what I mean:
(25) The mechanism of de-accentuation

a. Definition of N O N - F I N A L I T Y (PrWd')
The accented prosodie word must not be final in PrWd (Accent
must not be present in PrWd).
b. The family of N O N - F I N A L I T Y constraints
Mora: *[...μμμ'] [...μ μ'μ]
Syllable: *[...σσσ'] ->• [...σσ'α\
Foot: *[... F F F'] [...FF'F]
Prosodie Word: *[PrWd'] - [PrWd]
c. Morphological structure of compounds
PrWd' PrWd PrWd PrWd
/ \ / \ / \ / \
Root Root Root Root PrWd PrWd' PrWd' PrWd
*[[ousaká] [ben]] -> [[ousaka] [ben]] *[[black] [board]] [[bláck] [board]]

O s a k a dialect'
(and also A L I G N - R (PrWd, Σ ' ) ) prohibits the fi-

N O N - F I N A L I T Y ( μ ' , σ', F ' )
nal accent and requires it to fall on the penultimate constituent (in the
case of mora, syllable, and foot, two or more constituents are always avail-
able in the domain of a compound), whereas a violation of N O N - F I N A L I T Y
(PrWd') causes accent loss since there is no candidate constituent avail-
able in the domain. In section 2.2., I mentioned the way of foot construc-
tion which does not cross a morpheme boundary of a compound. This as-
sumption does not pose a problem with a 'single-PrWd hypothesis' here,
because the morpheme boundary in question is the one between the two
roots in (25c).16 On the other hand, any English compound involves more
than one PrWd, and compound accent falls on the initial PrWd. This dif-
ference implies that the members of a compound are more mutually in-
dependent in English than in Japanese, which is motivated by the fact that
even the final PrWd in English is assigned secondary accent, as in bláck
board.
In this way, the unaccented compounds in (23) are accounted for as typ-
ical cases of (24c) by minimally demoting M A X (accent) below N O N - F I -
NALITY (PrWd'):
(26) Unaccented compounds

NON-FINALITY ALIGN-R
MAX (accent) = ALIGN-L ( σ \ root)
(μ\ σ\ F', PrWd') (PrWd, σ')
a. /nihón + gó/
[nihon) + (gó)] PrWd
[nihón) + (go)] 1 ™" *! * * *
**
HE **
[nihon) + (go)]*™ *! *
^ [nihon) + (go)] PrWd
b. /on +in/
[on) + (in)] PrWd
[ón) + (in)] PrWd *! * *
**
[on) + (in)] PrWd
c. /yáku + sóku/
[yaku) + ( s ó k u ) ] ^ *
**
[yakú) + (soku)] PrWd *! * *
* **
[yáku) + (soku)] PrWd *! *
[yaku) + (soku)] PrWd
Thus, the de-accentuation phenomenon is a natural consequence of the

dominant ranking of the whole NON-FINALITY family over M A X (accent), in
which NON-FINALITY (PrWd') plays a crucial role among others, of course.
And this is the very reason why unaccented patterns are possible in the gen-
eral accentual system of Japanese.17 The accent loss in PrWd is carried out
at the cost of violations to the general constraint of 'culminativity,' which re-
quires that any word has one and only one accent somewhere in its domain
(Hayes 1 9 9 5 ) . In this system, the effect of culminativity works out as A L I G N -
R (PrWd, &), which is outranked by the other constraints.
We now see that even if we adopted NON-FINALITY (PrWd 1 ) into Kubo-
zono's ( 1 9 9 5 , 1 9 9 7 ) ranking in ( 5 ) , in a way like NON-FINALITY ( μ \ Σ ' ) »
1
PARSE (accent) » NON-FINALITY ( F 1 ) , A L I G N - C A » NON-FINALITY (PrWd ),
it would be very difficult to give a principled account for the de-accentu-
ation phenomena, because PARSE (accent) would take a rather long step
below NON-FINALITY (PrWd 1 ), even though more and more compounds
tend to undergo de-accentuation, as will be discussed in the next section.
The reason that A L I G N - C A would have to outrank N O N - F I N A L I T Y (PrWd')
is clear from the following example:
(27) ALIGN-CA » NON-FINALITY (PrWd'): sinzyúgai 'pearl'

a. /sinzyu + kái/ ALIGN-CA NON-FINALITY (PrWd')
sinzyú) + (gai) *
sinzyu) + (gai) *
*!
sinzyu) + (gai) *!
b. /sinzyu + kái/ NON-FINALITY (PrWd') ALIGN-CA
sinzyú) + (gai) *!
sinzyu) + (gai) *! *
*
sinzyu) + (gai)
This is because A L I G N - C A rules out an de-accented output candidate, and

this point is crucially different from our A L I G N - L ( σ \ root). 18 In contrast,
our approach is fairly easy to capture de-accentuation of compounds by
the minimal reranking of (24c).
4.2. Possible variations and change in progress
If NON-FINALITY (PrWd') and M A X (accent) are actually present at the po-

sition of (24c) in the constraint ranking, then the reranking possibilities of
MAX (accent) can be summarized in the following way:19
(28) Reranking possibilities

MAX (accent)
NON-FINALITY (PrWd') (c)
When MAX (accent) is located at the (28a) position, the ranking produces
the parsed-accent patterns of foreign heads like (18) and some archaic na-
tive and Sino-Japanese heads like (19); when it is at the (28b) position, the
ranking captures the most productive patterns in (20) and (21), in which
accent falls on the penultimate foot; and when it is at the (28c) position,
the ranking accounts for the unaccented compounds in (26). Since I as-
sume that NON-FINALITY ( μ \ σ\ F ' ) is just a single constraint, M A X (ac-
cent) cannot be ranked in between the two of NON-FINALITY (μ', σ\ F ' ) ·
The three possibilities in (28), then, also mean that logically, there should
be four types of accent variation as shown below:
(29) Possible variations

a. (28a) or (28b): Parsed or penultimate poot
b. (28b) or (28c): Penultimate foot or unaccented
c. (28a), (28b), or (28c): Parsed, penultimate foot, or unaccented
d. (28a) or (28c): Parsed or unaccented
Interestingly, these four types of variation are verified by the following

sets of examples:
(30) Examples of possible variations

a. Parsed / penultimate foot
densyo+ háto densyobáto / densyóbato 'carrier pigeon'
kóumori + kása • koumorigása / koumorígasa 'Western umbrella'
níwaka +áme niwakaáme / niwakáame 'sudden rainfall'
hidari + utíwa • hidariutíwa / hidariútiwa 'comfortable life'
kamí + omútu ->• kamiomútu / kamiómutu 'paper diaper'
nihón + zín ->· nihonzín / nihónzin 'Japanese people'
bésuto + tén ->• besutotén / besutóten 'best 10'
suizyou + sukíi suizyousukíi / suizyousúkii 'water skiing'
héyaa + buróo * heyaaburóo / heyaabúroo 'hair blow'
kurisúmasu + turíi ->· kurisumasuturíi / kurisumasutúrii 'Christmas tree'
b. Penultimate foot / unaccented
námida + mé -> namidáme / namidame 'tearful eye'
táion + kéi tainókei / taionkei 'thermometer'
kánsya + zyóu -> kansyázyou / kansyazyou 'letter of thanks'
suihei + sén suihéisen / suiheisen 'horizontal line'
ákita + inú akitáinu / akitainu 'Akita dog'
sóra + mimi -»· sorámimi / soramimi 'mishearing'
mawari + miti -> mawarímiti / mawarimiti 'roundabout way'
ki + kokóro -*• kigókoro / kigokoro 'disposition'
ni + sakana ->• nizákana / nizakana 'fish boiled with soy'
c. Parsed / penultimate foot / unaccented
sínzitu + mí • sinzitumí / sinzitúmi / sinzitumi 'truth'
kityou + mén -»• kityoumén / kityóumen / kityoumen 'methodical nature'
kasi + kin > kasikín / kasíkin / kasikin 'loan'
sekí + hán sekihán / sekíhan / sekihan 'rice boiled with red beans'
kubi + nekkó • kubinekkó / kubinékko / kubinekko 'scruff of the neck'
katí + tokí -»· katidokí / katídoki / katidoki 'shout of victory'
orí + túru orizúru / orízuru / orizuru 'folded-paper crane'
háto + múgi hatomúgi / hatómugi / hatomugi 'adlay, oats'
d. Parsed / unaccented
sankou + syó -> sankousyó / sankousyo 'reference book'
keisatu + syó -»· keisatusyó / keisatusyo 'police station'
sáiban + syó ->· saibansyó / saibansyo 'court of justice'
sió + tokí -»· siodokí / siodoki 'favorable tide, good chance'
wáru + tié ->• waruzié / waruzie 'serpentine wisdom'
wanpaku + monó ->· wanpakumonó / wanpakumono 'naughty boy'
warai + kóe waraigóe / waraigoe 'laughing voice'
wari + hási -> waribási / waribasi 'half-split chopsticks'
The 'local' variations in (30a-c) are captured as the minimal reranking (pro-
motion and/or demotion) of MAX (accent) and taken to be strong evidence
for its position between NON-FINALITY (μ1, Σ 1 , F11 ) and NON-FINALITY (PrWd ' )
in the present grammar (cf. (5)). The mimetics in (31) and the acronyms in
(32) also exhibit the same variation as (30b) and can be seen as arguments
for the position of MAX (accent) adjacent to NON-FINALITY (PrWd'): 20
(31) Mimetics (reduplication) with 'manner' and 'state' readings

a. Penultimate foot
run ->· rúnrun (to) 'cheerfully'
gyúu gyúugyuu (to) 'squeezingly'
bèta ->• bétabeta (to) 'stickily'
kira kirakira (to) 'brilliantly'
húra ->• húrahura (to) 'unsteadily'
gótya gótyagotya (to) 'promiscuously'
b. Unaccented
run -»• runrun (ni) 'cheerful state'
gyú gyuugyuu (ni) 'squeezing state'
bèta ->• betabeta (ni) 'sticky state'
kira kirakira (ni) 'brilliant state'
húra -»• hurahura (ni) 'unsteady state'
gótya gotyagotya (ni) 'promiscuous state'
(32) Acronyms
a. Penultimate foot
patoróoru + káa patókaa 'patrol car'
zéneraru + sutoráiki ->· zenésuto 'general strike'
haráguti + syóusuke harásyou 'Shosuke Haraguchi (person name)'
hasimoto + ryuutarou hasíryuu 'Hashimoto Ryutaro (person name)'
b. Unaccented
páasonaru + konpyúutaa ->• pasokon 'personal computer'
sékusyaru + harásumento • sekuhara 'sexual harassment'
kimura + tákuya kimutaku 'Takuya Kimura (person name)'
toyókawa + étuzi toyoetu 'Etsuji Toyokawa (person name)'
On the other hand, the 'non-local' variations in (30d) might at first sight
appear to be a case of non-minimal reranking, but I claim that their
reranking is nothing but minimal, because NON-FINALITY (μ', σ\ F ' ) and
NON-FINALITY (PrWd') tend to behave uniformly as members of the NON-
FINALITY family; namely, the data in (30d) show that the family members
are rerankable as a whole with MAX (accent), although it is true that the
variations in (30d) are relatively rare cases.
Since compounds with parsed accent within the final foot are limited in
number (except for foreign heads, as stated in section 3.1.), we can safely
say that the patterns with accent on the penultimate foot and/or without
any accent are more dominant than their parsed-accent counterparts in
(30a, c, d). Hattori (1991, 1998, 1999) also concludes from her historical
survey of 821 nouns that the general accentual change has converged by
90% at the two patterns concerned, which are much more prominent
than other patterns in the present state of Japanese grammar. She found
that words with the penultimate-foot accent account for 31 % while unac-
cented words amount to 58 %, although her survey contains not only com-
pounds but also simplex nouns.21 In fact, there are many compounds
where unaccented patterns are more dominant than their penultimate-
foot or parsed counterparts. (33) exemplifies the dominance relation
among the three variable patterns based on the examples in (30b-d),
where the one indicated by '>' or '<' is considered now as "more appro-
priate as a pronunciation of Standard Japanese than the other(s)" accord-
ing to NHK's (1998: 9) Dictionary of Japanese Pronunciation and Accen-
tuation:
(33) Dominance
a. Penultimate foot / unaccented
taiónkei < taionkei 'thermometer'
kansyázyou < kansyazyou 'letter of thanks'
suihéisen < suiheisen 'horizontal line'
akitáinu > akitainu 'Akita dog'
sorámimi < soramimi 'mishearing'
mawarimiti < mawarimiti 'roundabout way'
kigókoro > kigokoro 'disposition'
nizákana > nizakana 'fish boiled with soy'
b. Parsed / penultimate foot / unaccented
sinzitumí < sinzitúmi < sinzitumi 'truth'
kityoumén > kityóumen < kityoumen 'methodical nature'
kasikín > kasikin < kasikin 'loan'
sekihán > sekíhan < sekihan 'rice boiled with red beans'
kubinekkó < kubinékko > kubinekko 'scruff of the neck'
katidokí > katidoki < katidoki 'shout of victory'
orizúru < orízuru > orizuru 'folded-paper crane'
hatomúgi > hatómugi < hatomugi 'adlay, oats'
c. Parsed / unaccented
sankousyó < sankousyo 'reference book'
keisatusyó > keisatusyo 'police station'
saibansyó < saibansyo 'court of justice'
siodoki < siodoki 'favorable tide, good chance'
waruzié < waruzie 'serpentine wisdom'
wanpakumonó < wanpakumono 'naughty boy'
waraigóe > waraigoe 'laughing voice'
waribási < waribasi 'half-split chopsticks'
These data show that in the near future, more and more compounds are
likely to become de-accented, which means that MAX (accent) is going to
demote below N O N - F I N A L I T Y (μ', o\ F ' , PrWd') as a direction of accent
change in progress.
In addition to de-accentuation of compounds I have been concerned
with so far, see also Tanaka (2001, to appear) and Tanaka & Yamane
(2000) for de-accentuation of lexical classes and their different applicabil-
ity and variation from compounds.
5. Ranking summary and theoretical implications
In this paper, I have argued that Japanese grammar at present has a sys-
tem of assigning compound accent to the penultimate foot, and that is
why there are some compounds or their variants which may either parse
their lexical accent or undergo de-accentuation, as was illustrated in (28).
In other words, M A X (accent) must now be located between N O N - F I N A L I -
1
TY (μ , σ', F 1 ) and N O N - F I N A L I T Y (PrWd'), which, however, tend to demote
below N O N - F I N A L I T Y (PrWd 1 ) as a recent change in progress. Thus, my
proposed system allows us to give a principled account for what patterns
and variations are possible in the accentual system of Japanese compounds
as well as why they can tend toward accent loss. Needless to say, the exist-
ence of N O N - F I N A L I T Y (PrWd 1 ) is crucial in the system.
Note also that the movement of MAX (accent) through the constraint
hierarchy plays two important roles in characterizing the grammar of Jap-
anese: it characterizes not only the synchronic accentual systems based
mainly on word classes but also their diachronic change of the core sys-
tem from the past, as summarized in (34) (DEP (accent) is also ranked
here for the reason stated below):
(34) Grammar of Japanese accentuation
Synchronic Diachronic
Ranking Head
MAX (accent) J <SR PARSED Past

Foreign
1
(Native and Sino-Japanese)
N O N - F I N A L I T Y ( μ \ σ\ F')
I
Ç MAX (accent) J) «· PENULTIMATE FOOT Present

Native, Sino-Japanese, and Mimetics
1 (Acronyms)
NON-FINALITY (PrWd')
1
MAX (accent) J is· UNACCENTED In Progress

Mimetics and Acronyms
1 (Native & Sino-Japanese)
ALIGN-L (σ\ root)
I
1
ALIGN-R (PrWd, σ')
(DEP (accent))
The diachronic characterization is based on the fact that some nativized

foreign compounds place their accent on the penultimate foot while ar-
chaic native and Sino-Japanese compounds preserve their lexical accent,
and also that more and more compounds are favoring unaccented forms
over their parsed and/or penultimate-foot variants.
It is true that the role of MAX (accent) is important in the ranking sys-
tem in that it characterizes both synchronic and diachronic aspects of the
grammar by moving its position through the other constraints. Now a fun-
damental question may be addressed, however, if we adopt the assump-
tion that only a member of the FAITHFULNESS family (here, MAX (accent))
can move through the hierarchy (cf. note 19): Why has MAX (accent) de-
moted from the periphery (i.e. foreign) to the core (i.e. other word classes)
or from the past to the present? This assumption stems from the following
two hypotheses: one is 'Ranking Invariance,' which states that in the un-
marked case, there is a single constraint ranking for the whole lexicon (Itô
& Mester 1995: 183); and the other is 'FAITHFULNESS Conversion,' which
states that a FAITHFULNESS constraint promotes or demotes through the
fixed constraint ranking so as to be unified with the other members in the
family. In the case of (34), MAX (accent) at the highest position demotes
through the otherwise invariant hierarchy of MARKEDNESS constraints so

as to converge at DEP (accent), which is ranked at some lower position.
To put it differently, the demotion of MAX (accent) here is ascribed to its
'gravitation' to DEP (accent). The reason why DEP (accent) is much lower
in the ranking is known by the fact that originally unaccented words usu-
ally get compound accent, as was discussed in (13c) and (14).
This property of FAITHFULNESS constraints in Japanese grammar seems
to be observed also in the segmental phonology, for example, in Itô &
Mester's (1995:184) model of lexical stratification, which is schematized
in (35). Again, MAX demotes through the fixed ranking of MARKEDNESS
constraints, which are not relevant here. Note also that the DEP constraint
is taken to have a lower ranking, because loanwords (alien, foreign, and
Sino-Japanese) systematically undergo vowel epenthesis and consonant
gemination by the requirements of other constraints when they are bor-
rowed into Japanese:22
(35) Japanese segmental phonology
Ranking Word Class Status in Grammar

S Y L L A B L E STRUCTURE
( M A X J ) Alien Periphery
I
A
N o VOICED GEMINATES
Ç MAX Foreign
I
N o SINGLE [ p ]
MAX ^ Sino-Japanese
P O S T - N A S A L VOICING
MAX Yamato Core
(DEP)
Unlike (34), the demotion of MAX in (35) does not reflect the diachronic
change of the segmental system; rather, recent cultivated or younger-gen-
eration Japanese have MAX at a relatively higher position in their gram-

mar, influenced by the modern tendency to borrow foreign words on a
large scale. In spite of such a difference, Ranking Invariance and FAITHFUL-
NESS Conversion, though they are null hypotheses, seem to hold true for
both segmental and accentual phonology at least in Japanese grammar.
Acknowledgements
This paper is a radically revised version of my talk delivered at the monthly meeting of the Pho-
nology Association in Kansai (PAIK), which was held on March 23 in 1999 at Kobe University.
I would like to thank the following people for their comments and suggestions on an earlier ver-
sion: Junko Ito, H a r u o Kubozono, John Mc Carthy, Armin Mester, Akio Nasu, Noriko Ya-
mane, and Yuko Yoshida. I am also very grateful to Edward T.W. Haig for stylistic improve-
ment. Special thanks go to Jeroen van de Weijer, Tetsuo Nishihara, and an anonymous
reviewer, who greatly helped me elaborate my idea on the present topic through this book
project. Any remaining inadequacies or misconceptions are my responsibility alone, of course.
Notes
1. What I call "a c o m p o u n d " in this p a p e r can be defined as a polymorphemic (i.e. non-sim-
plex) word without any 'pitch valley (pitch fall + pitch rise)' in its domain, consisting of
two (or more) words or roots. In this sense, Sino-Japanese words with two roots like
doubutu 'animal' ( d o u 'moving' + butu 'thing') are taken to be c o m p o u n d s by definition,
as well as mimetic or reduplicative words like betabeta (to) 'stickily' (beta 'the sound or
m a n n e r of sticking something', although dou, butu, and beta are bound m o r p h e m e s and
do not occur alone.
However, I will not be concerned in the present discussion with the accentual behavior
of the so-called 'superlong head compounds,' which contain five or m o r e moras in their
heads, as given below:
(i) Superlong head compound

a. kita + oosutória > kitaoosutória ' N o r t h e r n Austria'
síritu + syougákkou siritusyougákkou 'private elementary school'
b. ninami + kariforunia -»• ninamikariforunia 'Southern California'
nankyoku + tankentai -> nankyokutankentai 'Antarctic expedition'
c. tihóu + saibansyó • tihousaibansyó 'district court'
gakusyuu + sankousyó ->• gakusyuusankousyó 'study b o o k ' Kubozono (1995: 23)
In these cases, it is surprising and extraordinary that even the unaccentedness and the fi-
nal accent of the original words in (ib) and (ic) are carried over in the whole compound,
just as the usual non-final accent remains intact in the case of (ia). The accentuation of
such superlong head c o m p o u n d s may involve the dominant ranking of IDENT (accent) (cf.
MAX (accent) in section 3.1.), but I will leave this issue to f u r t h e r research.
2. The reason that the accent of N2 is preserved rather than that of N1 may be attributed to
the fact that N2 is the morphological head of a compound.
3. Kubozono, Itô & Mester (1997) argue that in Japanese, N O N - F I N A L I T Y at the syllabic lev-
el works for loanword truncation as well as compound accent.
4. The violations are computed at the syllabic level and not at the moraic level because ac-
cent-bearing units are syllables in Japanese. In this respect, it may be restated as A L I G N -
R (PrWd, σ 1 ), which I will adopt in section 3.1. (see also Tanaka 2001, to appear).
As for A L I G N - C A , unaccented compounds will always violate it, since they do not bear
accent at all on the 'around-the-boundary' syllable. This constraint appears to require a
rather curious alignment in that it has a double-edge disjunction in it (either at the left
edge of N2 or the right edge of N l ) . In section 3.1., I will adopt a more straightforward
version of a single-edge alignment constraint that has different effects from A L I G N - C A .
5. The traditional view of distinguishing in mora length between short and long head com-
pounds is supported in Sato (1989), Uwano (1997), and so on. See Yoshida (1995) for a
treatment of compound accent in the framework of Government Phonology, which also
makes it possible to dispense with the length of N2 for its computation.
6. Recently, Alderete (1999) has proposed a ranking system of Japanese compound accent
in OT, which is somewhat different from (4). For a non-derivational analysis of foreign
accent (simplex words), which exhibits similar pattern to compound accent, see Kataya-
ma (1995,1997).
7. In Kubozono (1997: 274, this volume), the accent on the non-final rightmost foot is
dubbed 'default compound accent' or 'emergence of the unmarked' and provided with
some generality in the accentual system of Japanese, but with a less dominant status
than the preserved accent stated in (7). I am claiming here that default accentuation
should be more dominant in general than full preservation of input accent, as will be
clear from N O N - F I N A L I T Y ( F ' ) » M A X (accent) in (9a).
8. See also note 20 for related discussion.
9. As Kubozono (1995: 36,1997: 286, this volume) suggests, there may be a possibility of
considering the original N2 words in (11a) as final-accented and hence shifted, because
their final vowels i and u are all epenthetic or underlyingly absent. For criticism of such
an analysis, see Tanaka & Yamane (2000).
As for the data in ( l i b ) , neither N O N - F I N A L I T Y (μ') » M A X (accent) » N O N - F I N A L I T Y (σ',
F ' ) nor N O N - F I N A L I T Y ( μ ' , σ') » M A X (accent) » N O N - F I N A L I T Y ( F 1 ) does not account for
the accent preservation of such cases as in meisyó, sinzitumí, and so on, because their
accent is final at the moraic level. Thus, there is no doubt that the necessary reranking
must be M A X (accent) > N O N - F I N A L I T Y ( « ' , Σ ' , F ' ) , as indicated in (9b).
10. The compound accent in (12a) is final at the syllabic level. As is well-known, there are
no final-accented loanwords at the moraic level in Japanese.
11. In section 3.2., I will present an analysis which allows the variation of either the pre-an-
tepenultimate or antepenultimate accent within the foot concerned.
12. I am not concerned here with unaccented compounds such as urá + yamá ' urayama
'hill at the back' and gárasu + tamá -> garasudama 'glass ball,' which will be discussed
in section 4.1.1 also ignore some Sino-Japanese compounds whose input accent on the
final mora is parsed as in ( l i b ) (meisyó 'place of scenic beauty,' kyuusyó 'vital spot,' and
sinzitumí 'truth'), since they are possible but rare cases, and typically have another ac-
centual pattern as their variation. See section 4.2. for details of their variation.
Note also in (17) that there are some cases which involve hiatus between N l and N2, as
in sinobíasi, niwakáame, and the like. The vowel sequences emerged after compound
formation never form a diphthong or a long vowel in Japanese, because the word
boundary prevents a syllable as well as a foot from crossing it.
1 3 . An anonymous reviewer points out that since we already had a free ranking of N O N - F I -
N A L I T Y (JU\ Σ\ F ' ) and M A X (accent) in ( 1 9 ) , we now have six possible subrankings (i.e.
3 x 2 x 1 possibilities) and wonders whether all the subrankings are motivated. This
does not matter empirically, however, because the optimal candidate does not change
if we take A L I G N - L (σ1, root) into consideration in (19).
14. A n exception to (23) is úbu + ke ubuge 'downy hair,' a trimoraic word where ke is not
final-accented but unaccented: it always forms unaccented compounds such as wakige
'armpit hair,' munage 'chest hair,' and the like, even though they do not become quadri-
moraic. But it is generally the case that unaccented compounds emerge only from the
environments in (23a-c), where N2 is monomoraic or bimoraic (although ki + kokóro
-* kigokoro 'reliability' is a quadrimoraic compound exceptionally with a trimoraic
head). In other words, de-accentuation of compounds can apply only if their input head
is short or their output form is quadrimoraic. I will leave the question of how it is ac-
counted for to further research.
15. I am grateful to Armin Mester for pointing out the idea of N O N - F I N A L I T Y (PrWd').
16. Otherwise, since a compound consists of two (or more) words, it might be possible that
it contains two PrWd's, not two roots, and that the prosodie wordhood of N1 and N2
always percolates upward to the whole compound. That is, the prosodie wordhood that
matters here might be determined by the so-called 'A-over-A principle.' I do not adopt
such an assumption, because it does not hold for the Sino-Japanese cases in (23c) or the
mimetics in (30b) below, which are made up of two bound morphemes and not of two
PrWd's. Moreover, this principle does not apply to the case of English compounds.
17. By using bidirectional extrametricality, Tateishi (1992) offers a derivational account of
why trimoraic and quadrimoraic simplex words are prone to be unaccented; however,
it does not explain the cases in (23a) since the unaccented compounds are more than
four moras long, which he is not concerned with. Thus, it may be the case that unaccent-
edness of simplex words is triggered by different factors from that of compounds, even
though they have in common the fact that they have some length-effect on unaccentu-
ation (see note 1 4 ) and that their accent loss is due to N O N - F I N A L I T Y (PrWd 1 ).
18. See section section 3 . 1 and note 4 for the difference between A L I G N - C A and A L I G N - L
(σ1, root).
19. I assume here that only the members of the FAITHFULNESS family (i.e. M A X , D E P , and
IDENT) can promote or demote through the fixed constraint ranking. For detailed dis-
cussion on this topic, see section 5. Given this assumption, we focus on the rerankings
in (24a) and (24c) and ignore the one in (24b) that is mainly for quadrimoraic heads
with the antepenultimate accent as in (21). See also Tanaka (2001, to appear) and Tana-
ka & Yamane (2000) for a lexical-class-based account of how individual compounds se-
lect the specific reranking possibilities.
20. Note that in the case of mimetics in (31a), trochaic patterns such as bèta) + (beta) and
kira) + (kira) are produced. This fact might appear to pose a certain problem with the
proposed system, which would generate *betá) + (beta) and *kirá) + (kira) as it stands.
However, we must recall the definition of MAX (accent), which requires that the accent
of a head root has a correspondent in the compound. In usual compounds, the N2 accent
is respected because N2 is their morphological head, but this is not the case with mimet-
ics: I assume that the head of mimetic words is N l , and that is why MAX (accent) favors
bèta) + (beta) and kira) + (kira), excluding *betá) + (beta) and *kirá) + (kira). A L I G N - L
(ισ', root) also favors the trochaic patterns, because the head root is N l .
21. Strictly, Hattori refers to the former type not as "penultimate-foot accent" but "accent
on the syllable containing the third mora from the end." Since the dominance of the lat-
ter implies that of the former, I consider them virtually equivalent here.
22. As is well-known, native and mimetic words also undergo vowel epenthesis and conso-
nant gemination, respectively, as in kak + hazimer-u kakihajimeru 'begin to write,' yar
+ oer-u yarioeru 'finish doing,' beta-ri bettari 'very stickily,' biku-ri bikkuri 'very
surprisingly,' and so on.
References
Alderete, John
1999 Morphologically Governed Accent in Optimality Theory. Ph.D. dissertation,
University of Massachusetts at Amherst.
Hattori, Noriko
1991 Mechanisms of Word Accent Change: Innovations in Standard Japanese. To-
kyo: Shinozaki Shorin.
1998 "Base Transparency in Suprasegmental Changes: Ongoing Changes in Japa-
nese and English," Language, Variation, and Change 10: 2, 85-96.
1999 "On the Convergence of Accentual Patterns: 'Drift' Observed in Variation,"
Phonological Studies 2, ed. by the Phonological Society of Japan, 37-44. To-
kyo: Kaitakusha.
Hayes, Bruce
1995 Metrical Stress Theory. Principles and Case Studies. Chicago: University of
Chicago Press.
1995 "The Core-Periphery Structure of the Lexicon and Constraints on Rerank-
ing," University of Massachusetts Occasional Papers in Linguistics 18 (Papers
in Optimality Theory), 181-209.
Itô, Junko, Yoshihisa Kitagawa, and Armin Mester
1996 "Prosodie Faithfulness and Correspondence: Evidence from a Japanese Ar-
got," Journal of East Asian Linguistics 5, 217-94.
Katayama, Motoko
1995 "Loanword Accent and Minimal Reranking in Japanese," Phonology at Santa
Cruz 4,1-12.
1997 Optimality Theory and Japanese Loanword Phonology. Ph.D. dissertation,
University of California at Santa Cruz.
Kubozono, Haruo
1995 "Constraint Interaction in Japanese Phonology: Evidence from Compound
Accent," Phonology at Santa Cruz 4, 21-38.
1997 "Lexical Markedness and Variation: A Nonderivational Account of Japanese
Compound Accent," Proceedings of the 15th West Coast Conference on For-
mal Linguistics, 273-87.
(this volume) "Epenthetic Vowels and Accent in Japanese: Facts and Paradoxes"
Kubozono, Haruo, Junko Itô, and Armin Mester
1997 "Nonfinality in Japanese Phonology," Proceedings of the 16th International
Congress of Linguistics (CD-ROM), ed. by B. Caron. Amsterdam: Elsevier.
- McCarthy, John and Alan Prince
"Generalized Alignment," Yearbook of Morphology 1993, ed. by Geert Booij
and Jaap van Marie, 79-153. Dordrecht: Kluwer.
McCawley, James D.
1968 The Phonological Component of a Grammar of Japanese. The Hague: Mouton.
1977 "Accent in Japanese," University of Southern California Occasional Papers in
Linguistics 4 (Studies in Stress and Accent), 261-302.
N H K (ed.)
1998 A Dictionary of Japanese Pronunciation and Accentuation. Tokyo: N H K
Shuppan.
Poser, William
1990 "Evidence for Foot Structure in Japanese," Language 66, 78-105.
Sato, Hirokazu
1989 "The Accent Rule and Sequential Voicing in Compounds," The Japanese Lan-
guage and Japanese Education 2, ed. by Miyoko Sugito, 233-65. Tokyo: Meiji
Shoin.
Tanaka, Shin-ichi
1992 "Accentuation and Prosodie Constituenthood in Japanese," Proceedings of
the 5th Annual Meeting of the Tokyo Linguistic Forum, 195-216.
2001 "Japanese Accent in Optimality Theory: The Mechanism of De-accentuation
and its Different Applicability among Lexical Classes," The Interface between
Meaning and Form: A Festschrift for Dr. Minoru Nakau on the Occasion of his
Sixtieth Birthday, ed. by Shuichi Takeda, Atsuro Tsubomoto, Yukio Hirose,
Koichi Takezawa, and Nobuhiro Kaga. Tokyo: Kuroshio Shuppan.
(to appear) Accent and Rhythm: From the Basics of Phonology to Optimality Theory.
Tokyo: Kenkyusha.
Tanaka, Shin-ichi and Noriko Yamane
2000 "Paradox or Reranking?: Accent and Epenthetic vowels in Sino-Japanese,"
Proceedings of the 24th Annual Meeting of the Kansai Linguistic Society (KLS
20).
Tateishi, Koichi
1985 Accent of Compound Nouns in the Standard Japanese: A Nonlinear Analysis.
MA thesis, International Christian University, Tokyo.
1989 "Theoretical Implications of the Japanese Musicians' Language," Proceedings
of the 8th West Coast Conference on Formal Linguistics, 384-98.
1992 "Maximal Non-Exhaustive Invisibility," Proceedings of the 16th Annual Meet-
ing of the Kansai Linguistic Society (KLS 12), 134-44.
Uwano, Zendo
1997 "The Accent of Some Dialects in Japanese with Special Reference to Com-
pound Nouns," Accent, intonation, Rhythm, and Pause, ed. by Tetsuya Kuni-
hiro, Hajime Hirose, and Morio Kouno, 231-70. Tokyo: Sanseidou.
Walker, Rachel
1995 "Mongolian Stress: Typological Implications for Non-Finality in Unbounded
Systems," Phonology at Santa Cruz 4, 85-102.
1996 "Prominence-Driven Stress," Ms., University of California at Santa Cruz.
[Available at ROA-172-0197 from the Rutgers Optimality Archive.]
Yoshida, Yuko
1995 On Pitch Accent Phenomena in Japanese. Ph.D. dissertation, SOAS, Universi-
ty of London. [Published by Holland Academic Graphics, The Hague, in
1999.]
An Element-Based Analysis of Affrication
in Japanese*
Shohei Yoshida
Yokohama National University
1. Intoduction
In this paper I discuss the process of affrication in Japanese in the frame-

work of element theory (Kaye, Lowenstamm and Vergnaud 1985, 1990;
Harris 1990, 1994; Harris and Lindsey 1995). In element theory, affrica-
tion is viewed as a process whereby a plosive's Root Node is decomposed
into two to which a plosive element and a fricative element are linked, re-
spectively (Harris 1994:130-131). In standard Japanese, there are two dif-
ferent types of affrication processes both involving coronal stops. The
first type of affrication occurs when the articulatory place of dental-alve-
olar stops before /i/ and /]/ changes to palato-alveolar as a result of pala-
talization (e.g. /toti/ [totji] 'land', /tja/ [tja] 'tea'). The second type of af-
frication occurs when dental-alveolar stops are followed by a high-back
vowel (e.g. /tetu/ [tetsui] 'iron'). Deemed to be a universal tendency, the
type of affrication with concomitant palatalization is crucial for phono-
logical studies in the sense that it provides phonologists with some in-
sights into representations of coronals, palatalization and affrication. In
the following sections, I attempt to explore a non-arbitrary relationship
that is expected to exist between the process of affrication and the context
in which it takes place. In particular, I show that an element-based analy-
sis makes it possible to account for the tongue-raising of dental-alveolars
and the tongue-fronting of velars as manifestations of a unitary spreading
process, i.e. the spreading of the I element.
2. Data
There are four types of affricates in standard Japanese: the voiceless and
voiced dental-alveolar affricates [ts, dz] and the voiceless and voiced
palato-alveolar affricates [tj, d3]. The voiceless affricates [ts, t j ] are allo-
phonic variants of It/, /t/ becomes [ts] before a high-back vowel, while it
194 Shohei Yoshida
becomes [tj] before a high-front vowel and a palatal glide. Likewise, /d/
becomes [dz] and [d3] before a high-back vowel and a high-front vocoid,
respectively. It should be noted at this point that the phoneme /z/ is usu-
ally pronounced as the affricate [d3] before a high-front vocoid and as
[dz] otherwise. It follows that the contrast between Idi and /z/ is neutral-
ized before a high-front vocoid and a high-back vowel in modern stan-
dard Japanese. Before /j/, however, there is one context where the phone-
mic status of [d3] is obvious. When the /w/ in the sequence /de+wa/ 'in that
case' (CONNECTIVE + TOPIC) is deleted, the resulting /dea/ becomes /djaa/
[d3a] or [d3aa] as a result of glide formation and optional compensatory
lengthening (Poser 1985). In this case, the source of [d3] is clearly /d/. In
addition, all the voiced affricates, regardless of their phonemic status, are
prone to be lenited to the fricatives [z, 3] intervocalically in rapid speech.
(1) shows the distributions of the allophonic variants of Iti, Id/ and /z/ fol-
lowed by a palatal glide and the five vowels of Japanese.1
(1)
Iii Iii Id /a/ loi lui
Iti ítjvi m [te] [ta] [to] ftsui]
Idi [de] [da] [do]
[d3V~3V] [d3i~3i] [dzui-zui]
/ζ/ [dze~ze] [dza~za] [dzo~zo]
Examples:
/tjoo/ [tjoo] 'butterfly' /kati/ [katji] 'value'
/mate/ [mate] 'Wait!' /kata/ [kata] 'shoulder'
/ato/ [ato] 'later' /natu/ [natsui] 'summer'
/haiDjo/ [haid3o]~[hai3o] 'exclusion' /kaDi/ [kad3Í]~[ka3Í] 'fire'

/made/ [made] 'up to' /hada/ [hada] 'skin'
/nado/ [nado] 'et cetera' /kaDu/ [kadzui]~[kazui] 'fire'
(where 'D' denotes a segment whose phonemic status is controversial)
The context-sensitive affrication in the underlined syllables in (1) is of in-

terest here. It is worth differentiating between the type of affrication that
accompanies palatalization, and the type of affrication that occurs with-
out concomitant palatalization.
Affrication triggered by palatalization is far more common than that
without palatalization and its instances abound in languages. To cite but
a few, in informal spoken English alveolar stops optionally become pala-
An Element-Based Analysis of Ajfrication in Japanese 195
to-alveolar affricates when followed by the palatal glide /j/ across a word
boundary, e.g. last year [laistjia], did you [did3u:]; in Korean stem-final
dental stops become palato-alveolar affricates when followed by ¿-begin-
ning suffixes, e.g. kath+i [katji] 'together', path+i [patJi] 'plowed
field+suBJECT'; in Brazilian Portuguese the dental stops /t, d/ become
palato-alveolar affricates [tj, d3] before /i/, e.g. noite [noitJi] 'night', tarde
[tahd3i] 'afternoon'; and female Cairene Arabic speakers palatalize and
affricate dental-alveolar stops when the following vowel is III, e.g. dilwaiti
[d3ilwa?tj"i] 'now', balad+i [balad3i] 'my country'.
In addition to dental-alveolar stops, the velar stops /k, gl are often coro-
nalized and affricated to [tj", d3] before front vowels, if not pronounced
with a secondary palatal articulation as in [kJ, gi], in many languages. Affri-
cation that accompanies the coronalization of velar stops is found in Aca-
dian French (Flikeid 1988), Russian (Kiparsky 1982:124), Slavic (Chom-
sky and Halle 1968: 421-422), and Eastern Arabic dialects (Johnstone
1967). In Japanese, velar stops are dialectally fronted before HI and 1)1
(Fujiwara 1997:55-59). For example, in the Tohoku provinces, Ikl before
HI is pronounced either as the palato-alveolar affricate [tjï] or the velar-
palatal affricate [kçï], e.g. /kirei/ [tjïree]~[kçïree] 'beautiful'. These cases,
together with the rarity of the palatal stops [c] and [j.] in languages (Cat-
ford 1988: 94), indicate that in the palato-alveolar-palatal region the affri-
cates [t J, d3] are preferred over dental-alveolar stops followed by a glide
[ti, di], and the palatal stops [c, j], pointing to the unstable status of palatal
closure, and a correlation between palato-alveolar-palatal articulation and
affrication.
On the other hand, it appears to be the case that affrication that does
not involve palatalization is much rarer. Such a case is found in Quebec
French. In Quebec French, coronal stops are affricated before the high-
front vowels HI and lyl without concomitant palatalization (Kaye 1989:
29), e.g. type [tsip] 'type', dur [dzyif] 'hard'. Strictly speaking, however,
this case should be distinguished from the true case of affrication without
palatalization as in Japanese in that the environment in the Japanese case
involves a high-back vowel whereas that in the Quebec French case in-
volves a high-front vowel though palatalization does not manifest itself.
This does not mean, however, the affrication without palatalization is
non-existent in other languages. An instance of affrication genuinely par-
alleling the Japanese case comes from Korean, in which the impermissible
sequence of a stem-final dental stop and a suffix-initial liquid or a nasal is
broken up by the epenthetic vowel [i], and the dental stop becomes a den-
tal-alveolar affricate, e.g. path+ro [patsiro] 'toward the plowed field',
196 Shohei Yoshida
path+n [patsin] 'plowed field+topic', k\th+ro [kitsiro] 'to the end',

muth+ro [mutsiro] 'to the land'. Bhat (1978: 60-61) provides similar cases
in other languages, which remain to be attested.
As such, it is possible to draw the conclusion that affrication induced by
palatalization is a recurring process and is therefore deemed a universal
tendency, whereas the affrication without palatalization is a language-
specific phenomenon. The orthodox rule-based generalization that in
Japanese coronal stops become affricates before high vocoids is superfi-
cial for three reasons. First, despite its simplicity, such a generalization is
purely descriptive and does not take into account the aforementioned
asymmetry between the two types of affrication regarding their typologi-
cal frequencies. Second, no consideration is given to the fact that the trig-
gering feature of affrication in the context of the high-front vocoids is in-
ternal to the target, i.e. a triggering feature spreads and becomes part of
the target, whereas the triggering feature of affrication before a high-back
vowel is external to the target, i.e. it affects the target while being con-
fined to an adjacent segment. Finally, as a general point, in rewrite rules
the relationship between the process and the environment in which it
takes place is essentially arbitrary.
3. Element theory and representations of vowels
In element theory, the phonological primes are called elements. Since el-
ement theory originated as a sub-theory of government phonology (Kaye,
Lowenstamm and Vergnaud 1985, 1990), the possibility of representing
both vowels and consonants in terms of univalent elements, rather than
binary-valued features, has been pursued by researchers such as Renni-
son (1990), Harris (1990,1994), Cyran (1997), Harris and Lindsey (1995),
Williams (1998) and authors in Cyran (1998).
Elements are monovalent, being specified for their presence only. Vow-
els are constructed from the elements I, U and A. Elements have the
property of interpretational autonomy (Harris and Lindsey 1995: 40), and
are interpretable and pronounceable in isolation, unlike binary-valued
features, which, with value specifications, contribute to the realization of
a segment only in combination with several other features. A phonologi-
cal expression (henceforth 'PE') is either simplex, composed of a single
element, or complex, composed of two elements or more, and has one
head. When linearly represented, the head of a PE is indicated by locating
it at the right edge of the PE by convention. Thus, in the PE (X.Y), Y is

head, X being operator. PEs either have active elements as their heads or
may be empty headed, the distinction between them being roughly map-
pable to the ATR/non-ATR contrast. The head of empty-headed PEs is
metaphorically called the neutral element (Harris and Lindsey 1995; Har-
ris 1994). In a metaphoric statement of Harris and Lindsey (1995: 60) the
neutral element is described as "a blank canvas to which the colors repre-
sented by A, I, U can be applied", and has no marked articulatory or
acoustic property. The neutral element is omnipresent in PEs and mani-
fests itself only when it is head; when it is operator it is overridden by oth-
er element(s). For example, in many vowel systems the PEs (I), (U) and
(A) are realized as [i], [u] and [a], whereas their empty-headed counter-
parts (I._), (U._) and (A._) are realized as [ι], [υ] and [a].
Complex vocalic PEs potentially yield, in addition to the element-
headed/empty-headed contrast, an η-way contrast (η = the number of el-
ements that constitute a PE) because each of the elements may be head.
For example, the two elements I and A give rise to the empty-headed
(I.A._) [ε], the I-headed (A.I) [e] and the A-headed (I.A) [as], where op-
erators are randomly ordered. By the same token, U and A may generate
(U.A._) [O], (A.U) [o] and (U.A) [D]. However, the fact remains that the
vowel inventories of languages contain only subsets of segments which
can logically be generated. In order to define individual systems, licensing
constraints (Charette and Göksel 1994, 1998; Cyran 1997; Cobb 1995,
etc.) have been proposed. Licensing constraints are stipulatory state-
ments that generate all and only segments present in languages by re-
stricting the combinatorial possibilities of elements and have so far been
tested mostly against representations of vowels.
When the resonance elements I, U and A were proposed by Kaye, Lo-
wenstamm and Vergnaud, they were no more than theoretical constructs
motivated by phonological reasoning. Elements now enjoy some empiri-
cal weight as researchers such as Harris (1994), Harris and Lindsey
(1995), Ingleby, Chalfont and Brockhaus (1994), Williams and Brockhaus
(1992) and Williams (1998) have established the correlation between
some elements and their acoustic signatures that are sufficiently invariant
across languages and speakers. According to these authors, elements are
strictly cognitive categories mappable in the first instance onto acoustic
patterns, not onto articulations. Harris and Lindsey (1995: 55) claim that
"when a given element is input to speech production mechanisms, the
speaker will marshal whatever articulatory resources are necessary or
available for the spectral realization of the target element pattern". The
198 Shohei Yoshida
reader is referred to Harris (1994), Harris and Lindsey (1995) and Will-
iams (1998) for acoustic signatures of independent elements in simplex
and complex PEs.
For the present purposes, the articulatory implementation of I, U, A
taken from Harris and Lindsey (1995: 55-56) are given below. The paren-
thesized comments are supplied by the present author.
(2) Element Articulatory implementation

I maximal expansion of the pharyngeal cavity and
maximal constriction of the oral cavity (palatality)
U a trade-off between maximal expansion of both the oral
and pharyngeal tubes (labiality, velarity)
A maximal expansion of the oral tube and constriction of
the pharyngeal tube (low as head, non-high as operator)
The internal structures of the five vowels of Japanese are as follows:
(3) i (I) ui (U)

e (A.I) o (A.U)
a (A)
A word must be said about the unroundness of the high-back vowel. In bi-
nary feature theory such a vowel must be assigned the property [-round]
and represented differently from a high-back rounded vowel. In element
theory both [u] and [ui] can be represented by the U element since a labial
activity is but one way of expanding the oral cavity. Specifically, I propose
that the unrounded high-back vowel of Japanese is represented as an emp-
ty-headed PE containing the U element.
4. Consonants
When the resonance elements I, U and A occur in non-nuclear positions
either on their own or in combination with other elements, they define
palatality, labiality (and velarity in this paper), and pharyngeality, respec-
tively. Thus the phonological objects I and U, which in isolation are real-
ized as i and u in nuclear positions, characterize palatal and labial (and ve-
lar in this paper) articulations in non-nuclear positions. For example, the
vowels i and u and their non-syllabic counterparts j and w differ only in
their syllabic affiliation.
Also found in representations of consonants are the laryngeal or source

elements L and H, which in nuclear positions specify low tone and high
tone, respectively. In consonants, L defines voicing of fully voiced ob-
struents whereas H defines voicelessness. Vowels, glides and liquids are not
assumed to contain the L element, hence they are devoid of phonologically
significant voicing. Different languages employ L and H in different ways
to make the so-called voiced-voiceless contrast. For example, the voiced-
voiceless contrast of English obstruents is due to the absence of a source el-
ement in the so-called voiced series and the presence of H in the voiceless
series. In French and Spanish, on the other hand, the voiced obstruents con-
tain L and the so-called voiceless obstruents contain no source element.
Three manner elements ?, h and Ν have been proposed. The ? element
is associated with a closure in the vocal tract and present in oral and nasal
stops, affricates, and laterals. The h element characterizes noise or a nar-
rowed stricture that produces a turbulent airflow, and is present in re-
leased oral stops as well as in fricatives and affricates. When a PE consists
either of ? or h alone, devoid of a resonance element, it is realized as a
glottal lacking a supralaryngeal gesture. The Ν element is responsible for
nasality of nasal consonants and nasalized vowels.2
The linear representations of the main allophonic variants of Japanese
consonants I propose are as follows:
(4) Labial
w (U)
ρ (h.?.U) b (L.h.?.U)
Φ (h.U)
m (L.N.7.U)
Dental-alveolar
r U
t (h.?._) d (L.h.?._)
s (h. J
ts (?._); (h._) dz (L.?._); (L.h.J
η (L.N.?. J
Palato-alveolar
S (h.I.J
t j (?._); (h.I._) d 3 (L.?._); (L.h.I.J
ni (L.N.?._); (I)
Palatal
j (I)
ç (h.I)
200 Shohei Yoshida
Velar
k (h.?.U._) ki (h.?.U._); ( I )
g (L.h.?.U._) i] ( L . N . 7 . U J
Glottal
h (h)
Moraic nasal3
iîi (Ν)
(where the semi-colon denotes that the segments containing it con-
sist of two Root Nodes.)
The representations in (4) differ from Kaye, Lowenstamm and Vergnaud

(1990), Harris (1990), Harris (1994), Harris and Lindsey (1995) and Will-
iams (1998) in two crucial aspects. The first difference is the representa-
tion of coronality. It should be pointed out here that in element theory
coronality is defined as a property shared by dental, alveolar, palato-alve-
olar articulations to the exclusion of palatal articulation. Kaye, Lowen-
stamm and Vergnaud (1990), Harris (1994), and Harris and Lindsey
(1995) assume an independent coronal element, 'R'. In an attempt to re-
duce the total number of elements, Williams (1998), Broadbent (1991),
Cyran (1997), Bloch-Rozmej (1998) and Lee (1998) have analyzed coro-
nality as a manifestation of the A element occurring in non-nuclear posi-
tions. However, I adopt a proposal associating coronality with the lack of
specification for place feature as argued by Avery and Rice (1988) and
Cho (1991), couched in underspecification theory, and Backley (1993),
couched in an earlier version of element theory. Translated into element
terms, coronals are represented as empty-headed PEs occupying non-nu-
clear positions, as shown in the representations of dental-alveolars and
palato-alveolars in (4). Representing coronals as segments devoid of a
place-defining element has the advantage of accounting for well-known
special properties of coronals cited in Paradis and Prunet (1991), e.g. their
transparency effects in vowel spreading, their readiness for assimilation,
and their inventory, typological and occurrence frequencies.
The second difference is the representation of velarity. In the previous
studies of element theory, empty-headed PEs occurring in non-nuclear
positions, which in my analysis are coronals, are claimed to be velars.
However, I propose that velarity and labiality are two different manifes-
tations of the same U element. Assigning the dual function to U enables
us to capture close relations holding between labial and velar articula-
tions such as the existence of labio-velar stops in West African languages,
and the implicational universale noted by Maddieson (1987: 40) that if a
language has Ipl then it will also have Ik/, and if a language has Ik/ then it
will also have It/. Specifically, I propose that when occupying non-nuclear
positions, the U element as head is manifested as bilabial articulation
whereas it is realized as velar articulation as operator. The first implica-
tional universal can be accounted for if we assume that the presence of
the U element in stops as head ((...?.U) = a bilabial stop) entails its pres-
ence as operator ((...?.U._) = a velar stop) but not vice versa. The second
implicational universal also stems from the present elemental representa-
tions of dental-alveolars and velars. That is, the presence of empty-head-
ed stops containing U (,..?.U_), which are velars, implies the presence of
empty-headed PEs without U (...?._), which are dental-alveolars.
5. Contour structures
Following Harris (1994), I assume that affricates are represented as contour

structures. The representations of [ts] and [tj] are given in (5), where heads
are underlined. The representation of [t] is also given for comparison.
(5) t ts tj
X X X
I / \ / \
ROOT «C T T T T
l \ / I l \ / I I \
RESONANCE · ? ? · · h ? · · h
\ / \ / \
I
[t] is represented as an empty-headed PE containing the occlusion ele-

ment ? and the noise element h that are linked to Root Node. Its affricat-
ed counterpart [ts] is represented as a contour structure where a stop and
a fricative share a single place-defining property (i.e. an empty head). In
light of this representation, the affrication of [t] is construed as the de-
composition of the plosive's single Root Node into two (Harris 1994:
130). The sequentially ordered two Root Nodes are associated with the
occlusion element ? and the noise element h respectively, and the two
Resonance Nodes are linked to a single place-defining property, i.e. an
empty head. This contour structure ensures a friction phase after the re-
lease of the closure and the homorganicity of the plosive and fricative
phases. The difference between [ts] and [t|] lies in that the latter has an
202 Shohei Yoshida
additional I element as operator attached to the second Resonance Node.

One would wonder why the two Resonance Nodes of affricates, in partic-
ular of [ts], should not be a single node as a static effect of the OCR The
motivation for assuming two Resonance Nodes will be given in the fol-
lowing analysis.
In element theory the voicing of obstruents in intervocalic position, the
devoicing of obstruents in coda positions, and other processes all denoted
by the cover term lenition or weakening are formally defined as instances
of decomposition whereby the number of element(s) constituting a seg-
ment is reduced (Harris 1990, Brockhaus 1995). The decomposition of a
plosive into an affricate is understood to involve a reduction of the num-
ber of elements per Root Node without element loss. The change of affri-
cates to fricatives in intervocalic position as in the Japanese examples in
(1) involves the delinking of the first Root Node, hence a decrease in the
total number of elements.
6. Palatalization and affrication: a universal perspective
We now have to understand why the decomposition of dental-alveolar

stops takes place before the high vowels at all.
In addition to the rarity of palatal stops, Catford (1988: 94-95) also
notes that sounds traditionally labeled as the palatal nasal [ji] and the pal-
atal lateral [Á] are not always pronounced as genuine palatals, but rather
as the dental-alveolars with a secondary palatal articulation [ni], [li]. It
should also be recalled that as the result of tongue fronting, velar stops
usually become palato-alveolar affricates rather than palatal stops or af-
fricates. All these are indicative of the general incompatibility of the I el-
ement and the occlusion element ?, which can be stated in terms of the
following constraint:
(6) /-? Constraint

? cannot c-command I.
The I-? Constraint prevents the I element from being dominated by the
first branching node (Root Node) dominating the occlusion element ?,
forbidding the co-occurrence of the two elements under the same Root
Node. This constraint may apply to oral stops, nasal stops and laterals, as
they all contain the ? element. 4 When the spreading of the I element from
a nuclear position to a preceding onset position is obligatory, i.e. when I
is required to coexist with the ? element under a single Root Node, in lan-
guages where this constraint is operative, the Root Node and the Reso-
nance Node it dominates are fissioned into two, and I is linked to the sec-
ond Resonance Node to avoid being c-commanded by ?. It can be argued
that the fission of stops in the palato-alveolar-palatal region is attribut-
able to the fact that palatal articulation usually involves a considerably
large constriction area. The change from [t] to [tj~] is given in (7).
(7) t i tr
χ χ —^ X X
I I / \ I
ROOT RC ? r ? ?
I \ I / I l \ I
RESONANCE ·- ? · ? · « h ·
The I-? Constraint in (6) forbids the spreading I to be interpreted within a

PE containing the occlusion element ?. As a result, the plosive's Root and
Resonance Nodes are fissioned into two, and the spreading I and the noise
element h are interpreted in a PE under the second Root Node. In the re-
sulting structure, the ? element c-commands the empty head but not I.
One may wonder why the following structure should not be selected.
(8) tj
χ
ROOT
\
RESONANCE
In this structure, the ? element does c-command the I element since the
first branching node dominating ? (Root Node) also dominates I. In ef-
fect, two Root Nodes share a single Resonance Node. However, I reject
this structure because the initial Root Node as well as the second Root
Node contains the I element and therefore this structure does not capture
our basic insight concerning the general incompatibility of palatality and
occlusion.
204 Shohei Yoshida
In the representation of [t J] in (7), the spreading I does not replace the

empty head of [t] but becomes operator, adding palatality to the existing
coronality defined by the empty head, yielding a palato-alveolar articula-
tion under the second Root Node. It should be clear that in the present
analysis palatals and palato-alveolars are viewed as being phonologically
distinct, i.e. palatals are I-headed PEs whereas palato-alveolars are emp-
ty-headed PEs containing the I element as operator. There is a different
view in the literature (e.g. Lahiri and Blumstein 1984) which does not rec-
ognize a phonological distinction between palatals and palato-alveolars.
Let us compare the present analysis with a feature-based analysis of pal-
atalization in Lahiri and Evers (1991). For Lahiri and Evers the spreading
feature of Iii and 1)1 is [-anterior] within the Coronal Node. Upon spread-
ing, [-anterior] replaces [+anterior] within dental-alveolars. However, it is
not very clear why dental-alveolars are palatalized to palato-alveolars af-
fricates, rather than palato-alveolar stops or genuine palatal stops or af-
fricates, which are non-anterior. To account for the affrication and the
change of place to palato-alveolar, one needs to assume that palato-alve-
olar articulation is unmarked in the palato-alveolar-palatal region, and
that affricates are unmarked in the palato-alveolar region, treating [-high]
and [+continuant] in the representations of surface affricates as being
supplied by means of redundancy rules.
The I-? Constraint in (6) does not forbid the formation of palatal frica-
tives, since fricatives do not contain the ? element by definition. Thus the
change of /hi to [ç] before HI and /j/ in Japanese (e.g. /hi/ [çi] 'day', /hjaku/
[çakui] 'a hundred') is straightforwardly accounted for as in (9). Ihl has no
resonance element and it is not headed, either. Lacking a specification for
place, it is palatalized to a genuine palatal.
(9) h Ç
χ χ > χ χ
ROOT
\
RESONANCE • h ·
I
Recall that the palatalization of coronals is characterized as [-anterior]

spreading in Lahiri and Evers (1991). However, the spreading of the same
feature to a glottal fricative, along with the purported markedness of pal-
An Element-Based. Analysis of Affrication in Japanese 205
atal articulation, incorrectly predicts that the resulting fricative is palato-

alveolar.
The element-based analysis predicts that the raising of dental-alveolars
may also be triggered by non-high front vowels such as /e/ (A.I) and /ae/
(I.A) because they contain the I element. Shibatani (1990:202) states that
in Japanese the palatalization of the dental-alveolar fricative /s/ before Id
as well as HI occurred until the fifteenth century. He also notes that such
palatalization still occurs in such areas as Tohoku, Hokuriku, Shikoku
and Kyushu where the standard form /sensei/ [seuisee] 'teacher' is pro-
nounced as [JeûiJeeJ.The representation of the syllable [Je] in this word
is as follows:
(10) s e
χ χ —^
ROOT
RESONANCE
I/ \ A
7. The fronting of velar stops
As noted above, velar stops are palatalized to palato-alveolar or velar-

palatal affricates in dialects spoken in the Tohoku provinces, e.g. /ki/
[tj"i]~[kçi] 'tree'. Given the I-? Constraint and the representation of coro-
nals and velars as empty-headed PEs, it is also possible to give a unified
autosegmental account of the tongue-raising of dental-alveolars and the
tongue-fronting of velars. In either case, the spreading property is the I el-
ement. The change of /k/ to [tf] in the context of a high-front vocoid is
shown in (11).
(11) k i tj i
X ^^ h X -;
ROOT ? ·
RESONANCE ·-
/ -χ I
υ I
206 Shohei Yoshida
In this case, the spreading I element replaces the operator U in the internal
structure of /k/, suppressing velar articulation. The result is identical to the
output of the affrication of It/ before a high-front vocoid. To see why U
within /k/ is replaced by the spreading I, let us consider cases of palataliza-
tion in nuclear positions, i.e. the spreading of I in nuclear positions.
In nuclear positions, the I and U elements may combine, resulting in
front rounded vowels, though the combination is subject to parametric
variation. Japanese does not permit the combination of I and U in nuclear
positions, as manifested in the nonexistence of front rounded vowels, and
in some morphological processes. For example, when the adjectival suffix
consisting of the I element alone spreads to the final nuclear position of an
adjectival stem in dialects spoken in the Tohoku and Kanto districts, dif-
ferent results obtain depending upon the elemental content of the stem-
final position. If the stem-final position is occupied by A, the two elements
form a complex PE as the combination of I and A is allowed, whereas if the
stem-final nuclear position contains U, the spreading I replaces it.
X X X
\ /
taka+i [takee] 'high' (A) «κ (I) -- (A.I)
samu+i [samii] 'cold' (¥) < (I) - (I)
sugo+i [suigee] 'terrific' ( A . ¥ ) <t (I) -- (A.I)
As mentioned above, the combination of I and U in nuclei is subject to

parametric variation. However, given the absence of labio-palatals in the
world's languages, it seems to be the case that the same combination is
universally barred in non-nuclear positions attributable to the following
constraint:
(13) /-Í7 Constraint

I and U do not combine within a PE in non-nuclear positions
This does not deny the presence of labials with a secondary palatal artic-
ulation such as [pi], [b»j] and [mi]. These sounds are contour segments
where labial PEs and the I element are linked to two distinct Root Nodes,
which will be discussed shortly.
To accord with this constraint, the combination of the U element, which
is already present in /k/, and the I element added through spreading into
this segment is avoided by one of the following strategies:
(14)
a.(h.?.¥._) « I (?._); (h.I.J k -1J
b. (h.?.U._) « I (?.U._); (h.I) k ^ kç
c. (h.?.U._) « I (h.?.U._); (I) k ki
The three strategies depict three different results of velar palatalization.

The effect of (14a) is seen in (11). The avoidance of the I-U combination
within a PE is achieved by replacing the operator U with I, while the emp-
ty head remains intact. If no further constraint applies, the resulting PE is
(h.?.I._), which would be a palato-alveolar stop. However, the I-? Con-
straint applies and (h.?.I._) is fissioned into the affricate [tj], (14b) de-
scribes the change of [k] to the velar-palatal affricate [kç] as shown in
(15). Unlike (14a), the spreading I does not replace any existing element.
Instead, I is linked to the second Resonance Node of the affricate induced
by the I-? Constraint.
(15) k i
χ h r
ROOT j^— ? ·
RESONANCE ν
/ I
υ _ ' ι
When (14c) is selected, the spreading of the I element to velars results in

the addition of a secondary articulation. Note that the I-U Constraint in
(13) is expected to apply to labials as well as velars since labials also con-
tain the U element. If U in labials is replaced by a spreading I, labials are
also coronalized. However, no such place appears to be attested in the
world's languages. This is presumably by virtue of the status of U as head
in labials. This prediction is compatible with Paradis and Prunet's (1991)
claim that "palatalization of noncoronals (labials and velar) only adds a
secondary articulation" (pl5), where 'palatalization' excludes the coro-
nalization of velars. This type of palatalization of velar and labial stops is
shown below:5
208 Shohei Yoshida
(16) k i
h χ χ
• I
ROOT ?
RESONANCE · ·
/ \ I
U I
(16b) ρ i
h χ χ
I• I
ROOT ?
RESONANCE
υ I υ I
Given the representations of front vowels, which all contain the I ele-
ment, we expect to find languages where the fronting of velars is triggered
by nonhigh-front vowels such as /e/ (A.I) and /ae/ (I.A). Arabic dialects
spoken in Eastern Arabia (e.g. Kuwait, Bahrain, Qatar, Oman, etc.), pro-
vide such cases. In these dialects, velar and uvular stops are fronted in the
contiguity of mid- and low-front vowels as well as /i/, e.g. /kam/ [tjaem]
'how much', /halqa/ [fraltjae] 'ring' (Johnstone 1967). In these cases, too,
the spreading property is the I element.
8. Affrication without palatalization
As pointed out before, the type of the affrication without palatalization

as seen in Japanese (/-tu-/ [-tsui-]), Quebec French (/-ti-/ [-tsi-], /-dy-/
[-dzy-]) and Korean (/-ti-/ [-tsi-]) differs from the type of affrication trig-
gered by palatalization in that the triggering property is external to the
target. That is to say, the high vowels affect the preceding coronal stop
without spreading to a preceding onset position. This type of affrication
is challenging to any autosegmental framework that attempts to capture
a non-arbitrary relationship between a phonological process and a con-
text in which it occurs. The present framework is no exception. There
seems to be no logical connection between the vowels above (U) [ui], (I)
[i], (U.I) [y], (_) [i] and affrication. 6 Therefore, I shall limit myself to Jap-
anese and present a working hypothesis that is possible at a current level
of understanding.
As noted in Shibatani (1990:161), the high-back unrounded vowel [ui]
is centralized after /s/, /z/, /t/ and pronounced as in [süi, dziïi, tsüi]. I as-
cribe this centralization to the spreading of the empty head of the coro-
nals to the following vowel.7 Consider (17) in which the centralized high
vowel [ώ] is represented as the empty-headed PE (U._) occurring in a nu-
clear position.
(17) s UI s UI
X X X X
ROOT
RESONANCE
l\ l\
• h • h
U U
The U element is the head of [ui]. However, the spreading empty head be-
comes head and U is demoted to operator, generating [tu].8 If this account
of the centralized vowel is correct, then it follows that the affrication of
It, d/ before HI and lui have one thing in common. That is, in both cases an
onset and a following nucleus share a property under the Resonance
Node, i.e. the type of affrication with concomitant palatalization involves
the I element shared by an onset and a following nucleus, whereas the
type of affrication without palatalization involves an empty head shared
by the two positions. To unify the two types of affrication, the I-? Con-
straint in (6) can be modified as follows
(18) Sharing Constraint

? cannot c-command a property shared with a following nuclear po-
sition.
The change of /tu/ to [tsüi] in light of this constraint is as follows:

210 Shohei Yoshida
(19) t UI
ui t Iii
X X -> X X
ROOT T -h h •
· .
•v—
l \ 1
RESONANCE •· ?? ψ · ? · · h ·
·
I \ / \
U υ
The Sharing Constraint disallows the empty head of t to be shared with

the following nuclear segment. As a result, the Root and Resonance
Nodes of t are fissioned into two. Each of the two resulting Resonance
Nodes has its own empty head, and the second Resonance Node shares
its empty head with the following vowel. In the resulting structure of [ts],
the ? element does not c-command the shared empty head.
It should be pointed out that empty headed PEs that can occur in non-
nuclear positions are not limited to coronals but velar consonants (.. .U._)
are also empty headed. Notwithstanding, the vowel [ui] is not centralized
after velars. The difference between coronals and velars is the presence of
the U element in the latter consonants. Thus the situation is that the head
U in [ui] is not demoted to operator if a preceding consonant also contains
U. This can be accounted for by assuming that the headship of the U ele-
ment in [ui] is sanctioned by what Charette and Göksel (1994) call bridge,
a well-established concept in the element theory literature (e.g. Cyran
1997; Bloch-Rozmej 1998; Lee and Yoshida 1998) whereby some status or
ability of an element is licensed by the same element in an adjacent posi-
tion. As shown in (20), the headship of U in [ui] is protected by the same
element in k and so the empty head of k has no effect on the following vow-
el since an empty operator is phonetically uninterpretable.
(20) k ui
h χ χ
ROOT ?— * *
RESONANCE · _ ·
/ \ VI
υ _' υ
ι t
U-bridge
9. Conclusion
The element-based representations of segments and the I-? Constraint al-
lows us to analyze the various types of palatalization in Japanese as a uni-
tary process, i.e. I-spreading. The spreading I element is either linked to a
preceding non-nuclear position, adding a secondary articulation to a seg-
ment that already occupies the position (e.g. kJ, gj, pi, bi), or is incorporat-
ed into the preceding segment. When I is incorporated into dental-alveo-
lar stops, the place-defining property of dental-alveolars (i.e. the empty
head) and I combine to yield a palato-alveolar articulation. Further, si-
multaneous affrication takes place to avoid the violation of the I-? Con-
straint. The resulting segments are palato-alveolar affricates. When the I
element is incorporated into velar stops, it either replaces their U element
or becomes the place-defining property of the second Resonance Node
without affecting their U element and empty head, yielding palato-alveo-
lar affricates and velar-palatal affricates, respectively.
Unlike the type of affrication with palatalization, a cross-linguistic ac-
count of the type of affrication without palatalization is yet to be discov-
ered. I have tentatively proposed a constraint to account for the two types
of affrication in Japanese.
Notes
* I thank Mohammad Badawy, Hee Gyong Kang, Kyong-Young Lee, Manjae Lee, Shin-
Mee Park and Olivier Tesson for providing and confirming the Arabic, French and Ko-
rean data. I gratefully acknowledge the extremely helpful comments and suggestions of
Wiebke Brockhaus, Duck-Young Lee and Jeroen van de Weijer on an earlier draft of this
paper. Infelicities which remain despite their help are only my responsibility.
1. The effects of the five vowel segments on Iti, Id/ and /z/ are unchanged whether the vowel
segments are short, long or the first elements of diphthongs.
2. The assumption that nasality and voicing are expressed by two different elements is of the
standard element theory. Nasukawa (1998) has claimed that voicing and nasality are ex-
pressed by a single element. However, the following analysis does not hinge on this issue.
3. The representation and IPA transcription given here are of the moraic nasal that is fol-
lowed by pause, vowels, glides and fricatives. I claim that the apparent velar closure of
the moraic nasal before pause (e.g. hon [horj] 'book') is due to an unreleased glottal stop
frequently heard in prepausai position, e.g. te [te 7 ] 'hand', hai [hai ? ] 'yes'. As no one, I
believe, would include the prepausai glottalization as part of the lexical representations
of these words, the closure of the prepausai moraic nasal should not be deemed part of
its lexical representation, either, [ΰι] immediately followed by the glottal closure is indis-
tinguishable from [g], When followed by an onset filled with a P E containing the ? ele-
ment, that P E spreads to the moraic nasal, making a homorganic nasal-stop/affricate
cluster.
212 Shohei Yoshida
4. I assume that the representation of the alveolar lateral approximant [1] is (?); (_).
5. It goes without saying that nothing prevents the I element from becoming a secondary
palatal articulation of dental-alveolars in languages.
6. The representation of the Korean vowel [i] as an empty nucleus is taken from Lee (1998).
7. In the GP literature, empty-headed PEs in nuclear positions have been viewed as repre-
sentations of lax vowels (e.g. Charette and Göksel 1994), while headed PEs have been
seen as those of tense vowels.
8. The term 'spreading' here is a metaphor for 'the alignment of two heads'. See Kaye, Lo-
wenstamm and Vergnaud (1985) and Lee and Yoshida (1998) for details.
References
Avery, P. and K. Rice
1989 Segment structure and coronal underspecification. Phonology 6.2,179-200.
Bhat, D. N. S.
1978 A general study of palatalization. In J. H. Greenberg, C. A. Ferguson and E.
Moravcsik (eds.), Universals of Language, Vol. 2: Phonology, 47-92. Stanford
University Press.
Bloch-Rozmej, A.
1998 In E. Cyran (ed.), 41-54.
Broadbent, J.
1991 Linking and intrusive r in English. UCL Working Papers in Linguistics 3,281-
302.
Brockhaus, W.
1995 Final Devoicing in the Phonology of German. Tübingen: Max Niemeyer Ver-
lag.
Catford, J. C.
1988 A Practical Introduction to Phonetics. Oxford: Oxford University Press.
Charette, M. and A. Göksel
1994 Vowel harmony and switching in Turkish languages. SOAS Working Papers in
Linguistics and Phonetics 4,31-52, University of London.
1998 Licensing constraints and vowel harmony in Turkic languages. In E. Cyran
(ed.), 65-88.
Cho, Y. Y.
1991 On the universality of the coronal articulator. In C. Paradis and J.-F. Prunet
(eds.), 159-180.
Chomsky, N. and M. Halle
1968 The Sound Pattern of English. New York: Harper and Row.
Cobb, M.
1995 Vowel harmony in Zulu and Basque: The interaction of licensing constraints,
h-licensing and constituent structure. SOAS Working Papers in Linguistics
and Phonetics 5, 23-40, University of London.
Cyran, E.
1997 Resonance Elements in Phonology: A study in Munster Irish. Lublin: Wy-
dawnictwo Folium.
Cyran, E. (ed.)
1998 Structure and Interpretation: Studies in Phonology. Lublin: Wydawnictwo Fo-
lium.
Flikeid, K.
1988 Unity and diversity in Acadian phonology: An overview based on compari-
sons among the Nova Scotia varieties. Journal of the Atlantic Provinces Lin-
guistic Association 10: 64-110.
Fujiwara, Y.
1997 Nihongo-hougen no jissou 'The facts of Japanese dialects'. Zoku Shouwa ( •
Heisei) Nihongo-hougen no Sougou Kenkyuu Ά comprehensive study of Jap-
anese dialects in Showa (-• Heisei)' Vol. 4. Tokyo: Musashino Shoin.
Harris, J.
1990 Segmental complexity and phonological government. Phonology 7, 255-300.
1994 English Sound Structure. Oxford: Basil Blackwell.
Harris, J. and G. Lindsey
1995 The elements of phonological representation. In J. Durand and F. Katamba
(eds.), Frontiers of Phonology: Atoms, Structures and Derivations. Harlow:
Longman.
Ingleby, M., C. Chalfont and W. Brockhaus
1994 Robust techniques for recognition of new knowledge-based speech primi-
tives. In R. Togneri (ed.), The Proceedings of the Fifth Australian International
Conference on Speech Science and Technology. Canberra: Australian Speech
Science and Technology Association, vol. 1, 374-380.
Johnstone, Τ. M.
1967 Eastern Arabian Dialect Studies. Oxford:Oxford University Press.
Kaye, J., J. Lowenstamm and J. R. Vergnaud
1985 The internal structure of phonological elements: A theory of charm and gov-
ernment. Phonology Yearbook 2, 305-328.
1990 Constituent structure and government in phonology. Phonology 7,194-232.
Kiparsky, P.
1982 Explanation in Phonology. Dordrecht: Foris Publications.
Lahiri, A. and S. E. Blumstein
1984 A re-evaluation of the feature coronal. Journal of Phonetics 12:133-146.
Lahiri, A. and V. Evers
1991 Palatalization and coronality. In In C. Paradis and J.-F. Prunet (eds.), 79-100.
Lee, D.-Y.
1998 Korean Phonology: A Principle-based Approach. Munich: Lincom Europa.
Lee, D.-Y. and S. Yoshida
1998 Α-head alignment: the case of vowel harmony in Korean. In E. Cyran (ed.),
195-204.
Nasukawa, K.
1998 An integrated approach to nasality and voicing. In E. Cyran (ed.), 205-226.
Paradis, C. and J.-F. Prunet
1991 Introduction: Asymmetry and Visibility. In C. Paradis and J.-F. Prunet (eds.),
The special Status of Coronals: Internal and External Evidence. San Diego:
Academic Press, 1-28.
Poser, W.
1985 Japanese evidence bearing on the compensatory lengthening controversy. In
L. Wetzels and E. Sezer (eds.), Studies in Compensatory Lengthening. Dor-
drecht: Foris Publication, 167-186.
Rennison, J.
1990 On the elements of phonological representations: the evidence from vowel
systems and vowel processes. Folia Linguistica 24,175-244.
214 Shohei Yoshida
Shibatani, M.
1990 The Languages of Japan. Cambridge University Press.
Williams, G.
(1998) The Phonological basis of speech recognition. Ph.D. dissertation, School of
Oriental and African Studies, University of London. Also in Williams (forth-
coming).
(forthcoming) The Phonological Basis of Speech Recognition. The Hague: Holland Ac-
ademic Graphics.
Williams, G. and W. Brockhaus
1992 Automatic speech recognition: a principle-based approach. SOAS Working
Papers in Linguistics and Phonetics 2, 371-401.
The Accent System of the Kyoto Dialect of Japanese:
A Study on Phrasal Patterns and Paradigms
Yuko Z. Yoshida and Hideki Zamma

Doshisha University and Kobe City University of
Foreign Sudies
Introduction
This paper aims to explore the cross-categorical characteristics of pitch

accent in the Kyoto dialect of Japanese (KJ). The discussion centres on
the detailed study of various classes of nouns, case-marking particles,
verb paradigms and phrasal interactions. Some sections will show inter-
esting contrasts between KJ and Standard Japanese (SJ) in terms of
pitch accent behaviour. To identify the contrasts between the two dia-
lects, we will focus on pitch phenomena both in single phrases from var-
ious grammatical categories and in the merged phrases found in con-
nected speech.
First, the pitch pattern of a noun plus a case-marking particle in KJ
demonstrates that there are two classes of so-called 'accentless noun'.
This discussion of accentless nouns leads directly to a phrasal analysis, in
which we observe interesting accentual behaviour. We will claim that KJ
utilises a pseudo accent—an accent which is phonologically assigned in a
given accentless domain—which should be distinguished from a lexical
accent. A further claim involving the inaccessibility of the domain-initial
position lends support to the idea that there are two types of 'accentless'
classes which are found cross-categorically.
In terms of typological classification, KJ is presented here as an ac-
cent language, as distinct from a tone language. In some earlier studies,
the pitch patterns of this dialect, like those of SJ, were treated as tone
melodies (Haraguchi 1977), grouping the two dialects together as tone
languages (which also make use of accentual information). Our propos-
al of the inaccessibility of the domain-initial position places KJ solely
into the accent language group, having no recourse to tone melodies of
any kind.
216 Yuko Ζ. Yoshida and Hideki Zamma
1. The Kyoto dialect and its typological grouping
1.1. Examples of noun + case-marking particle

As set out below, the KJ data of case-marked nouns reveal some interest-
ing pitch properties in the dialect. The contrast shown below between KJ
and SJ should highlight the properties in question. Note also a lexical vari-
ation: KJ contains lengthened words, i.e. bimoraic 1 words, which in other
dialects (e.g. SJ) correspond to monomoraic words (1). For example, ha
'teeth', ki 'tree' and e 'handle' in SJ are lengthened to haa, kii and ee, re-
spectively. Pitch patterns in the two dialects do not necessarily correlate,
i.e. the pitch patterns of some lexical items in one dialect cannot always be
deduced from those in the other dialect. In the examples given below, a bar
over a segment indicates that the relevant part is perceived with high pitch
(other segments have default pitch) and a * denotes a lexical accent.
Kyoto Dialect Standard Japanese

(1)
a. ha a ha a -ga 'teeth (-nom.)' ha ha -ga
b. k i i ki i-ga 'tree (-nom.)'
c. e e e e -ga 'handle (-nom.)' e e-ga
(2)
* * * *
a. ka ki ka ki -ga 'fence (-nom.)' ka ki ka ki -ga

* * * *
b. ka ki (i) 2 ka ki -ga 'oyster (-nom.)' ka ki ka ki -ga

c. ta ne ta ne -ga 'seed (-nom)'
d. ka ki ka ki -ga 'persimmon (-nom.)' ka ki ka ki -ga
(3) * * * *
a. ma ku ra ma ku ra -ga 'pillow (-nom.)' ma ku ra ma ku ra -ga

* * * *
b. ta ma go ta ma go -ga 'egg (-nom.)' ta ma go ta ma go -ga

* *
c. non applicable 'head (-nom)' a ta ma a ta ma -ga

d. u sa gi u sa gi -ga 'rabbit (-nom.)' u sa gi u sa gi -ga
e. sa ku ra sa ku ra -ga 'cherry (-nom.)' sa ku ra sa ku ra -ga
The data above show that in both SJ and KJ, a noun in isolation, or a se-
quence of a noun plus a particle, may form a phonological domain of pitch
accent. Otherwise, we should find pitch patterns where the noun portion
and the particle behave independently in terms of pitch assignment. For
The Accent System of the Kyoto Dialect of Japanese 217
example, a sequence of a trimoraic word and a particle may show the

pitch pattern, HOOH (H stands for a high-pitched mora, and 0 denotes a
mora without high pitch). That is, the particle portion dominates no inde-
pendent accent domain, except for special styles of speech.
Forms without any lexical accent divide into two classes in KJ, whereas
SJ allows only one type of pitch pattern for accentless words. In SJ, all the
morae up to the rightmost edge of the given phonological domain are
high-pitched in a phrase which lacks a lexical marking. In a phrase (a pho-
nological domain) with a lexical accent, all the morae up to the accented
one appear high-pitched except for the initial mora (if the initial mora it-
self does not bear the lexical accent). To put this in another way, we ob-
serve a sudden drop in pitch immediately following the accented mora. In
KJ, mono-moraic words are usually lengthened and there are three possi-
ble pitch patterns (see (1))—one pattern more than we find in SJ. Four
possible pitch patterns in bimoraic words include two types of 'accentless'
words: (2c) represents a word without a lexical accent, where high pitch
falls on the ultimate mora of the phonological phrase, while (2b) shows a
word with high-pitch extending throughout the whole phonological
phrase. Two other possibilities find the words accented either on the ini-
tial (2a), or the second, coinciding with the final position (2b) of the word.
KJ lacks examples of trimoraic words with the accent on the final mora,
unlike SJ, where this type is illustrated by forms such as atama'ga (3a).
The lexical accent may fall on the initial (3a) or the second (3b) mora.
Also for trimoraic words (see (3)), there are two types of accentless
words, as explained for bimoraic words. Longer words without internal
morphology do exist, though not in abundance. In fact, as we discuss later
(Section 2.2.), longer words are subject to accent assignment unlike words
with a lexical accent. (For convenience, the symbol for that assigned ac-
cent is indicated here by *, the same as that of the lexical accent.)
Kyoto Dialect Standard Japanese

(4) *
a. mu ra sa ki mu ra sa ki -ga 'purple ( nom)' mu ra sa ki mu ra sa ki -ga

* * * *
b. u gu i su u gu i su -ga 'bush warbler (-nom)' u gu i su u gu i su -ga
As for quadrimoraic morphologically simplex Yamato words, there are

only two patterns, both of which have an accent on the second (i.e. the an-
tepenultimate) mora. One type allows the high pitch to be shared with
morae to the left, whereas the other type does not.
Based on the observations above, the next section considers how KJ

should be categorised in terms of prosodie typology.
1.2. Typological grouping of the Kyoto dialect
The pitch patterns that are predictable from accentual information place
SJ and KJ properly within the accent-language group, comprising either
stress or pitch accent languages, rather than in the tone-language group.
From the location of the accent, those segments with high-pitch and those
without high-pitch can be predicted. In a tone language, each syllable
contains individual tonal information, high or low, and thus the location
of the high and low tones for a given word is lexically determined (i.e. ar-
bitrary). Nevertheless, the distribution of morae which are not high is pre-
dictable both in SJ and KJ, once we know the high pitch distribution. For
this reason, the present discussion of nouns in KJ does not consider a low
pitch or tone, but only deals with high pitch information.
Let us focus on the analysis by Haraguchi (1977) of the Osaka dialect,
which he chooses as being representative of the Kansai dialects, or the
Kyoto-type accent, 3 to show how lexical tone melodic information is irrel-
evant to the present discussion, and to support the typological claim
above. Haraguchi postulates two basic tone melodies, which are HL and
LHL, to differentiate the pitch patterns of the two groups of nouns and to
determine whether the word should carry high pitch on its initial mora or
not. Η of the tone melody is linked to the accented mora, or to the final
mora of an accentless phrase. The initial mora is high pitched, when the
word accent is on the initial mora, or if the pitch 'spreads' to the initial
mora from the accented one 4 or from the final mora in an accentless do-
main. Examples of the former type are more likely to be found in longer
words, such as compounds or quadrimoraic native nouns like uguisu (see
(4b)), where the accent is on the second mora of the word.
The analysis by Haraguchi accesses both accentual information and
tonal information for every lexical item, which means that the dialect is
questionably categorised both as an accent language and as a tone lan-
guage. Our claim in this paper, that KJ only refers to information relating
to accentuation and positions in the accent domain (see Section 2 for a de-
tailed analysis), makes tonal information redundant. That is, the pitch
patterns of KJ illustrated above can be accounted for without treating the
dialect simultaneously as an accent language and a tone language. We
would like to propose a simplified interpretation which need not assume
these tone melodies as lexical properties, and we argue that in this way KJ
may be properly defined as a member of the accent language group.
Section 2 is devoted to a detailed explanation of our analysis, and
broaches the issues of 1) how the diverse pitch patterns seen in the above
data appear, 2) lexical accent assignment on the antepenultimate mora,
and 3) the inaccessibility of the initial mora. To begin with, we explain
how various pitch patterns are possible in KJ, eschewing somewhat tem-
plate tone melodies.
2. Accents and domains
In this section, we propose a definition of pitch accent languages and

show how our analysis works. First, the source of high pitch is discussed
in relation to the headship of a phonological domain, and then the phe-
nomenon formerly treated as 'spreading' is re-analysed as part of the pho-
netic interpretation of the headship in the relevant phonological domain.
2.1. Headship of a phonological domain and high pitch
In both stress accent languages and in pitch accent languages, the head
portion (the head of a syllable or mora—or, more precisely the nucleus)
of a given phonological domain bears the primary accent. If a domain
contains a lexical marking, then the lexically marked mora is automatical-
ly identified as the domain head. In a pitch accent language, the high pitch
originating from a lexical accent may be shared by neighbouring seg-
ments. SJ demonstrates a relatively simple sharing mechanism: the high
pitch of the head mora is shared by all the morae to the left, except the
domain-initial one. The domain-initial mora (nucleus) is inaccessible for
high-pitch sharing, perhaps to indicate the domain or word boundary
(Yoshida 1990). KJ chooses to share the high pitch of the accented mora
with all the morae to the left, if the sharing is observed within the domain.
We say if because another lexically determined option is available for Ky-
oto words: whether or not to share the high pitch with the neighbouring
segments. Compare, for example, two quadrimoraic words with antepen-
ultimate accent, murasaki 0H00 and uguisu ΗΗ00 (see (4a, b)).
A question arises as to which mora in an accentless domain should bear
headship. If the high pitch originates from a lexical accent as explained
above, then the source for the high pitch in a lexically accentless domain
must be identified. Also, note that not only in accented quadrimoraic

words but also in lexically accentless words, the two options of pitch shar-
ing (to share or not to share in the sense explained above) are available.
Neither (3d) nor (3e) above bear a lexical accent; furthermore, the former
is of the type where only the word- (or phrase-) final mora is high-pitched,
whereas pitch sharing is observed in the latter.
An accentless domain, regardless of its morphological organization, al-
ways possesses a high-pitched portion. In other words, the accentless do-
main should have a head portion which behaves like a lexical accent to
'provide' the high pitch. In this regard, Kaye (1990) proposes the Licens-
ing Principle, which applies to all phonological domains, including pitch
accent domains.
(5) The Licensing Principle (Kaye 1990: p. 306,11.11-13)

All phonological positions save one must be licensed within a domain.
The unlicensed position is the head of this domain.
KJ shows clearly the location of the head portion of a domain that lacks
an inherent head. One type of accentless word clearly indicates that the
domain-final mora is the head, both in the domain of a noun in isolation,
and in the domain comprising a noun and a particle (Yoshida 1999):
(6) Note: [ ] indicate the phonological domain relevant for pitch

a. i) [ki i] ii) [ki i -ga] 'tree (-nom.)'
b. i) [a na] ii) [a na -ga] 'hole (-nom.)'
c. i) [u sa gi] ii) [u sa gi -ga] 'rabbit (-nom.)'
Only the rightmost mora is high-pitched, which assumes the headship of

the domain.
The other type of accentless word, which is high-pitched throughout
the domain, as in sakura-(ga) HHH(-H), should follow the same princi-
ples for selecting its head. The rightmost mora is the head of the domain,
where high-pitch is shared by all the other morae, 5 if the noun is lexically
determined as such. Our task here is to identify the lexical cue for the two
classes of accentless words, which will be discussed in detail later (see
3.2.1.).
The next subsection considers the default lexical accent location in KJ,
which supports our classification of the dialect as an accent language. A
recent change in KJ, whereby some features of stress accent languages are
becoming apparent, assigns accent on to predictable locations.
2.2. Antepenultimate accent

The following general tendency in present-day KJ affects quadrimoraic
words. The accent location in quadrimoraic words, both verbs and nouns,
is changing—word-initial accents are being replaced by accents on the an-
tepenultimate mora. Accents at fixed locations suggest that KJ belongs to
the group of accent languages which perhaps conform to metrical accent
assignment, as in the case of SJ, in which an accent is assigned on the an-
tepenultimate mora (Yoshida 1995).
Quadrimoraic nouns including morphologically simplex/complex
nouns and many proper nouns which were once accented on the initial
mora are now accented on the second mora.
(7)
a. 'nightingale' b. 'Kuwatani (a surname)' c. 'Kiyomizu (a place name)'
* * *
u gu i su ku wa ta ni ki yo mi zu
I 1 i
ψ ψ ή: % %
u gu i su / u gu i su ku wa ta ni / ku wa ta ni ki yo mi zu / ki yo mi zu
This change in accent location follows the default lexical accent assign-
ment process. A close observation of the accentual distribution of trimo-
raic words exemplified in (3) above clearly reveals that the default lexical
accent location is the antepenultimate mora. Sampling 115 morphologi-
cally simplex yamato (Japanese native) nouns from the word list in
Hirayama (1957), we report the following distribution.
(8)
Example number
*
initial accent (3a) 42
i ta ti 'weasel'
*
medial accent (3b) 20
ha ta ke 'farm'
final accent (3c) n.a 0
rising accentless (3d) 12

su zu me 'sparrow'
high accentless (3e) 41

ka su mi 'haze'
222 Yuko Z. Yoshida and Hideki Zamma
If lexically accented, 42 out of 62 accented trimoraic y amato words bear

word-initial (i.e. antepenultimate) accent. Trimoraic yamato words with
word-final accent are not attested in the dialect; in fact, this dialect avoids
placing lexical accents on the final mora of the word. Observation of bi-
moraic words confirms that a word-final accent is not favoured: those fi-
nal accented nouns are subject to lengthening (see note 2). The fact that
about 68 % of all accented words bear word-initial (antepenultimate) ac-
cent proves that the default location of the lexical accent in KJ is the an-
tepenultimate mora.
The change in accent location noted at the beginning of this section for
quadrimoraic nouns is cross-categorical. A change observed in the Kyoto
verbs also involves a shift in preference towards antepenultimate accent.
These old forms are taken from Hirayama (1957), while the new forms
are based on our own recordings of native speakers.
(9) OLD FORM NEW FORM

* *
a. tu ku t ta 'make-past' tu ku t ta
* *
b. u ka Ν da 'float-past' u ka Ν da
* *
c. na ra Ν da 'queue-past' na ra Ν da
* *
d. si zu Ν da 'sink-past' si zu Ν da
* *
e. ka wa i ta 'dry-past' ka wa i ta
* *
f. ku da i ta 'crush-past' ku da i ta
* *
g. ta ta i ta 'beat-past' ta ta i ta
* * *
h. a ro o ta 'wash-past' a ro o ta ~ a ro ta
* * *
i. u to o ta 'sing-past' u to o ta ~ u to ta
The data in (7) and (9) indicate that at least the modern form of the dia-
lect shows behaviour similar to that seen in stress languages: the accent
falls at a predictable location, the antepenultimate mora. The type of pho-
netic interpretation of the head portion determines whether the language
is a stress-accent language or a pitch accent language. The accented
(head) portion receives phonetic interpretation either as prominence/
strength for a stress accent, or as pitch-height for a pitch accent in a given
domain (Yoshida 1995).
In the quadrimoraic nouns (7a) and verbs given above, there are two
possibilities; either, pitch is shared between the initial mora and the ac-
cented penultimate mora (9b-i), or it is not (9a). Recall the data (1-3) for
a general overview of this issue. As for shorter nouns, bimoraic 6 and tri-
moraic Yamato nouns, only accentless ones can vary in this way. Logically,
we should be able to observe pitch sharing on lexically accented bimoraic
and trimoraic words if the lexical accent is on the second mora (sharing
with the initial one). This possibility is exemplified by the quadrimoraic
noun uguisu (HHOO). In order to identify which factors determine wheth-
er or not pitch sharing occurs in a given domain, we turn to the case of ac-
centless nouns, which allow both possibilities.
3. The special status of domain edge
3.1. Domain final position
3.1.1. Domain-final mora and 'accentuation'

In 2.1., we explained that the domain-final mora serves as the head of a
domain in which no inherent accent marking exists. As discussed above,
there are two types of domain: only one type bears high pitch on the head
mora (10a, c, e), while the other type shares the pitch with all the other
morae in the domain (10b, d, f).
(10)
a. ki i ki i -ga 'tree (-nom.)'
b. e e e e -ga 'handle (-nom.)'
c. ta ne ta ne -ga 'seed (-nom)'
d. ka ki ka ki-ga 'persimmon (-nom
e. u sa gi u sa gi -ga 'rabbit (-nom.)'
f. sa ku ra sa ku ra -ga 'cherry (-nom.)'
To investigate the possible reasons behind the two different pitch patterns
found in accentless domains, we must first verify whether these two types
both really lack 'accentuation' or not: if one has accentuation and the oth-
er does not, then a clear difference between them is immediately appar-
ent. These domains do not contain any information concerning the loca-
tion of lexical markings; however, it is not yet clear if the head mora, the
domain-final mora, receives accentuation as well as high pitch as a means
of indicating headship. The presence of accentuation is clearly shown, if it
appears word-medially; since the pitch drop is observed between the head
mora and the mora immediately following it. What is required is a reliable
test which reveals phrase-final accentuation in SJ. The juxtaposition of
two phonological phrases in connected speech exposes the accentuation
of the phrases involved, as demonstrated below in (11), where we observe
how the noun in isolation and with the case-marking particle react to be-
ing placed next to another phrase, e.g. a verb.
(11) Standard Japanese

* *
a. wa sa bi wa sa bi -o 'horseradish (acc.)'
* *
b. a zu ki a zu ki -o 'red bean (acc.)'

c. sa ka na sa ka na -o 'fish (acc.)'
*
a', wa sa bi - o ta be ta '(I) ate some horseradish'

*
b'.azuki-o ta be ta '(I) ate red beans'

*
c'. sa ka na - o ta be ta '(I) ate some fish'
In connected speech, as (lla'-c 1 ) above show, only if the noun plus a par-
ticle phrase contains a lexical accent will the accent of the verb not be per-
ceived (known as phrasal downstep), since the leftmost accent is the one
projected to the noun-verb domain. As in (11c'), if no accent is found in
the noun portion, the accent of the verb becomes the accent of the entire
phrase. Bearing this in mind, let us consider the Kyoto phrases.
(12)Kyoto
*
a. wa sa bi wa sa bi - o 'horseradish (-acc.)'
b. ki tu ne ki tu ne - o 'fried bean curd (-acc.)'
c. sa ka na sa ka na - o 'fish (-acc.)'
*
a', wa sa bi - o ta be ta
b'. ki tu n e - o- ta
. be
,(*)ta * ki tu n e - o ta be ta * ki tu n e - o ta be ta
* *
c'. s a k a n a - o ta be ta * sa ka na - o ta be ta * sa ka n a - o ta be ta
The pitch patterns in (12) show that all the noun phrases in Kyoto should
bear accentuation. All the examples are subject to phrasal downstep. This
phenomenon indicates that both (12b, c) show accentuation in the do-
main-final position (Section 2.1.). The domains do not have lexical ac-
cents (the lexical accent of a noun should remain intact following its con-
catenation to a nominative marker), but they do show 'accentuation' in
the domain-final position (the head position). Here we propose the fol-
lowing for KJ:
(13) Proposal 1
Nouns which belong to the classes (lb, c), (2c, d) and (3d, e) are all
LEXICALLY ACCENTLESS. The rightmost segment of such a
word is assigned a domain ACCENT, which we call a PSEUDO
ACCENT.
Referring to the SJ accentless word, in which the domain-final nucleus

(the head portion of the final mora) acts as the domain-head, we see that
the head receives the high-pitch interpretation but no accentuation. In
contrast to the SJ case, KJ interprets the headship in the domain-final nu-
cleus as accentuation, which naturally accompanies a high-pitch.
The question arises as to whether this interpretation as accentuation is
controlled via distinct dialectal parameters or whether it results from
some other condition. In both dialects it is true that, when the noun-
phrase constitutes a domain in its own right, headship is assigned. It is
only when the juxtaposition of a noun phrase and a verb occurs, that the
accentual difference between the two dialects comes to light. Let us
therefore compare the behaviour of the two dialects in this respect. If the
noun phrase in KJ acquires a pseudo accent, as proposed above, then the
noun phrase should occupy its own domain [[noun phrase] verb] or
[[noun phrase][verb]] rather than merging with the verb [noun phrase—
verb]. Let us consider this possible concatenation in more detail.
The difference in phrasal structure between SJ and KJ arises from syn-
tactic relations. By 'syntactic relations' we refer here to the way in which
phonological processes should apply to the phonological string—a pho-
nological domain which can be a noun, a noun phrase, a longer phrase or
a sentence. A domain must have a head (Kaye 1995), and the inherently
accented mora automatically serves as the head in a lexically accented
noun. The final mora of the accentless phonological word (domain) takes
over the headship of that domain. Note that in SJ this is not interpreted
as an 'accent', but merely as high pitch. Yoshida (1995) proposes that the
noun-phrase and the verb form independent domains (14a). When con-
catenated, the leftmost accent of the noun-verb string is projected to the
domain of AB as the accent of the entire phrase.
(14) a. [[A][B]] domains: A, B, AB

b. [[A] Β ] domains: A, AB
c. [ A Β ] domain: AB
What follows is a general formula describing SJ accent assignment in

phrases and compounds:
(15) [ C [A][B] C ]
In SJ, where both A and Β contain lexical marking, i) if C is a lex-
ical category, Β is strong, ii) if C is a phrasal category, A is strong,
and the marked nucleus of the strong element is the head of the
domain. (Yoshida 1995:194)
Of relevance here is (ii) in (15), when the category is phrasal. Below in

(16) are examples repeated from (11) with bracketing to show the gener-
alization in (15).
(16)
*
a. [[wa sa bi -o] [ta be ta]] '(I) ate some horseradish'

* ^
b. [[a zu ki -o] [ta be ta]] '(I) ate red beans'

*
c. [[sa ka na -o] [ta be ta]] '(I) ate some fish'
Merging of two domains with lexical markings into one forces the output
domain to select one of the two markings as the head of the output do-
main. As is generalised in (15), the leftmost lexical marking is chosen as
the head of the merged domain, since the domains involved here are both
phrasal. In SJ, as (16a-c) show, only if there is a lexical accent within the
noun-phrase domain will the accent of the verb not be perceived. As in
(16c), if no lexical accent is found in the noun portion, the accent of the
verb becomes the accent of the merged domain.
Referring back to (12a'-c'), to compare with the situation in SJ (16a-c),

we see that even the lexically accentless domains of noun phrases ( l i b ' ,
c') trigger phrasal downstep, which means that the domains of noun
phrases behave as if they are accented (pseudo accent (13)).
(17b) shows the instruction for assigning pitch to noun-verb sequences
in KJ. The phonological parsing of the noun-verb domain supercedes that
of the verb domain, where the verb does not constitute its own domain.
Note, as the asterisked bracketing illustrates, if both the noun and the
verb dominate one domain each, as in the SJ case, then the lexical accent
of either domain should be projected in connected speech. The only do-
main with a lexical accent to be projected to the noun-verb domain is that
of the verb. This bracketing would predict that high pitch occurs only on
the accented mora of the verb, which is unattested.
(17)
*
a. [[wa sa bi - o ] ta be ta]
*
b. [[ki tu ne - o ] ta be ta] * [[ki tu ne - o ] [ta be ta]] *[[ki tu ne - o ] [ta be ta]]
(*)ta]
c. [[sa ka n a - o ] ta be * [[sa ka n a - o ] [ta be ta]] *[[sakana-o] [ta be ta]]
When the noun lacks the overt case-marker, the bracketing remains the
same.
(18)
-1 (*)
a. [[wa sa bi] ta be ta ]
_ (*)
b. [[ki tu ne] ta be ta] * [[ki tu ne] [ta be ta]] * [[ki tu ne] [ta be ta]]
* *
(*)
c. [[sa ka na] ta be ta] * [[sa ka na] [ta be ta]] * [[sa ka na] [ta be ta]]
The properties of the inner domain are projected to the outer domain, and
in this [[A]B] structure, the head of A, the inner-domain, is projected as the
head of the domain AB. The inner domain lacks a lexical marking in kitune
and satana (18b, c)—unlike the inner domain of wasabi 'm (18a). Notwith-
standing, the domain A has to nominate the potential head for the outer
domain AB, thus the head of A, the domain-final nucleus, is projected to
the merged domain AB, to be interpreted as the pseudo accent.
In SJ too, there exist cases where a phrase such as a noun-particle se-
quence acquires an accent when followed by another case-marker 7 . The
location of the accent is indicated by the drop in pitch on the mora adja-
cent to the right of the accented mora, as observed in the phrase in (19c).
The fact that the dative noun phrase sakura-ni is not accented anywhere
in the phrase is demonstrated by the data in (19d), where the noun-phrase
is followed by an accented verb.
(19) SJ (Noun -Particle 1 -Particle 2)

sa ku ra sa ku ra -ni sa ku ra -ni - m o
'cherry tree' 'cherry tree -DAT.' 'cherry tree -DAT. -also'
d. * e. (*)
sa ku ra -ni tu ke ru sa ku ra -ni - m o tu ke ru
'to attach (something) to a cherry tree' 'to attach (something) also to a cherry tree'
Similarly in Kyoto, in the same sequence of particles an accentuation ap-

pears immediately in front of the second particle:
(20) Kyoto (Noun-Particle 1-Particle 2)

i) a. b. c.
ki tu ne ki tu ne -ni ki tu ne -ni - m o
'fox' 'fox -DAT.' 'fox -DAT. -also
ii) a. b. c.
sa ku ra sa ku ra -ni sa ku ra -ni - m o
'cherry tree' 'cherry tree -DAT.' 'cherry tree -DAT. -also'
However, the accentual behaviour of KJ in the sequence consisting of a

case-marked noun and a verb shows a departure from the SJ pattern. In
SJ, the phrase-final mora does not acquire an accent (11c'), whereas an
accent appears on the final mora of the noun-particle sequence in Kyoto
(12b', c').
3.1.2. Evidence for pseudo accent and domain organisation

In support of the acquired accent that we will label, 'pseudo accent', the
following evidence may be cited. There exists a phenomenon in SJ which
is sensitive to the domain-final accent. If, in KJ, we observe a similar kind
of phenomenon that reacts to a phrase-final pseudo accent, then we may
confidently claim the existence of such an accent.
The Genitival marker -no has the effect of deleting a noun's accent
when in juxtaposition. Precisely, when the particle follows a noun with
word-final accent, the accent of the noun is not perceived. Yoshida (1995)
analyses the Genitival marker as a lexically accented particle (see also

footnote 7) that induces a reaction—Accent Clash Avoidance—to elimi-
nate adjacent accents. For example, the accent of azuki in SJ falls on the
final mora; when -no is suffixed to the noun, however, the accent of the
noun is not perceived (21b). If the accent falls elsewhere, word-initially or
medially, the accent of the noun remains (21a). Note that the behaviour
of -no is unique in comparison to other case markers such as -ga (nom.),
which always respect the lexical accent of the nouns.
(21) Replies to the question: 'Which flavour did you choose?'

* *
a. wa sa bi [[wa sa bi] -no] 'horseradish (-Gen.)'

* s *
b. a zu ki [[a zu ki] -no] 'red bean (-Gen.)'

*
c. sa ka na [[sa ka na] -no] 'fish (-Gen.)'
Accent Clash Avoidance is only relevant in this [[A]B] domain structure.

In SJ, the syntactic relation between a noun and a particle is as illustrated
in (21), that is, [[noun] particle], where by default the lexical property of
the noun is projected to that of the noun-particle sequence. Compare this
to those cases where a verb with initial accent follows a word-finally ac-
cented noun without any overt case-marker: the two accents are adjacent.
(22) Accent Clash Avoidance—only in noun-particle sequences in S J

[[A] B] —not in phrasal sequences [[A][B]]
*
a. [[wa sa bi] [ta be ta]] '(I) ate some horseradish.'

*
b. [[a zu ki] [ta be ta]] '(I) ate red beans.'

*
c. [[sa ka na] [ ta be ta]] '(I) ate some fish.'
As in (22b), this avoidance phenomenon does not apply to the structure

[[A][B]]. Also note that in SJ, the domain-final mora only bears the head-
ship of the lexical-accentless domain (Yoshida 1995,1999), and does not
result in an actual accent. In SJ, headship in the domain-final position
only serves to support high pitch, so the segments to the right of that mora
(in phrasal sequences, etc.) are not subject to any loss of high pitch; the
sudden drop in pitch is only observed in the post-accented position.
In fact, Accent Clash Avoidance in KJ offers clear evidence for the sug-
gested domain structure and for the proposed pseudo accent. In this dia-
lect, phrases (where the noun phrase is represented as A and the verb as
B) are organised into [[A]B] and Accent Clash Avoidance is observed
following the juxtaposition of two accents. Note that in Κ J, as with other
Kansai dialects, overt case-marking particles tend to be omitted. Recall
that there is no noun class with a lexical accent on the word-final mora
(see (3)). So if there is actually a pseudo accent on the word-final mora of
the noun (domain A), then the accent should react to the presence of a
verb-initial accent.
(23) Pseudo Accent: χ

*
a. [[wa sa bi] mo ro ta]

(x) *
b. [[ki tu ne] mo ro ta] Accent Clash Avoidance -»· delete χ
(χ) _ * _*
c. [[sa ka na] mo ro ta] * sa ka na mo ro ta
Accent Clash Avoidance -»• merge χ & *
There are two types of Accent Clash Avoidance, one deleting the pseudo
accent in the noun (23b) and the other merging the pseudo accent with
the accent of the verb. Indeed, the location of the accent in the verb holds
the key to this phenomenon. Compare this with the data presented in
(18), which is repeated for convenience in (24). The accent is on the sec-
ond mora of the verb, and is not adjacent to the pseudo accent in the noun
portion.
(24)
*
a. [[wa sa bi] ta be ta ]
*
b. [[ki tu ne] ta be ta] * [[ki tu ne] [ta be ta]]

*
c. [[sa ka na] ta be ta] * [[sa ka na] [ta be ta]]
As seen above, the domain-final pseudo accent in KJ reacts when placed

next to another accent. This reaction supports the claim that, in KJ, there
is an acquired accent in the domain final position, which we call here a
pseudo accent.
3.2. Domain-initial position
3.2.1. Inaccessibility of domain-initial position
So far, we have examined two points. Firstly, both the usagi type (25a) and
the sakura type (25b) of noun are lexically accentless; secondly, those ac-
centless nouns acquire a pseudo accent in the domain-final position. They
do belong to the same lexically accentless class, although the actual pitch
patterns of the two types are different. In this section, we pursue the ques-
tion of how the differences in pitch patterns arise.
(25) _ _
a. u sa gi u sa gi -ga 'rabbit (-nom.)'
b. sa ku ra sa ku ra -ga 'cherry (-nom.)'
Basically, the difference between the sakura type and the usagi type of
noun rests on whether the high pitch is shared by other morae or not. The
span of high-pitch sharing in an accented word extends from the accented
mora to the leftmost mora of the noun. For example:
(26) High-pitch sharing in the Kyoto dialect —from the accented mora to
the leftmost mora of the relevant domain —
u gu i su ~ u gu i su
The form in (26a) shows that the high pitch on the accented mora can be
shared by an adjacent mora. Alternatively, the same noun can be pro-
nounced as in (26b), depending on individual speaker choice. This form
manifests itself only when the accented mora bears the high pitch without
extending it to other morae. For most quadrimoraic nouns, the same re-
sult emerges: there are two alternative pitch patterns (see (24a, a')).
Note that in SJ, the initial nucleus (the head of the initial mora) is inac-
cessible for high-pitch sharing (see Section 2.1.). The same is true in KJ,
although this system differs from SJ by marking it at the lexical level. In
SJ, where high-pitch sharing is obligatory for all classes of words where
sharing is possible, any phonological domain is subject to this inaccessibil-
ity. However, KJ chooses to treat this as lexical information, so the shar-
ing of high-pitch comes about as a result of referring to lexical properties.
(27) PROPOSAL 2
In the Kyoto dialect, the domain-initial mora may be lexically inac-
cessible for high-pitch.
In KJ, we do not observe high-pitch sharing if the initial mora is lexically

marked as being inaccessible8.
(28)
a. u gu i su ku wa ta ni ki yo mi zu
'nightingale' 'Kuwatani (a surname) 'Kiyomizu (a place n a m e ) '
*
a', u gu i su ku wa ta ni ki yo mi zu
'nightingale' 'Kuwatani (a surname) 'Kiyomizu (a place n a m e ) '
b. χ χ
to ri 'bird' sa ku ra 'cherry'
c. _x _x
a na 'hole' u sa gi 'rabbit'
The variations found in (28a, a') may be attributed to idiolectal differenc-

es—specifically, whether the domain-initial mora is deemed inaccessible
(28a) or not (28a1). The words without a lexical accent may contain this
lexical marking to indicate the inaccessibility, viz. whether to allow pitch
sharing or not. The unmarked case is observed in tori or sakura (28b),
where high-pitch sharing occurs. If marked (28c), no pitch sharing is ob-
served, leaving only the final position of the word (domain) high-pitched.
If the distribution of the two word types in (28b) and (28c) are compared,
referring to the table in (8), the sakura type occurs three times more fre-
quently than the usagi type.
If this analysis—that is, an analysis which utilizes initial inaccessibility
to account for the L-initial pitch pattern—is on the right track, then we
can adequately account for the system of accentuation in KJ, in that we
can attribute the difference between the two unaccented pitch patterns to
the presence or the absence of initial inaccessibility, rather than to the ba-
sic tone melodies, HL or LHL, as has been done in previous studies such
as Haraguchi (1977). This is a desirable result for the analysis of the Jap-
anese accentual system, because it conforms to the widely-held view of
Japanese as an accent language, as we argue in Section 1.2.
Moreover, this analysis can account for the data relating to verbal in-
flection and to noun accentuation in KJ. In what follows, we will discuss
these two issues, to show the suitability of the analysis.
3.2.2. Evidence for initial inaccessibility: Accentuation in verbal inflection

To outline the accentuation system of verbal inflection in KJ, let us first
compare it with the one in SJ, which is well-studied in the literature (cf.
Bloch (1946), McCawley (1968), Haraguchi (1977, 1991), Poser (1984),
etc.):
(29) Standard Japanese

a. Accentedness (i.e. accented or unaccented) is determined by the stem:
* *
tabe-ru 'eat (indicative)' tabe-ta 'eat (past)'

asob-u 'play (indicative)' ason-da 'play (past)' 9
b. There is only one pitch pattern for the accented verb; in Haragu-
chi's (1977) terms, it is the HL melody.
As shown in (29a), the lexically accented verb stem tabe- is always accent-
ed, whichever suffix is attached to it, although the placement of the accent
changes according to the choice of suffix. On the other hand, the lexically
unaccented verb stem asob- is always accentless. Moreover, the melody
for the accent is always HL, using the term employed in Haraguchi
(1977); that is, pre-accent morae all have high pitch (except for the first
mora, which is pitchless because of initial inaccessibility (cf. Yoshida
(1995)) or Initial Lowering (cf. Haraguchi (1977))), while the morae to
the right of the accent are all pitchless. When the word is accentless, on
the other hand, all the morae in the word have high pitch (again, except
for the initial one).
Both of these properties, however, differ dramatically in KJ, as shown
below:
(30) Kyoto Japanese

a. Accentedness is determined by the suffix.
b. There are two pitch patterns; i.e. HL and LHL melodies.
Recall that the accentedness of verbs in SJ is determined by the stem,

rather than by the suffix. In addition, while SJ has only one pitch pattern,
KJ has two. Let us exemplify these properties using concrete examples:
(31)
*
a.
agar-u 'rise (indicative)' agat-ta 'rise (past)'
ake-ru 'open (tr.) (indicative)' ake-ta 'open (tr.) (past)'

*
b.
aruk-u 'walk (indicative)' a ru i-ta 'walk (past)'
*
iki-ru 'live (indicative)' i ki-ta 'live (past)'
Compare the indicative forms and the past forms in (31). You will notice
that none of the verbs are accented in the former, while all are accented
in the latter (albeit with a difference in pitch pattern). This means that the
verb stem itself does not contain information about accentedness, and
that the latter is determined by the suffix attached to it; hence (30a). In
the cases above, the indicative suffix -(r)u is not accenting while the past
suffix -ta is accenting.
Moreover, the difference between (31a) and (31b) shows that there are
two pitch patterns both for accentless and accented words. The indicative
forms in (31a) and (31b) are both accentless, as we have just mentioned,
but only the final mora of the words in (31b) contain the high pitch. Sim-
ilarly, the accented past forms in (31b) have the initial pitchless mora,
which those in (31a) lack. This fact clearly suggests the appropriateness of
the generalization we gave in (30b). And, from the fact that the suffixes
do not play a crucial role in determining the pitch pattern, it is natural to
conclude that this is determined by the verb stems; thus in (31), the verb
stems agar- and ake- have, in Haraguchi's terms, the HL melody, while
aruk- and iki- have LHL.
3.2.2.1. Unaccented forms

Having clarified above that the accentedness of the verb is regulated by
the suffix attached to it, we now consider what kind of behaviour each suf-
fix shows. First, let us take the suffixes which make the form unaccented;
that is, indicative forms, polite forms, imperative forms, and hortative
forms. Examples of each are given below. (Henceforth, the examples are
represented in the following specific order: 2-mora verbs and 3-mora
verbs are those verbs whose indicative forms consist of 2 morae and 3 mo-
rae respectively, and C-final verbs and V-final verbs are those whose
stems end with a consonant and a vowel respectively. The examples of
group (a) begin with a high-pitched mora, and those of group (b) with a
pitchless mora. It is important to note that the examples given in this sec-
tion are merely representative forms, and thus all the verbs which fall into
the given category show similar pitch behaviour.)
2-mora verbs 3-mora verbs

C-final V-final C-final V-final
(32) indicative
a. iw-u10 ki-ru agar-u ake-ru
'say' 'wear' 'rise' 'open (tr.)'
b. aw-u mi-ru aruk-u iki-ru
'meet' 'look' 'walk' 'live'
(33) polite
a. iw-1-masu ki-masu agar-i-masu ake-masu
b. aw-i-masu mi-masu aruk-i-masu iki-masu
(34) imperative
a. iw-i ki-i agar-i ake 11
b. aw-i mi-i aruk-i i ki
(35) hortative
a. iw-o ki-yo agar-o ake-yo
b. aw-o mi-yo aruk-o i ki-yo
It is clear from the examples above that these suffixes do not assign an ac-
cent to the verb: all of the words are unaccented, with high pitch assigned
to all morae in group (a) words and to the final mora in group (b) words.
Moreover, the pitch pattern is consistent: in other words, if the initial
mora of the stem is high-pitched/pitchless in one of the forms with an un-
accented suffix, then the mora is also high-pitched/pitchless with other
unaccented suffixes. The various forms of the stem aw- 'meet', for exam-
ple, all have an initial pitchless mora, while those of iw- 'say' all have an
initial high-pitched mora.
There is, however, one exception to this generalization: the LHL and
HL patterns are neutralised in all forms of C-final 2-mora verbs, except
the indicative. Note that the stem mi- shares the same pattern with ki- in
(33)-(35). Because a highly idiosyncratic procedure would appear to be at
work here, we will ignore this class of words for the purposes of the
present discussion.
3.2.2.2. Accented forms

Next, let us turn to the other group of suffixes, i.e. those which place the
accent on the stem. The following forms show the accent on the mora im-
mediately preceding the suffix. (Note that there are several negative
forms in KJ.)
(36) negative (standardized)

*
iw-a-nai ki-nai agar-a-nai ake-nai

k * * * *
aw-a-nai mi-nai aruk-a-nai i ki-nai
(37) negative (archaic)

*
a.
iw-a-zu(-ni) ki-zu(-ni) agar-a-zu(-ni) ake-zu(-ni)
b * * *
aw-a-zu(-ni) mi-zu(-ni) aruk-a-zu(-ni) i ki-zu(-ni)
Note that the pitch pattern is consistent across all the forms, including
those in the previous section: iw- has an initial high-pitched mora in all
the forms from (32) through to (37), and similarly, aw- has an initial pitch-
less mora in all the forms. It is clear that the suffixes determine accentua-
tion, while the pitch pattern is determined by the stem.
3.2.2.3. Problematic forms and an analysis

So far we have considered regular forms—regular, in the sense that both
accentuation and pitch patterning are consistent throughout the forms.
There are, however, several forms in which the placement of the accent is
irregular. First, consider the provisional forms:
(38) provisional
£ * * Ί· -i·
iw -e-ba ki-re-ba agar-e-ba ake-re-ba
^ ^ Ψ ífc
aw-e-ba mi-re-ba a ruk-e-ba i ki-re-ba
In this form, the accent falls on the second mora preceding the suffix, ex-
cept in aw-e'-ba, where the accent falls on the mora immediately preced-
ing the suffix. Note that the forms in Section 3.2.2.2. all have the accent
on a designated mora—specifically, the mora immediately preceding the
suffix. The provisional forms are irregular in this respect.
The past forms show similar behaviour:
(39) past
^ ^ ^ ^
i t-ta ki-ta a gat-ta a ke-ta
^ Η· Ψ
oo-ta mi-ta a ru i-ta i ki-ta
The accent falls on the second mora preceding the suffix, except in oo-ta,
where no accent appears, and iki'-ta, where the accent falls on the mora
immediately preceding the suffix, (a'gat-ta is also an exception, but we
will ignore this because Η-initial 4-mora verbs show a unique pattern of
behaviour; e.g. i'w-as-i-ta (where -(s)as- is a causative suffix), ki'-sas-i-ta
vs. agar-a's-i-ta, ake-sa's-i-ta. Note, however, that the pattern illustrated
by a'gat-ta is changing to the expected one, resulting in aga't-ta, as is dis-
cussed in Section 2.2. (see (9)).)
Finally, consider the colloquial negative forms:
(40) negative (colloquial)

^ ^
iw-a-hen ki i-hin12 aga r-a-hen a ke-hen
aw-a-hen mi i-hin a ruk-a-hen i ki-hin

In this form, the accent falls on the second mora preceding the suffix, ex-
cept in aw-a'-hen and iki'-hin, in which the accent falls, once again, on the
mora immediately preceding the suffix.
In all of the forms seen above, there are several word classes in which
accent placement behaves in an exceptional manner. The problem to be
addressed, then, is why there are such exceptions in these forms. Recall
that, as we have seen in the preceding section, accent location in a verb is
the property of the suffix. This being the case, why do the suffixes above
not consistently place the accent on a designated mora?
If we assume the analysis presented here, the problem above can be ex-
plained in a simple way. Specifically, it becomes possible under such an as-
sumption to argue that the exceptions emerge because of initial inacces-
sibility marked on the stems. The explanation proceeds as follows.
As we proposed above, the first mora of the group (b) words in the ex-
amples is marked as inaccessible. If it is the case that this marking must
be respected, the accent which is required by a suffix on the first mora
must move elsewhere. If no mora is available on the left side, as is the case
238 Yitko Ζ. Yoshida and Hideki Zamma
in 3-mora words, then the accent moves rightwards. See the following di-
agrams:
(41) a lexical accent assignment * pitch interpretation *
a-e-ba -> a-e-ba ->• a-e-ba

by -ba
b. 1
a-e-ba
In (41a), the usual assignment of accent by the suffix -ba is carried out, ac-
cording to which the accent falls on the second mora preceding the suffix.
It means, however, that the initial inaccessibility fails to be satisfied when
assigning the accent to the initial mora. To respect the lexical marking of
initial inaccessibility, it is necessary to move the accent rightwards by one
mora, as shown in (42b).
Note that this exceptional behaviour is observed only when both of the
following conditions are met: (i) the stem exhibits the OHO pitch pattern;
and (ii) the suffix otherwise assigns the accent on the initial mora of the
verb. Recall that other exceptional forms besides a-e'-ba all conform to
these conditions: e.g. oo-ta (unaccented), iki'-ta (cf. aru'-i-ta)\ aw-a'-hen,
iki'-hin (cf. aru'k-a-hen). This fact clearly shows the suitability of the
present analysis.13
3.2.2.4. Accentuation of nouns
Another piece of evidence comes from the accentuation of nouns con-

taining three morae. Note that Hirayama (1957) lists only the following
two types of accented 3-mora noun:
(43) *
a. accent on the initial mora: e.g. a tama 'head'
b. accent on the second mora, *
where the initial one is pitchless: e.g. katana 'sword'
It will be recalled from Section 2.2. that there is a clear tendency in KJ to

place the accent on the antepenultimate mora in nouns. The pattern in
(43a) conforms to this generalization. The one in (43b), however, does
not: the accent falls on the penultimate mora. Do we have to mark all the
forms with this pattern as exceptions?
If we assume the analysis presented above, this pattern is not at all

problematic. The forms that follow the (43b) pattern are regarded as be-
ing lexically marked for initial inaccessibility. Although the accent should
be put on the antepenultimate mora, it cannot be because of this marking:
note that the antepenultimate mora is the initial one in this case. Thus, the
accent is carried down to the second mora, which is penultimate.
Moreover, the fact that there is no high-pitch-initial noun with the ac-
cent on the penultimate mora (i.e. HHO) supports this analysis. If the pen-
ultimate accent in (43b) is assigned as it is, it is also predicted that the pen-
ultimate accent occurs in nouns with an initial high-pitch-initial mora,
which is not the case. Thus, the natural conclusion is that the penultimate
accenting of nouns of the (43b) type is not specifically intended, but that
the accent is located there to preserve the lexical marking on the stem,
thereby violating the general pattern.
4. Conclusion
In this paper, we have developed a way of analysing the pitch assignment

system in KJ so that it conforms to the typology of accent languages. This
is achieved by means of the pseudo accent and the notion of inaccessibility,
both of which are attributable to the status of domain-edges. This analysis
is appealing for several reasons. First, the pseudo accent can account for
the phrasal accent phenomenon, which is unique to KJ. Second, the inac-
cessibility of the domain-initial position can account for the exceptional
behaviour of the verbal paradigm. Finally, these devices can account for
many phenomena in the dialect without resorting to any underlying low
pitch, which finds little empirical support in pitch accent languages.
Acknowledgements
We'd like express our gratitude to Phillip Backley, Jeroen van de Weijer and Tetsuo Nishi-
hara, and the anonymous reviewer of this volume for invaluable comments and suggestions
for this paper. Our thanks also go to the audience of the Phonology Forum 1999, held at the
Tokyo Metropolitan University (2-4 September) where an earlier version of this paper was
presented; especially Shosuke Haraguchi, Takeru Honma, Itsue Kawagoe, Akio Nasu, Koi-
chi Tateishi, Shin-ichi Tanaka, Noriko Yamane, and Teruo Yokotani. This paper is dedicated
to Jion.
Notes
1. The unit employed in this paper is the mora, which refers to any of the following: 1) a
vowel preceded by zero or one consonant e.g. [e] or [fa], 2) the initial half of a geminate
consonant [kippu], 3) the latter half of a long vowel [too] or a diphthong [da/], 4) a mo-
rale nasal [hoiV]. The issue of the unit, mora vs. syllable, is not central to our discussion.
Briefly, however, KJ abounds with lexical items with accents even on those special mo-
rae, 3) & 4), e.g. on the latter half of a diphthong koi 'carp' and on a moraic nasal as in
toNbi 'black-eared kite'. This contrasts with the SJ case, in which lexical accents are not
found on those morae, the non-head portion of a syllable.
2. A recent change in the dialect form is found in this type of noun. In Hirayama (1957),
the accented final mora in bimoraic words is lengthened when in isolation (i.e. without
a particle), although the preference of many speakers today is to use the final accented
form without lengthening.
3. Some minor dialectal variations of the data between the Kyoto and the Osaka dialects
should not affect the basic analysis; in other words, most of the Kyoto data fits well with
that of the Osaka dialect.
4. In KJ, no such words are attested—lexically accented bimoraic or trimoraic simplex Ya-
mato words which are subject to pitch spreading.
5. In SJ too, the rightmost mora serves as the head of the accentless domain (Yoshida
1999).
6. Here the term 'bimoraic' also refers to lengthened monomoraic forms such as kii 'tree'
(ki in SJ).
7. This is also pointed out by Koichi Tateishi and Teruo Yokotani, and we have always
agreed with the idea that the case-marked nouns obtain final-accents when followed by
another particle. We still believe the genitival -no, however, the case marker which is
discussed later in Section 3.1.2., cannot be generalised with other case-markers. Geniti-
val -no at the same time functions to 'nominalise' the phrase as a pronoun (equivalent
to one in English), as in the form wasabi-no, meaning 'one with wasabi\ This function
is unique to this genitival particle, and perhaps it is reasonable to assume a lexical accent
on the particle, unlike all other cases.
8. A correlate between the two types of lexical marking, lexical accent marking and the
inaccessibility of the initial mora, can be noted here. As section 1.1. illustrates, KJ con-
tains no lexically accented words which chooses to share high-pitch. Lexical accent
marking may have accompanied this inaccessibility in that lexical domain. However,
wherever we see high-pitch sharing we also observe accent assignment i.e. quadrimoraic
nouns (see 2.2.) and the lexically accentless nouns. Those nouns dominate a lexical do-
main without any markings for the lexical accent. In such a domain, where there is no
lexical-accent information, the inaccessibility information was not available from the
outset. Perhaps some of the accentless words, by analogy, set the lexical marking of in-
accessibility, along with accent assignment for quadrimoraic words, or along with the as-
signment of pseudo accent for lexically accentless trimoraic words. This correlate, how-
ever, requires more detailed consideration, and is therefore not pursued further in this
discussion.
9. The stem-final segment of C-final verbs undergoes some changes called onbin in the
past and the gerund forms. It is also observed in (31) and (39) below. The segmental
change to the vowel in oo-ta in (39b) is also classified as onbin in Japanese linguistics.
10. As is well-known, stem-final /w/ in Japanese only appears before /a/ and is deleted else-
where. The segment is represented in the examples only to show that the words contain
/w/ underlyingly.
11. The allomorph of the imperative suffix for V-final stems is φ (zero). The hi in 2-mora V-
final verbs in (34) is not regarded as the suffix, but as part of the preceding vowel, which
is lengthened to satisfy binarity (cf. Itô and Mester (1992)).
12. The vowel of this suffix has harmonised to the one in the stem. Moreover, the stem vow-
el in this form has lengthened, probably to satisfy the binarity (cf. Itô and Mester
(1992)) required by the suffix -hen.
13. In the case of oo-ta, it is accent deletion which is carried out in order to preserve the lex-
ical marking. This is consistent with our claim here because deletion is observed only in
those words where the initial mora would otherwise be accessible. This accent deletion
may have something to do with the fact that the oota class is consonant-ending, which
undergo onbin alternation (see note 9). Note that the consonant typically becomes the
first part of a geminate through onbin, which can hardly bear the high pitch by itself. In
order to avoid the pitch contour OHO for tat-ta 'stand (past)', for example, accent deletion
may take place. Non-geminate-ending stems may be aligned to this pattern.
References
Bloch, Bernard
1946 Studies in Colloquial Japanese I: Inflection, Journal of the American Oriental
Studies 66.
Haraguchi, Shosuke
1977 The Tone Pattern of Japanese: An Autosegmental Theory of Tonology.
Kaitakusya. Tokyo.
Haraguchi, Shosuke
1991 A Theory of Stress and Accent. Foris. Dordrecht.
Hirayama, Teruo
1957 Nihongo Oncho no Kenkyuu (A Study of Japanese Accent). Meiji Shoin. Tokyo.
1992 Weak Layering and Word Binarity. ms. University of California, Santa Cruz.
Kaye, Jonathan
1990 'Coda' Licensing, Phonology 7:2, 301-330.
1995 Derivations and Interfaces, in Frontiers in Phonology: Atoms, Structures, Der-
ivations, ed. by Jacques Durand and Francis Katamba. Longman. London.
McCawley, James D.
1968 The Phonological Component of a Grammar of Japanese. Mouton. The
Hague.
Poser, William
1984 The Phonetics and Phonology of Tone and Intonation in Japanese, Doctoral
Dissertation. MIT.
Yoshida, Yuko
1995 On Pitch Accent Phenomena in Standard Japanese. Doctoral dissertation.
SOAS, University of London. [Published in 1999 by Holland Academic
Graphics.]
1999 Binary Inter-nuclear Licensing for Accent Assignment in Relation to Phono-
logical Domains, in Phonologica 1996: Syllables, ed. by John Rennison. Hol-
land Academic Graphics. The Hague.
II. Studies in Japanese Morphology
Word Plus: The Intersection of Words and Phrases
Taro Kageyama
Kwansei Gakuin University
Introduction
The autonomy of morphology with respect to the other modules of the

grammar poses a perennial problem in morphological research. In partic-
ular, the relation between morphology and syntax, repeatedly debated
under the rubric of various versions of the lexicalist hypothesis, is so com-
plex that it is necessary to accurately pinpoint the real issues involved in
it before we enter into fruitful discussions of any particular topic. To my
mind, the major issues are roughly divided into two groups. One is what
we may call the "component problem", namely whether or to what extent
syntax and morphology interact with each other in the system of gram-
mar. There are a variety of proposals on this issue (see Borer (1998) for
an overview), ranging from the strong lexicalist position which grants
morphology its complete autonomy (Di Sciullo & Williams 1987) to the
strictly syntactic position which integrates all productive morphology into
syntax (Lieber 1992, Halle & Marantz 1993), with mixed theories in be-
tween allowing word formation both in the syntax and in the lexical com-
ponent (Baker 1988, Shibatani & Kageyama 1988). The other major issue
concerning the morphology-syntax interface is the problem of "lexical in-
tegrity" or "syntactic atomicity", namely whether or to what extent syn-
tactic operations can have access to the internal structure of a word. Here
again, diverse views have been expounded (Di Sciullo & Williams 1987,
Bresnan & Mchombo 1992).
The present paper attempts to shed new light on these two problems by
closely examining a certain type of complex expressions in Japanese
which has been largely ignored in the literature. The expressions we are
going to probe are exemplified by (1).
(1) a. zen | gaimu-daizin 'ex-foreign minister', han | taisei 'anti-establishment'

b. booeki-gaisya | syatyoo 'president of a trading company', L S A | nenzi-taikai
' L S A annual meeting'
c. moto I booeki-gaisya | syatyoo 'former president of a trading company'
246 Taro Kageyama
(la) illustrates prefixed expressions and (lb) compound expressions, and

the two types of expressions can be mixed up as in (lc). What is most
striking is that unlike regular lexical compounds which are pronounced
with a sequence of high-pitched moras, the expressions in (1) bear a
phrase-like accent with a slight pause or minor phrase boundary (indicat-
ed by "I") after the prefixes or between the members of compounds. An
obvious question is whether such expressions are identified as words or
phrases.
In section 1, I will argue, along with Poser (1990), Kubozono (1995),
and Kageyama (1982; 1993), that those phrase-like expressions indeed
constitute words because of their syntactic atomicity. Section 2 demon-
strates, however, that despite their word status, those expressions are
highly similar to phrases in terms of the internal semantic relations and
syntactic behavior allowing sentence-level anaphora. In a nutshell, ex-
pressions like those in (1) are identified as words on the morphological
side but behave like phrases otherwise. To resolve this contradiction, sec-
tion 3 proposes a novel category called "Word Plus" (W + ), which is larger
than Word or X o but nonetheless constitutes a morphological object dis-
tinct from a syntactic phrase like X'. Section 4 suggests the universal va-
lidity of the proposed W + category by considering English compounds
with rear stress like looker-on and well-known.
The proposed Word"1" category will have non-trivial implications for the
architecture of morphology. It has been customarily assumed that words
constitute the category of X o , and that morphology concerns the internal
structures of X o or below whereas syntax builds up phrases and sentences
on the basis of Xo. In other words, X o is held to be the boundary that sep-
arates morphology from syntax. I will call this the "category problem".
The category problem has not been questioned seriously even by re-
searchers who have attempted to make the sublexical categories more
precise. For example, Selkirk (1982), who proposes the category "Root"
which occurs below "Word", and Ackema (1999), who develops an X-bar
morphology postulating X 2 and X 4 below Xo, nonetheless hold X o or
Word to be the maximal domain of morphology. This paper will show that
the proposed W + category rather than X o is the maximal projection of
morphological structure.
1. The lexical integrity of phrase-like words

This section will demonstrate that phrase-like expressions like those illus-
trated in (1) are equipped with syntactic atomicity and thus qualify them-
selves as "words". Before proceeding to the demonstration, we will first
delimit the scope of the data to be handled in this paper.
It is well known that Japanese has a peculiar set of prefixes which ex-
hibit a phrasal pitch contour instead of the usual lexical accent (Kageya-
ma 1982, Poser 1990). Representative examples of these prefixes (dubbed
"Aoyagi prefixes" by Poser (1990) after the Japanese grammarian S. Aoy-
agi) are given in (2), where the moras indicated by capital letters are pro-
nounced with a high pitch.
(2) Phrase-like prefixes:

hon- 'this, the present': Η θ η | dalGAKU 'this university'
moto- 'former, ex-': MOto | soORIDAizin 'a one-time Prime Minister'
zen- 'ex-, immediately preceding': ZEn | soORIDAizin 'the ex-Prime Minister'
gen- 'current': GEn | kalTYOO 'the current president'
kaku- 'each': KAku | dalGAKU 'each university'
boo- 'a certain': BOo | tyoMEESAKka 'a certain famous writer'
doo- 'above-mentioned': D O o | dalGAKU 'the above mentioned university'
zen- 'all, whole': ZEn | kuMIAI-in 'all the union members'
ryoo- 'both': RYOo | dalGAKU 'both universities'
ko- 'deceased': KO | uEMURA-si 'the late Mr. Uemura'
han- 'anti-': HAn | talSEE 'anti-establishment'
hi- 'non-' : HI | yoOROPPA-GEngo 'non-European languages'
These prefixes are phonologically peculiar not only in bearing an accent

on their own but also in being followed by a slight pause or minor phrase
boundary, indicated by the symbol
While Poser (1990) focused his attention on the prefixes, the phrasal ac-
cent contour and the concomitant minor phrase boundary are not limited
to the prefixed expressions but are shared by compound expressions like
(lb). Compound expressions of this type are highly productive, as one can
freely make examples like those in (3).
(3) Compound words pronounced with phrasal accent

zenkoku | nodoziman-taikai 'All-Japan amateur singing contest', sin-
ki I kakutyoo-keikaku 'new expansion plan', teiki | kenkoo-sindan
'regular physical checkup', mission-kei | siritu-daigaku 'private uni-
248 Taro Kageyama
versities founded by missionaries', gengogakka | kyoodoo-kenky-

uusitu 'common room of the linguistics department', kigenzen | 226-
nen 'the year 226 B.C.', ryooriten | keieisya 'restaurant manager',
tokkyuu I teisya-eki 'station at which limited express trains stop',
(kono hikooki no) Heathrow-kuukoo | tootyaku-zikoku 'the sched-
uled time of (this airplane's) arriving at Heathrow Airport', syugo |
kuriage-kisoku 'the subject raising rule', naizyu| kakudai-seisaku
'policy to boost domestic demand', California-syuu | Sacramento
'Sacramento, California'
Such prefixed and compound expressions not only have the phrase-like
phonology in common but can also be freely embedded inside each other
to produce more complex expressions with iterated minor phrase bound-
aries, as exemplified by (4).
(4) a. [[kokuritu-daigaku] | [zen | gakutyoo]]

national-university former president
'former president of a national university'
b. [boo I [[kokuritu-daigaku] | [zen | gakutyoo]]]
certain [national-university former president]
'a certain ex-president of a national university'
The indiscriminate mixture of prefixed and compound expressions indi-

cates that the prefixation and the compound formation are of the same
morphological nature, most probably being situated at the same level (in
the sense of level-ordered morphology). Thus we will treat both prefixed
and compounded expressions in a unitary way.
It might be noted that the minor phrase boundary observed in the
above examples is reminiscent of an analogous phonological break (indi-
cated by ":") inherent in what Shibatani and Kageyama (1988) call "post-
syntactic compounds", namely N-VN compounds which are created most
plausibly at S-structure after the transformational cycle. Kageyama
(1993) points out that despite the phonological similarity, compounds like
those in (3) are different from post-syntactic compounds.
First, post-syntactic compounds are sanctioned only by grammatical case
relations (nominative, accusative, and sometimes dative), whereas the (3)-
type compounds allow various semantic relations between the two mem-
bers, as shown by the time, place, and other adverbial relations in the first six
examples in (3). Second, subject honorification can be applied to the head
of post-syntactic compounds, but not to that of the (3)-type compounds.
(5) subject honorifics on the head

a. post-syntactic:
sensei-ga [Yooroppa : go-taizai] no ori
teacher-NOM [Europe : HONORIFIC-stay] GEN occasion
'while the teacher stayed in Europe'
b. (3)-type compound:
*?tyooki I go-taizai-keikaku
long-term HONORIFIC-stay-plan 'the plan for a long-term stay'
Third, post-syntactic compounds may sometimes allow phrases in the first

member, whereas the (3)-type compounds strictly exclude phrases from
appearing inside.
(6) phrasal expressions inside

a. post-syntactic:
[[kono zikken] : syuuryoo]-go ni 'after this experiment is completed'
[[this experiement] : finish] after DAT
b. (3)-type compound:
*[sono ryooriten] | keieisya 'manager of that restaurant'
that restaurant manager
(6b) is ungrammatical on the intended, reading where the determiner

sono 'that' modifies only ryooriten 'restaurant' within the compound.
The grammatical case relations, honorifics, and phrasal expressions in-
side post-syntactic compounds constituted the primary motivation for
Shibatani and Kageyama (1988) to locate their formation after the syntax
rather than in the lexicon. It is reasonable to assume that the (3)-type
compounds, which do not exhibit those properties, are lexical rather than
(post-)syntactic.
We are now in a position to prove the word status of the phrase-like
complex expressions. The proof will be based upon the syntactic and lex-
ical properties of those expressions. First, both prefixed and compound
expressions obey the "No Phrase Constraint" (Roeper and Siegel 1978),
which is a manifestation of the general principle of lexical integrity or syn-
tactic atomicity (Di Sciullo and Williams 1987). Thus, neither the head
nor the non-head of the phrase-like expressions can accommodate NPs
involving genitives, adjectives, or other kinds of modifiers (Poser 1990,
Kageyama 1993).
(7) a. *boo I [ NP yumeina haiyuu] 'a certain famous actor'

certain [ famous actor]
250 Taro Kageyama
b. *kaku I [ NP tihoo-no tosi] 'each provincial city'

each [ province-GEN city]
(8) a. *zenkoku | [ NP gassyoo-no konkuuru] 'All-Japan chorus contest '
All-Japan chorus-GEN contest
b. * [ NP huransu-ryoori-no mise] [ keieisya 'proprietor of a French restaurant'
French-cuisine-GEN shop proprietor
Particularly noteworthy about (7) and (8) is the fact that the ungrammatical-
l y is removed if their NP parts are replaced by compound nouns, as in (9).
(9) a. boo I [N yumei-haiyuu] (same meaning as (7a))

b. kaku I [N tihoo-tosi] (same meaning as (7b))
c. zenkoku | [ N gassyoo-konkuuru] (same meaning as (8a))
d. [N huransu-ryoori-ten] | keieisya (same meaning as (8b))
The only exception to the No Phrase Constraint that I am aware of is the

prefix moto- 'formerly, ex-', which may take a genitive phrase as in moto
[puro-yakyuu no sensyu] 'formerly a professional baseball player'. The
idiosyncrasies of this prefix will be explicated in section 3.
Second, the syntactic atomicity of the expressions at issue is shown by
the rule of identity deletion in coordinate sentences. This rule can delete
up to part of phrases but does not intrude into words, as illustrated in (10)
(Kageyama 1989; 1999).
(10) a. Ken-wa sukiyaki-o tabe, Naomi-wa susi-o tabe-ta.

K-TOP sukiyaki-ACC eat N-TOP sushi-ACC eat-PAST
'Ken ate sukiyaki, and Naomi sushi.'
b. Ken -wa [ NP Furansu-no kuruma] o kai, Naomi-wa Itaria-no kuruma-o kat-ta.
K-TOP France-GEN car-ACC buy N-TOP Italy-GEN car-ACC buy-PAST
'Ken bought a car made in France, and Naomi, one made in Italy.'
c. *Ken -wa [ N inu-goya] o tukuri, Naomi-wa [ N usagi-goya]-o tukut-ta.
K.-TOP [dog-house]-ACC make N.-TOP [rabbit-house]-ACC make-PAST
'Ken made a doghouse, and Naomi a rabbit-house.'
The deletion rule, which can erase the head noun of the NP structure in
(10b), is forbidden to affect only the second member of the compound
noun inu-goya 'doghouse' in (10c). In the same vein, the deletion cannot
destroy the integrity of the complex expressions involving the phrase-like
prefixes. The same degree of ungrammaticality as (10c) is exhibited by
(11a) below, where only the base is deleted leaving the prefix behind. This
should be contrasted with the full grammatically of ( l i b ) , where the pre-

fixed words in (11a) are replaced by semantically similar NPs.
( 1 1 ) a. *A-san-wa [ N gen | kaityoo] to siriai de, B-san-wa

A-Mr-TOP [current-president]-with acquainted-is, B-Mr-TOP
[ N zen I kaityoo]-to siriai-da.
[ex-president]-with acquainted-is
'Mr. A is acquainted with the current president, and Mr. Β with the
ex-president.'
b. A-san-wa [ N P genzai-no kaityoo] to siriai de, B-san-wa
A-Mr-TOP [current-GEN president]-with acquainted-is, B-Mr-TOP
[ N P mae-no kaityoo]-to siriai-da.
[former-GEN president]-with acquainted-is
A third indicator of the syntactic atomicity of the phrase-like expressions

is the impossibility of modifying a word-internal element with an external
adjective or relative clause. Compare (12a) with (12b).
( 1 2 ) a. Tookyoo-ni iru [booeki-gaisya | syatyoo]

Tokyo-in is(animate) trading-company president
'the ex-president of a trading company, who is in Tokyo'
b. *Tookyoo-ni aru [booeki-gaisya | syatyoo]
Tokyo-in is(inanimate) trading-company president
intended meaning: 'the ex-president of a trading company located in Tokyo'
c. Tookyoo-ni aru [booeki-gaisya]-no syatyoo
Tokyo-in is(inanimate) trading-company-GEN president
(12a) is impeccable, since the relative clause with the verb iru 'be', said of
an animate subject, modifies the following human noun as a whole. In
contrast, (12b) is completely ruled out because the relative clause with
the verb aru 'be', selecting an inanimate subject, is intended to modify
only the first member (booeki-gaisya 'trading company') of the com-
pound. This is parallel to the infelicity of, for example, *a very [dark-
room], where the adverb very cannot modify the adjective dark in the
compound noun darkroom. The word status of booeki-gaisya\syatyoo
'trading-company president' is confirmed by the fact that (12b) becomes
grammatical if booeki-gaisya 'trading company' gains independence as a
phrase with the genitive marker, as in (12c).
The preceding three observations clearly indicate that the prefixed and
compound expressions with a minor phrase boundary make up syntacti-
cally inviolable units and are therefore qualified as words. Additional ev-
252 Taro Kageyama
idence for the word status is provided by the fact that the complex expres-
sions in question are amenable to lexical or morphological conditions that
generally pertain to word formation.
As is well known, word formation processes are lexically governed, re-
stricted notably by the vocabulary strata determined by the origin of in-
dividual words. In Japanese there is a pronounced tendency for a native
morpheme to combine with another morpheme of native origin and for a
Sino-Japanese morpheme to combine with another Sino-Japanese mor-
pheme, although words composed of mixed vocabulary strata are occa-
sionally encountered. This combinative restriction is illustrated by (13).
(13) a. roodoo-sya 'laborer' *roodoo-mono

labor [S-J] - person [S-J] labor [S-J] - person [native]
a'.hataraki-mono 'hard worker' *hataraki-sya
work [native] - person [native] work [native] - person [S-J]
b. dai-kyoositu 'big classroom' *oo-kyoositu 'big classroom'
big [S-J] - classroom [S-J] big [native] - classroom [S-J]
b'.oo-beya 'big room' *dai-beya 'big room'
big [native] - room [native] big [S-J] - room [native]
Now the same restriction is observed in the phrase-like expressions as

well. Since all of the phrase-like prefixes except the native moto- 'former'
are Sino-Japanese, they are most comfortably attached to Sino-Japanese
(or Western) loan words.
(14) a. kaku | gakkoo 'each school'

each [S-J] school [S-J]
a'.*kaku | manabiya 'each school'
each [S-J] learning-place [native]
b. doo I syuukai 'the above-mentioned assembly'
above-mentioned [S-J] assembly [S-J]
b'.*doo I atumari 'the above-mentioned assembly'
above-mentioned [S-J] gathering [native]
It seems that only the native moto- exceptionally meshes with both Sino-
Japanese and native bases.
(15) a. moto | kyoositu 'formerly a classroom'

former classroom [S-J]
b. moto I sigotobeya 'formerly a work-room'
former work-room [native]
It will be observed in section 3 that this exceptional behavior of moto- has

syntactic repercussions.
The lexical governedness of phrase-like words is further illuminated by
the idiosyncratic restrictions on the selection of combined words.
(16) a. zen | syokuin 'all the staff

b. itibu I syokuin 'part of the staff
c. *hanbun | syokuin 'half the staff'
cf. hanbu no syokuin 'half of the staff {no = genitive particle)
d. *kahansuu | syokuin 'majority of the staff
cf. kahansuu no syokuin 'majority of the staff
In (16), zen- 'all' and itibu- 'part' are eligible for word formation, but seman-
tically similar words, hanbun 'half and kahansuu 'majority', resist comound-
ing and instead call for a phrase structure supported by the genitive particle.
Lexical idiosyncrasies of this kind are characteristic of word formation.
This section has demonstrated the word status of the peculiar phrase-
like expressions on the basis of phenomena involving lexical integrity and
lexical idiosyncrasies. Because of the word status, we will refer to the
phrase-like words, both prefixed and compounded, as Word"1" (Word
Plus). This is intended as a new morphological category that is larger than
ordinary words but still belongs to the morphological as opposed to the
syntactic domain. For example, the morphological structures of zen \ dai-
tooryoo 'ex-President' and Reagan | zen | daitooryoo 'ex-President Re-
agan' will be represented as follows.
(17) a. N* b. N+
prefix
zen Ν Ν prefix
'ex-' daitooryoo Reagan zen Ν
'president' 'ex-' daitooryoo
'president'
In (17a), the phrase-like prefix zen- 'ex-' selects N + as its base. This c-se-
lection property need be specified in the lexical entry of the prefix, along
254 Taro Kageyama
the lines of Lieber's (1980) subcategorization theory of affixes. The struc-

ture in (17b) illustrates a more complex W + involving the compounding
of Reagan and zen\daitooryoo. Here, Reagan, which normally has the sta-
tus of N, is projected to N + to be compounded with another noun of the
same category.
In this section we have established the "Word" part of Word"1". In the
subsequent section, we will turn our attention to the "Plus" part of Word"1"
and elucidate what extra properties W + has which are not shared by ordi-
nary words.
2. The phrasal character of Word plus
Having established the lexical integrity of W + expressions, we now en-

counter a seeming contradiction that those expressions, albeit morpho-
logically identified as words, display phrasal behavior with respect to their
internal phonology, semantics, and syntax.
First, the phonological properties. As has been repeatedly mentioned,
W + expressions carry phrasal accent, with a slight pause inside. This
phrasal accent, as opposed to the so-called lexical accent observed with
regular derived or compound words, provides an ostensible indication
that W + might belong to the syntax. Rendaku or sequential voicing is an-
other phonological phenomenon that is observed inside lexical words
(18a) but can never apply to what we identify as W + (18b).
(18) a. lexical compound: booeki-gaisya [<- Â:aisya 'company']

b. Word + : kaku | &aisya 'each company', zen | /caisya 'all companies'
The phrase-like accent contour and the absence of sequential voicing sug-
gest that the constituent parts of W + are independent units connected
with each other very loosely.
The phonologically loose connection of W + is correlated with the inter-
nal semantics. Compare the meanings of a regular compound (19a) and a
W + compound (19b).
(19) a. regular compound: sokkyuu-toosyu 'fastball pitcher'

b. W + : Yomiuri-Kyozingun | toosyu 'a pitcher of the Yomiuri Giants'
Inasmuch as the modifier in a lexicalized compound generally designates a

typical characteristic of the head noun, sokkyuu-toosyu 'fastball pitcher' in
(19a) is a name for a type of baseball pitcher characterized by the high

speed of the balls he throws. Since the typical characteristic need not always
hold true, it is not contradictory to say that Mr. A, though sokkyuu-toosyu,
throws slow balls in today's game. On the other hand, the modifier ("of the
Yomiuri Giants") of the W + compound in (19b) is not so much a character-
ization of the pitcher's typical behavior as the predication that must factu-
ally hold for the head noun. It is an obvious contradiction to say Mr. Β is Yo-
miuri-Kyozingun | toosyu but belongs to the Tigers. In this sense, the
difference between (19a) and (19b) seems to correspond to the distinction
Rappaport Hovav and Levin (1992:133) draw between "event" and "non-
event" interpretations of -er nomináis. For example, the compound noun
life-saver has a "non-event" interpretation in that it does not necessarily im-
ply that the person so named has actually had an experience of saving a life,
whereas the NP a saver of lives induces the "event" interpretation that the
person so described has been involved in saving lives. This "event" inter-
pretation appears to be induced by all W + expressions.
The preceding cursory observations will be sufficient to indicate that
W + has internal semantics and phonology comparable to noun phrases.
An even more striking demonstration of the phrase-like character of W +
is provided by the prefix doo- 'the above-mentioned', which is shown to
participate in sentence-level coreference.
For English it is well established that the applicabilities of Postal's
(1969) Anaphoric Island Constraint differ in two types of anaphora, in-
bound and outbound. As shown by Ward, Sproat and McKoon (1991),
outbound anaphora of sublexical elements, as in (20), is made possible by
pragmatically evoking an appropriate referent corresponding to a noun
contained in a complex word, whereas inbound anaphora like (21) is
strictly disallowed by general morphosyntactic principles.
(20) outbound anaphora:

Patty is a definite Kal Καηλ cat. Every day she waits for it¡.
(Ward, Sproat and McKoon 1991: 451)
(21) inbound anaphora:

*Bill was a McCarthyjite, and Fred was also a Μτη{ite.
(Sproat1988: 293)
Here we are concerned only with inbound anaphora. Sproat (1988) and
Ward et al. (1991) resort to different morphosyntactic principles to ac-
count for the impossibility of inbound anaphora. For Sproat (1988), in-
256 Taro Kageyama
bound anaphora is ruled out because of the categorial mismatch between

the DP structure of English pronouns and the morphological categories
below Xo; for Ward, Sproat and McKoon (1991:450), the prohibition stems
from the fact that pronouns form a closed class and therefore cannot par-
ticipate freely in the formation of new words. As far as English is con-
cerned, the two views appear equally plausible. If we turn our attention to
Japanese, however, the categorial approach seems more to the point.
A conspicuous case of inbound anaphora is presented by the prefix
doo-. This prefix has two meanings, illustrated by (22a) and (22b), but
only the latter meaning concerns us here.
( 2 2 ) a. Α-san to B-san wa doo-nenrei desu.

'Mr. A and Mr. Β are of the same age.'
b. Nagai-san wa Washington-syuu no kookoo-de 3-nen-kan osie,
Nagai-Mr TOP Washington-state GEN high-school at 3-years taught,
ima wa doo | syuu no daigaku-de manande iru.
now TOP said-state GEN university at studying-is
'Mr. Nagai taught at a high school in the state of Washington for three years,
and is now studying at a university in the above-mentioned state.'
The doo- in (22a) means 'same, identical' and the complex word involving
it carries the regular lexical accent. On the other hand, the homophonous
prefix in (22b), our present concern, has entirely different properties: it is
pronounced with a phrasal accent and has the function of referring to a
previously mentioned entity. Thus, doo\syuu 'the said state' is bound by
Washington-syuu 'state of Washington' in the preceding clause.
Before delineating the referential properties of the second instance of
doo-, we should first show that this prefix is truly referential rather than
deictic. This point is significant, because if the function of this prefix were
deictic, then its apparent referentiality could be reduced to pragmatic in-
ferences just like the English outbound anaphora in (20). We prove the
point by comparing doo- with two other W + prefixes which also have ref-
erential functions: hon- 'this, the present' and too- 'the said, in question'.
(23) a. [In a book preface] hon | syo no mokuteki wa ...

this I book GEN purpose TOP
'The purpose of this book is ...'
b. [a cashier's notice] too | ten wa syoohi-zei o itadaite-orimasen.
this I store TOP sales-tax ACC collect-not
'We (i.e. this store) do not collect sales tax.'
These two prefixes are purely deictic, as is shown by the fact that sentences
like (23a, b) can be used merely by pointing to the entities in question,
without reference to any linguistic antecedent. If we follow Sag and
Hankamer's (1984) distinction between "deep anaphora" and "surface
anaphora", the referential function of these two prefixes is characterized
as deep anaphora, which is determined pragmatically or deictically. In
stark contrast to these two is the prefix doo-, which turns out to participate
in surface anaphora. This prefix invariably requires an overt linguistic an-
tecedent and cannot be used to point deictically to an entity placed before
the speaker's eyes or pragmatically to an entity evoked in the brain.
(24) [Watching the police arresting a congressman on TV news]

*Doo I daigisi no sizisya wa okotte iru daroo.
'I'm sure the supporters of this congressman are angry.'
The ill-formedness of (24) might be partly due to the stylistic incompati-

bility where the literary and written usage of doo- conflicts with the col-
loquial spoken language. However, even if written in literary style, doo-
is totally inappropriate if there is no linguistic antecedent. Observe a de-
cided contrast between the deictic hon- (25a) and the pronominal doo-
(25b).
(25) a. [At the conclusion of an article]

hon I ronbun no simekukuri to site ...
'In concluding this present article, ...'
b. [At the conclusion of an article]
*doo I ronbun no simekukuri to site ...
'In concluding the above-mentioned article,...'
It is thus found that the W + prefix doo- is endowed with the function of
creating a referential expression which presumably corresponds to pro-
nominals. While the English personal pronouns are distinguished by gen-
der, person, and number, the Japanese doo- can be prefixed to an unlim-
ited number of Sino-Japanese head nouns to give rise to pronominal
expressions of various semantic types. For example, doo \ sya 'the said
company' can refer to any antecedent NP identifiable as a business firm,
doo I daigaku 'the said university' to any NP identifiable as a university,
and doo \ koku 'the said country' to any NP identifiable as a country.
Having clarified the pronominal function of the prefix doo-, we now
provide examples of inbound anaphora involving this prefix inside.
258 Taro Kageyama
(26) a. New South Wales-syuu¡ o osotta yamakazi wa yooyaku tinka no ki-

zasi o misete-kita. [Doo | syuu¡ |syooboosyo-honbu] wa kinkyuu-
zitai-sengen o kaizyo-sita.
'The forest fire which raged in New South Wales, Australia, is fi-
nally going to be contained. The fire department headquarters of
the above-mentioned state has lifted the emergency measures.'
b. Daitooryoo wa asu yuukoo-zyooyaku¡ ni tyoo'in-suru. [Doo |
zyooyakui | saisyuuan] ni yoruto ...
'The President is going to sign the amity treaty. According to the
final version of the said treaty,...'
In these examples, adapted from newspaper articles, the complex expres-

sions in brackets make up W + compounds, the first members of which are
anaphoric to the preceding antecedents. Such complex compounds in-
volving doo- inside are extremely common in journalistic writings, and in
the same style, examples of outbound anaphora and of mixtures of out-
bound and inbound anaphora are observed no less commonly.
(27) a. Kodai no iseki ga [Nara-zyosi-daigaku{ |koonai] de mitukatta to,

doo I daigaku no hakkutu-tyoosa-iinkai ga happyoo-sita.
'"Ancient ruins were discovered on the campus of Nara Women's
University," the excavation committee of the said university an-
nounced.'
b. [Egypte |kaku | si] ni yoruto, doo | koku, no gyoosei-saibansyo wa ...
'Every Egyptian newspaper reports that the administrative court-
house of the said country
(28) [K.-syotent j syatyoo] niyoru kokain-mituyu-ziken de, [moto | doo

I sya¡ I kameraman] ga beikoku ni syuttyoo-si...
'Concerning the case of cocaine smuggling by the president of K.
Publishing Company, an ex-cameraman of the said company went
on a business trip to America and ...'
The interpretations of the rfoo-pronominals in (27) and (28) are not

achieved by pragmatic inferences like Ward et al.'s English example (20)
but directly by the linguistic antecedents that precede them. This means
that in (27a), for example, the antecedent (Nara Women's University)
contained in a complex W + is syntactically active, and consequently that
the whole W + does not constitute an anaphoric island. The waiver of the
Anaphoric Island Constraint here is due solely to the referential function
inherent in the prefix doo-, which itself belongs to the W + category. Thus,
if ordinary DP pronouns are employed, the islandhood of W + expressions
is restored with respect to both inbound and outbound anaphora. Witness
the following examples where the doo-words in (27a) and (26b) are re-
placed by ordinary pronouns.
(29) a. Kodai no iseki ga [Nara-zyosi-daigaku | koonai] de mitukatta to,

sono daigaku no / sono hakkutu-tyoosa-iinkai ga happyoo-sita.
'"Ancient ruins were discovered on the campus of Nara Women's
University," the excavation committee of the university announced.'
b. Daitooryoo wa asu yuukoo-zyooyaku ni tyoo'in-suru. *[Sore /
Sono zyooyaku | saisyuuan] ni yoruto ...
'The President is going to sign the amity treaty. According to its fi-
nal version ... '
In the case of outbound anaphora in (29a), the referent of sono daigaku

'that university' is vague, and the possibility of designating Nara Women's
University comes about only inferentially in the same way as Ward et al.'s
English examples. As for inbound anaphora, sore 'it' and sono-zyooyaku
'that treaty' are totally impossible inside the W + expression in (29b).
The question we now ask is: To what extent is the internal structure of
W + visible to the syntax? The subsequent observations will reveal that the
internal structure of W + is available to the construal of the doo-pronom-
inale in a manner parallel to the internal structure of noun phrases.
We begin by noting that doo-derived words are not "anaphors" in the
sense of Condition A of the Binding Theory. Like the personal pronouns
and unlike the reflexive pronouns, doo-pronominals must not be bound
in a single clause.
(30) a. *Gaimu-daizinl wa doo \ daizin¡ o semeta.

lit. 'The Foreign Minister blamed the said minister;.'
b. Gaimu-daizinj wa {zibun-zisin¡ / *kare¡} o semeta.
'The Foreign Minister; blamed {himself; / *him¡}.'
(31) a. *Koosei-syooi wa doo \ syoo{ o dame ni sita.
lit. 'The Ministry of Welfare; spoiled the said ministry;.'
b. Koosei-syoo; wa {zibun-zisin¡ / *sore¡} o dame ni sita.
'The Ministry of Welfare; spoiled {itself¡ / *it¡}.'
Rather, doo-words behave like pronominals in complying with Condition

Β of the Binding Theory (Pronouns must be free within a governing cat-
260 Taro Kageyama
egory). The ungrammaticality of (30a) and (31a) should be contrasted

with the full grammaticality of (32a, b), where doo-words are contained
in NPs and are free.
( 3 2 ) a. Gaimu-daizinl wa [ N P doo | daizini no hisyo] o semeta.

Foreign-Minister, TOP [said-minister¡ G E N secretary] ACC blamed
lit. 'The Foreign Minister; blamed the said minister's secretary.'
b. Koosei-syooi wa [ N P d o o | .vyoo, no syokuin] ni tuutatu o dasita.
Welfare-Ministry¡ TOP [said-ministry, G E N staff] DAT notice ACC sent
lit. 'The Ministry of Welfare, sent notice to the staff of the said ministry;.'
Crucially, parallel structural relations obtain when doo-pronominals are

embedded within a larger W + (without the mediation of the genitive
marker).
( 3 3 ) a. Gaimu-daizin, wa [ N moto | [doo \ daizin¡ | hisyo]] o semeta.

Foreign-Ministe^ TOP [former [said-minister; | secretary] ACC blamed
lit. 'The Foreign Minister; blamed the former secretary of the said minister;.'
b. Koosei-syooi wa [ N doo \ syooi | syokuin] ni tuutatu o dasita.
Welfare-Ministry; TOP [said-ministry; | staff] D A T notice ACC sent
lit. 'The Ministry of Welfare; sent notice to the staff of the said ministry;.'
The two sets of examples in (32) and (33) demonstrate the parallelism be-
tween NP structure and W + structure. Regardless of the presence or ab-
sence of the genitive particle, doo-pronominals are accepted if they are
embedded within a larger branching structure and are not directly gov-
erned by the antecedent. This is a strong indication that the morphological
branching inside W + is visible to the syntax, at least for the purpose of in-
terpreting the reference of doo-words. It should be stressed, however, that
we are not claiming that the syntax can freely access the internal structure
of W + . Were this the case, our arguments in section 1 for the syntactic ato-
micity of W + would lose ground. Rather, our claim is that the islandhood
is removed by virtue of the inherent referential property of the prefix doo-.
3. The morphological structure of W+
Sections 1 and 2 brought to light two conflicting sets of properties ob-

served with W + . On the one hand, W + expressions are shown to possess
lexical integrity; on the other hand, they are found to share with noun
phrases certain phrasal properties related to accentuation and the con-
strual of doo-. To resolve this conflict, I propose a new morphological cat-

egory "Word + ". Before doing so, however, it will be in order to touch
upon two other plausible analyses that have been proposed in the litera-
ture, one semantic and the other prosodie.
Kubozono (1995) analyzes what we call W + compounds in terms of the
semantic relationships between the two members, including (34a-e).
(34) a. organization + official position

kaisya | syatyoo 'company president'
b. personal name
Miki I Takesi 'Takesi (first name) Miki (family name)'
c. name + status
Katoo I kyoozyu 'Professor Kato'
d. region + part
Kyuusyuu | nanbu 'southern part of Kyushu'
e. argument relation
naikaku | kaizoo 'reshuffle of the Cabinet'
It is immediately apparent that such a semantic characterization is far

from adequate. For one thing, it is next to impossible to enumerate the ex-
act semantic relations that exhaustively comprise the phrase-like com-
pounds, as there are miscellaneous adverbial relations.
(35) a. time: 2000-nendo | nyuugaku-siken

'entrance examination for the year 2000'
b. locative: zenbei | gorufu-toonamento
'All U.S. golf tournament'
For another, there are a myriad of examples which have one and the same
semantic relation and yet bear different accentual patterns. The examples
in (34d) and (34e) should be compared with those in (36), which have the
same semantic relations but display the regular lexical accent.
(36) a. region + part: niHON-ZEnkoku 'all parts of Japan'

b. argument relation: iSIKI-TYOosa 'attitude survey'
We thus conclude that it is difficult to delimit the range of phrase-like

compounds in terms of their internal semantic relations.
Another plausible analysis is suggested by Poser (1990: 286). Drawing
on Inkelas's (1989) theory of dual prosodie and morphosyntactic subcat-
egorization, Poser assumes that the phrase-like prefixes morphologically
subcategorize a stem but prosodically subcategorize a minor phrase. The
262 Taro Kageyama
notion of subcategorization, though tenable for affixes, does not hold for
compound words like those we brought out above. In order to account for
those compounds, it will be necessary to set up a prosodie rule that com-
bine two minor phrases into a compound. This, however, begs the funda-
mental question of exactly what causes the difference between regular
compounds and minor-phrase compounds.
Our approach is neither semantic nor prosodie, but morphological. It is
proposed that the accent patterns are direct reflections of the morpholog-
ical architecture of compound and affixed words. To prove this point, we
leave out semantic and prosodie considerations, concentrating our atten-
tion on the morphological composition.
It is known that Japanese words are grouped into several classes, de-
pending on their internal makeup (Kageyama 1982). The basic units of
word formation are single morphemes, which may be either native or
Sino-Japanese. In the case of Sino-Japanese morphemes, the smallest
building block is represented by single Chinese characters like hoo 'visit'
and bei 'America', which in turn are combined to form two-character
words like hoo-bei 'a visit to America', which may be further expanded to
bigger words like hoo-bei yotei 'a schedule of a visit to America'. The or-
derly processes of word formation observed in such examples will be eas-
ily formalized in terms of the level-ordered morphology. For example, the
process of compounding single Sino-Japanese morphemes is located at an
earlier level than that of compounding two-character words.
(37) level 1: [hoo] + [bei]

visit America
level 2: [hoo-bei] + [yo-tei]
visit-America schedule
Separation of these two levels accounts for the fact that the level 1 com-
pounding is strictly limited to single morphemes and cannot apply to
words written with two or more Chinese characters. Compare hoo-bei
with the ungrammatical makeups in (38).
(38) a. *hoo-[bei-koku] 'a visit to America'

visit- [America-country]
b. *hoo-[Amerika-gasyuu-koku] 'a visit to the United States of America'
visit-[America-united.states-country]
The preceding exposition is a simplified view of how Japanese word

formation proceeds from smaller units (which have a higher degree of in-
ternal cohesion) to larger units (which have a morphologically and pho-

nologically looser connection inside). Technically, this compositional ar-
rangement will be formalized in a variety of ways such as the strict level
ordering as suggested above, the use of different types of boundary mark-
ers (Kageyama 1982), or X-bar morphology using certain diacritics. Elab-
orating on Selkirk's (1982) idea of distinguishing Stem from Word,
Kageyama (1993) explores a system of Japanese word formation which
sets up three types of building blocks: Root, Stem, and Word. This system
is illustrated here with Sino-Japanese words.
(39) a. Root: single morphemes like hoo- 'visit' and bei 'America'
b. Stem: two-morpheme words like hoo-bei 'a visit to America'
c. Word: words that can be used independently, typically with more
than two morphemes
These building blocks are combined with each other by rules like (40) to
make larger and larger units.
(40) a. Root (+ Root) -»· Stem

b. Stem (+ Stem) Word
c. Word (+ Word) -»· Word
According to these rules, the internal compositions of hoo-bei 'visit to

America' and hoo-bei-yotei 'schedule for a visit to America' will be rep-
resented as in (41a) and (41b). (Word, Stem, and Root are respectively
abbreviated as W, S, and R.)
( 4 1 ) a. hoo-bei b. hoo-bei-yotei C. hoo-bei-bi
visit America schedule 'day of a visit to U.S.'
We need not concern ourselves with the technical details of these repre-
sentations. What is important is that the difference in 'size' among W, S,
264 Taro Kageyama
and R is not just phonological or semantic but has a morphological reality.

This is substantiated by the possibility of inserting certain coordinate con-
junctions inside the complex words. As observed by Kageyama (1993:17),
the conjunctions naisi 'or' and oyobi 'and' have a peculiar capacity to
show up inside W-level complex words, but not inside R- or S-level words.
(42) Word-level complex words with conjunctions inside

[kokuritu naisi koorituj-daigaku 'national or prefectural/munici-
pal universities', [geruman-kei oyobi romansu-kei]-gengo 'Ger-
manic and Romance languages', gaikokuzin-fkyoosi naisi koosi]
'foreign teacher or lecturer', [hoo-bei oyobi hoo-tyuu]-yotei
'schedules of a visit to U.S. and a visit to China'
(43) Stem-level complex words with conjunctions inside
*[tyuu-goku naisi kan-koku]-zin 'Chinese or Korean people', *[ni-
hon oyobi taiwan]-zaru 'Japanese and Taiwan monkeys', *hu-
[yoozyoo naisi sessei] 'carelessness or negligence about one's own
heath'
(44) Root-level complex words with conjunctions inside
*hoo-[bei naisi tyuu] 'a visit to U.S. or China', *[koo oyobi ryoku]-
tya 'black tea and green tea'
Faced with examples like (42), one might wonder whether the coordinat-
ed elements as a whole are embedded inside the complex words or they
are derived by conjunction reduction from a fuller structure as in (45).
(45) [kokuritu-daigaku] naisi [kooritu-daigaku]

national university or prefectural/municipal university
->• [kokuritu-0] naisi [kooritu-daigaku]
Although the reduction analysis cannot be dismissed off hand, it has some
unfavorable consequences. First, the suggested reduction rule deleting
part of a word results in a blatant violation of the lexical integrity princi-
ple. Second, some additional conditions will have to be introduced to ac-
count for the fact that R- and S-level words do not allow the deletion.
Third, the deletion analysis claims that the resultant structure is some-
thing like kokuritu- 0 | naisi kooritu-daigaku, where the morphophono-
logical break falls right after the first element kokuritu- 'national', but the
native intuition dictates that the coordinated part [kokuritu- naisi koori-
tu-] 'national or prefectural/municipal' is a cohesive unit.
Besides naisi and oyobi, which are of native origin, there is a Sino-Jap-
anese morpheme which typically concatenates two or more words: ken
'and, cum', as in soori-daizin ken gaimu-daizin 'Prime Minister and (con-
currently) Minister of Foreign Affairs' and sinsitu ken ima 'a bed-cum-sit-
ting room'. In the light of this conjunctive morpheme, it is not unreason-
able to treat naisi 'or' and oyobi 'and' as sublexical coordinators that
directly conjoin words.
As expected, these coordinators are able to appear inside the W + ex-
pressions as well. Below are some examples from Kageyama (1993).
(46) zen | [soori-daizin oyobi gaimu-daizin] 'ex-Prime Minister and ex-

Foreign Minister', kaku | [kokuritu-daigaku oyobi kooritu-daiga-
ku] 'each national university and each prefectural/municipal uni-
versity', gaikokuzin | [kyoosi naisi koosi] 'foreign teacher or for-
eign lecturer'
Insofar as the examples in (42) and (46) are concerned, the W + expres-
sions are no different from ordinary compounds with lexical accent.
However, a notable difference is detected if another conjunction, to
'and', is taken into account. The canonical function of this conjunction is
to connect NPs rather than mere words. Substitution of to for naisi and
oyobi in W-level compounds (42) yields outrageous results like (47).
(47) *[kokuritu to koorituj-daigaku 'national and prefectural/munici-

pal universities', *[geruman-kei to romansu-kei]-gengo 'Germanic
and Romance languages', *gaikokuzin-[kyoosi to koosi] 'foreign
teacher and foreign lecturer'
The last example in (47), gaikokuzin kyoosi to koosi, is acceptable only

on the reading 'a foreign teacher and a lecturer', where the second con-
junct ('lecturer') is not included in the scope of the modifier ('foreign').
Now the ungrammaticality of (47) should be contrasted with the exam-
ple in (48) which I found in a newspaper headline.
(48) ko I Hasegawa-Kazuo-si to Uemura-Naomi-si, kokumin-eeyosyoo ni kimaru.

'The late Mr. Kazuo Hasegawa and Mr. Naomi Uemura are to be given the Nation-
al Honor Award.' (The Asahi, April 1984)
Poser (1990:283) briefly mentions this phenomenon, judging a similar ex-

ample (49) to be unacceptable in the intended meaning where the prefix
takes the whole coordination in scope.
266 Taro Kageyama
(49) moto syusyoo to daitooryoo (Poser 1990: 283)

former Premier and President
In view of the attested example (48), however, I am inclined to assume

that the coordination by to is actually well-formed inside W + , especially
when a rather long pause is put after the prefix. The analogous examples
below seem fully acceptable.
(50) zen | [bun-soo to hoo-soo] 'ex-Education Minister and Justice Min-

ister', gen I [kaityoo to huku-kaityoo] 'current president and vice-
president', dai I [3-syoo to 4-syoo] 'Chapters 3 and 4', boo | [kyoo-
zyu to zyokyoozyu] 'a certain professor and associate professor',
syuusyoku | [1-nenme to 2-nenme] 'first and second years of em-
ployment', kookyookyoku | [dai-5ban to dai-6ban] 'Symphonies
No. 5 and No. 6'
The coordinated parts given in square brackets cannot be analyzed as NPs,

since we already established in section 1 that W + expressions obey the No
Phrase Constraint. In order to maintain the lexical integrity of W + , it is
necessary to assume that the coordinated elements in (48) and (50) consti-
tute some kind of morphological object, namely a W + structure like (51b).
(51) a.
prefix prefix
deceased Ν
gaimu daizin Hasegawa-si to Uemuar-si
foreign minister Mr. H. and Mr. U.

Following Lieber (1980), we presume that each affix specifies in its lexical
entry what kind of morphological category it selects as the base. For ex-
ample, the prefix hu- 'in-, un-', which can be attached to a Root (e.g. hu-
an 'uneasiness') and a Stem or Word (e.g. hu-giri 'ingratitude'), has a lex-
ical entry like (52a), whereas ko- 'deceased', always accompanying a mi-
nor phrase boundary, is specified as in (52b).
(52) a. hu- 'in-, un-': [ [Root/Stem/Word]]

b. ko- 'the late': [ [Word+]]
Given morphological structures like (51), the appropriate accent patterns

can be obtained by assuming that the lexical accent rule assigning a se-
quence of high pitches applies inside the Word category and does not ex-
tend to W + . This ensures that a W + carries two or more layers of lexical
pitch contours, with a minor phrase boundary in between.
Although we have established that W + obeys the No Phrase Con-
straint, a caveat will be necessary on the idiosyncratic behavior of the pre-
fix moto- 'former, ex-'. As previously mentioned, this prefix is the only
native morpheme among the W + prefixes. It is also peculiar in that it can
be used as an independent word, as in (53).
(53) Kare wa moto no monbu-daizin desu.

he TOP one-time GEN education-minister is
'He was formerly a Minister of Education.'
Somehow related to these peculiarities is the fact that moto- can take
what appears to be a genitive NP. Observe the following contrast between
the two semantically similar prefixes, moto- and zen- 'ex-'.
(54) a. zen | [keizai-kikakutyoo | tyookan]

ex- [Economic Planning Agency | director general]
a'.*zen | [keizai-kikakutyoo no tyookan]
ex- [Economic Planning Agency GEN director general]
b. moto I [keizai-kikakutyoo | tyookan]
ex- [Economic Planning Agency | director general]
b'.moto I [keizai-kikakutyoo no tyookan]
ex- [Economic Planning Agency GEN director general]
Only moto- allows a genitive phrase after it. More examples follow in
(55), where the genitive particle no seems relatively acceptable, with the
same meaning as the compound versions without the genitive.
(55) a. moto | [massaazi-ten no keieisya]

former [massage-parlor GEN manager]
b. moto I [gaikoku-zin no kyoryuuti]
former [foreign-people GEN settlement]
c. moto I [kyozingun no toosyu]
former [the-Giants GEN pitcher]
268 Taro Kageyama
It is not the case, however, that moto- can attach freely to any type of NPs.
Descriptive modifiers like adjectives and relative clauses are totally ruled
out.
(56) a. *moto | [yuumei-na toosyu]

former [famous pitcher] 'formerly a famous pitcher'
b. *moto I [[ima Oosaka ni sundeiru] Koobe-simin]
former [[now Osaka-in living] Kobe-citizen]
'formerly a citizen of Kobe, who now lives in Osaka'
Moreover, there seems to be a semantic restriction on the genitive phras-

es that are allowed after moto-. The well-formed examples in (55) above
present a sharp contrast to the ill-formed ones below.
(57) a. *moto | [kanemoti no keieisya]

former [rich GEN manager] 'formerly a rich manager'
b. *moto I [gaikoku kara kita hitobito no kyoryuuti]
former [foreign-country FROM came people GEN settlement]
'formerly a settlement of the people coming from foreign countries'
Comparison of (55) with (57) demonstrates that the eligible genitive

phrases are limited to those which denote conventionalized concepts more
or less comparable to lexicalized compound nouns (without the genitive).
Just as the compound gaikoku-zin kyoryuuti 'foreign settlement' is a name
designating a particular historic site, so is the genitive phrase gaikoku-zin
no kyoryuuti when it is used after moto-. This semantic characterization
strongly suggests that the whole complex with moto- is a word rather than
a phrase, because only words can have a naming function.
Note that the genitive phrases in (55) are clearly distinct from lexical-
ized expressions like (58)—the Japanese equivalents of what Taylor
(1996) calls "possessive compounds" (e.g. a girls' school and a woman's
magazine).
(58) haha no hi 'Mother's Day', nomi no iti 'flea market', ama no zyaku
'devil's advocate', ama no gawa 'the Milky Way', mago no te 'back
scratcher'
Despite the superficial occurrence of the genitive particle, those expres-

sions are completely lexicalized and enjoy the freedom to participate in
lexical word formation, yielding compounds pronounced with a lexical
accent, like haha no hi purezento 'Mother's Day present' and nomi no iti
biyori 'fine weather suitable for flea markets'. Such lexical compounding
is not possible with the genitive phrases in (55). This leads us to conjec-
ture that the exceptional occurrence of the genitive in (55) might be due
to the contamination of the two different usages of the prefix moto- as a
W + prefix and as the phrasal determiner taking an NP structure (53). Put
another way, the no-phrases in (55) are attached directly to moto- (by mis-
taken analogy) and are consequently subjected to the general principles
of word formation like the naming function.
This conclusion implies that there is no necessity to postulate move-
ment either on the part of the prefix moto- or on the part of its base at any
level of derivation, like the LF movement of affixes proposed by Pesetsky
(1985) or the syntactic incorporation of Ν to D suggested by Miller (1993:
104). The absence of movement will be ascertained by the scope relation
of moto-. It is a general rule that the prefix takes the base (i.e. the material
it attaches to) in its scope. In (59a), moto- takes scope over syusyoo
'Prime Minister', whereby the proper name (Thatcher) that appears to
the left of moto- is outside the scope. This is reasonable, since 'ex-' desig-
nates only the official position.
(59) a. Thatcher | moto | syusyoo 'ex-Prime Minister Thatcher'

b. *moto I Thatcher | syusyoo 'formerly Prime Minister Thatcher'
The scope relation makes (59b) meaningless. This example could make
sense only in a hypothetical situation where Prime Minister Thatcher
changed her family name after leaving her office.
To sum up, we have observed that probably due to its native status, mo-
to- has the exceptional property of expanding its base from Word + to a
genitive N'. The N', however, must be restricted to conventionalized
nameworthy concepts and cannot include descriptive modifiers, because
the whole moto- complex counts as a "word".
4. Possible candidates for W + in English
The discussion so far has been confined to Japanese. In this section, we

will bring forth evidence from English to suggest the universal applicabil-
ity of our proposed W + category.
It has been frequently observed in English that most compound words
have primary stress on the first members, while some carry primary stress
270 Taro Kageyama
on the rear (Fudge 1984, Ladd 1984). Obviously, the former compounds
correspond to the Japanese compounds with a lexical accent, and the latter
to the Japanese W + compounds with a phrasal accent (Kubozono 1995).
(60) world fámous, crystal cléar, waist high, student rebéllion,

government fúnding, police investigátion, Arnold Pálmer,
Mr. Jones, President Clinton
The phrase-like stress pattern is also observed with such prefixes as non-,
ex-, and anti- which have their own stress (Bates 1988).
(61) pseúdo-scíentific, nón-Indo-Européan, ánti-[government

intervéntion], éx-[electrical enginéer]
The similarity with the Japanese W + becomes even more striking when
we notice that those complex expressions allow internal coordination.
(62) socio-linguistics and economics, anti-abortion and segregation,

ex-housewife and homemaker
Expressions like these, which have hitherto been analyzed simply as com-
pound words on a par with regular Word-level compounds, are now iden-
tified as candidates for the W + category.
It is interesting to note, as an anonymous review pointed out to me, that
the prefixes in (61) and (62) are borrowings from Latin or Greek, as all
the Japanese phrase-like prefixes except moto- 'former' are borrowed
from Chinese. Presumably, this will not be a mere coincidence but follow
from the nature of loan morphemes which is as yet to be explicated. Here
I can only speculate that native morphemes naturally fuse together into
tight units, forming words of relatively low levels (Stem, Root, Word),
whereas borrowed morphemes tend to resist total assimilation into cohe-
sive word units and some are merely added to the outermost shell of word
structure as an extra layer, Word + . An in-depth study is obviously neces-
sary on the nature of loan morphemes.
The Word + analysis can also provide an effective explanation for the
occurrence of inflectional elements inside complex words. Consider the
position of the plural -s in two types of compound words:
(63) a. pushups (*pushes-up) (Williams 1997), chéckups (*checks-up),

breakthroughs (*breaks-through)
b. lookers-ón, passers-by, hangers-on, diners-out, whippers-in,

runners-up, washers-up, brothers-in-law
The compounds in (63a), which undergo the regular Compound Stress

Rule, have the plural -s on the right edge, whereas those in (63b), carrying
a phrase-like stress pattern, allow the plural inflection inside them. Inter-
estingly enough, the reversal of order brings back the regular front stress
and sets the plural marker on the right edge.
(64) onlookers
If we assume that the plural inflection creates the Word"1" category, the in-
ternal plural marker in the compounds in (63b) can be explained by the
following derivation.
(65) look + -er -> [looker] N > [[looker] N -s]N+ - [ [[looker] N -s]N+ [on] P+ ] N+
A similar phenomenon is observed with complex words involving the suf-

fix -fui. A s pointed out by Allen (1978), this suffix has two manifestations
in (66a) and (66b).
(66) a. careful, powerful, delightful, thoughtful

b. mouthful, spoonful, cupful, bagful
The adjective-forming suffix -fui in (66a) is stressless with its vowel re-
duced to a schwa, whereas the -fui in (66b), denoting the amount held by
a container, carries a stress on its own and its vowel is fully pronounced
as [-ful]. Allen (1978) accounts for the distinction of the two types of -ful
by assigning the adjectival -ful to Level 2 and the fully pronounced -ful to
Level 3 in her extended level-ordering hypothesis. Our alternative analy-
sis is to treat Allen's Level 3 as constituting a distinct category of W + . This
analysis is supported by the fact that the expressions in (66b) can have the
plural inflection inside them, although a recent tendency is to put the plu-
ral on the right edge.
(67) spoonsful, bagsful (cf. spoonfuls, bagfuls)
Another candidate for W + in English is found in certain types of com-

pound adjectives which realize comparative and superlative inflections
on their first members.
272 Taro Kageyama
(68) a. widespread, more widespread, *wider-spread/*widest-spread

easy-going, more easy-going, *easier-going
b. well-known, better-known/best-known, *more well-known/*most
well-known good-looking, better-looking/best-looking, *more/
*most good-looking well off, better off/best off, *more well off/
*most well off
long-running, the longest-running musical comedy (Namiki 1985:144)
kind-hearted, kinder-hearted (Namiki 1985:144)
The examples in (63b), (66b), and (68b) have their internal structures ac-
cessible to the syntax, so that the plural and comparative/superlative in-
flections, which are generally held to take place in syntactic structure,
sneak into the branching W + structure.
Once W + is acknowledged as a legitimate, albeit marginal, morpholog-
ical category, a systematic analysis can be offered to examples like (70)
which have been slighted as simply idiosyncratic.
(69) unaccounted-for, uncared-for, uncalled-for, well-cared-for,

unaccountable-for, liveable-with, put-up-able-with {OED)
These examples are idiosyncratic in that prepositions are attached after

derivational suffixes, but the fact that the prepositions in them bear stress
with unreduced vowels is a strong indication of the W + status.
It is thus assumed that the Word + category opens up a new perspective
to make order out of morphological constructs that have been regarded
as disorderly or exceptional. Although our present focus has been on Jap-
anese and English, it is likely, as an anonymous reviewer has suggested,
that the same category has a universal applicability to phrase-like word
units in Hebrew, Turkish, Polish, and other languages which have been
problematic for the traditional theories that hold Word or X o as the max-
imal domain of morphology.
5. Conclusion
Recently, Bisetto and Scalise (1999) have taken up Italian complex ex-
pressions which look like compound words but are analyzed as syntactic
phrases because of the insertability of lexical material. From this they
conclude that morphology and syntax can be distinguished as different
components. The phrase-like expressions discussed in this paper pose a
similar problem of the morphology-syntax boundary but lead us to a dif-

ferent conclusion from Bisetto and Scalise's.
We have argued that different kinds of building blocks participate in
forming various "sizes" of words in Japanese. In particular, the new mor-
phological category of W + is proposed which is larger than Word but
nonetheless belongs to the realm of morphology. It might appear that the
formation of W + can be accommodated in the framework of level-or-
dered morphology by allocating it to the last level of derivation in the lex-
ical component, just as Kiparsky (1982) proposed that the regular inflec-
tions take place at the last stratum in his lexical phonology theory.
However, if the lexical and syntactic components are sharply separated
from each other, with the bracket erasure convention wiping out all mor-
phological structures before the syntax, the behavior of the referential
prefix doo- observed in section 2 remains unaccounted for.
Instead of level ordering, we postulate different types of morphological
categories to capture the real nature of the phenomena. The repertoire of
morphological and syntactic categories will now be hierarchically repre-
sented as follows.
(70)
syntactic XP
phrasal accent, reference by prefix doo-,
structure X'
plural and comparative/superlative inflections
Word +
morphology Word
structure Stem lexical accent
Root
Syntax generates the structures of XP and X' whereas morphology deter-

mines the structures of Root, Stem, Word, and Word + . Apart from the
word formation in syntax (e.g. incorporation) and phonology (e.g. clitici-
zation), this distinction in category types corresponds to the one between
the syntactic and the lexical components. The "component problem",
however, does not match the "atomicity problem" perfectly, because the
internal composition of Word + , a morphological object, can be made vis-
ible for particular syntactic and phonological purposes.
On the phonological side, the lexical accentuation in Japanese, apply-
ing inside Word, allows W + to have two or more lexical accents, thereby
274 Taro Kageyama
giving W + a phrase-like appearance. On the syntactic side, the referential

function of the prefix doo- serves to cancel the anaphoric islandhood, ren-
dering the internal structure of W + available to the interpretation of ana-
phoric relations at the sentence level. By the same token, the plural and
comparative/superlative inflections in English, which are normally at-
tached to the right edge of Word, can be embodied inside W + , as in look-
ers-on and better-known.
We have thus seen that the component problem (the lefthand side of
the diagram in (70)) and the atomicity problem (the righthand side in
(70)) are not strictly co-extensive but diverge at the border occupied by
the category W + . In a way, W + is a Janus-faced category at the intersec-
tion of morphological and syntactic structures. It is a special category
which might have originated from phrase structure but is now grammati-
cized as a morphological object. The reason we regard W + as special is
that it manifests itself only when the designated prefixation or com-
pounding is executed, and in unmarked cases the maximal projection in
word structure should be deemed Word or X o as has been customarily as-
sumed. For, if the Word category were automatically projected to Word"1",
then all words would have to be pronounced with a phrasal accent. This
makes a decided contrast to the category projections in syntax, where all
phrases are assumed to project automatically to the maximal projection
XP. We thus claim that the way of projection in word structure is different
from the one in syntactic structure, contrary to Lieber (1992) and Acke-
ma (1999), who maintain that identical principles apply to both structures.
Though erratic at first glance, the existence of W + will not be so unnat-
ural if we recall that the reverse case is also attested, namely syntactic in-
corporation. Since the unmarked locus of compounding is the lexicon,
syntactic incorporation can be viewed as a process which is lexically orig-
inated but is established as a grammatical operation in syntax. Viewed
from this perspective, the Word"1" category and syntactic incorporation
could be looked upon as epiphenomena deriving from a more fundamen-
tal principle of modularity in grammar: The modules that comprise the
grammar are basically independent, but are given leeway to have access
to each other under certain conditions.
Acknowledgements
I am grateful to the editors of this book for inviting me to contribute, and to two anonymous
reviewers for suggestive comments on the draft.
References
Ackema, P.
1999 Issues in morphosyntax. Amsterdam: John Benjamins.
Allen, M.
1978 Morphological investigations. Ph.D. dissertation, University of Connecticut.
Baker, M.
1988 Incorporation. Chicago: University of Chicago Press.
Bates, D.
1988 Prominence relations and structure in English compound morphology. Ph.D.
dissertation, University of Washington.
Bisetto, A. and S. Scalise
1999 Compounding: morphology and/or syntax? In L. Mereu (ed.), Boundaries of
morphology and syntax. Amsterdam: John Benjamins. 31-48.
Borer, H.
1998 Morphology and syntax. In A. Spencer and A.M. Zwicky (eds.), The hand-
book of morphology. Oxford: Blackwell. 151-90.
Bresnan, J. and S. Mchombo
1995 The lexical integrity principle. Natural Language & Linguistic Theory 13,181-
254.
Di Sciullo, A.-M. and E. Williams
1987 On the definition of word. Cambridge, Mass.: MIT Press.
Fudge, E.
1984 English word-stress. London: George Allen and Unwin.
Halle, M. and A. Marantz
1993 Distributed morphology and the pieces of inflection. In K. Hale and S.J. Key-
ser (eds.), The view from building 20. Cambridge, Mass.: MIT Press. 111-76.
Inkelas, S.
1989 Prosodie constituency in the lexicon. Ph.D. dissertation, Stanford University.
Kageyama, T.
1982 Word formation in Japanese. Lingua 57, 215-58.
1989 The place of morphology in the grammar. In G. Booij and J. van Marie (eds.)
Yearbook of Morphology 2. Dordrecht: Foris. 73-94.
1993 Bunpoo to gokeisei [Grammar and word formation], Tokyo: Hituzi Press.
1999 Word formation. In N. Tsujimura (ed.), The handbook of Japanese linguistics.
Oxford: Blackwell. 297-325.
Kiparsky, P.
1982 Lexical morphology and phonology. In The Linguistic Society of Korea (ed.),
Linguistics in the morning calm. Seoul: Hanshin Pub. Co. 3-92.
Kubozono, H.
1995 Gokeisei to on'in-koozoo [Word formation and phonological structure]. To-
kyo: Kurosio Publishers.
Ladd, D. R.
1984 English compound stress. In D. Gibbon and H. Richter (eds.), Intonation, ac-
cent and rhythm. Berlin: Mouton de Gruyter. 253-66.
Lieber, R.
1980 On the organization of the lexicon. Ph.D. dissertation, MIT.
1992 Deconstructing morphology. Chicago: University of Chicago Press.
Miller, D. G.
1993 Complex verb formation. Amsterdam: John Benjamins.
276 Taro Kageyama
Namiki, T.
1985 Gokeisei [Word formation], Tokyo: Taishukan.
Pesetsky, D.
1985 Morphology and logical form. Linguistic Inquiry 16,193-246.
Poser, W.
1990 Word-internal phrase boundary in Japanese. In S. Inkelas and D. Zee (eds.),
The phonology-syntax connection. Chicago: University of Chicago Press. 279-
88.
Postal, P.
1969 Anaphoric islands. CLS 5, 205-39.
Rappaport Hovav, M. and B. Levin
1992 -er Nomináis: implications for a theory of argument structure. In T. Stowell
and E. Wehrli, (eds.), Syntax and semantics 26: Syntax and the lexicon. New
York: Academic Press. 127-53.
Roeper, T. and M. Siegel
1978 A lexical transformation for verbal compounds. Linguistic Inquiry 9,199-260.
Sag, I. and J. Hankamer
1984 Towards a theory of anaphoric processing. Linguistics and Philosophy 7, 325-
45.
Selkirk, E.
1982 The syntax of words. Cambridge, Mass.: MIT Press.
Shibatani, M. and T. Kageyama
1988 Word formation in a modular theory of grammar. Language 64, 451-84.
Sproat, R.
1988 On anaphoric islandhood. In M. Hammond and M. Noonan (eds.), Theoreti-
cal morphology. San Diego: Academic Press. 291-301.
Taylor, J.
1996 Possessives in English. Oxford: Clarendon Press.
Ward, G., R. Sproat and G. McKoon
1991 A pragmatic analysis of so-called anaphoric islands. Language 67, 439-72.
Williams, E.
1997 Lexical and syntactic complex predicates. In A. Alsina et al. (eds.), Complex
predicates. Stanford: CSLI Publications. 13-28.
Further Evidence in Support of the Righthand Head
Rule in Japanese*
Takayasu Namiki
Ibaraki University
Introduction
The aim of this paper is to discuss further evidence in favor of the Right-
hand Head Rule (henceforth RHR) in Japanese and similar evidence in
English. After reviewing the arguments for the RHR in English by Wil-
liams (1981), I adduce three additional types of evidence for the RHR in
Japanese on the basis of Japanese compounds: (1) recursive compounds,
(2) reversible compounds, and (3) the abundance of complex elements in
the lefthand position in contrast to the scarcity of complex elements in the
righthand position in Japanese compounds. Furthermore, I show similar
evidence in English. I touch on questions such as whether there is a dif-
ference in expandability between the lefthand constituent and the right-
hand constituent of a compound in lefthand head compound languages as
there is in righthand head compound languages. Finally, I suggest replac-
ing the term "Righthand Head Rule" with the term "Final Head Rule", a
more general term than the former.
1. Williams' (1981) arguments for the RHR
In his well-known paper, Williams defines head as in (1) and proposes the
Righthand Head Rule (2) on the basis of two arguments from English,
which are given in (3) and (4).1
(1) The Definition of Head

If both X and the head of X are eligible members of category C,
then X e C = head of XeC. (Williams 1981: 247)
(2) The Righthand Head Rule

In morphology, we define the head of a morphologically complex
word to be the righthand member of that word. (Williams 1981:248)
278 Takayasu Namiki
(3) "It is generally the case that a suffix determines the category of a
word of which it is a part. Thus, for example, in English we have
the following:
X-ism Ν X-ize V X-ish A
V-ist Ν X-fy -»> V
V-ion ->· Ν
V-er Ν
A-ness Ν
Prefixes do not in general determine the category of the words

they attach to; rather, the category of a word of the form prefix X
is determined by the category of X. Thus:
counter+revolution N is a Ν
counter+sink v is a V
counter+productive A is a [sic] A" (Williams 1981: 248)
(4) "A similar situation holds for most compounds - the righthand
member determines the category of the whole:
[[offjp [white]A ]A [[dry]A [dock]N ]N [[bar]N [tend] v ]V"
(Williams 1981: 249)
Although there have been many books and articles dealing with the R H R
or the notion "head of a word" (Lieber 1980, 1992, Kageyama 1982,
Namiki 1982, 1985, 1994, Selkirk 1982, Zwicky 1985, 1993, Trommelen
and Zonneveld 1986, Di Sciullo and Williams 1987, Scalise 1988, 1992,
and Nishihara et al. this volume), I will not summarize them here. I will
simply assume that we can conclude that the RHR generally holds in
English on the basis of Williams' arguments alone. I will also assume that
we can conclude that the RHR generally applies to Japanese on the basis
of arguments in Kageyama (1982) and Namiki (1982), which are similar
to Williams'. In the following sections, I present three more types of evi-
dence from Japanese compounds that the RHR holds in Japanese.
2. Recursive compounds in Japanese
In languages where compounding is productive, compound nouns are by

far the most numerous. Compound nouns can also be coined freely. Fur-
thermore, compound nouns have the unique property that they are recur-
Further Evidence in Support of the Righthand Head Rule in Japanese 279
sively formed; in other words, there is no principled limit to the number

of constituents compounds may have, due to the fact that a compound
noun freely becomes the base of another compound noun. Following are
some examples of "recursive compounds" in Japanese (heads in capitals).
(5) a. kokkai -taisaku -IINCHOU

'Diet' 'measures' 'CHAIRPERSON' = 'chairperson for negotiation in the Diet'
b. yatou -yontou -kokkai -taisaku -iinchou -KAIGI
'opposition party' 'four parties' 'Diet' 'measures' 'chair' 'MEETING'
(Namiki 1985: 81)
= 'a meeting of the chairs of four opposition parties for negotiation in the Diet'
c. chuukounen -roudou -idou -shien -tokubetsu
'aged and middle-aged' 'labor' 'movement' 'support' 'special'
-JOSEIKIN
'GRANT-IN-AID'
= 'a special grant-in-aid for supporting aged and middle-aged people who have
changed their jobs' (The Asahi, morning edition, 6/10/99 (Thu), p. 4)
d. chihou -koumuin -seido -chousa -kenkyu -kai -HOUKOKU
'local' 'civil servant' 'system' 'investigation' 'research' 'meeting' 'REPORT'
= 'a report of the meeting for the investigation of local civil servant system'
(The Asahi, morning edition, 6/10/99 (Thu), p. 1)
e. minami -kantou -chiiki -shinsai -oukyuu -taisaku -katsudou
'south' 'Kantou' 'area' 'natural disaster' 'emergency' 'measures' 'action'
-YOURYOU
'MANUAL'
= 'a manual for emergency measures for natural disasters in the South
Kantou area' (Jouhou Chishiki Imidas, 1990, Shuueisha Publishing Co., p. 878)
f. Mitoshi -Mataguma -chiku -sangyou -haikibutsu -saishu -shobunjou
'Mito City' 'Mataguma' 'area' 'industrial' 'waste' 'final' 'disposal factory'
-kensetsu -hantai -UNDOU
'construction' 'opposition' 'MOVEMENT' = 'a movement against the construction
of a final disposal factory of industrial waste at Mataguma in Mito City' (Pamphlet
of Annual Meeting in 1996, Union of Faculty and Staff, Ibaraki University, p. 21)
These are all examples of so-called "Sino-Japanese compounds" which

are particularly productive in Japanese. In all of the above compounds the
head of the compound occurs in the rightmost position, regardless of the
internal structure of the compound.2 By the head of a compound I mean
the constituent of a compound which [1] determines the lexical category
280 Takayasu Namiki
of the whole compound (cf. (1) above), and [2] has a "kind of" or "IS A"
relation between the whole compound and itself (cf. Namiki 1994). How-
ever, since every constituent of the recursive compounds in (5) is a noun,
it should be noted that the first criterion for defining the head of a com-
pound (i.e. the determination of the lexical category of the whole com-
pound) is vacuously satisfied, and that the second criterion is necessary in
deciding which noun is the head in the compounds in (5).
3. Reversible compounds in Japanese
There are many examples of "reversible compounds" (Scalise 1992:179)

in Japanese. These are compounds where both the form XY and the form
YX exist. In most cases they refer to different entities (English examples:
housedog and doghouse). In either order, the righthand constituent is the
head of the compound (written in capitals below). Consider the following
(cf. Namiki 1994: 50)):3
(6) a. hachi-MITSU mitsu-BACHI

'bee' 'SYRUP' = 'honey' 'syrup' 'BEE' = '(honey) bee'
b. tsutsumi-GAMI kami-ZUTSUMI
'wrapping' 'PAPER' = 'wrapping paper' 'paper' 'WRAPPING'
= 'something wrapped in paper'
c. mizu-DEPPOU teppou-MIZU
'water' 'GUN' = 'water pistol' 'gun' 'WATER' = 'flash flood'
d. taru-ZAKE saka-DARU
'cask' 'SAKE' = 'sake in a cask' 'sake' 'CASK' = 'cask for sake'
e. doku-GUMO kumo-DOKU
'poison' 'SPIDER' 'spider' 'POISON'
= 'poisonous spider' = 'spider poison'
f. w a i n - G U R A S U gurasu-WAIN
'wine' 'GLASS' = 'wine glass' 'glass' 'WINE'
= 'a glass of wine esp. served in a restaurant'
g. e i s e i - H O U S O U housou-EISEI
'satellite B R O A D C A S T I N G ' 'broadcasting' 'SATELLITE'
= 'satellite broadcasting' = 'broadcast satellite'
h. shokuhin-KAKOU kakou-SHOKUHIN
'food' 'PROCESSING' 'processing' 'FOOD'
= 'food processing' = 'processed food'
i. basu-TSUUGAKU tsuugaku-BASU
'bus' 'GOING TO SCHOOL' 'going to school' 'BUS'
= 'going to school by bus' = 'bus for going to school'
j. kyouiku-KANKYOU kankyou-KYOUIKU
'education' 'ENVIRONMENT' 'environment' 'EDUCATION'
= 'educational environments' = 'environmental education'
These are typical examples of reversible compounds in Japanese, and

many more such pairs can be found easily.
In each pair of compounds in (6), the XY and YX forms have different
referents. For example, a referent of "wain-GURASU (XY)" is a kind of
"gurasu (Y)" ('glass') and a referent of "gurasu-WAIN (YX)" is a kind of
"wain (X)" ('wine') in (6f). Furthermore, the difference in referents can
be shown by consideration of the classifiers which each member of these
pairs takes. For example, since mitsu-bachi refers to a kind of insect, the
classifier -hiki can be used with it, and so ni-hiki no mitsu-bachi ('two
bees') is a well-formed phrase. On the other hand, since hachi-mitsu is a
kind of sweet liquid, the classifier bin ('bottle') can be used with it, and so
hito-bin no hachi-mitsu ('a bottle of honey') is a well-formed phrase in
Japanese (cf. Namiki 1992:67). Likewise, the classifier -mai (as in san-mai
no tsutsumi-gami 'three sheets of wrapping paper') can be used with tsut-
sumi-gami, while the classifier -ko (as in san-ko no kami-zutsumi 'three
things each of which is wrapped in paper') can be used with kami-zutsumi.
Indeed there are a few pairs where the XY form and the YX form denote
the same entity (for example, wa-gomu and gomu-wa (both meaning
'rubber band'), keshi-gomu and gomu-keshi (both meaning 'eraser'), and
karee-raisu and raisu-karee (both meaning 'curry and rice')), but these are
clearly exceptional and easy to list.
In my opinion it is a significant property of compound nouns that their
constituents (when there are two) can be reversed in many cases. Com-
pounds are, by definition, formed with two or more independent words,
and independent words generally do not have a restriction on their order
in compounds (such that some words can be used in the lefthand or initial
position only, or that some words can be used in the righthand or final po-
sition only). This property is clearly different from that of prefixes and
suffixes.
I suspect that a large number of reversible compounds can be seen only
in languages which have righthand head compounds, though it may be the
case that a handful of reversible compounds can be seen in languages
which have lefthand head compounds.
282 Takayasu Namiki
4. The abundance of complex elements in the lefthand

constituent of compounds
In syntax we usually make a distinction between a modifier and a head

and one between a head and a complement. Such distinctions are also
made in morphology. In syntax it is usually a modifier and a complement
rather than a head that are made more expanded by coordination and/or
modification. Let us turn to the question of whether such a difference in
expandability exists between a modifier/complement and a head in Japa-
nese morphology.
On the basis of the arguments made in sections 2,3, and 4, we can safely
conclude that Japanese compounds have their heads in the righthand po-
sition. Since the head position in Japanese is in the righthand side both in
a modifier-head relation and in a complement-head relation in morphol-
ogy as well as in syntax, it is predicted that there is a similar difference in
expandability between a modifier/complement and a head in Japanese
morphology. In what follows, I discuss various types of compounds where
complex elements occur as a lefthand constituent, and show that this pre-
diction is borne out.
4.1. Compounds as lefthand constituents and simple words

as righthand constituents
There are a number of examples of Japanese compounds in which the

lefthand constituent is a compound. Some of them are given below
(brackets are mine, unless otherwise stated):
(7) a. [[Porutogaru -chuubu] [tokuyuu]] no areta kouchi

'Portugal' 'midland' 'particular' G E N (= genitive) 'barren' 'plateau'
= 'a barren plateau particular to Midland Portugal' (Namiki 1993:16)
b. [[sumiyaki -gama] [dokutoku]] no nioi
'charcoal making' 'kiln' 'particular' G E N 'scent'
= 'the scent particular to a charcoal kiln' (Namiki 1993:16)
c. [[chiteki -koukishin] [ousei]] na dokusha
'intellectual curiosity' 'full o f 'readers'
= 'readers full of intellectual curiosity' (Namiki 1993:16)
d. [[kiso-kenkyuu] [hitosuji]] de ikitekita Nakahara w a , . . .
'basic research' 'be devoted only to' 'having lived' 'Nakahara' TOP (= topic)
= 'Nakahara, who has devoted himself only to basic research,'(Kunio Yanagida,

1981, Gan-kairou no Ashita (Gekan), Koudansha Publishing Co., p. 246)
e. [[sanuki-fujin] [ichiryuu]] no kokoronimo nai kenson no poozu
'Sanuki' 'woman' 'specific' GEN 'in her mind' 'absent' 'modesty' GEN 'pose'
='just a pose of modesty which was specific to women in Sanuki.' (Namiki 1993:16)
All the righthand constituents in the compounds in (7) are "adjectival

nouns" ("keiyou-meishi" in Kageyama (1993:23)), which have traditional-
ly been called "keiyou-doushi" ("adjectival verbs") in Japanese grammar.
There is, however, a distinction between these adjectival nouns. The
first type is those to which ni or ga (so-called "joshi" (particles)) can be
pre-attached optionally, and the second type is those to which no particles
can be pre-attached. Observe the following contrast:
(8) a. Porutogaru-chuubu ni tokuyuu 'particular to Midland Portugal'

(a phrase and not a compound)
b. sumiyaki-gama ni dokutoku 'particular to a charcoal kiln'
(a phrase and not a compound)
c. chiteki-koukishin ga ousei 'full of intellectual curiosity'
(a phrase/clause and not a compound)
(9) a. *kiso-kenkyuu ni/ga hitosuji
b. *Sanuki-fujin ni/ga ichiryuu
Furthermore, the hitosuji type of adjectival nouns in (9) are particularly

interesting in that the meanings (or submeanings) of these adjectival
nouns used in compounds in (7d) and (7e) are different from those of the
same adjectival nouns used independently: hitosuji in (7d) means "be de-
voted only to ..." and ichiryuu in (7e) means "specific to ...", while hito-
suji used independently means "one (straight) line" and ichiryuu used in-
dependently means "first class or first rate". In other words, hitosuji and
ichiryuu have special submeanings only when they are used as the right-
hand constituents of compounds. These special submeanings can be
called "compound-specific submeanings." It should be stressed that hito-
suji and ichiryuu are marked cases and the majority of adjectival nouns
(such as tokuyuu, dokutoku, ousei, batsugun, souou) are unmarked cases,
that is, the former do not allow ni or ga to pre-attach to themselves and
they have the compound-specific submeanings, while the latter allow ni
or ga to do so and they have the same meanings that they have in non-
compounds.
284 Takayasu Namiki
4.2. Conjuncts as lefthand constituents and simple words

There are examples of Japanese compounds where coordination occurs in

the lefthand constituent. Consider the following:
( 1 0 ) a. [[okkusufoodo to kenburijji] tokuyuuno] "saijo" taipu

'Oxford' 'and' 'Cambridge' 'particular' 'talented women' 'type'
= 'a type of talented women particular to Oxford and Cambridge Universities'
(Akane Kawakami, 1995, Watashi no Okkusufoodo, Shoubunsha Publishing Co.,
p. 182)
b. [[koukishin to makenki] ousei] datta kanojo
'curiosity' 'and' 'competitive spirit' 'full of' 'was' 'she'
= 'she who was full of curiosity and a competitive spirit' (The Asahi, evening
edition, 10/3/95 (Tue), p. 15)
c. [[ehon to jidousho] hitosuji] de yatte kiteita kara,
'picture book' 'and' 'books for children' 'be devoted only to' 'have done' 'because'
= 'because I was devoted only to picture books and books for children' (Kunio
Yanagida and Hideko Ise, 1998, "Hajimari no Kioku," Hon, Koudansha, Vol. 23,
No. 5, p. 5)
d. [[shufu to hanayome-shugyou no ojousan] aite] no ryouri-gakkou
'housekeeper' 'and' 'training for marriage' 'young women' 'for' 'cooking school'
= 'a cooking school for housekeepers and young women training for marriage'
(The Asahi, morning edition, 4/12/92 (Sun), p. 12)
e. shikashi genjitsuwa [[juku ya yobikou] gayoi] de ...
'but' 'in reality' 'cram school' 'and' 'prep school' 'going to' 'because of'
= 'but, in reality, because of going to a cram school and/or a prep school,...'
f. [karaoke to geemu] taikai (Kageyama 1993: 327)
'karaoke' 'and' 'game' 'competition'
= 'karaoke and game competitions'
g. [shokuji to miyage] tsuki (Kageyama 1993: 327)
'meal' 'and' 'present' 'with'
= 'with a meal and a present'
h. [[yuumoa to sesou -bunseki] afureru] koramu-kiji
'humor' 'and' 'social conditions' 'analysis' 'be full of' 'column' 'article'
= 'a column which is full of humor and good at analyzing social conditions'
(The Asahi, evening edition, 11/28/92 (Sat.), p. 7)
In (10), the righthand constituents of the compounds in question are all

words: adjectival nouns in (10a), (10b), and (10c); nouns in (lOd), (10e),
(lOf), and (10g); a verb in (10h). Other lexical categories have not been
found in my data. On the other hand, the lefthand constituents in the
compounds under consideration vary: in (10a), (10b), (lOf), and (10g),
two nouns are coordinated; in (10c), (10e), and (10h), a noun and a com-
pound noun are coordinated; in (lOd) a noun and a noun phrase are co-
ordinated. Notice that hitosuji, whose characteristic was emphasized in
and around (9a), is used again in (10c), where coordinated nouns are used
as its lefthand constituent.
In all the examples in (10), it seems impossible to assume that some
omission or deletion process is involved in forming the compounds at is-
sue (the bracketed parts), since such an assumption would lead to the
claim that, for example, *yuumoa afureru to sesou-bunseki afureru kora-
mu-kiji (cf. (10h)), and %shufu aite to hanayome-shugyou no ojousan aite
no ryouri-gakkou (cf. (lOd)) are synonymous well-formed expressions.
But the former expression, in fact, is ill-formed. The latter is non-synon-
ymous with (lOd), because the latter expression, though well-formed,
means that there are two cooking schools, one for housekeepers and the
other for young women, but (lOd) means that there is one school for both
housekeepers and young women.
4.3. Noun phrases as lefthand constituents and simple words

In the next examples, noun phrases occur as the lefthand constituent of a

compound. Let us observe the following:
( 1 1 ) a. [[konpa no soshikiryoku] N P [batsugun]] no Y-san

'party' G E N 'organizing ability' 'distinguished' G E N 'Mr. Y'
= 'Mr. Y, who is distinguished in the ability to organize a party'
(Eiichi Chino, 1993, Sekai Kotoba no Tabi, p. 12)
b. [[[jibun ga motte umareta] s sainou] N P [souou]] no shigoto
'he' N O M (= nominative) 'innately gifted' 'talent' 'suitable' G E N 'job'
= 'the job which is suitable to the talent with which he was innately gifted'
(Hiroyuki Agawa, 1993, "Shiga Naoya, No. 69" Tosho, No. 527, p. 51.)
c. [[[kotosarani jibun no bibou o ishiki shiteiru] s josei] N P [tokuyuu]]
'especially' 'her own' G E N 'beauty' OBJ 'conscious' 'woman' 'particular'
no tsumetasa
G E N 'coldness'
286 Takayasu Namiki
= 'the coldness particular to women who are especially conscious of their own
beauty' (Namiki 1993:16)
d. [huruhonya no nyoubou] NP goroshi (Kageyama 1993: 327)
'secondhand book seller' GEN 'wife' 'killing'
= 'the killing of the wife of a secondhand book seller'
e. [[anokoro no chuuryuu -katei] NP [sodachi]] no hitotsu no pataan
'those days' GEN 'middle-class' 'family' 'being raised' GEN 'one' 'pattern'
= 'one type which was typical of the people who were raised in middle-class
families those days' (Teruhiko Kuze, 1996, Furemo Sede, Koudansha, p. 20)
f. [[rajio no buhin] NP zukuri] ya...
'radio' GEN 'parts' 'making' 'and'
= 'the making of radio parts and ...' (Souji Shimada, 1986, Kakei Toshi, Koudansha,
p. 257)
The righthand constituents of the compounds above are adjectival nouns

(batsugun in (11a), souou in ( l i b ) , and tokuyuu in (11c)) and a noun {gor-
oshi in ( l i d ) , sodachi in ( l i e ) , and zukuri in ( l l f ) ) . We can clearly see
from (8) to (12) that adjectival nouns are particularly productive in form-
ing compounds whose lefthand constituents are complex. For details of
such adjectival nouns, see Namiki (1993).
It should be clear from the preceding arguments that lefthand constit-
uents of compounds are easily expandable. By "expandable" I mean here
that constituents can contain compounds, coordination, phrases, or claus-
es. How about the righthand constituents then? This is the topic dealt with
in the next section.
5. The scarcity of complex elements in the righthand

constituents of compounds
Kageyama (1993) states that phrases are generally excluded in words and
he furthermore maintains that
... the principle of the exclusion of phrases in words is occasionally

loosened with respect to modifiers, but it is strictly observed with re-
spect to heads, (my translation - TN) (Kageyama 1993: 215)
showing the following contrast:

( 1 2 ) a. [[sunakku no mama] goroshi] (Kageyama 1993: 215)

'snack bar' GEN 'proprietress' 'killing' =
'the killing of the proprietress of a snack bar'
b. *[hito [dasuke to tsukiai]] (Kageyama 1993: 215)
'others' 'helping' 'and' 'acquaintance'
c. *[abura [azayakana e]] (cf. [abura e]) (Kageyama 1993: 328)
'oil' 'vivid painting' 'oil' 'painting'
He also gives other examples similar to (12b), as shown below.
( 1 3 ) a. »[[karaoke] [taikai to kontesuto]] (Kageyama 1993: 328)

'karaoke' 'party' 'and' 'contest'
b. *[[kokuritsu-daigaku] [kyouju to jokyouju]] (Kageyama 1993: 339)
'national university' 'professor' 'and' 'associate professor'
c. *[[gaikokujin] [koushi to kyoushi]] (Kageyama 1993: 339)
'foreigner' 'lecturer' 'and' 'teacher'
Judging from his examples in (12b) and (13), it seems that the Japanese
coordinating conjunction to ('and') cannot coordinate two words in a
righthand constituent of a compound. But he points out that other coor-
dinating conjunctions oyobi ('and' (formal)) and naishi ('or' (formal))
can do so, on the basis of the following examples (Kageyama 1993: 339):
( 1 4 ) a. [[kokuritsu-daigaku] [kyouju oyobi jokyouju]]

'national university' 'professor' 'and' 'associate professor'
b. [[gaikokujin] [koushi naishi kyoushi]]
'foreigner' 'lecturer' 'or' 'teacher'
Oyobi and naishi are formal counterparts of to ('and') and matawa ('or'),
respectively.
Two more examples show that coordinating conjunctions other than to
can coordinate two words in the righthand constituent of a compound.
Consider the following:
( 1 5 ) a. hinshu-kairyou dewa nai ga, [[tane-nashi] [suika ya

'species improvement' 'be' 'not' 'but' 'seedless' 'watermelon' 'and'
budou]] no youni
'grape' 'like'
= 'it is not the improvement of species, but like seedless watermelons and grapes
...' (Akihiko Okabe, 1996, "Kagaku no Kisetsuhuu," Gakutou, Vol. 93, No. 7, p. 65)
288 Takayasu Namiki
b. bengoshi o yousuru [[wagakuni] [saidai de yuiitsu]] no zenkoku

'lawyers' OBJ 'include' 'Japan' 'largest' 'and' 'only one' 'nation-wide'
-soshiki
'organization' = 'it was the only nation-wide organization consisting of lawyers,
which was also the largest in scale ...' (Hiroyuki Nakajima, 1994, Kensatsu-sousa,
Koudansha Publishing Co., p. 44)
The coordinating conjunction ya ('and' with a non-exhaustive meaning)

implies that there may be more things involved than those conjuncts ex-
plicitly mentioned, and de ('and') coordinates adjectival nouns.
Kageyama (1993: 261) states that it is difficult for to ('and') to coordi-
nate two words in the lefthand constituent and those in the righthand con-
stituent of compounds. See also Kageyama (this volume). It is true that to
generally fails to coordinate two words in the righthand constituent of
compounds, as shown in (12b) and (13). But I have found an instance
where to does coordinate two words in the righthand constituent. Ob-
serve the following:
(16) [[kakuseizai] [shiyou to shoji]] de juuikkagetsukan keimusho

'stimulant drug' 'use' 'and' 'possession' 'because o f 'for 11 months' 'prison'
ni ita.
'in' 'was' = '(He) was in prison for 11 months because he had possessed and used
stimulant drugs.' (The Asahi, morning edition, 5/30/96 (Thu), p. 34)
We have just seen a few examples where conjunctions coordinate two

words in the righthand constituent of compounds, but it should be
stressed that they are clearly exceptional. Furthermore, it is possible to
assume that some omission or deletion process is involved in forming
such exceptional examples: for example, kokuritsu-daigaku kyouju oyobi
kokuritsu-daigaku jokyouju (cf. (14a)), tanenashi suika ya tanenashi bu-
dou (cf. (15a)), and kakuseizai shiyou to kakuseizai shoji (cf. (16)).
I do not think, however, that it is difficult for to to coordinate two
words in the lefthand constituent of a compound, because I already gave
some such examples in (10). In particular, adjectival nouns seem to ac-
cept such examples fairly easily (see (10a-d)). Judging from the discus-
sions so far, it can be concluded that a lefthand constituent in Japanese
compounds is easily expandable, while a righthand constituent is limited
in expandability.
In other words, the lefthand constituent can contain compounds, coor-
dination, phrases, and clauses under certain conditions, but the righthand
constituent can contain only some of these. Specifically, the righthand

constituent of a compound can be a compound, since it is still a word,
though it is complex in the sense that it is formed out of two or more
words. As to the coordination of the righthand constituents, it is a gray
area in that the coordination of the righthand constituents is basically not
allowed, as Kageyama (1993) points out, although a handful of such ex-
amples have been found. It should be stressed, however, that examples
have not been found where a phrase or a clause occurs in the righthand
constituent.
I would like to introduce a new symbol here in order to express the dif-
ference in expandability between the lefthand constituent and the right-
hand constituent of compounds in Japanese. The symbol in question is
"X". By X I mean that a lexical category of a word can be ex-
panded to a category which constitutes a phrase or a clause under certain
conditions. If we use this symbol, we can generally express the internal
structure of Japanese compounds, as follows:
(17) the internal structure of Japanese compounds: [X Y]4
From this it follows that in Japanese, the righthand constituent is the

head of a compound, since the lefthand constituent is expandable into a
phrase or a clause under certain conditions and so cannot be a head of a
compound word by definition, but the righthand constituent has to be a
word.
Therefore, I maintain that the prediction made in section 4—that there
is a difference in expandability between a modifier/complement (i.e. the
lefthand constituent of compounds) and a head (i.e. the righthand constit-
uent of compounds) in Japanese morphology—has been borne out by the
arguments made in sections 4 and 5.
To summarize, I have argued that there are three more pieces of evi-
dence in support of the RHR in Japanese on the basis of various com-
pounds, apart from those given in Williams (1981): (1) recursive com-
pounds, (2) reversible compounds, and (3) the abundance of complex
elements in the lefthand constituent in contrast to the scarcity of complex
elements in the righthand constituent of Japanese compounds. I also have
argued that there exists a difference in expandability between a lefthand
constituent and a righthand constituent of Japanese compounds, and that
this difference can be attributed to the internal structure of such com-
pounds in (17).
290 Takayasu Namiki
6. The existence of similar evidence for the RHR in English
In this section I briefly deal with evidence for the RHR in English similar
to that for the RHR in Japanese given so far.
6.1. Recursive compounds in English
Some examples of recursive compounds in English are given below:
( 1 8 ) a. parcel post customs DECLARATION (Namiki 1985: 81)

b. Massachusetts Water Resources Research CENTER (Namiki 1985: 81)
c. farm produce delivery truck repair SHOP (Allen 1978:188)
d. student film society committee scandal INQUIRY (Spencer 1991: 41)
e. bathroom towel rack designer TRAINING (Selkirk 1982:15)
f. bathroom towel rack designer training program COMMITTEE (Namiki 1991:524)
In every case the head of the compound occurs in the rightmost position.
These recursive compounds constitute one type of evidence for the RHR
in English, as in Japanese.
6.2. Reversible compounds in English
There are many examples of reversible compounds in English such as the

following:
(19) a. house D O G dog HOUSE (Namiki 1993:19)

b. sugar MAPLE maple SUGAR (Namiki 1993:18)
c. piano PLAYER player PIANO (Namiki 1993:19)
d. association FOOTBALL football ASSOCIATION
e. horse RACE race HORSE
f. grammar SCHOOL school GRAMMAR
g. oil LAMP lamp OIL
h. lunch BOX box LUNCH
It is semantically clear that in each pair the righthand constituent is the head
of compounds. The difference in referents in each pair can also be seen, for
example, in the fact that sugar maple is a countable noun, as maple is, and
maple sugar is an uncountable noun, as sugar is. This is similar to the differ-
enee in classifier selection by Japanese reversible compounds (e.g. hito-bin

no hachimitsu and ni-hiki no mitsu-bachi) discussed in section 3.
6.3. The abundance of complex elements in the lefthand

constituent of compounds in English
There are many examples where complex elements appear in the lefthand
constituent of English compounds. They are divided into a few subtypes,
as in section 4.
6.3.1. Compounds in the lefthand constituent of compounds

Let us consider the following:
(20) a. dishwasher-proof (Namiki 1985: 80)

b. ginger ale bottle green (Selkirk 1982: 48)
c. motor car accident prone (Selkirk 1982: 48)
These compound adjectives have compound nouns as lefthand constitu-

ents.
6.3.2. Coordination in the lefthand constituent of compounds

Examples are given where coordination of words occurs in the lefthand
constituent of English compounds. Observe the following:
(21) a. the fire- and water-proof building

b.... in customs and immigration lines (Paul Gilbert, Jr. and Itsuo
Shirono, 1993, American Life in Focus, Yumi Press, p. 26)
c. Within the sentence we can indicate clause and phrase boundaries
by commas (,),... (Geoffrey Leech and Jan Svartvik, 1994, A Com-
municative Grammar of English, 2nd ed., p. 18.)
d. How can you become a blood or organ donor? (Jeri Lyn Rieken
and Gail House, 1986, Succeeding on Your Own, HBJ, p. 361)
e. the Charles and Di syndrome (Lieber 1992:11)
f. a pipe and slipper husband (Lieber 1992:11)
These examples seem to be superficially the same in structure, but two

types should be distinguished. The first type is related to omission or de-
letion of the repeated words, and the second type is not. Let us look at the
following contrast:
292 Takayasu Namiki
(22) a. the üre-proof and water-proof building (synonymous with (21a))

b. ... in customs lines and immigration lines (synonymous with (21b))
c. ... clause boundaries and phrase boundaries (synonymous with (21c))
d.... a blood donor or organ donor (synonymous with (21d))
e. *the Charles syndrome and Di syndrome
f. *a pipe husband and slipper husband
Lieber (1992: 11) calls examples like those in (21e, f) "phrasal com-
pounds."
6.3.3. Phrases and clauses in the lefthand constituent of compounds
There are relatively many instances where phrases occur in the lefthand
constituent of English compounds. Consider the following:
(23) a. a ground-to-air missile (Fabb 1984:136)

b. two-pack-a-day smoker (Morita 1985: 49)
c. ready-for-departure signs (Namiki 1985:155)
d. an under-the-stars concert (Shimamura 1986: 24)
e. over the fence gossip (Lieber 1992:11)
f. the adjuncts-as-heads analysis (Fujihara 1995:175)
There are not so many examples where clauses occur in the lefthand con-
stituent. Some of them are given below:
(24) a. a don't-tell-me-what-to-do look (Bauer 1983: 164)

b. a who's the boss wink (Lieber 1992:11)
c. the get-married-by-25-no-matter-what compulsion
(Fujihara 1995:180)
It should be clear that the lefthand constituents of compounds are as eas-

ily expanded in English as in Japanese. Next, I examine whether the same
happens with respect to the righthand constituent of compounds.
6.4. The scarcity of complex elements in the righthand

constituents of compounds in English
I do not know of any explicit statement about the scarcity of complex el-
ements in the righthand constituents of compounds in English. But, to my
knowledge, it seems harder to find the compounds whose righthand con-
stituents are expanded. I have found only one such example. 5 Consider
the following:
(25) ... psycholinguistics, which studies the psychological processes un-

derlying
[[speech] [production and comprehension]]. (Andrew Radford,
1997, Syntax: A Minimalist Introduction, Cambridge University
Press, p. 2)
Concerning this example it is appropriate to assume that some deletion

process is involved in forming this compound:
(26) ... underlying speech production and speech comprehension
Probably such examples are limited to superficial coordination of the

righthand constituents of compounds after the deletion of the repeated
words, as in the Japanese compounds discussed in section 5.
It seems reasonable to conclude that there is a difference in expand-
ability between a modifier/complement and a head in English morpholo-
gy as well as in Japanese morphology.
In conclusion, it has been shown that there are various types of evi-
dence in support of the RHR in the morphology of Japanese and English.
In other words, Japanese and English are typical languages with right-
hand head compounds.
7. Residual problems
There are many languages in the world which have been claimed to have
righthand head compounds. They include not only Japanese and English
but also Korean, Dutch, German, etc. Of course, there are languages
which have been claimed to have lefthand head compounds, such as mod-
ern Italian (Scalise 1988, 1992), Vietnamese (Lieber 1980), and Indone-
sian (Subandi 1998). But I suspect that there are differences between the
former type of language and the latter type of language, that is, differenc-
es in the kinds of evidence in support of the righthand head compounds
and evidence in support of the lefthand head compounds. Specifically, I
would like to raise questions such as whether there are as many recursive
compounds in the latter type of language as in the former type of lan-
guage,6 whether there are as many reversible compounds in the latter as
in the former, and whether there is a difference in expandability between
294 Takayasu Namiki
the lefthand constituent and the righthand constituent, though in a mir-

ror-image direction, in the latter type of language. Further research is
necessary to show whether there are such differences and, if there are, to
clarify why there are.
Finally, I would like to touch on the term "Righthand Head Rule." In
my opinion, this term seems inappropriate, since it presupposes that let-
ters in words and sentences are generally written from left to right in nat-
ural languages. In some languages, however, letters are written from right
to left, and in others, letters are written from top to bottom. 7 A more ap-
propriate term would be "Final Head Rule", since this does not depend
on the direction of writing systems, and so is a more general term than
"Righthand Head Rule." The "Final Head Rule" will get support if we
can find languages where letters are written from right to left and the
head of a compound occurs on the lefthand side. For the head of a com-
pound in such languages is a final constituent, though on the lefthand
side, and it can be covered by the Final Head Rule, but not by the Right-
hand Head Rule. This new term is more consistent with the parametric
view of compound heads: head-initial or head-final. 8
Notes
* I wish to express my gratitude to the editors of this book, Jeroen van de Weijer and Tet-
suo Nishihara, for offering me a chance to contribute this paper to the present volume. I
would like to thank Edward Quackenbush, Brent de Chene, Jeroen van de Weijer, Nori-
aki Yusa, and two anonymous reviewers for giving me valuable comments and sugges-
tions on my earlier drafts, and Edward Quackenbush and Brent de Chene for improving
my drafts stylistically. I am also grateful to my students at Ibaraki University and Ibaraki
Christian College for collecting part of my data on compounds in Japanese and English.
Any remaining errors and inadequacies are my own.
1. This definition in (1) means that if both X and the head of X are eligible members of cat-
egory C, then X's membership of C is equivalent to the head of X's membership of C. Wil-
liams himself notes that there are exceptions to the R H R : a prefix en- and exocentric
compounds such as push up and run down (Williams 1981: 249-250). Namiki states that
there are at most seven prefixes in English which are category-determining and so con-
stitute exceptions to the R H R : a- (as in afire and aloud), be- (as in behead and belittle),
de- (as in defrost and demilitarize), dis- (as in disbar and disable), en- (as in encage and
enrich), out- (as in outdistance and outsmart), and un- (as in unleash) (Namiki 1982: 27-
28, and Namiki 1985: 21-22).
2. Kubozono (1995: 104-105) states that right-branching compounds are prosodically
marked and left-branching compounds are prosodically unmarked in Japanese and Eng-
lish. H e also maintains that the number of left-branching compounds (as in a. and b. be-
low) is much larger than that of right-branching compounds (as in a', and b'. below) in
Japanese and English.
a. [[seiji seido] kaikaku]

'politics' 'system' 'innovation' = 'innovation of a political system'
a', [[kouhaku [uta gassen]]
'red and white' 'song' 'battle' = 'song competition between men and women'
b. [[computer class] instructor]
b'. [evening [computer class]]
In all the examples above, the head of compounds appears as the rightmost constituent.
Kubozono (1999: 140), furthermore, claims that the phonological markedness of right-
branching structure is observed in languages such as Chinese and Italian. See also Akasa-
ka and Tateishi (this volume).
3. Italicized initial consonants in the righthand constituents of these reversible compounds
indicate that so-called "rendaku" (sequential voicing) has occurred in the initial conso-
nants at issue. For details of rendaku, see Otsu (1980) and Akasaka and Tateishi (this vol-
ume), among others. Saka- in (5d) is the variant form of sake which is specific to the left-
hand constituent of compounds in Japanese (cf. saka-gura and saka-mushi).
4. An anonymous reviewer has suggested to me a similar structure ([XP] Y), attributing the
difference in expandability to the difference in categorial status between the lefthand
constituent and the righthand constituent, and bringing Cinque (1993) to my attention.
5. Ed Quackenbush has given me some more examples like (27): automobile design and
manufacture, lawmakers and breakers, and missile launching and recovery. The deletion
of the repeated words has applied to these examples, too.
6. Scalise (1992) denies the existence of recursive compounds in modern Italian, stating "In
general Italian compounds are not recursive ... Compounds such as ... capostazione can-
not serve as constituents of other compounds." (Scalise 1992:196)
7. In older Japanese, letters of words and sentences were written from top to bottom, and
even in contemporary Japanese, letters are still printed from top to bottom in newspapers
and novels.
8. An anonymous reviewer has pointed out to me that the Righthand Head Rule is not a
rule but a parametric value (are compound heads on the right or the left?).
References
Akasaka, Yukiko and Koichi Tateishi
this volume "Heaviness in Interfaces," this volume.
Allen, Margaret
1978 Morphological Investigations, Doctoral Dissertation, University of Connecti-
cut.
Bauer, Laurie
1983 English Word-Formation, Cambridge University Press, Cambridge.
Cinque, Guglielmo
1993 "A Null Theory of Phrase and Compound Stress," Linguistic Inquiry, 24,239-
297.
Di Sciullo, Anna-Maria and Edwin Williams
1987 On the Definition of Word, The MIT Press, Cambridge, Massachusetts.
Fabb, Nigel
1984 Syntactic Affixation, Doctoral Dissertation, MIT.
296 Takayasu Namiki
Fujihara, Chiaki
1995 "Eigo ni okeru 'Hyphenated Phrasal Word' ni tsuite [On 'Hyphenated Phras-
al Word' in English]," Echuudo, 25,169-202.
Fukazawa, Haruka and Mafuyu Kitahara
this volume "Domain-relative Faithfulness and the OCP: Rendaku Revisited," this vol-
ume.
Kageyama, Taro
1982 "Word Formation in Japanese," Lingua, 57, 215-258.
1993 Bunpou to Gokeisei [Grammar and Word Formation], Hituji Shobou, Tokyo.
this volume "Word Plus: The Intersection of Words and Phrases," this volume.
Kubozono, Haruo
1995 Gokeisei to On'inkouzou [Word Formation and Phonological Structures],
Kurosio Shuppan, Tokyo.
1999 Nihongo no Onsei [Speech Sounds of Japanese], Iwanami Shoten, Tokyo.
Lieber, Rochelle
1980 On the Organization of the Lexicon, Doctoral Dissertation, MIT.
1992 Deconstructing Morphology, The University of Chicago Press, Chicago.
Morita, Junya
1985 "X-bar Morphology and Synthetic Compounds in English," English Linguis-
tics, 2,42-59.
Namiki, Takayasu
1982 "The Notion 'Head of a Word' and Core and Periphery Word Formation: In-
teractions between Affixation and Subcategorization," Studies in English Lin-
guistics, 10, 21-41.
1985 Gokeisei [Word Formation (in English)], Taishukan Publishing Company, To-
kyo.
1991 "Fukugou-Keiyoushi to Kakudai-Junjozuke no Kasetsu [Compound Adjec-
tives and the Extended Ordering Hypothesis]," Gendai Eigogaku no Shosou
[Aspects of Contemporary English Linguistics], ed. by Shuji Chiba et al.,
Kaitakusha Publishing Co. Ltd., Tokyo.
1992 "Eigo no Zougoryoku [The Productivity of English Word Formation ]," Ni-
hongogaku [Japanese Linguistics], 11, 5, 66-74.
1993 Fukugou-Keiyoushi no Keitai-Tougoteki Kenkyuu - Fukugougo no Fuhenteki
Tokuchou o Motomete - [Morpho-Syntactic Studies of Compound Adjectives
in English: Toward Universalities of Compounding], a Report of a Grant-in-
aid for Scientific Research (C) (Grant No. 03610235), Ibaraki University.
1994 "Heads and Subheads of Compounds," Synchronic and Diachronic Ap-
proaches to Language: A Festschrift for Toshio Nakao on the Occasion of His
Sixtieth Birthday, ed. by Shuji Chiba et al., Liber Press, Tokyo.
Nishihara, Tetsuo, Jeroen van de Weijer, and Kensuke Nanjo
this volume "Headedness in Morphology: Tendencies in the Truncation of Loanword
Compounds in Japanese," this volume.
Otsu, Yukio
1980 "Some Aspects of Rendaku in Japanese and Related Problems," MIT Work-
ing Papers in Linguistics, 2, 207-227.
Scalise, Sergio
1988 "The Notion of 'Head' in Morphology," Yearbook of Morphology, ed. by
Geert Booij and Jaap van Marie, 229-245, Foris, Dordrecht.
1992 "Compounding in Italian," Rivista di Linguistica, 4,1,175-199.
Selkirk, Elisabeth
1982 The Syntax of Words, MIT Press, Cambridge, Massachusetts.
Shimamura, Reiko
1986 "Lexicalizatio of Syntactic Phrases," English Linguistics, 3,20-37.
Spencer, Andrew
1991 Morphological Theory, Blackwell, Oxford.
Subandi
1998 "Nichi-I-go no Fukugou-meishi no Kouzou ni tsuite [On the Structures of Com-
pound Nouns in Japanese and Indonesian]," Kokusai Shinpojium Hikaku Goi
Kenkyuu [Proceedings of an International Symposium on Comparative Studies
of Vocabulary], ed. by Ikudou Tajima, 66-74, Nagoya University.
Trommelen, Mieke and Wim Zonneveld
1986 "Dutch Morphology: Evidence for the Righthand Head Rule," Linguistic In-
quiry, 17,147-169.
Williams, Edwin
1981 "On the Notions 'Lexically Related' and 'Head of a Word'," Linguistic Inqui-
ry, 12, 245-274.
Zwicky, Arnold
1985 "Heads," Journal of Linguistics, 21,1 -29.
1993 "Heads, Bases and Functors," Heads in Grammatical Theory, ed. by Greville
G. Corbett et al., 292-315, Cambridge University Press, Cambridge.
Against Headedness in Compound Truncation:
English Compounds in Japanese
Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo
Miyagi University of Education, Leiden University, and
St. Andrew's University
1. Introduction
In this paper, we will examine the truncation (or 'clipping', 'contraction')

of loanword compounds in Japanese. First, we will explore the issue of
headedness in morphology in general, and pay particular attention to the
side at which heads are located in complex morphological structures. We
will examine the well-known Right-hand Head Rule (Williams 1981), and
discuss exceptions to this generalization. The issue of headedness is rele-
vant because it is a natural assumption that in truncation processes the
'head' of the original structure is retained. However, we will see that the
truncation of English compounds in Japanese is not governed by left-
headedness at all, but that rather a prosodie account is called for.
The Right-hand Head Rule (RHR) stipulates that heads are located at
the right-hand side of complex morphological structures in English. How-
ever, it has turned out that the RHR by no means represents a linguistic
universal (see Lieber 1980, Scalise 1988, Becker 1988, etc. for discussion).
The compound morphology of Romance languages, for instance, shows
that particular structures can also be left-headed. Vogel (1990) argues that
loanword truncation patterns in Italian, whose native compound morphol-
ogy is also left-headed, corresponds to Italian truncation patterns, in which
the left-hand element is retained (e.g. English night club ->• Italian night).
In general then, as Vogel suggests, loanword truncation patterns seem to
depend on headedness in the native morphology.
In this article, we test the validity of Vogel's hypothesis with respect to
Japanese (see also Namiki, this volume). This language presents some
problems for Vogel's idea (as also pointed out by Kubozono 1993). Al-
though Japanese is a right-headed language (i.e. well-behaved according
to the RHR), there are a number of loanword compounds in which the
left-hand element is retained under truncation, e.g. shorudaa bakku ->·
shorudaa ('shoulder bag'). In this example, the left-hand member is re-
300 Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo
tained, unlike what might be expected on the basis of the RHR. The ques-
tion arises how truncation is governed, if not (or only partly) by the RHR.
Another pattern of truncation of loanword compounds in Japanese,
which raises the same question, is also observed: in this pattern, both
halves of loanword compounds in Japanese are truncated, as in waado
purosessaa waa puro ('word processor'). In fact, this pattern of trunca-
tion is the most common in Japanese (see Itô 1990, Kubozono 1993), al-
though it is not common in other languages, such as English or Dutch. 1
Facts like these again raise the question of what constraints there are on
truncation, and what the role is that headedness plays.
In our analysis of the different truncation patterns, we suggest that, in
the framework of Optimality Theory, the pattern of truncation of loan-
word compounds in Japanese is governed by a language-particular hier-
archy of universal constraints on truncation. Constraints that play a cru-
cial role rally for the retention of the leftmost part of a compound,
retention of material from all parts of the input, and the formation of a
proper prosodie word.
2. Preliminaries
Japanese has a number of loanwords from English and other European

languages. Here we will mainly deal with loanword compounds from En-
glish. There are several ways in which compounds are truncated in Japa-
nese: (a) double truncation, in which both halves of the loanword com-
pound are partly retained, (b) back truncation, in which the first element
of the loanword compound is retained, and (c) front truncation, in which
the second element of the loanword compound is retained. Representa-
tive examples are given below:2
(1) Truncation patterns in Japanese

a. double truncation (Itô 1990: 220)
waado purosessaa waa-puro 'word processor'
hebii metaru hebi-meta 'heavy metal'
rajio kasetto rekoodaa • raji-kase 'radio cassette recorder'
sukeeto boodo •suke-boo 'skateboard'
paasonaru koNpyuutaa • paso-koN 'personal computer'
paNtii sutokkiNgu paN-suto 'panty stockings'
b. back truncation (Shibatani 1990: 254ft, Kubozono 1993: 187)
suupaa maaketto suupaa 'supermarket'
Against Headedness in Compound Truncation: English Compounds in Japanese 301
mini sukaato mini 'mini skirt'

sunakku baa -*• sunakku 'snack bar'
sukotchi uisukii ->• sukotchi 'Scotch whisky'
c. front truncation (Shibatani 1990: 254ff., Kubozono 1993:187)
purattohoomu ->• hoomu 'platform'
gooru kiipaa -> kiipaa 'goal keeper'
sooiNgu mishiN ->· mishiN 'sewing machine'
The truncation of loanword compounds in Japanese is part of the mor-

phology of the language.3 In general, Japanese tends to shorten lengthy
words and compounds. It is generally accepted that double truncation of
loanword compounds is most common in Japanese (Kubozono 1993; see
also below). The reason why double truncation is most productive in Jap-
anese might be explained as a result of analogy with truncation patterns
in Sino-Japanese (Ishiwata 1993: 99). Sino-Japanese compounds com-
posed of four Chinese characters are frequently truncated by retaining
the first morpheme of each member of the compound and deleting the
rest. This is illustrated in (2):
(2) Truncation of Sino-Japanese compounds

Word -> Word
/ \ / \
Word Word Word Word
/ \ / \
stem stem stem stem stem stem
gai + koku shi + hoN gai shi
gaikoku shihoN gai shi 'foreign capital'
The example above represents a very productive process, and other ex-
amples of this type are presented in (3).
(3) Truncation of Sino-Japanese compounds

(Itô 1990: 229, Yonekawa 1989:122)
ka + tei + sai + ban + sho -> ka + sai 'family court'
koku + yuu + tetsu + d o o -* koku + tetsu 'state railway'
g e N + shi + baku + daN ->· g e N + baku 'atomic bomb'
daN + tai + k o o + shoo -> daN + k o o 'collective bargaining'
It should be noted that these patterns are observed in Japanese, but not
in Chinese. In other words, these Sino-Japanese compounds were truncat-
ed after they had been borrowed into Japanese. Umegaki (1963) suggests
that the primary unit of word formation in Japanese is based on a two-syl-
lable prosodie unit; secondary units are four-syllable units (2 + 2) and
three-syllable units (2 + 1 or 1 + 2). The term 'syllable' that Umegaki
(1963) uses corresponds to the concept of 'mora', which is nowadays
more commonly used in Japanese phonology. From a phonological view-
point, it is obvious that the output of truncation of both loanword com-
pounds and Sino-Japanese compounds in Japanese is basically composed
of four moras (2 + 2): all examples so far conform to this pattern (where
of course the mora nasal counts as a single mora). Moreover, Itô (1990)
and Kubozono (1995) suggest that the basic unit of prosodie structure in
Japanese is a two-mora structure, and truncations of loanword com-
pounds are based on a four-mora pattern (2 + 2), which we might regard
as a foot. We will return to this insight below, and try to give a more pre-
cise account of these patterns then. First, let us turn to a brief discussion
of headedness.
3. Head of a word
In this section, we will examine the notion of headedness in detail, be-

cause Vogel (1990) suggests that the process of compound truncation in
Italian can be adequately accounted for by the role which headedness
plays in the morphology of Italian in general.
The notion of head of a phrase in syntax has played a crucial role in
generative grammar, and the notion of head is also used in morphology.
For example, since most English compounds are endocentric, they have
heads. Williams (1981) argues that in such compounds the right-hand
member is the head because it determines the category of the complex
word. Williams' generalization—the so-called Right-hand Head Rule, is
presented in (4):
(4) Right-hand Head Rule (Williams 1981:248)

In morphology, we define the head of a morphologically complex
word to be the righthand member of that word. (...) Call this defi-
nition the Righthand Head Rule.
The lexical class membership of the English compounds in (5) is correctly

explained by the RHR.
(5) English compounds (Katamba 1993: 311)

[ bird N watch N ] N [ over P react v ] v [ sugar N daddy N ] N
[ blue A book N ] N [ blue A black A ] A [ wind N screen N ] N
William's RHR was based on the morphology of English. In fact, Will-

iams (1981) does not state explicitly whether the RHR is intended to be
a universal statement or not. Some researchers have argued that the
RHR can also be applied to other languages, for instance, Japanese
(Kageyama 1982), Dutch (Trommelen and Zonneveld 1984,1986), Swa-
hili (Schultink 1988), Greek (Ralli 1992), and so on. Some examples from
some of these languages are given in (6):
(6) a. Japanese compounds (Kageyama 1982: 221)

[ naga A banasi N] N 'long talk'
[ tati A eri N ] N 'standing collar'
[ denki N sutoobu N] N 'electric heater'
[ huru A hoN N ] N 'second-hand book'
b. Dutch compounds (Trommelen and Zonneveld 1984:181,1986:149,157)
[ kans N arm A ] A 'having few chances'
[kalmAeerv]v 'to calm down'
[piepvjongA]A 'very young'
[diepAzeeN]N 'deep sea'
Other researchers, however, argue against the universality of the RHR.

Lieber (1980) points out that Vietnamese has left-headed compounds, as
illustrated in (7). In these cases, the left-hand member determines the ma-
jor class membership of the compound as a whole:4
(7) Vietnamese compounds (Lieber 1980: 99)

a. nguoi o
person be-located 'servant'
b. nha thuong
establishment be-wounded 'hospital'
c. lam viec
do-make matter-affair 'to work'
d. lam ruong
do-make rice-field 'to engage in farming'
Based on the cases which we have seen so far, languages generally seem
to choose either the left-hand member or the right-hand member as the
determinant of major class membership of the compound as a whole (see

also Selkirk 1982).5·6
Scalise (1988,1992) suggests that Italian has both left-headed and right-
headed compounds. The examples in (8) are from Scalise (1988).
(8) Italian compounds (Scalise 1988: 243)

a. Latinate compounds
terre MOTO 'earthquake'
sangui SUGA 'bloodsucker' (= leech)
b. modern Italian compounds
NAVE traghetto 'ferry-boat'
DIVANO leeto 'divan-bed'
Right-headed compounds come from Latin, which is an SOV language,

while left-headed compounds arose in modern Italian, which is an SVO
language. Scalise thus relates the position of the head to the basic word
order pattern in these languages. Scalise (1988) points out that a similar
situation obtains in Somali. Thus, in languages such as Italian and Somali
the head can be located at the left-hand or right-hand side, whereas in lan-
guages such as English or Dutch the side at which the head is located is
fixed. This can also be related to the basic word order within NPs, as in [N
A (modifier)] in the Romance languages, and [(modifier) A N] in the
Germanic languages, which have right-headed compounds. The corre-
spondence between syntactic order and compound headedness is summa-
rized in the following paradigm (see also Greenberg 1963):
(9) Syntactic word order and morphological headedness (Scalise 1992:182)

Latin Italian
syntactic order SOV SVO
order in compounds OV VO
position of the head right left
inflection right left
So far, we have investigated the notion of headedness from the viewpoint

of the determination of the lexical category of the entire compound. How-
ever, there are also other kinds of arguments on which to base headedness
in compounds, or, alternatively, the individual members of a compound
may be regarded as 'heads' in other senses than the lexico-morphological
one that we have discussed so far. Thus, Namiki (1994) gives three differ-
ent kinds of arguments to base headedness on (see also Zwicky 1985):
(10) Definition of 'head' (Namiki 1994: 270)

a. the semantic argument
b. the morphological determinant
c. the morphosyntactic locus
As to (10a), Zwicky (1985) suggests the following:
We could take the head / modifier distinction to be at root semantic:

in a combination X + Y, Y is the 'semantic head' if, speaking very
crudely, X + Y describes a kind of the thing described by X.
(Zwicky 1985: 4)
This semantic consideration clearly applies to compounds like ham sand-

wich: the head is sandwich, because a ham sandwich is a kind of sandwich,
not a kind of ham?
(10b) refers to the fact that the lexical category of the whole word is de-
termined by that of the head, i.e. this is the criterion we employed above.
It also applies to suffixation: for example, in the noun happiness, the cat-
egory is determined by the suffix -ness, which has the feature [+N], This
case is explained by the RHR: the element on the right-hand side deter-
mines the category of the whole structure. However, there are also cases
in English in which the left-hand element determines the category of the
whole word, e.g. the feature [+V] of the prefix en- determines the lexical
category of the whole word, as in [en[+V]-i-richA]v.
Finally, (10c) looks at the position where inflectional morphemes are
attached within a complex form, and argues that the head will bear the
phonological marks of inflection. For example, the head in ham sandwich
is sandwich by the definition in (10c), because the correct plural form is
ham sandwich+es, not *ham+s sandwich.
Hence, there is some discussion on how heads are determined in com-
plex syntactic and morphological structures. For this reason, some re-
searchers have doubted the fact whether the concept plays a role at all (e.g.
Becker 1990) or whether the notion can be operationalized in a meaning-
ful way in morphology (e.g. Zwicky 1992). We will demonstrate that head-
edness does not play a crucial role in compound truncation processes in
Japanese, although its role in other parts of morphology (such as the deter-
mination of lexical class membership, cf. (6) above) stands unchallenged. 8
We now turn to Vogel (1990), who argues that headedness plays a crucial
role in morphology in general and in the truncation of loanword com-
pounds in particular. Let us consider this suggestion in the next section.
4. Truncation of loanwords: Vogel (1990)
Vogel (1990) discusses universal constraints on the truncation of loan-

word compounds on the basis of data from truncations of English com-
pounds in Italian. These show that the second element of English com-
pounds in Italian is truncated, so that the form used in Italian consists of
only the first element of the English compounds. This is shown in (11).
(11) Truncation of English compounds in Italian (Vogel 1990: 99)9

a. night club -> night
b. scotch tape ->• Scotch
c. water closet ->• water
d. plaid blanket ->• plaid
To explain this pattern of truncation, Vogel (1990) proposes the following

three hypotheses:
(12) Three hypotheses to account for truncation (Vogel 1990:103)

HI: Truncation is due to some basic linguistic strategy.
H2: Truncation is due to some property of English compounds.
H3: Truncation is due to some property of Italian compounds.
Let us briefly examine these three hypotheses. According to Vogel

(1990), H I refers to the fact that there is something salient about the first
member of borrowed compounds, regardless of the morphological char-
acteristics of either the source language or the borrowing language. In
such an approach, truncation consists of a strategy of deleting everything
but the first word. However, Vogel suggests that this runs counter to the
observation that right-hand members play an important role and are most
salient in morphology (pace the RHR) and syntax. Furthermore, Vogel
notes that the endings of words tend to be more salient in child language
acquisition than the beginnings of words, according to Operating Princi-
ple A for language acquisition, suggested by Slobin (1973): "Pay attention
to the ends of words". On the basis of these considerations, Vogel deems
H I to be inadequate. However, in defence of HI we note that Hawkins &
Cutler (1988) is but one of many studies that point out the psycholinguis-
tic importance of word onsets vis-à-vis other parts of the words. We return
to this observation below.
The second hypothesis refers to the fact that the Italian borrowing pat-
tern is governed by the structure of the English compounds. Vogel (1990)
also discards this hypothesis because monolingual Italian speakers, who

are of course unaware of English morphological patterns, apply the cor-
rect truncation pattern of English compounds in Italian, i.e. as in (11).
Moreover, the English stress pattern does not seem to play a role either,
as the forms in (13) show; while in the forms in (13a) main stress is on the
left-hand side of the compound, and in the forms in (13b) main stress is
on the right-hand side, in both cases the left-hand member is retained af-
ter truncation (see (11) above):
(13) English compounds, with stress marks (Vogel 1990:104)

a. 'night club b. scotch 'tape
'water closet plaid 'blanket
For this reason, Vogel (1990) argues that neither the morphology nor the
phonology of the English compounds plays a decisive role. She therefore
concludes that the second hypothesis must also be rejected.
Finally, we examine the third hypothesis, H3. According to Vogel, Ital-
ian mainly has left-headed compounds: this is the productive pattern (cf.
(8) above). Vogel concludes that the left element of English compounds
in Italian is retained because of the structure of Italian and thus that H3
is adequate. This predicts that it makes no difference from which lan-
guage compounds are borrowed: regardless of the source language, the
left-hand element will be retained. In fact, Vogel shows that compounds
of German origin have the same pattern of truncation as the English com-
pounds, as illustrated in (14):
(14) Truncation of German compounds in Italian (Vogel 1990:107)

Volkswagen -> volks
Blitzkrieg blitz
With this background, we will now turn to an account of truncation of En-

glish compounds in Japanese.
5. Truncation patterns in loanwords in Japanese
If Vogel's Hypothesis 3 in (12) above is correct, and also holds for lan-
guages other than Italian, the right-hand element of English compounds
in Japanese is expected to be retained, leading to front truncation, be-
cause Japanese has right-headed compounds (see (6b) above; Kageyama
1982). Against this prediction, however, front truncation (exemplified in

(lc)), which retains the second element, is rare in Japanese, while there
are many examples of back truncation (lb) (see also Kubozono 1993). In-
deed, the most common and productive pattern, surprisingly perhaps
from the discussion of truncation so far, is double truncation, which re-
tains both first halves of English loanword compounds (la). Here, the no-
tion of morphological headedness does not seem relevant at all, and an
analysis in different terms seems to be called for. According to the data
we collected from a dictionary of truncations (Motwani 1993), the relative
frequency of occurrence (in terms of the number of input compounds that
successfully can undergo truncation) of double truncation, back trunca-
tion and front truncation in Japanese, respectively, is as follows (see also
(16) below):
(15) Relative frequency of different truncation patterns in Japanese

double truncation > back truncation » front truncation
Some more examples are given in (16) (cf. also (1) above).
(16) a. double truncation (47%)

maikuro koNpyuutaa mai-koN 'micro computer'
sekusharu harasumeNto seku-hara 'sexual harassment'
purofeshonaru resuriNgu puro-resu 'professional wrestling'
pawaa suteariNgu pawa-sute 'power steering'
afutaa rekoodiNgu afu-reko 'post-recording'
b. back truncation (39%)
paato taimu paato 'part-time job'
homo sekusharu homo 'homosexual'
bideo dekki bideo 'video deck'
tekisuto bukku tekisuto 'textbook'
shorudaa bakku shorudaa 'shoulder bag'
c. front truncation (14%)
nyuusu kyasutaa kyasutaa 'news caster'
mootaa baiku baiku 'motor bike'
fashon moderu moderu 'fashion model'
kafe oore oore 'café au lait'
Kubozono (1993) suggests that back truncation, which is the second most
common pattern, can be explained as a result of HI suggested by Vogel
(1990) (see (12) above). Kubozono (1993) gives the following examples:
(17) Back truncation in Japanese (Kubozono 1993:187)

a. Scotch whisky Scotch
b. mini-skirt mini
c. snack bar -»• snack
d. volleyball ->• volley
However, this approach does not deal with double truncation, and hence
does not achieve an integrated analysis of truncation. We will attempt this
in the next section.
6. Analysis
To analyze the Japanese facts, we assume the basic principles of OT

(Prince & Smolensky 1993 et seq.). In OT, Universal Grammar is com-
posed of a set of universal, though violable constraints.10 It is assumed
that each language has a language-particular constraint hierarchy. A con-
straint may be violated only in order to satisfy a higher ranked constraint
in the grammar. The role of the grammar is to select the output form from
a set of candidates. An analysis of truncation in Japanese therefore
amounts to finding the relevant constraints and their relative ranking. We
propose that an interaction of phonological and semantic constraints
plays a role in determining the attested output of every loanword com-
pound. Thus, it is not necessary to stipulate for each and every compound
what the relating truncation is. Certain truncations, however, may have
been lexicalized completely, so that these have to be stated explicitly in
the Japanese lexicon.
In this paper, we will follow Benua (1995: 116-25), who analysed Japa-
nese hypocoristic patterns as so-called templatic truncation, in which "the
truncated output is a consistent prosodie unit, usually a minimal word (a
foot)" (Benua 1995: note 33). While she deals with a different area of trun-
cation in Japanese, which we will not deal with here, we share the basic ideas
regarding the relevant constraints and their ranking with her. As it turns
out, there are two major differences between the patterns of hypocoristics
and those of loanword compound truncation: (1 ) long vowels are permitted
at the end of derived prosodie words in the former, but not in the latter, (2)
no more than one mora of the base can be retained and then lengthened to
conform to the bimoraic constraint in hypocoristics, but lengthening of the
vowel is not allowed in truncations. We will assume that the first fact is ac-
counted for by a constraint preventing long vowels from occurring at the
end of prosodie words, which belongs exclusively to the "foreign" stratum

(Itô & Mester 1995) and that the second fact results from the dominance of
the constraint IDENT-BT[V-Iength] over the bimoraic requirement in the
"foreign" stratum, while, as assumed by Benua (1995:119), the bimoraic re-
quirement is ranked higher than iDENT-BT[V-length] in the hypocoristic
phonology, which allows lengthening of the base vowels.
First, let us examine some of the phonological constraints that play a
role. We tentatively exclude front truncations from consideration, such as
mootaa baiku baiku 'motor bike'. These truncations make up only a
small percentage of the total number of truncated forms, and therefore
seem less central to a prosodie account (see also Kubozono 1993).11 In-
stead, some of these front truncations may have resulted from a 'special-
ization of meaning' phenomenon. For example, kaado 'card' is the
clipped form of kurejitto kaado 'credit card', which was originally used
only in particular contexts. Although what kaado means varies from con-
text to context, in the appropriate situation it is immediately taken for a
credit card. The same holds in the case of puuru 'pool', which could come
from either suimiNgu puuru 'swimming pool' or mootaa puuru 'car park',
but is usually used only in the former meaning.
With respect to the back and double truncations, it is important to observe
that the output of the truncation patterns of Japanese almost always take on
the following structure, consisting of two bimoraic prosodie words, which
form, recursively, another prosodie word, as argued for by McCarthy &
Prince (1993a: 84-85,1993b: 146), Booij (1995:144), Selkirk (1996:190), etc.:
(18) Structure of truncation outputs

[ [PrWd J [PrWd 2 ]] PrWd
PrWd = prosodie word
where both PrWd, and PrWd 2 consist of two moras
For the sake of convenience, we tentatively refer to the word-level PrWd

(i.e. PrWd 1 and PrWd 2 in (18)) as minor PrWd and compound-level PrWd
as major PrWd whenever a distinction needs to be made. Of course, fol-
lowing standard analysis, the nasal counts as one mora if it occurs in the
coda of the syllable, (e.g. maikuro koNpyuutaa mai-koN 'micro com-
puter'), and a long vowel counts for two koras (e.g. kaa sutereo kaa-sute
'car stereo').
This templatic requirement must be regarded as a conglomerate of two
more basic constraints, the first demanding that each prosodie word con-
sists of two moras, and the second demanding that even a small fraction
of very prosodie word which composes the compound must be retained in

truncation.
(19) a. MINWD : truncated outputs are minimal words, i.e. bimoraic

b. PARSE-PRWD : even a small fraction of every prosodie word which
composes the compound must be parsed into the truncated output 12
As McCarthy and Prince (1994, 1995a: 322) claim, the concept of mini-
mal word no longer has an actual status as a primitive template in any
language in the OT era; instead its effect is derived from the interaction
of such metrical constraints as F T B I N , PARSE-SYLL, A L I G N - F T - L (see Be-
nua 1995: 118-19). In our paper, however, we will for the sake of conve-
nience continue to refer to the minimal word, which is equivalent to a bi-
moraic foot in such quantity-sensitive languages as Japanese. The two
constraints together demand that outputs have a structure conforming to
that in (18).
In what we think is a relatively recent development, truncations of
three moras are becoming increasingly popular, e.g. Misu-Do from Mis-
utaa Doonatsu 'Mister Doughnut [a doughnut franchise]', tere-ka from
terefoN kaado 'phone card'. Thus, the long vowels in the second member
of the original compound are shortened, which shows that forms which
have a monomoraic syllable structure with a short vowel are better
formed than bimoraic structures with a long vowel in the second PrWd.
Below, we will propose a constraint ranking accounting for this.13 Howev-
er, three-mora outputs are also found even when there is no long vowel
in the input, such as in rabu-ho from rabu hoteru 'lit. love hotel', Roi-Ho
from Roiyaru Hosuto 'Royal Host [the name of a restaurant]', and Bura-
Bi from Burakku Bisuketto 'Black Biscuit [the name of a singer]', where
there is no phonological reason to avoid *rabu-hote, *Roi-Hosu, and *Bu-
ra-Bisu. Another template, consisting of three moras, is probably more
appropriate for these forms, which we have to ignore here (cf. Itô (1990:
237, note 29) for a similar view).
An additional constraint that plays a role demands that the leftmost
two moras are preserved in the output, and not the rightmost two moras,
or just any two moras, in double truncation. Furthermore, even when
doubly truncated forms are not preferred, the retained element usually
turns out to be the leftmost (minor) PrWd of the major (i.e. compound-
level) PrWd (e.g. suupaa from suupaa maaketto). This "leftmostness" that
double truncation and back truncation share should be generalized by a
uniform constraint. To capture this uniformity, we will propose (20),
which seems to have an effect similar to Benua's ( 1 9 9 5 : 1 2 3 ) ANCHOR-

L(FT) constraint.
(20) LEFTMOST: The leftmost element of the constituent is retained in

truncation.
The retained leftmost element in (20) varies depending on the constitu-

ent. If the relevant prosodie word is a (minor) PrWd, it is the leftmost
mora that is retained. If it is a major PrWd (i.e. a compound), it is the left-
most (minor) PrWd that is retained. Note that this constraint per se does
not require that the leftmost two moras be retained in double truncation;
it is independently handled by another constraint, M I N W D . Since M I N W D
is ranked lower than LEFTMOST, monomoraic prosodie words can occur
(violating M I N W D ) if they satisfy LEFTMOST and other higher-ranked con-
straints to be discussed below.
To see how the three constraints in (19) and (20) that are relevant so far
work, consider the truncation pawa-sute from pawaa suteariNgu 'power
steering'. The moraic structure of the input is given in (21):
(21) Moraic structure of pawaa suteariNgu 'power steering'

σ α ο oo σ σ
μ μμ μ μμμμ μ
pawaa suteariNgu
The constraints are ranked as in (22)
(22) LEFTMOST > M I N W D , P A R S E - P R W D
The optimal output reflects the first two moras of the first input word and
the first two moras of the second input word. On the other hand, a form
in which the last two moras of the first word are taken, and—for in-
stance—the first two moras of the second word (resulting in *waa-sute)
would also result in a well-formed moraic structure (i.e. pass both M I N W D
and P A R S E - P R W D ) but would not pass higher-ranked LEFTMOST. The form
*pawa-su is ill-formed, because the second output word su fails to satisfy
the M I N W D requirement. Finally, the form *pawa satisfy the constraints
LEFTMOST (because it is the leftmost element (a minor PrWd) of the major

PrWd) and M I N W D but fails to pass the constraint P A R S E - P R W D . These
considerations are expressed in the following sample tableau:
(23) Tableau for pawaa suteariNgu
/pawaa suteariNgu/ LEFTMOST MINWD PARSE-PRWD
pawa-sute
waa-sute *!
pawa-su »!
pawa * !•
As we assume in ( 2 2 ) , LEFTMOST must be ranked higher than M I N W D and

P A R S E - P R W D , because, as we saw above, outputs violating M I N W D and
P A R S E - P R W D are sometimes attested, but exceptions to LEFTMOST are
hardly attested. (There are a few exceptions such as BuN-Taa from SebuN
Sutaa 'Seven Star (a cigarette brand)' and bifu-teki from biifu suteeki
'beef steak': these two examples are exceptional with respect to other
constraints as well—note that the latter may have been a direct borrowing
from French.).
Furthermore, we could explain a number of examples as the result of
the fact that some of the Japanese truncations may have reflected the
original English truncated forms. For instance, the back-truncated
shukotchi 'Scotch' may be regarded as exceptional if it was formed on the
basis of the full form shukotchi uisukii 'Scotch whisky' (where *shuko-ui
would be expected), but is perfectly well-formed if we assume that the
English truncated form Scotch was borrowed into Japanese along with the
full form Scotch whisky.
The three phonological constraints discussed so far together account
for almost all forms in (1) and (16) above. We will now turn to some coun-
terexamples which force us to refine our analysis. First, note that there are
a number of regular exceptions to the general pattern, all of which involve
geminate consonants:
(24) Japanese: second element has a geminate

bideo dekki -> bideo 'video deck'
tekisuto bukku -> tekisuto 'textbook'
shorudaa bakku ->• shorudaa 'shoulder bag'
sutereo dekki ->· dekki 'stereo deck'
Buraddo Pitto Bura-Pi 'Brad Pitt'
poteto chippusu pote-chi 'potato chips'
To account for this subregularity, which was noted independently by

Kubozono & Ota (1998) and Kuwamoto (1998), we assume, uncontrover-
sially, that forms like bukku have the moraic structure in (25):
(25) Moraic structure of bukku 'book'

α σ
μμ μ
bukku
If the truncation pattern for tekisuto bukku were, as predicted, teki-buk,

this would violate another constraint of Japanese, CODA-COND (Itô 1986,
1989, Itô and Mester 1994, 1999), which forbids all final consonants ex-
cept "placeless nasals (written [N]), geminates and nasals homorganic to
a following stop" (Benua 1995: note 35). This overrides all the constraints
relevant to truncation. One way to repair the ill-formed buk would be
skip the mora on the [k] and include the final vowel [u], resulting in *teki-
buku. However, the skipping of segmental or prosodie material is also
heavily penalized by the "no-skipping" requirement (McCarthy & Prince
1986), which is expressed in the current framework by the constraint
14
CONTIGUITY (McCarthy and Prince 1993a, 1995b). These two constraints
are defined as follows:
(26) a. CODA-COND : *Place]

b. CONTIGUITY: "segmental material that is contiguous in the input
must also be contiguous in the output."
Hence, CODA-COND and CONTIGUITY and, as we argue above, LEFTMOST

are higher-ranked than M I N W D and P A R S E - P R W D , as in ( 2 7 ) :
(27) CODA-COND, CONTIGUITY, LEFTMOST > M I N W D , PARSE-PR W D
In all cases where the regular 2 + 2 mora structure cannot be formed

due to the interplay of the constraints mentioned so far, another strate-
gy is selected: in the first three forms in (24) the first word of the com-
pound (consisting of three or four moras) is selected, in the fourth it is
the last part, and in the final two examples parts of both words in the
original compound are used. We assume these different choices are lex-
ically determined, though all are driven by the fact that no regular, pho-
nological truncation is possible. Thus, importantly, the fact that the trun-
cation of tekisuto bukku is tekisuto is, in our opinion, not a result of the
fact that this compound is somehow marked as exceptional, but rather
of the fact that the normal pattern of truncation fails for phonological
reasons.
As was already mentioned above, long vowels are normally shortened
at the end of the second prosodie word in truncation outputs. We can put
this down to a phonological constraint against final long vowels, which
operates in structures like (18) above. The constraint could be defined as
follows:
(28) NOFINALLONGVOWELS ( N F L V ) : * V V ] MAJOR PRWD
Examples include furi-ma from furii maaketto 'flea market', where Sfuri-
mela would be incorrect, depa-ga from depaato gaaru 'lit. department
store) girl' (*depa-gaa), hai-ka from haiwei kaado 'lit. highway card'
(*hai-kaa), as well as Misu-Do from Misutaa Doonatsu (*Misu-Doo) and
tere-ka from terefoN kaado (*tere-kaa), both of which were mentioned
above. Unlike the CODA-COND, this constraint NFLV holds of loanword
truncated structures only, not of underived loanword compounds (e.g.
rabu retaa 'love letter', dansu paatii 'dance party'), Sino-Japanese com-
pounds (both underived and truncated) (e.g. dantai koosyoo 'collective
bargaining' and its truncated form dan-koo) nor hypocoristics (e.g. Maa-
chan, Taa-kun, Koo-chan). To account for the restricted nature of NFLV,
we first assume that this constraint belongs to the 'foreign' stratum of Jap-
anese lexicon (Itô and Mester 1995). This assumption appropriately ex-
cludes Sino-Japanese vocabulary and hypocoristics from the NFLV effect.
Next, we will ascribe the fact that NFLV applies only to truncated forms
to the classical concept of "derivedness". The application of this lexical-
phonological idea to the loanword phonology was, to our knowledge, first
proposed by Itô (1990: 218-19). According to her proposal, such a con-
straint is enforced on derived forms only, not on underived forms and,
since abbreviated loanwords are derived from their base forms, they sat-
isfy this criterion.
Of course, the concept of "derivedness" cannot be directly incorporat-
ed into our non-derivational analysis. One of the solutions, which we will
follow here, is proposed by Benua (1995) in the framework of Correspon-
dence Theory (McCarthy and Prince 1995b). Under Correspondence

Theory, she proposes that truncation involves two correspondence rela-
tions: an input-to-output relation between the input and the base (IO-
Faith) and an output-to-output relation between the base and the truncat-
ed form (BT-Identity). As Benua (1995: 82) mentions, "[t]he input is
mapped to the base by ΙΟ-correspondence, and BT-correspondence re-
lates the base to the truncated form," as illustrated in (29):
(29) Truncation (= Benua 1995: 82)

BT-Identity
Base < > Truncated Form
ΙΟ-Faith I
Input
Following her proposal, we will assume the following ranking among IO-
Faith (LDENT-IO[V-length]), BT-Identity (LDENT-BT[V-length]) and the
phonological constraint NFLV.
(30) LDENT-IO[V-length] » NFLV » LDENT-BT[V-length]
Since LDENT-IO[V-length], which demands that the length of correspon-

dent vowels in the input and the output is identical, dominates NFLV,
NFLV is not enforced in non-truncated (underived) bases, while truncat-
ed forms must satisfy NFLV because it dominates LDENT-BT[V-length],
which requires that the length of vowels in truncated form be identical to
that of the corresponding vowels in its base. This is what McCarthy and
Prince (1994) call "the emergence of the unmarked" ranking.
One of the striking differences between the patterns of hypocoristics and
those of loanword compound truncation is that, in the former, it is possible
for even one mora of the base to be retained and then lengthened to con-
form to the bimoraic constraint, while lengthening of the vowel is not al-
lowed in truncations. 15 This difference is adequately accounted for by as-
suming the following rankings between M I N W D and I D E N T - B T [ V - L E N G T H ] ,
(31) a. Hypocoristics: M I N W D > LDENT-BT[V-length]

b. Truncations: LDENT-BT[V-length] » M I N W D
As is illustrated in ( 3 1 ) , in the hypocoristic phonology (Benua 1 9 9 5 : 1 1 9 ) ,

the bimoraic requirement (our M I N W D ) is ranked higher than IDENT-
BT[V-length], which results in the lengthening of the base vowels, while
in truncation, vowel lengthening never occurs because the length of cor-
respondent vowels in the base and its truncated form must always be
identical even if the output fails to satisfy the bimoraic requirement.
In the following, we will concentrate on phonological constraints and
leave LDENT-IO[V-length] and LDENT-BT[V-length] out of consideration.
The six phonological constraints mentioned above are ranked as in (32):
(31) CODA-COND, CONTIGUITY, LEFTMOST, N F L V » MINWD, PARSE-PRWD
TO illustrate how these constraints and their ranking explain our data, we
will take waa-puro (from waado prosessaa), suupaa (from suupaa maa-
ketto), and pote-chi (from poteto chippusu) as examples.
(33) a. Tableau for waa-puro

/waado purosessaa/ CODACD CONTIG LEFTMOST NFLV MINWD PARSE-PRWD
•®· waa-puro
waa-pu *!
waado *!
b. Tableau for suupaa

/suupaa maaketto/ CODACD CONTIG LEFTMOST NFLV MINWD PARSE-PRWD
suu-maa *!
•S· suu-ma *
suupaa *
c. Tableau for pote-chi

/poteto chippusu/ CODACD C O N T I G I LEFTMOST NFLV MINWD PARSE-PRWD
pote-chip *!
ra pote-chi *
ra· poteto 1 *
In the case of suupaa maaketto, the four-mora doubly truncated candidate

*suu-maa triggers the violation of the higher-ranked constraint N F L V , but
the other candidates suupaa and *suu-ma satisfy this constraint (the candi-
date suupaa does not violate NFLV, because this constraint holds of only
Major PrWd; suupaa itself is a minor PrWd). Note that suupaa does not vi-
olate LEFTMOST, because it is the leftmost element (a minor PrWd) of the
major (compound-level) PrWd. Since the candidates suupaa and *suu-ma
violate the lower-ranked constraints P A R S E - P R W D and M I N W D , respective-
ly, they are on a par with each other with respect to their potential of being
realized as the actual output. The rest is outside of the OT grammar. 16
As expected, cases like poteto chippusu can be handled in the same way
as suupaa maaketto. The four-mora doubly truncated candidate *pote-chip
triggers the violation of the highly-ranked constraint CODA-COND, but the
other candidates *poteto and pote-chi pass this constraint, though they vi-
olate the lower-ranked constraints P A R S E - P R W D and M I N W D , respectively.
Semantic reasons undoubtedly also play a role in truncation. It is often
the case that the leftmost element in compounds indicates a specific prop-
erty, while the rightmost element indicates a classlike, general character-
istic. We believe this constraint is related to the tendency that Vogel
(1990) observed when she formalized Hypothesis 1 above (12). This is ob-
vious in the following examples from English, in which the rightmost ele-
ment is often lost nowadays or is optional:
(34) English: rightmost element lost

barber's (shop), dry-cleaner's (shop), five-and-dime (store), phys-
ical (exam), midterm (exam), Thanksgiving (Day)
These compounds consist of [modifier] + [head] where the head is a noun

which is quite common, like shop, store, day, etc. Thus, a constraint ex-
pressing the fact that the contrastive element in compounds is preserved
is high-ranked in English.17 In Japanese, the same tendency is also clearly
present, although in this language other constraints, such as the prosodie
constraints we saw above, also play a role. Consider a set of examples like
the following:
(35) Japanese: rightmost element lost

daburu (ruumu) 'double room'
shiNguru (ruumu) 'single room'
kurooku (ruumu) 'cloakroom'
ribiNgu (ruumu) 'living room'
basu (ruumu) 'bath room'
dainiNgu (ruumu) 'dining room'
tsuiN (ruumu) 'twin room'
In all these examples, the right-hand, general part may be truncated,

while the left-hand, specific part is invariably preserved. Note that in no
way does the output conform to the prosodie template we proposed
above. We assume that the semantic pressure to preserve the specific over
the general, especially in groups of related words such as those in (35), fa-
cilitates this tendency.
7. Conclusion
In this paper we have discussed the truncation of compounds in various

languages, focusing on Japanese, which shows a number of variable pat-
terns. We found that there were constraints of various types which gov-
erned truncation: phonological ones, such as the tendency toward a four-
mora pattern, the avoidance of final codas and final long vowels, and se-
mantic constraints, in which the contrastive force of the separate parts of
the compound played a role, with a marked tendency to preserve the spe-
cific, contrastive part of the whole. In no way did right-headedness, as de-
manded by the Right-hand Head Rule, play any role in the formation of
the truncated form. Rather, variation in the output was seen to be the re-
sult of phonological differences in the input.
Acknowledgements
We thank Laura Benua, Bert Botma, Ellen Broselow, Jan Kooij, Nancy Kula, Sang Jik Rhee,
Grazyna Rowicka, Norval Smith and Erik Jan van der Torre for valuable discussion of ear-
lier versions. Remaining errors are our own.
Notes
1. Szymanek (1989) suggests that back truncation is the most common in English. This is
also the case in French, German and Dutch. See Wiese (1996:62-63) for German, Ni-iku-
ra et al. (1996: 384) for French, and Hamans (1997:1734) for Dutch.
2. In some languages, double truncations are common. For example, see Shin (1997) for Ko-
rean, and Crowley (1997) for English in Papua New Guinea. In standard English, exam-
ples are FedEx (< Federal Express), sci-fi (< science fiction), Amtrak (< American Track)
and AMEX (< American Express).
3. Hock (1991: 196) suggests that shortening is a common tendency in linguistic change.
Lass (1987:211) also points out examples as early as from the 18th century. Scheler (1977)
suggests that the oldest examples are from the middle of 16th-century, although these are
but few.
4. Other problematic examples for the R H R concern Russian diminutives (Lieber 1992),
and French diminutives (Jaeggli 1980).
5. Van Beurden (1988) claims that in language acquisition children assume that the right-
hand element decides the whole category and thus consider the R H R as the unmarked
option, while later they have to decide the parameter about the position of head (left or
right) according to the data of their native language.
6. Huang (1998) argues that Chinese is a headless language in the morphology, for neither
the right-hand nor the left-hand element in a compound determines the lexical category
of the whole.
7. There are exceptions, of course, such as lexicalised compounds like pickpocket, which is
not a kind of pocket.
8. Kubozono (1990,1996) points out that the notion of phonological head also plays a role
in patterns of blending and nominal compounds.
9. In Italian native compound forms, back truncations such as (11) are more frequent than
front truncations. Double truncation is unknown in Italian (Piermarco Bertinetto, p.c.)
10. Haspelmath (1993) also argues for a constraint that tolerates exceptions in morphology,
antedating mainstream OT.
11. Still another reason for excluding front truncation is that, as Yonekawa (1996) suggests
in his book on the vocabulary of the young, front truncation is often adopted to create
secret languages. This suggests that they know this is the least common and thus least
understandable way of word-formation.
12. The constraint P A R S E - P R W D also accounts for such English acronyms as AIDS, PET,
WASP, etc. (see also Itô 1990) and shows that Japanese double truncation and English ac-
ronym formation are very similar phenomena. Since they are both common patterns in
both languages, this analysis shows that seemingly different languages adopt similar pat-
terns, regardless of the location of the syntactic head (right in Japanese, left in English).
13. Exceptionally, we suggest, the long vowels in suke-boo f r o m sukeeto boodo 'skate-
board', RoN-Buu from RoNdoN Buutsu 'London Boots [a pair of comedians], bata-
pii from bataa piinattsu 'buttered peanut', suke-paa from sukai paafekuto (TV) 'Sky
Perfect T V ' are retained, and sometimes there is variation in this respect. Interesting-
ly, all these exceptions happen to have bilabial stops at the beginning of the second
PrWd.
14. There are a few examples that we know of that involve skipping: one of them is paso-
koN, which somehow has replaced regular paa-koN, which was indeed attested at an
earlier stage (Kubozono and Ota 1998:190-91), and others include Ame-futo (which has
replaced the regularly back-truncated AmerikaN) from AmerikaN futtobooru 'Ameri-
can football', supa-koN from suupaa koNpyuutaa 'super computer', which might have
been formed on the analogy of paso-koN, and bifu-teki from biifu suteeki 'beefsteak',
which also violates LEFTMOST, as we noted above.
15. Itô (1990:236, note 19) gives pii-goro from pitchaa goro 'pitcher grounder' as a counter-
example, which does not sound like a familiar word to two of the three present authors.
16. The candidate suupaa does not violate MINWD, because this constraint is satisfied as
long as the output is minimally two moras long; it does not prevent the retention of an
original longer than two moras. See Itô (1990: 234).
17. Aigeo (1991: 9) treats double clipping (truncation) in English as Innovative clipping (e.g.
junior college * juco, aviation gasoline > avgas, capsule communicator • capcom).
References
Aigeo, J.T.
1991 Fifty years among the new words. Cambridge: Cambridge University Press.
Becker, T.
1990 D o words have heads? Acta Linguistica Hungarica 40, 5-17.
Benua, L.
1995 Identity effects in morphological truncation. In J.N. Beckman, L.W. Dickey
and S. Urbanczyk (eds.), University of Massachusetts Occasional Papers 18,
77-136.
Booij, G.E.
1995 The phonology of Dutch. Oxford: Oxford University Press.
Crowley, T.
1997 An introduction to historical linguistics. Auckland: Oxford University Press.
Di Sciullo, A. -M. and E. S. Williams
1987 On the definition of word. Cambridge, Mass.: MIT Press.
Greenberg, J.H.
1963 Some universals of grammar with particular reference to the order of mean-
ingful elements. In J. H. Greenberg (ed.), Universals of language. Cambridge,
Mass.: MIT Press. 58-90.
Hamans, C
1997 Clipping in modern French, English, German and Dutch. In R. Hickey and S.
Puppel (eds.), Language history and linguistic modelling: A festschrift for
Jacek Fisiak on his 60th birthday. Berlin: Mouton de Gruyter. 1733-41.
Haspelmath, M.
1993 The diachronic externalization of inflection. Linguistics 31, 279-309.
Hawkins, J. A. and A. Cutler
1988 Psycholinguistic factors in morphological asymmetry. In J.A. Hawkins (ed.),
Explaining language universals. Oxford: Blackwell. 280-317.
Hock, H.H.
1991 Principles of historical linguistics. Berlin: Mouton de Gruyter.
Huang, S.
1997 Chinese as a headless language in compounding morphology. In J. Packard
(ed.), New approaches to Chinese word formation. Berlin: Mouton de Gruyter.
261-83.
Ishiwata, T.
1993 Waseieigo to kokusaitsuuyoogo [Japanese-English and international common
languages]. In Gairaigo [Foreign words], Tokyo: Kawade-syoboo Shin-sha.
89-104.
Itô, J.
1986 Syllable theory in prosodie phonology. PhD dissertation, University of Mas-
sachusetts, Amherst [Published by Garland, New York, 1988].
1989 A prosodie theory of epenthesis. Natural Language and Linguistic Theory 7,217-
60.
1990 Prosodie minimality in Japanese. In M. Ziolkowski et al. (eds.), CLS 26-2. Chi-
cago: Chicago Linguistic Society. 213-39.
Itô, J. and R.A. Mester
1994 Reflections on C O D A - C O N D and alignment. Ms., University of California, San-
ta Cruz.
1995 Japanese phonology. In J.A. Goldsmith (ed.), The handbook of phonological

theory. Oxford: Blackwell. 817-38.
1999 The phonological lexicon. In N. Tsujimura (ed.), The handbook of Japanese
linguistics. Oxford: Blackwell. 63-100.
Jaeggli, O.A.
1980 Spanish diminutives. In Contemporary studies in Romance languages. Bloom-
ington, Indiana: Indiana University Linguistics Club.
Kager, R.
1999 Optimality theory. Cambridge: Cambridge University Press.
Kageyama, T.
1982 Word formation in Japanese. Lingua 57,215-58.
Katamba, F.X.
1993 Morphology. London: MacMillan.
Kubozono, H.
1990 Phonological constraints on blending in English as a case for phonology-mor-
phology interface. In G.E. Booij and J. van Marie (eds.), Yearbook of mor-
phology 3. Dordrecht: Foris. 1-20.
1993 Review of W.U. Dressier et al. (eds.), Contemporary morphology (Berlin:
Mouton de Gruyter). In Gengo Kenkyu [Journal of the Linguistic Society of
Japan], The Linguistic Society of Japan. 183-90.
1995 Gokeisei to on-in-koozoo [Word formation and phonological structure], To-
kyo: Kuroshio.
1996 Eigo no fukugoogo kyoosei ni tuite [On stress in English compounds]. In M.
Amano et al. (eds.), Gengo no sinsoo wo tazunete [Inquiries into the depth of
language]. Tokyo: Eichosha. 3-17.
Kubozono, H. and S. Ota
1998 On-in koozoo to akusento [Phonological structure and accent]. Tokyo: Ken-
kyuusha.
Kuwamoto, Y.
1998 Nihongo ni okeru fukugoogo ryakugo no on-in koozoo [The phonological
structure of compound clipping in Japanese]. On-in Kenkyuu [Phonological
Studies] 1,161-68. Tokyo: Kaitakusha.
Lass, R.G.
1987 The shape of English: Structure and history. London: J.M. Dent and Sons.
Lieber, R.
1980 The organization of the lexicon. Ph.D. dissertation, MIT, Cambridge, Mass.
1992 Deconstructing morphology. Chicago: University of Chicago Press.
McCarthy, J.J. and A.S. Prince
1986 Prosodie morphology. Ms., University of Massachusetts, Amherst and Bran-
déis University.
1993a Generalized alignment. In G.E. Booij and J. van Marie (eds.), Yearbook of
morphology 6,79-153.
1993b Prosodie Morphology I: Constraint interaction and satisfaction. Ms., Univer-
sity of Massachusetts, Amherst and Rutgers University.
1994 The emergence of the unmarked: Optimality in prosodie morphology. In M.
Gonzalez (ed.), Proceedings of ΝELS 24, 333-79.
1995a Prosodie Morphology. In J.A. Goldsmith (ed.), The handbook of phonological
theory. Oxford: Blackwell. 318-66.
1995b Faithfulness and reduplicative identity. In J.N. Beckman, L.W. Dickey and S.
Urbanczyk (eds.), University of Massachusetts Occasional Papers 18, 249-384.
Motwani, P.
1993 A dictionary of abbreviations and truncations. Tokyo: Maruzen.
Namiki, T.
1994 Heads and subheads of compounds. In S. Chiba et al. (eds.), Synchronic and
diachronic approaches to language: A festschrift for Toshio Nakao on the oc-
casion of his sixtieth birthday. Tokyo: Liber Press. 269-85.
Niikura, S. et al.
1996 Furansugo handobukku [A handbook of French], Tokyo: Hakusuisha.
Prince, A.S. and P. Smolensky
1993 Optimality theory: Constraint interaction in generative grammar. Ms., Rut-
gers University and University of Colorado.
Ralli, A.
1992 Compounding in modern Greek. Rivista di Linguistica 4,143-74.
Scalise, S.
1988 The notion of 'head' in morphology. In G.E. Booij and J. van Marie (eds.),
Yearbook of morphology. Dordrecht: Foris. 229-45.
1992 Compounding in Italian. Rivista di Linguistica 4,175-99.
Scheler, M.
1977 Der englische Wortschatz. Berlin: Erich Schmidt Verlag.
Schultink, H.
1988 Morphological heads: Evidence from Swahili. In G.E. Booij and J. van Marie
(eds.), Yearbook of morphology. Dordrecht: Foris. 247-58.
Selkirk, E.O.
1982 The syntax of words. Cambridge, Mass.: MIT Press.
1996 The prosodie structure of function words. In J.L. Morgan and K. Demuth
(eds.) Signal to Syntax: Bootstrapping from speech to grammar in early aquisi-
tion. Mahwak, NJ: Lawrence Erlbaum Associates. 187-213.
Shibatani, M.
1990 The languages of Japan. Cambridge: Cambridge University Press.
Shin, S.-H.
1997 Correspondence in Kyungsang Korean truncation. In S. Davis (ed.), Optimal
viewpoints. Bloomington, Indiana: Indiana University Press. 115-38.
Slobin, D.
1973 Cognitive prerequisites for the acquisition of grammar. In C. Ferguson and D.
Slobin (eds.), Studies of child language development. New York: Holt, Rine-
hart and Winston. 175-208.
Szymanek, B.
1989 Introduction to morphological analysis. Warszawa: Wydawnictwo Naukowe.
Trommelen, M. and W. Zonneveld
1984 The right side won, the right side won. In H.J. Bennis and W.U.S. van Lessen
Kloeke (eds.), Linguistics in the Netherlands 1984. Dordrecht: Foris. 179-87.
1986 Dutch morphology: Evidence for the right-hand head rule. Linguistic Inquiry
17,147-69.
Umegaki, M.
1963 Nihongairaigo no kenkyuu [Studies of foreign words in Japanese], Tokyo: Ken-
kyuusha.
van Beurden, L.
1988 Bantu heads on the lefthand side. In G. de Haan and W. Zonneveld (eds.),
Formal parameters of generative grammar IV (Yearbook 1988). Dordrecht:
ICG Printing. 1-13.
Vogel, I.B.
1990 English compounds in Italian: The question of the head. In W.U. Dressler et
al. (eds.), Contemporary morphology. Berlin: Mouton de Gruyter. 99-110.
Wiese, R.
1996 The phonology of German. Oxford: Oxford University Press.
Williams, E.S.
1981 On the notions 'lexically related' and 'head of a word'. Linguistic Inquiry 12,
245-74.
Yonekawa, A.
1989 Shingo to ryukoogo \New words and vogue words], Tokyo: Nan-undoo.
1996 Gendai wakamono kotoba koo [A study of the vocabulary of the young]. To-
kyo: Maruzen.
Zwicky, A.M.
1985 Heads. Journal of Linguistics 21,1-29.
1992 Some choices in the theory of morphology. In R. Levine (ed.), Formal gram-
mar: Theory and implementation. Oxford: Oxford University Press. 327-71.
III. Studies in Contrastive Japanese-
English Phonetics and Phonology
Two Different Kinds of Rhythm:
Japanese and English*
Yosihiro Masuya
Konan University
Introduction
Isochronicity of English has been discussed by quite a few phoneticians

and phonologists. Among them, there are Shen and Peterson (1962) who
denied it, basing their conclusion on their measurements, but ignoring the
question as to what the variation of their measurements looked like. Mea-
suring the duration of feet of David Abercrombie's reading of "The
North Wind and the Sun", Uldall (1971) concluded that "there is a strong
tendency to isochronism in this speaker's reading style". ("Foot" here
means an interval between a stressed syllable (or a silent stress1) and the
next stress, i.e. a stressed syllable or a silent stress, exclusive.) Lehiste
(1977) confirmed the existence in English of the "tendency toward isoch-
rony in production as well as in perception" and even went so far as to say
that isochrony should be integrated into English syntax. [Italics mine in
both quotations.] A few years later, Dauer (1983) investigated the dura-
tion of inter-stress intervals of English, Thai, Spanish, Greek and Italian,
of which English and Thai (conversational) were said to be stress-timed
and Spanish syllable-timed—the others were rhythmically unlabelled. As
part of her conclusion she said that the two different rhythmic types,
which were said to exist, had not been detected.
What makes a stretch of spoken words rhythmic? In English, stress,
whether uttered or silent, works as a beat and groups unstressed syllables
which follow; thus it composes a rhythmic unit, or a "foot", as defined
above, and repeating feet produces rhythm. The rhythm of a language
does not need to be as precise as metronomic rhythm. As Lehiste (1977,
p. 258) suggests, a stretch of spoken words is not arhythmic, as far as the
durational differences are not large enough to make the listener unable
to impose the rhythmic structure on the spoken words sequence.
In passing, the term "syllable" is too ambiguous and I have been using
it so far in the sense of a chest pulse, following Abercrombie (1967, chap.
3). If better is uttered in one chest pulse (see Abercrombie), the word con-
328 Yosihiro Masuya
sists of one syllable. When I was at Edinburgh University in Scotland, I

heard very in "That's very good" uttered in one chest pulse more often
than not; I even remember David Abercrombie once demonstrating be-
fore the class this manner of pronouncing the word. But most scholars
would not doubt that both better and very are structurally of two syllables
and the use of "syllable" in the sense of chest pulse causes a little confu-
sion. I, therefore, call syllable in this sense "chestable" 2 and keep "sylla-
ble" for the sense of C 0 . n VC 0 _ n ( V n " is the number of C's).
In Japanese, the repeating of chestables makes its rhythm. Chestables
are not difficult to hear in perception and to feel in production, especially
to a trained expert. Different speech styles may make chestable bound-
aries differ in Japanese. For example, the loan word "maati" [mBBtsJi]3,
which means "march", is said in one chestable in casual conversation and
in two chestables "ma}ati}" in deliberatively clear pronunciation; I use "}"
to show a chestable boundary.
Aim of this paper
Dauer (1983) drew a parallel between stress-timed and syllable-timed

languages, including rhythmically unlabelled ones, on one and the same
criterion, i.e. measuring the durations of the inter-stress period of both
types of language. As far as I recall, no one else has succeeded in making
a comparison between these two different types of speech rhythm on one
and the same criterion before. Her paper is very important in that she has
raised a question whether there is any genuine difference between the
languages which have been traditionally said to be rhythmically different.
Though I, intuitively, do not agree with her, I should like to try to an-
swer her question by comparing the rhythm of Japanese with that of En-
glish, the rhythm of both languages being familiar to me as a phonetician.
In order to answer it, one must draw a comparison on the same princi-
ple—comparing on one and the same criterion. The inter-stress period
does not work well in the examination of the Japanese rhythm because of
the behaviour of stress in Japanese—stress oftener goes with emphasis
there, which may form a considerably long inter-stress period. Apart from
the behaviour of the Japanese stress, I have been wondering whether it
would be proper to deal with inter-stress period as if the functional differ-
ence of stress—one working as a rhythmic beat and the other not working
as such—did not exercise any effect on the inter-stress duration. My pri-
mary aim in writing this paper is to show one possible way of comparison
on one and the same criterion without measuring the duration of the in-
ter-stress period as well as answering after my method the question Dauer
has raised.
Before I go on to the analysis of my data, I shall briefly review some of
the speech rhythm studies published in the 1980's and 1990's. Incidentally,
since I should also like those readers who are uninitiated into statistics to
follow my arguments, I shall insert brief explanations of statistics either in
footnotes or in the text hereafter.
1. Some comments on speech rhythm studies in the 1980's

and 1990's
Wenk and Wioland (1982) proposed the dichotomy of leader-timed (in
which timing was regulated group-initially) and trailer-timed (in which
timing was regulated group-finally) as an alternative typology of stress-
timed and syllable-timed. They based their arguments mainly on two
points, 1) that in a French sentence consisting of a sequence of a 12-sylla-
ble clause followed by a 6-syllable clause, the articulation time of the sec-
ond clause was much longer than half the articulation time of the first
clause, or rather somewhat nearer to that of the first; 2) that there oc-
curred group-final lengthening in French. Are these two points, however,
really counter-evidence to syllable-timing?
Any language has complex information carried by a stretch of linearly
arranged speech sounds. It is not surprising that some languages should
have part of that information conveyed by some rhythmic irregularity.
Regarding the first point, it might be that French tends to maintain the
balance of duration between the clauses, while those lengthened values of
the group-final syllables are supposed to be what should be excluded in
the statistical analysis4 as extraordinary values or outliers. What we have
to know is what the basic rhythm is like and whether any kind of irregu-
larity works as a signal of some kind of information. And there seems to
remain something to be examined by these researchers. "Leader-timed"
and "trailer-timed" are, to my mind, types of the rhythm unit—though, if
they are, the naming is not quite happy as labels for such units.
Fletcher (1991,199) made a statistically serious mistake in saying, "All
speakers' data show intercept values of over 100ms, ruling out the strict
syllable-timing principle that intercept values should approach zero. This
value probably represents the extra durational increment for the one or
two accented syllables in each prosodie word". Nearly five pages earlier
330 Yosihiro Masuya
she said something to the effect that if the relationship between the inter-
stress interval duration (in ms) (y below) and the number of syllables in
that interval (jt below) were "perfectly linear", the "regression 5 curve"
[sic] should have zero intercept and the correlation between them should
approach or equal 1.
When she said "perfectly linear", she seems to have had in mind such a
_y-to-x relation as y = α,*. 6 But simple linear regression is usually y = a0 +
axx. When we make a test to know which is the right one, we formulate
what is called "null hypothesis" 7 —H 0 : a 0 = 0 in this context. If the data
force us to accept the null hypothesis, y - a x x is our regression. A problem
is encountered at this point, however. Whether the data force us to accept
the null hypothesis or allow us to reject it, there is some probability for a
wrong judgement. Whenever a null hypothesis is formulated, an alterna-
tive hypothesis 8 is also formulated—//! : α 0 * 0 in this context. Let me call
"p," the probability of the wrong judgement on which the null hypothesis
that must not be rejected is rejected and "p 2 " the probability of the wrong
judgement according to which the null hypothesis which must not be ac-
cepted is accepted. 1) If the data allow us to reject the null hypothesis, the
alternative hypothesis is accepted, which means that the alternative hy-
pothesis is judged right. As p x exists, we accept the alternative hypothesis
at the risk of a given error level. This error level is what is called "signifi-
cance level".9
2) If the data do not allow us to reject the null hypothesis, the null hy-
pothesis is accepted, only because the grounds for rejecting the specifica-
tion the null hypothesis holds, are not found in the data. In this case, p2
exists, but it is not easy to calculate. Although the test of the hypothesis is
done on the principle that the rejection region 10 should be determined so
as to make p2 the least after px is fixed, there is no statistical procedure for
calculating p2, if the specification the alternative hypothesis holds is
something like a * 0.11
Fletcher's theory holds, only if H 0 : a 0 = 012 cannot be rejected and we
have no other choice than to accept it. But, as I mentioned just now, the
conditions of the significance test are completely different from those es-
tablished for the regression equation with the intercept: in the case of the
regression with the intercept, it is known how great (or how little) the risk
of the judgement error in adopting a given intercept is, whereas in the
case of the regression without the intercept, the greatness of the judge-
ment error in adopting zero intercept is not known—it may be consider-
able. The case with the intercept and that without the intercept are mat-
ters on different levels. Therefore, Fletcher's comment on French rhythm
based on her discussion of intercepts is meaningless. Nevertheless, some

people, regrettably enough, swallowed her theory uncritically.
Roach (1982) tried to test Abercrombie's (1967, pp. 97-8) claims that (i)
"there is considerable variation in syllable length in a language spoken
with stress-timed rhythm whereas in a language spoken with syllable-
timed rhythm the syllables tend to be equal in length" (p. 98) and that (ii)
"in a syllable-timed rhythm, the stress-pulses are unevenly spaced" (p. 97)
and made a comparison of the standard deviations calculated from the
measurements obtained from two groups of languages which Abercrom-
bie referred to—English, Russian and Arabic (stress-timed) and French,
Telugu and Yoruba (syllable-timed).
Here I take his examination of the claim (i) as an example. He showed
the standard deviations of the syllable duration from each language as
follows:
French: 75.5 English: 86

Telugu: 66 Russian: 77
Yoruba: 81 Arabic: 76
and said, "This set of figures does not appear to support claim (i)." (p. 74).
Unfortunately, very important values were ignored in his research:
showing a standard deviation by itself without any reference to the mean
value is meaningless. According to the Camp-Meidell theorem, over
95.1 % of the data lie within the range of χ ± 3 χ SD (χ: mean value of the
data; SD: standard deviation). Supposing mE= 4, sE = 0.8 for one language
and mF= 22, sF = 0.8 for another language (m for mean, s for standard de-
viation, E for one language and F for another), over 95.1 % of the data of
Language E lie within 4 ± 2.4, i.e. in the range from 6.4 to 1.6 and over
95.1 % of the data of Language F within 22 ± 2.4, i.e. in the range from 22.4
to 19.6. The ratio of the maximum value to the minimum value is 4.0 for
Language E and 1.1 for Language F: the data values are more or less sim-
ilar in Language F, whereas they are not in Language E, despite that sE =
s F . In order to examine the difference of variability between languages in
terms of the standard deviation, the comparison of the coefficients of
variation is required. The coefficient of variation is
CV = - or CV = - 100%
χ χ
(CV: coefficient of variation, s: standard deviation, x: mean value of JC).
With this hypothetical example, the coefficient of variation of Language
E is 20.0 % and that of Language F 3.6 %. I regret that Roach's efforts thus
332 Yosihiro Masuya
came to nothing and that his conclusion was nevertheless accepted and
quoted in an academic paper as if it had been meaningful.
Williams and Hiller (1994) constructed six types of unit and examined
whether English isochronicity was caused by chance and whether it was
linguistically-based. Their types of unit were:
1) "Full" unit—the one which starts "at every full (i.e. unreduced) vowel";
2) "Lexical(ly stressed)" unit—the one which begins "at every syllable
that bears the primary lexical stress of an orthographic word";
3) "Accented" unit—what begins "at every syllable that bears a pitch ac-
cent (i.e. where there is an easily-perceived peak in the pitch con-
tour)";
4) "Tense" unit—what begins with every syllable containing a tense vow-
el (i.e. diphthong or long monophthong);
5) "Random" unit—the one whose number of syllables is determined by
a random integer between 1 and 5, generated by a random number
generator;
6) "Arbitrary" unit—the one whose "number of syllables per unit [is] de-
termined by a fixed and arbitrary algorithm (e.g. every third syllable
starts a new foot)".
If some or all of 1)- 4) showed any more tendency towards isochronicity

than 5), they concluded that isochronicity in English was significantly
greater than chance and if linguistically-based units [1)- 4) were so called]
showed any more tendency towards isochronicity than 6), they concluded
that the tendency towards isochronicity was linguistically significant.
To reach their conclusions they had recourse to some statistical analyses.
Generally speaking, their R2 values were often too small. The R 2 value
shows how much of the data variation is explained by the calculated regres-
sion; therefore, if the R 2 value is too small, the regression is meaningless.
They calculated the regression line of the averaged syllable duration
against the number of syllables in a unit (abbreviated hereafter as "aver-
aged syllable regression (line)") and found that the averaged syllable du-
ration became shorter as the number of syllables in a unit increased 13 . It
is true that this compression of the syllable length is evidence of the iso-
chronous tendency. But they did not say which regression line they calcu-
lated, the ordinary one or the one for repeated measures.
There are a number of averaged syllable durations against the unit of a
given number of syllables—though there may be a single averaged sylla-
ble duration against a unit containing an unusually large number of sylla-
bles, e.g. a unit of 8 syllables; such a group should be excluded from the
calculation. Therefore, we have a number of groups of data, defined by
the number of syllables in a unit, such as the 1-syllable group, 2-syllable
group, 3-syllable group and now we have the variation of the data within
each group and the variation of the data between the groups. In such a
case as this, not in a case where the values of the two variables (x and y )
have a one/one correspondence, what is called "one-way analysis of vari-
ance" needs to be run to know how large the deduction of the variation
due to the regression14 from the variation between groups is, the latter
variation being the variation of the mean of each group. This deduction
shows the greatness of the "lack of fit" 15 of the regression line to the data.
If a lack of fit is judged to exist, the simple linear regression y = a0 + axx
is inapplicable.
After my attention was drawn to this problem in averaged syllable re-
gressions by a remark of Prof. Öta, I tried calculating three averaged syl-
lable regression lines from two of my dataseis, one from this paper and
the other from outside of it. In these calculations, following the authors of
the paper I am commenting on, I have excluded the units preceding an ut-
terance-final pause and in calculating the "lexical" regression, those units
preceded by a pause are also excluded because the authors' treatment of
them is unknown. Two of the regression lines are of genuine feet and one
is of a "lexical" unit. (The authors used the term " f o o t " in place of "unit".
But I should restrict the use of the term to what Abercrombie (1964b,
217) defines; a broadened use of a term in general makes things confused.
"Genuine feet" are " f e e t " in Abercrombie's sense.) The result is that the
existence of lack of fit is highly significant—sometimes extremely highly
significant16—with all of the three simple linear regression lines; these re-
gression lines are perfectly inadequate. It is most probable that the aver-
aged syllable regression lines Williams and Hiller calculated are also far
from adequate.
Here and there in the paper, they made a comparison of the results of
the regression analysis between the "linguistically-based" units and the
"random" unit even though the "random" regression line was not signif-
icant, and concluded that, if the results were favourable to the "linguisti-
cally-based" units, the tendency towards isochronicity in English was
larger than chance.
In arriving at such a conclusion, they were doing something meaning-
less. A regression line not being significant means that the inferred popu-
lation coefficient of the x-term17 is zero and that the x-term does not exist.
When they drew a comparison between the coefficients of χ of a signifi-
334 Yosihiro Masuya
cant regression line and a non-significant regression line, they were com-
paring a value which exists with a value which does not exist.
Earlier in the same paper, they showed the results of their calculation
of the simple linear regression of the unit duration against the number of
syllables per unit for each of their unit types and said, "The random and
arbitrary feet show generally higher r-squared values, indicating that this
factor (number of syllables) [sic] accounts for a higher proportion of the
variance than in the other foot types in the 'linguistic' foot types there
may be some other factor also at work in the determination of foot dura-
tion" (p. 431). Apart from a little obscureness of some phrases in this quo-
tation, they may have been right. So long as one judges from the calcula-
tions they made, the portion of the variation unexplained by their
regression seems to be larger in the "linguistic" units than in the "ran-
dom" and "arbitrary" units. But lack of fit had to be tested before any-
thing was said as a conclusion, and recalculations of the r-squared values
are advisable, for Prof. Öta (personal communication) has become aware
that the regular calculation of R 2 (i.e. dividing the variation due to the re-
gression by the total variation) is not happy for the regression for repeat-
ed measures, when a lack of fit does not exist. According to him, it is rea-
sonable that we should adopt for the coefficient of determination (i.e. R 2)
something which indicates how well the calculated regression equation
grasps the variation of the mean of every group; therefore, in the calcula-
tion of R 2, the variation due to the regression must be divided by the vari-
ation between groups, not by the total variation.
Van Zanten and van Heuven (1997) said in their paper, "Assuming that
the penultimate vowel of our Indonesian target words is stressed, we may
expect an anticipatory shortening effect for the stressed vowel in bisyllab-
ic as compared to monosyllabic words" and went on to say, "For longer
Indonesian words, the stress would still be on the penultimate vowel ..."
[italics mine]. It is surprising that whether a syllable is being stressed or
not is part of an assumption. To repeat Abercrombie's (1960) words,
stress is "something which is either present or absent" [italics in the origi-
nal]. If they listen to their informants' utterances, they can more or less
affirm that stress is there or that it is not there. I say "more or less" be-
cause I do not think the question Mrs Uldall (talk before the class) came
across has not been solved yet; she let several phoneticians, who were na-
tive speakers of English, listen to Abercrombie's reading of "The North
Wind and the Sun" and plot stresses on the text in order to prepare for
her (1971) paper and saw that they sometimes disagreed about the plot-
ting on some stressable unstressed syllables. Hence a few mis-plottings of
stress may happen; but it is not expected to be so serious that it gives so

great a damage as to cause a wrong result in so far as one keeps in mind
Abercrombie's (1967, 98) words:
"It is ... necessary to learn to listen differently [from the way one lis-
tens to everyday speech] in order to be able to analyse speech
rhythm, whether of one's mother tongue or another language, and to
describe it in general phonetic terms. Few succeed in doing this with-
out training" [italics in the original]
and has become familiar with the sounds of the target language. (The du-
ration of those wrongly constructed feet is expected to be buried in the
variation or discarded as extraordinary values or outliers.)
Now I shall return to the "linguistically-based" unit types which Will-
iams and Hiller casually constructed. If the speech rhythm is formed by
stress-beat in English, any other units than the stress-headed like "foot"
in Abercrombie's (1964b, 217) sense are irrelevant. Williams and Hiller
said, "A lexically stressed syllable is merely a candidate for sentence-level
stress, and may not receive stress in any given utterance" [italics mine].
Then how can a "lexical" unit when there is no stress in it, work as a com-
ponent of rhythmic construction? It can do nothing more than disturb it.
In the sentence following this, they said "As the word 'stress' has been
used to mean many different things, it is necessary to test more than one
definition of it". This is not, however, a good reason for including a factor
irrelevant to rhythm; what is important is to judge by ear and determine
the factor which forms rhythm. Otherwise, phonetics is only paper pho-
netics and any studies in paper phonetics are fruitless.
Incidentally, did they test the normality of the distribution with the data
from the unit types they constructed? The data of the "lexical" units I
have constructed to examine part of their theory, have been proved by the
Shapiro-Wilk test of normality not to be normally distributed when χ = 2
and χ - 4. From such data as these, viz. those which are not (approximate-
ly) normally distributed, statistical inference cannot be derived, because
statistical inference presupposes the (approximately) normal distribution
of the data. What is capable of being achieved in such a case is descriptive
statistics and nothing else. We can only show the results of measurement
in some informative ways—by calculating the mean, the standard devia-
tion, etc. and presenting a histogram, a scattergram, etc.; we could calcu-
late regression lines but cannot make any statistical tests. Therefore, most
of their discussions are likely to be statistically worthless. When one deals
336 Yosihiro Masuya
with those cases where one is not sure of the (approximate) normal dis-
tribution of the data, the test of normality is indispensable.
When they said that "accented" units began with every syllable that
bore a "pitch accent", it showed their ignorance of the fact that a stressed
syllable might be uttered in low pitch in English (see, for example, Halli-
day (1970)) and that stress may be silent (see above). Their "linguistical-
ly-based" types of unit are all irrelevant to the rhythm of English speech.
If one is unable "to listen differently from the way one listens to every-
day speech", how can one analyse the syllable boundary (or chestable
boundary) of the Standard Scottish English enter, for example—MacMa-
hon (1998, p. 21) says that the boundary lies in the middle of /t/ for some
people?
Nakatani et al. (1981) replaced an adjective-noun phrase in some En-
glish sentences with a reiteration of "ma"-syllable, e.g. "maMA MA" for
"absurd day" and trained the "talkers" [sic] who would help their exper-
iment, so that they could speak the sentence fluently with the proper
rhythm and the proper stress pattern on the "ma"-syllable phrase and
with the normal English manner of pronunciation on the rest of the sen-
tence, in order to examine an aspect of American English rhythm. The
sentences spoken by the talkers were recorded and the "ma"-syllable
phrase was excised from each target sentence. After they most carefully
reviewed the excised phrase, the experimenters selected what they judged
acceptable and measured the duration of each syllable of the excised por-
tion. The experiment was prepared with closest attention.
Who can be sure, however, that the talkers spoke the nonsense word
phrase of each sentence, being conscious (or subconsciously aware) that
they were reading it as a stretch of normal English sound sequences and
that the experimenters listened to the recorded sentences and excised
phrases, being conscious (or subconsciously aware) that they were listen-
ing to a stretch of normal English sound sequences? If the talkers and/or
the experimenters were conscious (or subconsciously aware) that they
were speaking and hearing non-speech phrases, the results obtained will
certainly be different from what is expected from the analysis of speech.
No one would be convinced that the switch in their brain fixed to the
speech mode even at the sequence of "ma"-syllables.
2. The dataseis for this paper
The datasets for this paper are obtained from "The North Wind and the
Sun" read by four English speakers and its translation read by as many
Japanese speakers; the texts are what is produced in phonetic transcrip-
tion in "The Principles of the International Phonetic Association", 1949
edition. The texts given to the readers were reproduced in the ordinary
writing system of both languages.
The speakers are two males and two females for each language. All the
Japanese speakers are speakers of the Tokyo accent. English speakers are
all alleged RP speakers and they have more or less common accentual fea-
tures—one of them (Em2) is a Scotsman born of Scottish parents in Scot-
land but no features of the Scottish accent are detected so far as this read-
ing is concerned: according to him, his mother was educated in English
schools; he may have been, too. E m i pronounced "traveller", which ap-
pears four times in the text, consistently in two chestables, Ef2 consistently
in three chestables and Em2 and E f l sometimes in two, sometimes in
three. The words "other" and "immediately" each occur once in the text
and E m i and E f l pronounced the former in one chestable and the latter in
three chestables, uttering "-diately" in one chestable, while Em2 and Ef2
pronounced the former in two chestables and the latter in five chestables.
All of these English speakers usually exploded the syllable-final stops and
when they exploded them, that exploded stop was pronounced in an inde-
pendent chestable, e.g. "winjd 3 ]", from time to time. Such a chestable was
regarded as an appendage to the word to prevent the situation too favour-
able to the isochronic tendency from being introduced; hence, "winjd 3 }"
was treated as a one-chestable equivalent. I should call these English
speakers "speakers of educated Anglo-English". The speakers are:
(1) Jml, male: born in Tokyo in 1934;

Jm2, male: born in Tokyo in 1969;
Jfl, female: born in Tokyo in 1966;
Jf2, female: born in Tokyo in 1971 and
Emi, male: born in Oxford in 1968;
Em2, male: born in Aberdeen in 1969;
E f l , female: born in Oxford in 1945;
Ef2, female: born in North London in 194518
Measurements of durations were made on the spectrograms in ms with

Sensimetrics Corporation's Speech Station 2™, v. 1.0.4 and Kay Elemet-
338 Yosihiro Masuya
ries Corporation's Multi-Speech™ Model 3700, v. 2.2; the system of

Speech Station 2 in which the energy envelope can lie below a spectro-
gram, both corresponding with each other, is especially helpful when the
boundaries of a chestable are plotted, though I do not mean that the ener-
gy envelope always helps, nor that it tells where the boundary lies. The best
signal with which to determine the boundary is my auditory impression
and feeling at the chest I get when I imitate the speaker's pronunciation.
If a sentence, a phrase or a word begins after a pause with a voiceless
or devoiced [+ voice] 19 stop, a sound loop of a repetition of the stop and
the following vowel has been made with Speech Station 2 and the starting
point of the loop where no sound break comes to be heard has been in-
terpreted as the beginning of the closure for the stop.
In Uldall (1971), David Abercrombie himself made the scansion, or
foot-boundary marking, of the text he had read and divided a long pause
after a sentence into two silent feet, i.e. feet filled with a silent stress
alone, e.g.
(2) the north wind gave up the attempt. | A | A | Then ....

(The caret is the symbol for silent stress.)
In Masuya (1994), Jack Aitken, who, at my request, did for me what Ab-
ercrombie did for Mrs Uldall, also made the scansion in the same way and
answering my question, confirmed that he felt two beats there. In the
scansion for this paper, I followed them; if I felt two beats, I inserted two
carets there and if three beats, three carets.
I made the scansion of the English readings myself—and the plotting of
chestable boundaries as well for the Japanese readings—partly because
Aitken showed that he did not make the scansion as a speaker but as a hear-
er by modifying the first scansion on his second listening and partly because
a phonetician, who was a native speaker of English, was unavailable.
Pauses are part of the rhythmic stretch of words and they do not break
the rhythm; therefore, I have dealt with unfilled feet, i.e. feet beginning
with a silent stress or those consisting of a silent stress alone, in the same
way as filled feet, which are composed of a stressed chestable followed by
either (an) unstressed chestable(s) or none of such. A silent stress has
been counted as one chestable-equivalent; hence, the pause in (2) is a se-
quence of two one-chestable feet and the second foot in (3) is a three-
chestable foot. If a pause is composed of the sequence of two feet consist-
ing of a silent stress alone, I divided the whole duration of the pause by
two and if the sequence is of three silent stress feet, by three, rounding off
the quotient to the first digit. But if the pause is of a silent stress followed
by another silent stress preceding (an) unstressed chestable(s), such as
(3) |J Λ Then the | ....
I have left them out as unmeasurable, for dividing the duration of the
words "then the" in (3) by two is meaningless.
Now I shall turn to the Japanese readings. In Japanese, chestable
boundaries may occur 1) after a vowel, e.g. arujtoki), 2) after the syllabic
"n", e.g. danjdanj, 3) in the midway through a long vowel, e.g. majati} or
4) in between a double consonant, e.g. sikjkari to}. In such a case as 4),
the duration of each chestable cannot be measured, though the whole du-
ration of the two chestables taken together can be. Two or three syllables
are usually spoken in one chestable, though not always.
I do not think pauses consist of chestables but they still form part of the
rhythm of Japanese utterances and their durations are also measured, be-
ing treated as equivalents to chestables. A short pause, between 28 and
90ms with the mean value of 64.6ms, occurs following a glottal stop; this
is regarded as part of the preceding chestable, not a genuine pause. This
is supposed to have something to do with the release of the glottal stop.
The duration of a longer pause has been divided either by the duration
value of the preceding chestable or by the mean value of the durations of
the preceding two chestables whenever there are two measurable ones
and the quotient, rounded off to the first digit, has been regarded as the
number of chestable-equivalents included in the pause.
My dataseis are indeed limited, but for time consuming work like this,
anyone's datasets might be. It is because they are limited that I have con-
ducted the statistical analysis. Statistics is the theoretical device to draw,
from some limited data, inferences about what the elements of a big
dataset, or technically a population, might be like—see note 17 about
"population". Without this theoretical inference, statistics would never
be worth using.
Statistical inference requires as a precondition that the elements of a
sample dataset should be normally distributed. When it is assumed quite
a few factors are involved in a phenomenon, such as in a speaker fixing
the so-called isochronical foot durations in an utterance, a normal distri-
bution of the measured values is generally expected, but, to make sure
and partly to find extraordinary values if any, I have run at the 5 % signif-
icance level (p < 0.05) both the Shapiro-Wilk test and the Geary test for
the English data and the Geary test for the Japanese data to discover
340 Yosihiro Masuya
whether the distribution of each set is normal—the sample size of the Jap-
anese data is too large for the former test. Every set of samples has been
found normally distributed.
The next thing to be tested is whether the datasets are independent of
one another; in other words, each set comprises samples from different
populations. If the inferred population mean and/or variance 20 of two
sample sets are (is) not regarded as the same in consequence of a statisti-
cal test, these sample sets are treated as derived from different popula-
tions; if these population parameters are statistically the same, the
datasets come from one population. Among my datasets, Jf2 and E m i are
independent as shown in Tables 1 and 2 below; Jml, Jm2 and Jfl are de-
rived from a third population and Em2, E f l and Ef2 from a fourth popu-
lation (the speakers' symbols being given to the datasets as their labels,
too). The sample data of Jml, Jm2 and Jfl, therefore, are gathered to-
gether into one dataset and those of Em2, E f l and Ef2 into another set
hereafter, which will be called "3 J's", "3 E's", respectively. Incidentally,
it is nothing but a coincidence that one is independent and that the other
three come from one population in the data of the two languages.
Table 1. 95 % confidence intervals21 of the population mean (μ)

and the population variance (σ 2 ) of the Japanese data.
The interval is expressed in the form of "(lowest value, highest value)".
m Jml Jm2 Jfl

μ (270.9, 290.3) (251.6,272.9) (249.3,271.5) (250.2, 269.6)
σ2 (3333,5114) (4061, 6231) (4372, 6707) (3343, 5128)
Table 2. 95 % confidence intervals of the population mean (μ)

and the population variance (σ 2 ) of the English data when = 3.
The interval is expressed in the form of "(lowest value, highest value)"
Emi Em2 Efl Ef2

μ (491.5,576.7) (547.8, 675.2) (588.9, 708.1) (578.3,727.7)
σ2 (4784,17645) (10730, 39577) (9210,33973) (14610, 53890)
In Table 1, the test at the significance level ρ < 0.05 has not indicated that
the population variance is significantly different between the four speak-
ers but suggested that Jf2's population mean is significantly different
from the others'—her lowest value of the confidence interval is almost
the same as the highest values of the other speakers. In Table 2, the pop-
ulation variance of E m i does not differ significantly from that of Em2
and E f l , but it does from that of Ef2, while the population mean of Emi
is significantly different from that of the other three speakers—Emi 's
highest value of the confidence interval is almost the same as the lowest
values of the other speakers. (When χ had a value other than 3, significant
differences were not detected.) The significance test has been carried out
at the level ρ < 0.05.
Now I shall turn to the comparison between Japanese and English and
at the same time I shall show in what way statistics makes a good contri-
bution to the description of speech rhythm construction, which is a sub-
sidiary objective of this paper.
3. Rhythm of Japanese and English
3.1. Scattergrams of the duration of the rhythmical units
Let us look at the scattergrams of four speakers, two each selected from
the Japanese and the English speakers (Figs. 1-4).
Jml
600 χ
% ^ 500
500 -
α Ë 400 * ~ •
duration (ms)
A
υ 100
0
0 5 10 15 20
order of occurrence
Figure 1. Scattergram of J m l as a sample of 3J's, showing the durations (in ms)

of chestables with pauses included in the order of occurrence. The order of
occurrence is translated into a sequence of numbers, 1.0,1.1,1.2, ... 1.9, 2.0,
342 Yosihiro Masuya
600
500
1α 'sε 400 Λ •
ω ο 300 <ν • . • • s s %* duration (ms)

ν ν . • y .*
χ> cd
C3 200
< — „ V '
% α
ω τί
υ 100
0 Ψ
5 10 15 20
order of occurrence
Figure 2. Jf2's scattergram showing the durations (in ms) of chestables with pauses
included in the order of occurrence. The order of occurrence is translated into a
sequence of numbers, 1.0,1.1,1.2,... 1.9, 2.0, . . . . (Jf2 is an independent sample
data.)
Emi
1400
1200
1000
a 800
o
• duration
600 -f
3
T3
400
O
.O
200
0 Ψ
0 1 2 3 4 5 6 7
foot size (no. of ches/foot)
Figure 3. E m l ' s scattergram which shows the durations of feet against the foot size,
defined by the number of chestables per foot. (Emi is an independent sample
data.)
Ef2
1200 -r
•
*
•
___ •
•
c 800 •1 1
o • 1
600 • X A duration
3
•• 1 $
•
-α $
400 I 1• •
o •
o •
200
i 1 1 1 1
0 1 2 3 4 5 6
Figure 4. Scattergram of Ef2 as a sample of 3E's, which shows the durations of feet
against the foot size, defined by the number of chestables per foot.
Interestingly enough, the Japanese scattergrams and the English scatter-

grams draw completely different pictures; this suggests that Japanese and
English rhythmic constructions have their own distinct features. The pic-
ture of the Japanese scattergrams is roughly a rectangle; for such a scat-
tergram, a regression does not work. On the other hand, the English scat-
tergrams form a slightly upward going slope higher at the right. This is a
visualization of so-called isochronicity with the slight lengthening of a
foot as the number of chestables in it increases. To this, a stretch of the
Japanese pseudo-feet exhibits a striking contrast. I shall return to this and
to the pseudo-feet in the next sub-section.
3.2. Regression
I have just said in the sub-section 3.1. that a regression does not work for the
Japanese dataseis. But in order to make a comparison between Japanese and
English on one and the same criterion and look at where the difference lies
and how different the rhythmic structures of the two languages are, I want a
regression for Japanese as well. So I have played a mathematical trick and or-
ganized a stretch of Japanese speech into a foot structure like that of English.
I have let a random number generating programme generate a sequence
of random integers between 1 and 6 inclusive chosen with a similar proba-
bility with which these integers, the representation of a foot size, occur in
344 Yosihiro Masuya
English. In the English readings used for this paper, six classes of foot size
exist with a silent stress regarded as a chestable equivalent: 1-, 2-, 3-, 4-, 5-,
and 6- chestable foot. The probability with which every foot size has oc-
curred in the speakers' readings is in Table 3. In the calculation of the prob-
ability, those feet of unmeasurable duration are also counted. Every inte-
ger of the random sequence is chosen with the mean probability, which is
shown in the p-row in Table 3.
Table 3. Probability of the occurrence of each foot size in the English datasets derived
from each population. The integers represent the foot size.
ρ: mean value of probability.
1 2 3 4 5 6
3E's 0.334821 0.276786 0.285714 0.066964 0.017857 0.017857
Emi 0.287879 0.212121 0.348485 0.136364 0 0.015152
Ρ 0.311350 0.244453 0.317100 0.101664 0.008929 0.016504
If the integer 1, for example, occurs with the probability of 0.311350, it oc-
curs 31.135 % of the time.
The sequence generated is
(4) 241111222343131214333223333421311231233
3 3 2 1 2 5 2 3 3 4 3 3 3 3 2 3 2 2 4 1 3 3 1 3 3 2 3 3 3 ....
There happened to be no integer of 6 among the first eighty-six random

integers of the sequence.
On the supposition that the foot size defined by the integers in the se-
quence of (4) occurred in this order in the Japanese readings, I have as-
signed pseudo-feet to each speaker's reading. If an integer happened to hit
a chestable or a sequence of chestables the duration of which is not mea-
surable, the integer was skipped unless the whole sequence of unmeasur-
able chestables fit the pseudo-foot or occupied part of it. Again if the num-
ber of the last chestable(s) in a speaker's reading was smaller than the
corresponding integer, this integer was discarded and the dividing of the
sequence of the chestables into pseudo-feet ended at the preceding inte-
ger; every speaker did not end at the same integer—Jml came to end at the
last integer written out in (4), Jm2 at the integer 3 immediately after the
last 1 and there were one chestable left. Jfl's end is at the integer 3 follow-
ing the last 2, Jf2's end at the integer 3 next to Jfl's end.
Now we have a similar formation for the Japanese readings to the
rhythmical structure of English, though I do not mean at all that this for-
mation makes any contribution to forming the rhythm of the Japanese
speech—far from it. But if the rhythmic structure of the Japanese speech
had such a formation, what would happen?
Below are the scattergrams of the Japanese pseudo-foot sequences de-
rived from each population compared with that of 3E's.
pseudo-feet of 3J's
^ 1600
¿ 1400
0 1200
S 1000
î 800 • duration
1 600
S 400
o 200
o
0
0 1 2 3 4 5 6
Figure 5. Scattergram of Japanese foot structure derived from 3J's.
Foot size is defined by the number of chestables per foot.
pseudo-feet of Jf2
^ 1600
j | 1400
"g 1200
te
JH
1000
•a 800 • duration
§ 600
400
« 200
£ 0
0 1 2 3 4 5 6
Figure 6. Scattergram of Japanese foot structure derived from Jf2.

346 Yosihiro Masuya
3E's
1200
^ 1000
t/i
Β
~ 800
c
o
5 600 • duration
3
^ 400
o
6 200
0
0 1 2 3 4 5 6 7
Figure 7. Scattergram of 3E's.

The upward going slope higher at the right is steeper in the Japanese data
(Figs. 5 and 6) than in the English data (Fig. 7; see also Fig. 3). What does
this visual impression tell?
It is not difficult to calculate regressions for the Japanese data. Lack of
fit does not exist in either the data of 3J's and those of Jf2. In passing, as
I said in Section 1, one-way analysis of variance is required and it has been
carried out with both Japanese and English dataseis. Although all
dataseis do not satisfy the precondition of the analysis, i.e. the homogen-
ity of variance, the test which is to be made is what is called "F-test" and
it is, fortunately, so robust that the consequences of the test are not dam-
aged for want of this precondition. The regressions are:
(5) a. for 3J's, *j> = 39.668 + 243.559*,

b. for Jf2, *j> = 5.62634 + 280.029*.
But the intercept has been found insignificant at the 5 % significance level
for both (a) and (b) and no other choice can be exercised than to accept
zero intercept, though the greatness of the risk of the wrong judgement is
unknown, as I said in Section 1. Thus the regressions which can be adopt-
ed for the Japanese pseudo-feet data are:
(6) a. for 3J's, y = 243.559* = 0.9960),

b. for Jf2, y = 280.029* (R2 = 0.9996).
In any event, the significance test of regression is made on the coefficient

of χ (i.e. slope) alone with the intercept ignored; the intercept is not a mat-
ter of great import in statistics. (6) (a) and (b) are visualized in Fig. 8.
1800
duration
(ms) / Jf2
1600
/
/
1400 / /3J's
1200
κ
// /
1000 / /
800
/
/
Λ
/
600
//
/ /
/ /
400
200
-1 0 1 2 3 4 5 6 7
-200 foot size
Figure 8. Regression lines representing the Japanese pseudo-feet sequence.
The English data are not so simple, or rather considerably complicated.

Lack of fit exists for both datasets, 3E's and Emi. The former is really
very complicated and I shall discuss Emi first. In Fig. 3 (Emi scatter-
gram), there is a pillar against χ = 1, and on top of it two quarterly rotated
little squares which indicate the duration values of 799ms and 718ms, lie
at some intervals from the other squares. The longest duration next below
these squares is 559 ms. This top group of this data pillar consists of (1,
799)22 a n c [ 718), each twice repeated. They are of two long pauses at
the end of a sentence. Feet composed of such pauses are usually much
longer than feet within a sentence in English readings; therefore, in the
calculation of the regressions in this paper, they are regarded as extraor-
dinary values and discarded. Then the data pillars can be divided in two
groups—the first three pillars and the rest. The whole dataset, as seen in
Fig. 9, is approximated pretty well by the quadratic equation (7),
348 Yosihiro Masuya
(7) y = 440.29 - 47.847* + 30.307*2 (R2 = 0.6195),
but if the scattergram is divided into the two parts, how the foot duration
is lengthened is seen more clearly and somewhat differently.
Emi
1400
1200
j | 1000
o 800
C
)HS • duration
3
T3
o 400
a
200
0
0 2 4 6
foot size (no. ches/foot)
Figure 9. Scattergram and the quadratic regression for Emi.
For the x-range, 1 < χ < 3, the regression is (8), giving the predicted
values 23 (1,424), (2, 481) and (3, 539).
(8) j> = 366.777 + 57.517* (R 2 = 0.9550).
When 4 < x, the data I have are too few and I cannot declare anything
positively but there is no room for doubt that the increase of the foot du-
ration is greater. Fig. 10 shows the relation between the regressions (7)
and (8).
1800
duration
(ms)
1600
1400
1200
1000
800
600
400
200
-1 0 6 7
-200 foot size
Figure 10. Relation between the simple linear regression (1 < χ < 3)
and the quadratic regression in the whole x-range for E m i .
The scattergram of 3E's is approximated by a quintic equation, which

overlaps with the scattergram in Fig. 11: the equation is
(9) y = -372.79 + 1826.1* - 1397.3x2 + 494.6x3 - 78.519x4 + 4.5749x5

(R2 = 0.4494)
and the curve this equation draws well indicates in what way the foot du-
ration increases. The increase is greater in the range from χ - 3 to χ = 4
and at both sides of this 3-to-4 range the increase is smaller. Here again
the dividing of the scattergram into two groups gives a clearer picture.
(The long-pause feet at the end of a sentence, (1,864) and (1,766) are also
discarded, the former being three times repeated and the latter twice re-
peated.)
350 Yosihiro Masuya
3E's
1200
800
600 • duration
200
0
0 1 2 3 4 5
Figure 11. Scattergram and the regression of 3E's.
The regression for the range 1 <jc<3 is (10) which gives the predicted val-
ues, (1,470), (2,550) and (3,629) and the regression for the range 4 <x < 6
is (11) supplying the predicted values (4,813), (5,871) and (6,929).
(10) y = 390.284 + 79.665x (R2 = 0.9805).
(11) y = 581.703 + 57.905* (R 2 = 0.9998).24
The regression lines (10) and (11) are in Fig. 12.
4. Comparison of the Japanese rhythm with the English
How complicated the English rhythmic structure is, has been described in
Section 3.2 and the contrast it makes with the Japanese simple rhythmic
structure has also been stated there. This complexity of English speech
rhythm may have its source, phonologically, in the double-layeredness of
the rhythmic structure—one is the stratum of chestable and the other is
that of foot formed by the organization of chestables. Japanese has a sin-
gle layer—the stratum of chestable. Besides the stratum structure, there
seems to be various individual differences in English rhythmic construc-
tion, apart from the difference of talking speed among speakers. In the
1800
duration
(ms)
1600
1400
1200
1000
800
600
400
200
-1 0 6 7
-200 foot size
Figure 12. Lower and upper regression lines of 3E's.
data of Masuya (1994), in which the rhythm of Lowland Scottish English

is examined and compared with that of an RP speaker's reading, lack of
fit does not exist in any dataset and the regressions are simple linear ones.
Now let us look at the difference of increment of a foot duration as the

foot size becomes bigger. Table 4 gives the predicted values, which are cal-
culated by the regression and Table 5 the ratio by which the foot or pseu-
do-foot duration increases.
Table 4. Values predicted by the regressions. The integers 1-6 represent the foot or pseudo-
foot size. L: linear regression. Q: quadratic or quintic regression.
1 2 3 4 5 6
3J's L 244 487 731 974 1218 1461
3E's L 470 550 629 813 871 929
Q 477 537 637 818 886 963
Jf2 L 280 560 840 1120 1400 1680
Emi L 424 481 539
Q 423 465 567 730 952 1236
352 Yosihiro Masuya
Table 5. Ratio of the increase of the predicted values. The integers 1-6 represent the foot or
pseudo-foot size. L: linear regression. Q: quadratic or quintic regression.
1 2 3 4 5 6
3J's L 1.0 1.97 3.0 3.99 4.99 5.99
3E's L 1.0 1.17 1.34 1.73 1.85 1.98
Q 1.0 1.13 1.34 1.71 1.86 2.02
Jf2 L 1.0 2.0 3.0 4.0 5.0 6.0
Emi L 1.0 1.13 1.27
Q 1.0 1.10 1.34 1.73 2.25 2.92
For the Japanese pseudo-feet, the ratio multiplies as many times as the
number of chestables in foot, though in 3J's it is not exactly as many as the
number of chestables because of an effect given by the rounding off of the
predicted values at the first decimal place.25 On the other hand, with the
English feet the duration of the 6-chestable foot is nearly 3 times the du-
ration of the 1-chestable foot for E m i and approximately two times for
3E's. The ratio of E m i is not very reliable because the data are too few
for bigger feet, as I said earlier. The duration of the 3-chestable foot is
about 1.3 times for both 3E's and Emi. Interestingly, L-ratios tell us that
in 3E's the increase of the duration is smaller with bigger feet and in E m i
smaller with smaller feet; this matches well with the scattergrams. If a
scattergram is divided in parts and the regression is calculated for each
part, more precise inference can be made. This ratio difference between
the two languages clearly indicates the isochronous tendency of the En-
glish speech rhythm and at the same time, the chestable-timed nature (or
syllable-timed nature, to use the more popular term) of Japanese is equal-
ly beyond reasonable doubt.
The comparison of coefficients of variation (CV hereafter) of the

chestable durations confirms Abercrombie's (1967) remark quoted in
Section 1 and repeated here again that "there is considerable variation in
syllable length in a language spoken with stress-timed rhythm whereas in
a language spoken with syllable-timed rhythm the syllable tend to be
equal in length". And it also confirms the consequences of the regression
analysis just made of the foot structures. The CV's of Japanese and En-
glish are in (12) and (13), respectively:
(12) a. 25.6% for 3J's (mean: 278.5ms; standard deviation: 71.2ms),

b. 23.2% for Jf2 (mean: 297.4ms; standard deviation: 68.9ms),
(13) a. 62.4% for 3E's (mean: 225.3ms; standard deviation: 140.5ms),

b. 56.2% for Emi (mean: 198.0 ms; standard deviation: 111.2ms).
This difference of CV's between (12) and (13) suggests that variation in
chestable length in English is considerable whereas that in Japanese is fair-
ly small and also that Japanese shows a tendency to iso-chestabicity (or iso-
syllabicity in more popular terms) whereas English displays a tendency to
isochronicity. Incidentally, in the calculation of CV's only chestables prop-
er have been dealt with, partly because Abercrombie (1967) does not have
pauses in mind when he makes the remark just cited and partly because
what we need to know here is the relation between chestables.
The CV's and the increase ratios of the predicted values of foot dura-
tion calculated by regression explicitly display the rhythmical feature dif-
ference between Japanese and English.
Notes
* Acknowledgements are extended to Prof. Minoru Öta, former Professor of Statistics
(now Emeritus Professor after retirement), Konan University—quite a few considerably
long faxes were exchanged—to Prof. Tomoyasu Taguti, Konan University, a mathemati-
cian, for telling me about pseudo random numbers, to Prof. David Rycroft, Professor of
English Literature, Konan University, for checking for grammar and last but not least, to
Dr Keiichi Masuya for designing the computer programme " G R L I N E " for me to draw
some of the regression lines for this paper.
1. About silent stress, see Jones (19507, p. 227, note 1) and Abercrombie (1964a, 20; 1968).
2. Some may criticize this term as etymologically unjustified; but I preferred the sound and
adopted it when it was suggested by Mrs Uldall. A portmanteau word is not always ety-
mologically justified. About chestable, see Masuya (1979).
3. The phonetic fonts used in this paper are those distributed on the internet by IPA. [§]: as
for the use of [s] I am following J. C. Catford; he calls this sound "apico-infradentalized
lamino-postalveolar fricative" (personal communication). A sibilant "which occurs in
most N. W. Caucasian languages" and in the articulation of which "there is a narrow ar-
ticulatory channel between the blade of the tongue and the extreme back of the alveolar
ridge ... at the same time, the apex of the tongue is in contact with the back of the lower
front teeth . . . " In my articulation of the Japanese /s/, a narrow articulatory channel is
made between the blade of the tongue and the alveolar ridge and the tip of the tongue is
touching the lower alveolar ridge or the back of the lower front teeth; I should call this
sound "apico-infraälveolarized (or infradentalized) lamino-alveolar fricative", also fol-
lowing Catford. If this sibilant precedes H I , it is palatalized. About this Caucasian sibilant,
see Catford (1992).
4. For statistical calculations in this paper, I use Microsoft Excel 2000™ and HB for Win.
5. A regression is an equation statisticians calculate to look at the relation of one variable
to other variables. If the relation is between one variable and another, the regression is
said to be simple.
354 Yosihiro Masuya
6. y represents an inference of y. Such equation represents the variation of the data and is
called "regression". If the equation has one χ term and represents a straight line, it is
said to be "simple linear".
7. The null hypothesis is a hypothesis set up to make a statistical test of the matter in ques-
tion. It is symbolised as H0 and Greek letters, which are reserved for theoretical values,
are used for the specification of the hypothesis: a 0 is an inferred theoretical value for a 0
in the regression, which is to be calculated from the data.
8. The hypothesis which is to be accepted if the null hypothesis is rejected. Symbolised as
Hv Again Greek letters are used for the specification of the alternative hypothesis.
9. If the probability of the error being made is 5 %, the matter in question is said to be sta-
tistically significant at the 5 % significance level and the significance level is sometimes
expressed by ρ < 0.05.
10. Rejection region is the range of calculated values within which the null hypothesis is re-
jected.
11. If the alternative hypothesis is something like Hx\a = α, [α, Φ 0], p2 is calculable.
12. / / , : a 0 * 0.
13. About regression, see note 5. Some details are in Masuya (1997, chap. 13).
14. The variation due to the regression is the differences between the mean of the measured
values and the values of y, i.e. the values calculated by the regression equation. In cal-
culating this variation, each difference is squared and the effect of positive/negative is
avoided.
15. About "lack of fit", see Masuya (1997, chap. 13).
16. The significance levels from the dataset outside this paper, that from this paper and that
of "lexical" units are: ρ < 0.0001, ρ < 0.00000001, ρ < 0.0013, respectively.
17. A population is a large set of elements—too large to take a measurement of each ele-
ment on the variables of interest. In our case, for example, the data obtained from the
life time's reading of a text having done and to be done by a speaker forms a population
and (an) actually recorded reading(s) of the speaker is (are) a sample of the population.
Incidentally, "population coefficient" is the coefficient to be calculated by all the con-
cerned elements in that population.
18. The recordings of the Japanese readings were made in 1995 by Miss K. Mori, Mr Y. Fuji-
no and Mr K. Nanjö, the then graduate students of Konan University, thanks to the in-
termediary act of Prof. K. Imai, Gakushuin University and the English readings were
recorded at the Phonetics Laboratory, Oxford University in 1996 by Mr Neil MacPhee,
an instructor of English at Gakushuin University.
19. In this paper the more popular binary system of feature representation is employed.
20. Population mean, population variance: the mean, the variance, respectively, as a param-
eter in a population.
21. The 95% confidence interval is the interval somewhere of which the value of the in-
ferred parameter is believed to fall with the probability of 0.95.
22. Data values will be shown in the form of (*, y), as is done in mathematics.
23. The terms of a regression consist of the estimators of the parameters in the population
and the y-values are inferred predictions.
24. (11) is significant at the 8 % level, while the other regressions are significant at the 5 %
level. If ρ < 0.05, (11) is not significant, which means that the x-term does not exist and
that foot durations are expected to be somewhat constant in the range 4 < χ < 6.
25. If the acceptance of the null hypothesis about the intercept of the Japanese regressions
had been an error of judgement, the correct predicted values and the correct ratios—
the latter being in b r a c k e t s - w o u l d have been (1, 283) [1.0], (2, 527) [1.86], (3, 770)
[2.72], (4,1014) [3.58], (5,1257) [4.44], (6,1501) [5.30] for 3J's and (1,286) [1.0], (2,566)
[1.98], (3,846) [2.95], (4,1126) [3.94], (5,1406) [4.92], (6,1686) [5.90] for Jf2. The result
is not very different.
References
Abercrombie, David
1960 "'Stress' and some other terms". Paper given to the National Phonetic Collo-
quium, Leeds, March 1960. Produced in Work in Progress, 9 (1976,51-53) and
reproduced in Abercrombie (1991, chap. 10).
1964a "A phonetician's view of verse structure", Linguistics, 6. Reproduced in Ab-
ercrombie (1965, chap. 3).
1964b "Syllable quantity and enclitics in English", in Abercrombie et al. (1964,216-
222). Reproduced in Abercrombie (1965, chap. 4).
1965 Studies in Phonetics and Linguistics. Ronald Mackin and Peter Strevens (Gen-
eral eds.), Language and Language Learning, 10. London: Oxford Univ. Pr.
1967 Elements of General Phonetics. Edinburgh: Edinburgh Univ. Pr.
1968 "Some functions of silent stress", Work in Progress, 2. Reproduced in Aitken
et al. (1971,147-56) and Abercrombie (1991, chap. 9).
1991 Fifty Years in Phonetics. Edinburgh: Edinburgh Univ. Pr.
Abercrombie, David, D. B. Fry, P.A.D. MacCarthy, N. C. Scott and J. L. M. Trim (eds.)
1964 In Honour of Daniel Jones: Papers Contributed on the Occasion of his Eighti-
eth Birthday 12 September, 1961. London: Longmans.
Aitken, A. J., Angus Mcintosh and Hermann Pálsson (eds.)
1971 Edinburgh Studies in English and Scots. London: Longman.
Catford, J. C.
1992 "Caucasian phonetics and general phonetics", in Paris (1992,193-216).
Crystal, David (ed.)
1982 Linguistic Controversies: Essays in Linguistic Theory and Practice in Honour
of F. R. Palmer. London: Edward Arnold.
Dauer, R. M.
1983 "Stress-timing and syllable-timing reanalyzed", Journal of Phonetics, 11, 51-
62.
Fletcher, Janet
1991 "Rhythm and final lengthening in French", Journal of Phonetics, 19,193-212.
Halliday, M. A. K.
1970 A Course in Spoken English: Intonation. London: Oxford Univ. Pr.
Hammerich, L. L., Roman Jakobson & Eberhard Zwirner (eds.)
1971 Form and Substance: Phonetic and Linguistic Papers Presented to Eli Fischer-
J0rgensen 11th February, 1971. Copenhagen: Akademisk Forlag.
Jones, Daniel
19507 An Outline of English Phonetics. 7th ed. Cambridge: W. Heffer & Sons, Ltd.
Lehiste, Ilse
1977 "Isochrony reconsidered", Journal of Phonetics, 5, 253-63.
MacMahon, M. K. C.
1998 Basic Phonetics. Glasgow: Univ. of Glasgow.
356 Yosihiro Masuya
Masuya, Yosihiro
1979 "Syllable and 'chestable'", The Journal ofKonan University, Faculty of Letters,
34,1-6. Reproduced in Masuya (1997, chap. 6)
1994 "Isochronicity of Lowland Scottish English", The Journal ofKonan Universi-
ty, Faculty of Letters, 33 <English Studies>, 108-27. Reproduced in Masuya
(1997, chap. 13).
1997 Phonetics and Phonology: Selected Papers. Tokyo: Kobian-Shobo.
Nakatani, Lloyd H., Kathleen D. O'Connor & Carletta H. Aston
1981 "Prosodie aspects of American English speech rhythm", Phonetica, 38, 84-
106.
Odé, Cecilia & Wim Stokhof (eds.)
1997 Proceedings of the Seventh International Conference on Austronesian Linguis-
tics, Leiden, 22-27 August, 1994. Amsterdam: Rodopi.
Paris, Catherine (ed.)
1992 Caucasologie et mythologie comparée, Actes du Colloque international du
C.N.R.S.-/V'- de Caucasologie (Sèvres, 27-29 juin 1988). Paris: Peeters.
Roach, Peter
1982 "On the distinction between 'stress-timed' and 'syllable-timed' languages", in
Crystal (1982, 73-79).
Shen, Yao & Giles G. Peterson
1962 "Isochronism in English". Studies in Linguistics, Occasional Papers, 9. Buffa-
lo, New York: Department of Anthropology and Linguistics, University of
Buffalo.
Uldall, Elizabeth T.
1971 "Isochronous stresses in R. P.", in Hammerich et al. (1971, 205-210).
van Zanten, Ellen & Vincent J. van Heuven
1997 "Effects of word length and substrate language on the temporal organisation
of words in Indonesian", in Cecilia Odé & Wim Stokhof (eds.) (1997, 201-
213).
Wenk, Brian J. and François Wioland
1982 "Is French really syllable-timed?", Journal of Phonetics, 10,193-216.
Williams, Briony and Steven M. Hiller
1994 "The question of randomness in English foot timing: a control experiment",
Journal of Phonetics, 22,423-439.
sC Clusters as Complex Segments: Evidence from
the Contrastive Phonology of English and Japanese
Noriko Yamane
Nagoya Bunri University
1. Introduction
The purpose of this paper is to present new findings about the uniqueness of
sC clusters (i.e., Is/ and voiceless stop consonant) from the aspect of English
historical phonology and Japanese loanword phonology. It is well-known
that sC clusters behave differently from other 'normal' consonant clusters,
and instead show similar characteristics with those of complex segments.
The work on complex segments has been extensively developed by Sagey
(1986), Lamontagne (1993), van de Weijer (1996), among others, who share
a certain consent that a complex segment designates itself as a single seg-
ment which consists of a combination of manner and/or place of articulation,
whose sequence may sometimes violate the sonority hierarchy. Since sC
clusters consist of both [+ continuant] and [-continuant], and violate sonor-
ity hierarchy in the onset position,1 the latter two researchers assume that sC
clusters are categorized as complex segments. A lot more evidence for the
assumption has been presented both empirically and theoretically. Howev-
er, little is known about how sC clusters behave in historical phonology and
loanword phonology. I would like to address two independent claims, (i) In
English historical phonology, not all sC clusters behave uniformly; the di-
achronic survey on sC clusters reveals asymmetrical behavior according to
the point of articulation of the voiceless stop part, (ii) In Japanese loanword
phonology, the complex status of sC is not particular to Present-day English
(henceforth, PE) or the source language, but is also reflected in the target
language of Japanese; this will be clear from how sequences of consonants
are realized in gemination. Both findings will support the idea that sC clus-
ters in PE should be analyzed as complex segments.
2. sC clusters in English historical phonology
A number of researchers have assigned monosegmental status to all sC

clusters (Fujimura 1997: 95, Kohler 1976:151, Fudge 1969b: 268,273, Sel-
358 Noriko Yamane
kirk 1982: 347-349, among others), whether they are in the onset or the
coda position. This treatment seems to be reasonable for PE. What I
would like to argue in this section is that this is not always correct, from a
diachronic viewpoint. Specifically, the status of sC clusters has shown an
Cor/Lab asymmetry in the coda position.
(1) Inventory of sC as Complex Segments

Position sC OE ME ModE
Onset [st] / • /
[sp], [sk] / • /
Coda [st] / / ( / = observed)
[sp], [sk] / (Yamane 1999a)
As the above figure shows, sC clusters in the onset, whether they are coro-
nal sC clusters (i.e. st) or non-coronal sC clusters (i.e. sp and sk), have al-
ways behaved like complex segments from OE through PE, but they have
not in the coda position. In OE, no coda sC clusters show complex status,
but focusing on the diachronic analysis of sC in the coda position, coronal
sC clusters begin to show monopositional status, historically earlier than
non-coronal sC clusters. The chronological point when this Cor/Lab
asymmetry is observed is ME. It is not until ModE that this asymmetry
starts to disappear, or that sp and sk as well as st start to behave as one
unit. The observation outlined here will be clarified in the discussion from
section 2.1. through section 2.4.
2.1. Old English (ca. 7 c - c a . 1100)
OE alliteration is often adduced as evidence for the assumption that sC

clusters are one unit, but this is only useful for its status in onset position
(cf. Fujiwara 1990: 200-209). Furthermore, no conclusive proof has been
put forward that sC clusters in coda position behave like monosegmental
units similarly to those in onset position. Rather, according to Kim
(1986), phonological phenomena in OE, such as High Vowel Deletion
(henceforth, HVD), Resolution, and Metathesis, indicate that sC clusters
in coda show bisegmental status. Let us take up her examples of HVD
below.
At the level of affixation, high vowels are deleted only after stem-inter-
nal superheavy syllables, i.e., after three moras.
sC Clusters as Complex Segments 359
(2) H V D after 3 moras
(i) Input a. freni 'make' b. de:m 'judge'

Affixation [[f r e m ] i ] de [[d e : m ] i ] de
(Past sg.) 1 I M I
μμ μμμ
Output frem-í'-de de:m-< >-de
Input a. scip 'ship' b. word 'word'

Affixation [s c i ρ ] « (wordju
(Plural) II 11 1
μμ μμμ
Output scip-M word-< >
Thus the words in (2 i a) and (2 ii a) do not undergo HVD, while the words
in (2 i b) and (2 ii b) do.
As for high vowels after the sC clusters, they are also deleted.
(3) H V D after sC clusters

Input a. nest 'nest' b. nest 'nest'
Affixation [nest]« [nest]«
(Plural) IV III
μμ μμμ
Output *nest-u nest-< >
Input a. hysp 'scorn' b. hysp 'scorn'

Affixation [hysp]/ [hysp];'
(Imp. sg.) I \ι 1 11
μ μ μμμ
Output *hysp-i hysp-< >
(cf. Kim 1986:151)
If sC clusters are taken as monopositional segments, dominating one mo-

ra, as in the representations (3 i a) and (3 ii a), the stem-internal syllables
would be counted as only two moras, which wrongly predicts that these
words do not undergo HVD. But H V D must apply to these words, in par-
allel with those in (2 i b) and (2 ii b). This fact suggests that the analyses
in (3 i b) and (3 ii b), in which sC clusters occupy bisegmental positions
owning two moras, are correct. This is one of her arguments that sC clus-
ters in coda were still consonant clusters, not single segments, at least in
the period of O E .
360 Noriko Yamane
2.2. Middle English (ca. 1100-ca. 1500)

This section offers evidence that only st started to behave as a single seg-
ment in the coda in ME. As a start, let us look at the distribution of long
vowels before sC clusters in PE.
(4) Long Vowels before sC Clusters

a. _ st: a: 2 , i:, ei, ai, u:, ου, ao, oi
b. _sp: a:
c. _ sk: a:
Every sC cluster can cooccur with long vowel /a:/, but only st allows a
great variety of long high vowels and diphthongs. Examples are shown
below. 3
(5) Asymmetry of st and the other sC Clusters

(i) _st
a. [a:] blast, cast, caste, fast, last, mast, past, vast
b. [i:] beast, east, feast, least, priest, yeast
c. [ei] baste, chaste, haste, paste, taste, waist, waste
d. [ai] Christ, heist
e. [u:] boost, roost
f. [ou] boast, coast, ghost, host, most, post, roast, toast
g. [au] frowst, joust, oust
h. [oi] foist, hoist, moist
(ii) SP
[a:] clasp, gasp, grasp, rasp
(iii) _sk
[a:] ask, bask, cask, flask, mask, task
The appearance of [a:] before all sC clusters is a reflex of lengthening in

Modern English (henceforth, ModE), and will be discussed later. The
other vowels before st (i.e. (5b-h)) are all results of the Great Vowel Shift
(henceforth, GVS) which applied to long vowels in Late Middle English
(ca. 1350-ca. 1500).4 This fact suggests that vowels before st, different from
sp and sk, must have already been long at that period. But previous re-
search (Minkova 1982, 1985, Yamane 1997, 1998a) concludes that sC
words evaded Open Syllable Lengthening in M E (henceforth, M E O S L )
and were short uniformly. Let us take up a M E word gaspe ('gasp') as an
example. The syllabification is based on Hayes (1980,1989).
(6) Evasion of MEOSL in Words with Medial sC Clusters
a. Syllabification b. Schwa Loss c. Syllable

J Loss d. Stray Coda
Adjunction
g a s p
As the above illustration shows, words like gaspe did not take part in
MEOSL. The reason is that the stray mora μ is compensated by the pre-
ceding vowel, whose syllable is not open unlike tale.5
Following this argument, words with medial st are also expected to be
prevented from lengthening. However, unlike words with sp or sk, words
with st do take part in lengthening. 6 In order to derive lengthening of such
words, we need to posit the following process, which we call Complex
Segment Formation (henceforth, CSF) of /st/ as below.
(7) CSF of st
a. /s/ Delinking b. Stray Code Adjunction
μ μ μ μ
\s t! /I
s t
Based on the assumption that all segments in the rhyme bear a mora in
English, both /s/ and Iii in the coda position own a mora. But Is/ delinks
from the mora as in (7a), and it is reassociated with the coda-final mora
by Stray Coda Adjunction, as in (7b). Thus st changes its status from an
ordinary consonant cluster into a single complex segment. Note that
this procedure targets st only, which shares unmarked place feature
[cor] as well as [voice].7
Given that CSF applies after the original MEOSL triggered by schwa
loss, words with word-medial st will automatically undergo lengthening.
Let us call the original MEOSL 'the first MEOSL', whose application is
limited to the structure CVCV. Let us call the lengthening triggered by
CSF as 'the second MEOSL', which applies to the structure CVst.
362 Noriko Yamane
(8) The second MEOSL (words with st clusters are lengthened)
a. /st/-Complex Segment Formation b. Spreading

σ σ
Words with st do not undergo the first MEOSL, parallel to sp or sk, as

seen in (6). After the vacuous application, words with st only take part in
the second MEOSL. Here the trigger is not the schwa loss, but CSF in
(8a), which feeds spreading as in (8b). Lengthening of these words are not
exceptions any longer, and can be unified as a natural consequence of
Compensatory Lengthening, namely of the second MEOSL. Its output
form [ta:st] hereafter successfully undergoes GVS, showing up as [teist]
in Mod E.
The existence of the second MEOSL which feeds GVS therefore is an
evidence for the monosegmental behavior of st clusters. In ME, then, st
started to become a complex segment, unlike sp and sk.
2.3. Early Modern English (ca. 1500-ca. 1900)
The sC clusters in early ModE behave differently from those in ME.

Lengthening in ModE is also Compensatory Lengthening, a structure
preserving process in terms of the number of moras; however, unlike in
ME, all sC clusters, st, sp and sk, behave uniformly with regard to length-
ening. Lengthening in ModE is called Pre-Voiceless Fricative Lengthen-
ing (henceforth, PVFL), which targets vowels before voiceless fricative
consonants, as shown below.
(9) PVFL
(i) sC Clusters
a. _ st: blast, cast, caste, fast, last, mast, past
b. _ sp: asp, clasp, gasp, grasp, rasp
c. _ sk: ask, bask, cask, casket, flask, mask, task
(ii) Voiceless Fricatives
a. _ s: brass, class, glass, grass, mass, pass
b. _ / : calf, chaff, graph, half, laugh, staff

c. _ Θ: bath, path, wrath
(cf. Dobson 1968: 525-29, Nakao 1985:137-140)
The vowels preceding all sC clusters as well as voiceless fricatives such as

/s/, /f/ and /Θ/ emerged as [a:], as we described in (4) and (5). All sC clus-
ters behave like one unit, without any COR/LAB asymmetry, and they all
attain the status of complex segment. 8
(10) PVFL
(i) Before sC
a. Delinking b. Spreading c. Stray Coda Adjunction
o σ σ
f!\\
μ μ μ
r
μ / μ\ \ μ 7i\
μ μ μ
I ni
g a s p g a s p g a s p
(ii) Before single voiceless fricatives

a. Delinking b. Spreading c. Stray Coda Adjunction
a α σ
μ μ
Ά
μ μ
Ά
μ μ
I I+
p a s p a s p a s
PVFL consists of (a) Delinking of voiceless fricatives, (b) Spreading and

(c) Stray Coda Adjunction. In both examples (10 i) and (10 ii), Spreading
is fed by Delinking, where the stray mora is compensated through vowel
lengthening. And in the process of (c), the delinked Is/ is docked to the
final mora. Here in the example (10 i c), the final mora dominates both of
s and p, where we find sp as complex segment. Thus sC clusters attain
monomoraicity. 9
To summarize, all sC clusters act in a uniform fashion, in that they un-
dergo PVFL in parallel with single voiceless fricatives. On these grounds
I have come to the conclusion that all sC clusters start to acquire mono-
segmental status without COR/LAB asymmetry in ModE.
364 Noriko Yamane
2.4. Present-day English (ca. 1900-)
Evidence in favor of regarding all sC clusters as complex segments is pre-

sented by Lamontagne (1993). Following Sagey (1986), who illustrates
that place features of complex segments behave as if they are linearly un-
ordered, he reports that this uniqueness also holds true of sC clusters. In
other words, the place features of /s/ and those of the stop part of sC clus-
ters are visible to phonological phenomena from both sides. This charac-
ter is witnessed by a cooccurrence constraint in English monosyllables.
According to Lamontagne (1993: 272-73), "a segment cannot cooccur (in
monosyllables) with an .v-stop sequence if it agrees in both place and stric-
ture (i.e., continuancy) with a member of the sequence." For instance,
words with two [ - continuant] coronals such as sfVt and tVif, and words
with two [+ continuant] coronals such as sCVs and sVsC are infrequent.
This means that the OCP is operative on the consonantal tier. But what
should not be overlooked here is that the OCP takes effect even if a con-
sonant intervenes: the OCP is effective in tVsi even if Is/ intervenes, and
also effective in sCVs even if the stop consonant 'C' intervenes.
If we regard sC clusters as true clusters, the intervening consonant will
preclude the OCP effect. Let us clarify this point with the representation
below.
(11) True Cluster Analysis ->• No OCP violation due to linear ordering
a. C s C b. s C C
[ - cont] [+ cont] [ - cont] [+ cont] [ - cont] [+ cont]
Fj [cor] Fj [cor] F Fj
(F = [lab], [cor], [dor]; Fj = [cor])
In both structures above, the highlighted features, relevant to the OCP, do

not satisfy the adjacency condition that the OCP requires. Such structures
are not within the scope of the OCP, and thus the rarity of these words
cannot be predicted. The problem is that such true clusters are strictly lin-
early ordered.
Lamontagne solves this problem with the complex segment analysis as
follows.
(12) Complex Segment Analysis ->• OCP violation thanks to non-linear

ordering
a. *C sC b. *sC — C
PL PL PL PL
I / \ / \ I
[ - cont] [ - cont] [+ cont] [ - cont] [+ cont] [+ cont]
Fj Fj [cor] F [cir] Fj
e.g. *(p, b, m) V sp *sp V (s, ζ, Θ, ô, s, ζ)

*(t, d, η) V st *st V (s, ζ, θ, ö, s, ζ)
*
(k, g, r ] ) V s k *sk V (s, ζ, θ, ö, s, ζ)
In complex segments, two place features behave as if they are linearly un-
ordered. As for the sC clusters, the two place features with their head
[cont] features are assumed to belong to separate tiers, which are domi-
nated by a single place node. The adjacency condition is satisfied here
thanks to this non-linear ordering of the relevant features. Thus the aster-
isked examples below the representations militate against the OCP, and
the infrequency of the examples shown is appropriately explained. 10 We
can see here that this analysis supports the status of complex segments in
both onset and coda.
Based on this argument, it seems reasonable to conclude that sC clus-
ters in PE are monosegmental units in both onset and coda.
2.5. The need for local conjunction
It was observed in the preceding sections (2.1.-2.4.) that historically, in

coda position, sC clusters have gradually merged into one segment; Sp
and sk did not come up as complex segments until st did. Is this direction-
ality of change predictable? How can we capture such a gradual change
of sC? In this section, we argue that the Local Conjunction [henceforth,
LC] (Smolensky 1993,1995) is necessary to capture the variability and di-
rectionality sC clusters have revealed.
As a first approximation, let us introduce the universal markedness
constraint called *CoMPLEX.n This constraint prohibits complex associa-
tion within a certain phonological unit. *COMPLEX can be decomposed
into the following two constraints at least:
366 Noriko Yamane
( 1 3 ) a. * COMPLEX (CODA): Multiple consonants are prohibited in the coda.

b. * COMPLEX (SEG): Multiple place features are prohibited in one segment.
If sC clusters are true consonant clusters, the syllable codas will contain
two consonants, which violates *COMPLEX (CODA). On the other hand, if
the sC clusters are complex segments, the root nodes will contain two
place features, which violates *COMPLEX (SEG). The change of the status
of sC can be captured with the conflict of these constraints.
(14) a. O E b. P E
* COMPLEX ( S E G ) *COMPLEX ( C O D A )
I I
*COMPLEX (CODA) *COMPLEX ( S E G )
4 I
st, sp, sk [st], [sp], [sk]
I f the * COMPLEX (SEG) dominates *COMPLEX (CODA) as in ( 1 4 a ) , s C clus-

ters emerge as consonant clusters. But if the ranking is reversed as in
(14b), sC clusters emerge as complex segments. In other words, the hier-
archy (14a) reflects O E , and the reversed hierarchy in (14b) seems to re-
flect PE. How can this constraint interaction predict the COR/LAB asym-
metry we have observed in M E through ModE? The problem is that the
interaction of these constraints, as it stands, wrongly predicts that sC clus-
ters are uniformly converted into complex segments all at once. The con-
straint of *COMPLEX (SEG) needs to be revised so that the Cor/Lab asym-
metry can be captured.
Let us assume that *COMPLEX (SEG) consists of several constraint fam-
ilies, which prohibit certain combinations of place features. L C is "an
operation in U G by which two constraints governing substructures of a
given local domain are conjoined into a higher-ranked constraint" (Smo-
lensky 1995:1). On the basis of the COR/LAB asymmetry or the dominance
hierarchy of *PL/LAB » *PL/COR (Prince & Smolensky 1993:181), L C pro-
duces the following ranking with three locally-conjoined constraints,
which is universally fixed.
(15) Constraint Families of *COMPLEX (SEG)

* P L / L A B , LAB » * P L / L A B , COR » * P L / C O R , COR
Each constraint is defined as follows.

(16) a. * P L / L A B , LAB: TWO Labial Place features are not associated to one
place node.
b. * P L / L A B , COR: Both a Labial Place feature and a Coronal Place
feature are not associated to one place node.
c. * P L / C O R , COR: TWO Coronal Place features are not associated to
one place node.
The constraint in (16c) is violated when the output of st is a complex seg-

ment, and the constraint in (16b) is violated when the output of sp or sk
is a complex segment. The process making complex segment is motivated
by the promotion of *COMPLEX ( C O D A ) . It can then be captured by the
dominance of *COMPLEX ( C O D A ) over each family of *COMPLEX ( S E G )
(i.e., the constraints in (16a-c)).
Let us see how the constraint interaction succeeds in singling out the
optimal output in each system.
(17) LME: CSF of sí only
*PL/LAB, LAB * P L / L A B , COR »COMPLEX ( C O D A ) *PL/COR, COR
(i) Is ΰ
*
a."®· [st]
b. s t *!
(ii) Is pi
a. [sp] *!
b."®· s ρ *
([ ] indicates the formation of the complex segment.)
In the L M E system, *COMPLEX ( C O D A ) is ranked between * P L / L A B , COR,

and * P L / C O R , COR. When the input is st as in (17 i), candidate (b) as a
true cluster loses, because it violates *COMPLEX ( C O D A ) , but candidate
(a) as a complex segment, which observes * COMPLEX ( C O D A ) , wins. How-
ever, when the input is sp as in (17 ii), candidate (a) as a complex seg-
ment loses because of the violation of * P L / L A B , COR, but candidate (b)
as a true cluster wins. This correctly predicts that CSF is allowed only in
st not in sp or sk.
Next, let us observe the system of ModE as below.
368 Noriko Yamane
(18) ModE: CSF of st, sp, sk

*PL/LAB, LAB * COMPLEX ( C O D A ) * P L / L A B , COR * P L / C O R , COR
(i) Is t/
a.®" [st] *
b. st *!
(ii) Is ρ/
a.·2" [sp] *
b. s ρ *!
In ModE, * COMPLEX ( C O D A ) has been promoted, and dominates * P L / L A B ,

COR as well as * P L / C O R , COR. When the input is st as in ( 1 8 i), candidate
(b) loses because of the violation of *COMPLEX ( C O D A ) , but candidate (a)
as a complex segment wins. Similarly, when the input is st as in (18 ii), can-
didate (b) loses because of the violation of *COMPLEX ( C O D A ) , but candi-
date (a) as a complex segment wins. This constraint interaction correctly
predicts that CSF applies both in st and in sp or sk in ModE.
In summarizing, the diachronic change of sC is captured as follows.
(19) Diachronic Change of sC in Coda

OE LME ModE
* P L / L A B , LAB * P L / L A B , LAB * P L / L A B , LAB
I I
1 1 1
* P L / L A B , COR *PL/LAB, COR *COMPLEX ( C O D A )
I I
1 1 1
* P L / C O R , COR *COMPLEX ( C O D A ) * P L / L A B , COR
I I I
1 1 1
*COMPLEX ( C O D A ) * P L / C O R , COR * P L / C O R , COR
I I 1
st, sp, sk [st], sp, sk [st], [sp], [sk]
(Yamane 1999a)
In the system of OE, *COMPLEX ( C O D A ) is ranked below a block of * C O M -

PLEX ( S E G ) consisting of the three constraint families, so none of the com-
plex segments is allowed. In late ME, the lowest constraint family of
* COMPLEX ( S E G ) , i.e. * P L / C O R , COR, is overridden, by the promotion of
* COMPLEX ( C O D A ) , and only st emerges as a complex segment, leaving sp
and sk aside. In ModE, the second lowest family, i.e. * P L / L A B , COR, is
overridden, by the further upward shift of *COMPLEX ( C O D A ) , so that sp
and sk as well as st emerge as complex segments.12 Thus LC makes it pos-

sible to capture the gradual shift of the status of sC clusters.
The next section shows that the status of sC clusters in PE reflects Jap-
anese loanword phonology.
3. English word-final sC clusters in Japanese loanword

phonology
The status of complex segment is sometimes reflected in loanword phonol-

ogy (Especially the exeptional behavior of sC in loanword phonology is in-
vestigated in Broselow 1 9 8 7 , 1 9 9 1 ) . According to van de Weijer ( 1 9 9 6 : 1 8 5 -
89), sC clusters behave differently from other normal clusters in loanword
phonology in Modern Standard Hindi and Sinhalese, where normal conso-
nant clusters are broken up by epenthesis, while sC clusters receive a pro-
thetic vowel (or remain as they are in Sinhalese). He argues that the resis-
tance of sC clusters to epenthesis follows from their representation below.
(21) * C <V>
/X
[cont] ! [stop]
Place I
Place
This representation has a doubly-linked structure, in which two monova-

lent features, [cont] and [stop], are both associated to a single C-slot. How
the manner features are treated is not at stake here, but important is that
sC clusters owns a single C-slot and a single place node. Van de Weijer
notes that in such a representation, insertion of a vowel would violate the
Line Crossing Constraint proposed by Goldsmith ( 1 9 7 6 ) .
The following sections reveal that Japanese loanword phonology sup-
ports the integrity of sC clusters, which is in accordance with the fundamen-
tal idea as represented above. The analysis we adopt is Correspondence
Theory (McCarthy & Prince 1 9 9 5 ) , in which the resistance to vowel epen-
thesis results from the dominance of the faithfulness constraint below.
(22) DEP-V: Every vowel of the output has a correspondent in the input.
This constraint militates against insertion of vowels in general.

370 Noriko Yamane
As for the resistance to consonant epenthesis, the constraint below is

dominant.
(23) DEP-C: Every consonant in the output has a correspondent in the input.
In sections 3.1. and 3.2., we generalize the gemination pattern of Japa-

nese. Section 3 . 3 . shows that D E P - C interacts with faithfulness constraints
on consonant clusters. And Section 3.4. discusses the interaction of DEP-
V with other constraints.
3.1. The generalization of gemination: featural contrast
Crucial evidence is found in the behavior of English word-final consonant

clusters in Japanese loanword phonology. When English words are bor-
rowed into Japanese lexicon, one of the final consonant clusters is some-
times realized as a geminated form. The phonological condition for the
gemination will be generalized as follows:
(24) In word-final consonant clusters CjC2(C3)# ... it is always C, which

undergoes gemination. There is no example whose C2 or C3 is real-
ized as a geminate (e.g. script > *sukuripufto). Furthermore, C, has
to be 'a potential geminate consonant' ((Kawagoe 1995: 53) (my
translation, Ν. Y.))
This paper will focus on words with sequence C1C2# (C¡C2C3# is not taken
up here). Note that the 'potential geminate consonants' notated above in-
dicate stops, affricates and fricatives, which are all [ - sonorant] conso-
nants. The generalization above appropriately captures the fact that the
sequence whose C t is a sonorant consonant (either nasal, liquid or glide)
is not realized in gemination. However, this condition overgenerates the
gemination patterns; even if Q is a [- sonorant] consonant, it does not al-
ways turn up with geminated forms. So the sequence in question actually
falls into two types; Gemination CC and Anti-Gemination CC. In order
to predict this asymmetrical distribution, attention must be paid to the
number of contrasting features of place, [continuant] and [sonorant], be-
tween Cj and C2. The figure below presents the grouping of Gemination
CC and Anti-Gemination CC.
(25) Number of Featural Contrasts (= N) of Word-Final CC in English
(i) Gemination CC
C,C 2 Contrasting features Ν Examples (English > Japanese)
a. -Cs ks pi, [cont] 2 box > bofcusu, wax > wa£:usu

ps pi, [cont] 2 pops > pop:usu, chips > chip:usu
b. -CI kl pi, [son] 2 tackle > ta/c:uru, knuckle > na/c:uru
pi pi, [son] 2 apple > ap:uru, nipple > nip:uru
si [cont], [son] 2 castle > kyaj:uru, hustle > has:uru
fl pi, [cont], [son] 3 waffle > wa/iuru, raffle > ra/:uru
c. -Cn sn [cont], [son] 2 lesson > les:un, listen > lLs:un
fn [cont], [son] 2 fashion > faj:on, mission > miJ":on
(ii) Anti-Gemination CC
C,C 2 Contrasting features Ν Examples (English > Japanese)
a. -Ct kt p· 1 sect > seduto, act > tafcuto

pt pl 1 kept > keputo, script > sukuriputo
b. -Cl tl [son] 1 kettle > ketoru, little > riforu
c. -Cn tn [son] 1 mutton > maion, button > bofan
(pi = place, [cont] = [continuant], [son] = [sonorant])
Notice that even if Q is a [ - sonorant] consonant, Anti-Gemination CC

(25 ii) systematically avoids gemination (e.g. sect > seduto, *seA::uto), in
contrast to Gemination CC (25 i) (e.g. box > bo/c:usu). This fact reveals
that the gemination pattern is not simply determined by the feature of C l5
but by the the number of featural contrasts between Q and C 2 (for the de-
tails, see Yamane & Tanaka 2000).
The significance of the interaction between clusters is also demonstrated
in Quebec French (Côté 1997). In the word-final consonant clusters Q Q ,
the deletion of C2 depends on the number of featural contrasts between
them, where [voice], place, [continuant] and [sonorant] are counted.
In Quebec French, the fewer the contrasting features are, the more
likely C 2 is to drop: when the contrast is more than two, C 2 never drops,
when it is two, C 2 optionally drops, and when it is only one, C 2 automati-
cally drops. The deletion serves to maximize the acoustic salience of Q ,
so that C¡ will be easily perceived.
As for the Japanese gemination patterns, this kind of enhancement
seems also to be true. The difference between the Quebec French dele-
tion and the Japanese gemination is outlined below.
372 Noriko Yamane
(26)
Quebec French Deletion Japanese Gemination
Input ctc2#
Phonological Process C 2 Deletion Cj Insertion
Target non-salient C 2 salient Cj
Featural Contrast 5Ξ 2
Output Q< >
(#: word-boundary)
French deletes non-salient C2, when the featural contrast is two or less,
producing Cj< >. But Japanese inserts salient C t when the contrast is two
or more, producing C 1 <C 1 >C 2 .
The apparent distinction is unified in the theory of featural contrast.
Both phenomena serve to enhance the acoustic salience of Q . The anal-
ysis for Japanese will be presented in section 3.3.
3.2. Evidence for sC as complex segments
Based on this observation, let us consider the case of sC clusters. If we

treat sC as true consonant clusters, the contrasting number would result
as below.
(27)
Contrasting features Ν Type of CC
st [cont] 1 => Anti-Gemination CC
sk pi, [cont] 2 => * Gemination CC
sp pi, [cont] 2 => * Gemination CC
St would have one featural contrast, but sk and sp would have two. Using
the featural contrast approach, the number of contrast, two, is supposed
to trigger gemination.
However, sk and sp systematically avoids gemination as in the exam-
ples below.
(28)
Contrasting features Ν Examples (English > Japanese)
st [cont] 1 <= cast > kyasufo, test > te.suίο
sk [cont] 1 <= desk > deiufcu, task > tasu&u
sp [cont] 1 <= crisp > kumupu, grasp > gurasupu
This fact indicates that all sC clusters should be uniformly grouped into
Anti-Gemination CC whose featural contrast is less than two. If we adopt
the idea that sC clusters are complex segments, whose featural contrast is
only one (i.e., [cont]) as the apparent distinction of place features are nul-
lified by the single place node that they share, based on the representation
in (21) for instance, the gemination pattern is correctly predicted.
However, in contradiction to the representation in (21), sC clusters in
Japanese are finally broken up with vowel epenthesis, unlike Moderrn
Standard Hindi and Sinhalese. This point will be discussed in 3.4.
The next section will analyze the Gemination CC and the Anti-Gemi-
nation CC respectively in the framework of OT, where the status of sC
clusters will be further confirmed.
3.3. Maintain contrast as a faithfulness constraint
This section presents an analysis of gemination in the framework of OT,

in combination with the Dispersion Theory of contrast (Flemming 1995).
The following constraints are adopted:
(29) Maintain Contrast constraints

a. CONTRAST = N : A segment is differentiated from the following one
by Ν articulatory contrast(s).
b . CONTRAST = 1 » CONTRAST = 2 » . . . » CONTRAST = Ν
(cf. Côté 1997: 65)
These constraints proposed by Côté are articulatory-based ones, modi-

fied from the auditory-based "Maintain Contrast" constraints Flemming
originally proposed. Adopting articulatory-based constraints seems to
enable us to capture the more general notion of phonetic distinctiveness,
whose maximization serves to enhance acoustic salience. Then the con-
straint ranking in (29b) encodes that the less distinctive the contrast, the
greater the violation. Briefly, the less contrastive strings of segments such
as sC clusters are comparatively unmarked sequences for the part of
speakers, but marked for the part of hearers. 13
Gemination follows from the effect of the constraint below.
(30) MAXDUR(CJ): Maximize the duration of Q (cf. Flemming 1995: 94)
This was originally proposed as an auditory-based constraint, but as Flem-

ming (1995: 96) notes, "Prolongation of an articulatory gesture will result
374 Noriko Yamane
in auditory differences which may then be exaggerated for purposes of en-

hancement (emphasis mine, Ν. Y. )". This statement may enable us to ad-
mit the constraint (30) as one of the articulatory-based constraints, in tan-
dem with other "Maintain Contrast" constraints raised in (29).
First let us consider the interaction of M A X D U R ( C J ) and D E P - C ( = 2 3 ) .
The relevant two constraints are contradictory, as D E P - C prohibits C-in-
sertion, while M A X D U R ^ ) forces it.
(31) a. DEP-C b. MAXDI^CJ)
MAXDUR(CJ) DEP-C
i i
*C1<C1>C2 c , < c > c -2
If D E P - C dominates MAXDUR as in (31a), gemination is prohibited. If the

ranking is reversed as in (31b), the gemination is forced. Since Japanese
loanword phonology allows gemination, which always involves consonant
insertion, D E P - C must be overridden by M A X D U R ( C J ) . Thus, the ranking
should be set as in (31b).
The next question to be considered is the ranking between MAX-
DUR(CJ) and M A X . M A X is another faithfulness constraint, which prohibits
phonological deletion, opposed to DEP. In loanword phonology, informa-
tion in the input tends to be preserved in the output, as long as it conforms
to the syllable structure constraints in the target language. Then MAX is
considered to be relatively highly ranked. But how does it interact with
MAXDUR(CJ)?
(32) a. MAX b. MAXDURÍQ)
MAXDUR(Q) MAX
The ranking in (32a) only states that phonological deletion is prohibited

even when gemination is allowed. The reversed ranking as in (32b) states
that gemination is forced without phonological deletion. Both of the
rankings above hold true of Japanese loanword phonology. Thus there is
no clue from which to make a choice between them.
One possible solution to the problem is that they are unranked as be-
low.
(33) MAXDUR ( C J ) MAX
V
DEP-C
Even if the ranking problem is answered in such a way, this ranking over-
generates the gemination; contrary to the fact, every input cluster CjC 2 in
the input emerges as C j < C j > C 2 in the output. Furthermore, it does not
tell us what kind of information and how much information in the input
is preserved in the output of the loanword phonology.
A solution to these problems can be found with the reconsideration of
MAX. MAX plays a role in specifying contrasts, in that it favors preserving
contrasts of featural specification of inputs (see Kirchner 1998). Accord-
ing to Flemming (1995: 44), this kind of function of faithfulness is taken
over by the "Maintain Contrast" constraints.
(34) Correspondence Theory Dispersion Theory

MAX => Maintain Contrast
Thus MAX can be rephrased into Maintain Contrast constraints, or CON-

TRAST - Ν (= 29a). Taking into consideration the fact that Maintain Con-
trast includes the function of M A X , we can revise CONTRAST = Ν as follows:
(35) CONTRAST = Ν (revised)

Ν (= the number of featural contrasts of C, with regard to the follow-
ing consonant) in the input is preserved in the output.
This constraint is violated whenever gemination occurs. Let us observe

the prediction.
(36)
c,c2
1/ CONTRAST = (n)
(n)
a Q Q
- 1/
(n)
b. Ci C, C,
*
1/
(0)
The ungeminated candidate (36a), as input specification (n) is preserved

in the output, conforms to CONTRAST = (n). But the geminated candidate
376 Noriko Yamane
(36b), whose Q is followed by the reduplicated C 1 ; violates CONTRAST =

(n), because (n) in the input is changed into (0) in the output.
Given that M A X is replaced with CONTRAST = N, the ranking between
M A X D U R ( C J ) and the faithfulness constraint naturally emerges. Further-
more, it tells us what kind of phonological information in the source lan-
guage is preserved in the target language: minimal articulatory contrast.
The problem of the overgeneration is also solved.
As noted in the previous section, the susceptibility of gemination de-
pends on the number of articulatory contrasts between C, and C2. And re-
call that Maintain Contrast constraints are decomposed into subcon-
straints (= 29b), ranked as below.
(37) CONTRAST = 1
[ ' ] MAXDUR(CÌ)
I (cut-off point indicator)
CONTRAST = 2
CONTRAST = 3
Gemination does not occur when the contrast is one, and occurs when it
is two or more. So the cut-off point for the susceptibility of gemination
can be drawn between CONTRAST = 1 and CONTRAST = 2 . Overriding C O N -
TRAST = 2 and CONTRAST = 3 , M A X D U R ( C J ) serves as an indicator of the cut-
off point.
Based on the discussion above, the dominance relations among MAX-
D U R ( C ] ) , CONTRAST = Ν and D E P - C will emerge as below. (The ranking in
(31b) has been added to that in (37).)
(38) CONTRAST = 1
MAXDUR(CJ)
I \
CONTRAST = 2 DEP-C
CONTRAST = 3
Let us see if the dominance relation given here correctly produces the
output. (For graphic simplicity, the candidates are expressed with the rel-
evant part only, i.e. the realization of C,C 2 .)
Observe Gemination CC first.
(39) Gemination CC
CONTRAST = 1 MAXDUR(C!) CONTRAST = 2 CONTRAST = 3
(i) ' w a f f l e ' ( 3 )

a. fr (3) *!
*
b. •»- ffr (0)
(ii) ' b o x ' ( 2 )
a. ks (2) *!
*
b. kks (0)
The candidates here are all immune to CONTRAST = 1, because they have
no correspondence of Ν = 1 in the inputs: Candidates in (39 i) with Ν = 3
in the input are submitted only to CONTRAST = 3 for evaluation, and those
in (39 ii) with Ν = 2 in the input are submitted only to CONTRAST = 2 . As
for the next constraint M A X D U R ( C ] ) , it is violated in ungeminated candi-
dates (39 i a) and (39 ii a), while it is not by the geminated ones (39 i b)
and (39 ii b). Thus, the geminated candidates emerge as winners.
Next let us examine the cases of Anti-Gemination CC.
(40) Anti-Gemination CC
CONTRAST = 1 MAXDUR(C,) CONTRAST = 2 CONTRAST = 3
(i) 'tact' (1)

*
a. kt (1)
b. kkt (0) *!
(ii) 'test' (1)

*
a. «• st (1)
b. sst (0) *!
The ungeminated candidates (40 i a) and (40 ii a) satisfy CONTRAST = 1,

but the geminated candidates (40 i b) and (40 ii b) are marked in CON-
TRAST = 1 since Ν = 1 in the input is changed into zero in the output. The
fatal violation here rules out the geminated candidates, and makes the un-
geminated candidates (40 i a) and (40 ii a) winners.
The featural contrast analysis developed here solves the problems we
have addressed. The first question was about the dominance relation be-
tween M A X and M A X D U R ( C , ) : M A X D U R ( C T ) interrupts a series of decom-
posed constraints of M A X or CONTRAST = N. The second problem was the
overgeneration of gemination: the asymmetrical distribution of Gemina-
tion CC and Anti-Gemination CC can be captured appropriately. The
third question was the type of phonological information to be preserved:
378 Noriko Yamane
the minimum featural contrast (N = 1) in the input. Thus the featural con-
trast approach in OT not only captures the gemination patterns, but also
answers the problems.
Furthermore, the analysis here provides us with a significant advan-
tage: it elucidates the mechanism of gemination. Notice that the gemina-
tion is expressed as a process which turns two or more featural contrasts
into zero. The reason is clear: The presence of many featural contrasts re-
quires speakers to make much articulatory effort, so speakers attempt to
minimize their number. Gemination can then be considered as an articu-
latory-based phonological process which favors speakers.14
3.4. Vowel epenthesis

The treatment of sC as complex segment needs further discussion in light
of vowel epenthesis. The question is why sC is broken up not by a conso-
nant, but by a vowel (graphically, *s<s>C vs. s<V>C), and how this para-
dox on the integrity of sC is explained.
We assume that the apparent contradiction results from the conflict of
universal constraints ranked in a single hierarchy. In loanword phonology,
the illegal sequence is altered into a legal one, in order to conform to a
series of well-formed syllable structure constraints in the target language
(cf. Katayama 1998, Kitahara 1996). Then the Maintain Contrast con-
straints we have developed are outranked by syllable structure con-
straints as below.
(41) Syllable structure constraints > Maintain Contrast
Once a series of Maintain Contrast is invalidated in this way, even the

minimum featural contrast does not have to be maintained. Then the in-
tegrity of sC can be broken up with vowels, according to the requirement
of syllable structure constraints.
Among the syllable structure constraints with regard to coda, CODA-
C O N D plays a crucial role.
(42) CODACOND
*C]a
[place] (Itô 1986)
CODACOND is violated if the coda consonant has a singly-linked place.15

Japanese codas are restricted to either a placeless nasal (e,g. hoN) or a
first part of a geminate (e.g. kitto), or a nasal which is homorganic to the

following consonant (e.g. tombo) (cf. Katayama 1998:44). All of these co-
das conform to C O D A C O N D .
Recall that D E P - V ( 2 2 ) militates against vowel insertion. Since sC in
coda is broken up with vowel epenthesis, D E P - V is outranked by C O D A -
C O N D as below.
(43) CODACOND » D E P - V
Combining the constraint hierarchies (38) and (43), based on the assump-
tion in (41), the overall ranking emerges as below.
(44) CODACOND
DEP-V
CONTRAST = 1
MAXDUR^)
I \
CONTRAST = 2 DEP-C
CONTRAST = 3
Let us make sure if the constraint hierarchy above allows vowel epenthe-
sis to occur. (For the lack of space, the constraints below M A X D U R ( C , ) are
omitted.)
(45)
CODACOND DEP-V CONTRAST = 1 MAXDUR(C,)

*
a. test *!
b. te.su.t *! • *
* *
c. <*~ te.su.to
d. tes.sto *
*!
e. tes.su.to * *!
The candidates (45a) and (45b) violate C O D A C O N D , because their codas

own a single place. The rest of the candidates (45c-e) escape its violation,
and are submitted to D E P - V ( I assume here that the violation of D E P - V is
computed not by the number of epenthetic vowels but by the presence or
380 Noriko Yamane
absence of them. So D E P - V is considered to be a categorical constraint.).

But they all violate D E P - V , and are submitted to the next lower constraint,
CONTRAST = 1 . Candidates (45d) and (45e) lose here, and the ungeminated
candidate (45c) will win.
The constraint hierarchy (44) is thus assured by the correct result. The
reason why the sC is broken up with a vowel, while it is not with a conso-
nant, is answered in OT: the dominance of syllable structure constraint
above Maintain Contrast. 16
The validity of the constraint hierarchy may be further mirrored in the
distributional difference of phonetic variation: In casual speech, a vowel
is often deleted in sC clusters as in (46a), but is not for normal consonant
clusters as in (46b).17
(46) a. [tesuto]~[tesio], [tasuku]~[tas/cu], [kurisupu]~[kurispu].

b. [takuto]~*[taAto], [keputo]~*[ke/?io]
This distributional difference is correctly predicted from the constraint

hierarchy (44), as below. (In order to make this point clear, we make the
new candidate (f) run for the tableau.)
(47)
(i) 'test' CODACOND DEP-V CONTRAST = 1 MAXDUR(C,)

a. test *! *
b. te.su.t *! * *
c. te.su.to * *
d. tes.sto *
*!
e. tes.su.to * *!
f. <*• tes.to * *
(ii) 'tact' CODACOND DEP-V CONTRAST = 1 MAXDURÍCJ

a. takt *! *
b. ta.ku.t *! * *
c. ta.ku.to * *
d. tak.kto *
*!
e. tak.ku.to *
*!
f. tak.to *! * *
Multiple winners are allowed for sC as in (47 i), but are not for normal con-
sonant clusters as in (47 ii).18 This is thanks to the dominance of the sylla-
ble structure constraint or C O D A C O N D , which makes candidate (47 i f) op-
timal, in that /s/ in coda has a doubly-linked place associated with the
following I I I . Thus hierarchy (44) accords with the vowel deletion in casu-
al speech as well.
Vowel epenthesis therefore follows from the dominance of syllable
structure constraints. The constraint hierarchy subsumes the observation
that the integrity of sC is maintained against consonant insertion (thus
s<s>C is ill-formed), but not against vowel insertion (thus s<V>C is well-
formed). It is also advantageous that the analysis here captures the vowel
deletion in casual speech, which may suggest that the complex status of
sC survives at some level in the output.
4. Conclusion
We have discussed the unique behavior of sC clusters. In the former part,
dealing with English, the diachronic survey revealed the C O R / L A B asym-
metry. It was found that st started to behave like one unit, chronologically
earlier than sp and sk. Based on the observation that the status of sC var-
ies according to each synchronic system, we shifted our attention in the
latter part of this paper to another synchronic system, that of Japanese
loanword phonology. The investigation of gemination patterns forced us
to categorize sC as Anti-Gemination CC, whose featural contrast is only
one. This number is ascribed to the unique representation of the complex
segment, whose place features are nullified by a single place node, unlike
normal consonant clusters. The treatment of sC in such a way enabled the
correct prediction of gemination patterns, using the theory of featural
contrast in tandem with OT. The analysis developed here has additional
advantages. First, the mechanism of gemination is explained in terms of
minimization of articulatory effort and maximization of acoustic salience.
Second, the prima facie paradox, i.e. a consonant does not break up the
integrity of sC but a vowel does, is resolved in OT. Lastly, it makes it vis-
ible that the status of complex segment is mirrored in the output of Japa-
nese loanword phonology.
Acknowledgements
This work partially contains the basic insights of Yamane & Tanaka (2000), but is a revised and
extended version giving an OT-based account for gemination of word-final consonant clusters
in Japanese loanword phonology. I wish to express my gratitude to the following people for
their helpful discussions and suggestions on several points in the paper: John Kingston, John
382 Noriko Yamane
Me Carthey, Lisa Selkirk, Shin-ichi Tanaka, Jeroen van de Weijer, and two anonymous review-
ers. Thanks also go to James Landkamer, who kindly gave me helpful suggestions on grammat-
ical errors. Last but not least, I thank the editors, Jeroen van de Weijer and Tetsuo Nishihara,
for having let me join this valuable project. Of course, I alone am to blame for any remaining
inadequacies or misconceptions therein.
Notes
1. This point has made researchers assume that sC in the onset is 'extrasyllabic'. But a dis-
cussion on the extrasyllabic approach is beyond the scope of this paper, which focuses on
sC in coda position, not onset position. As for the 'extrametrical' approach, we need to
add some comments. It would have difficulty in treating the second MEOSL and PVFL
in section 2.2. through 2.3.; e.g. the bimoraic tas<t> would turn out to be the trimoraic
ta:s<t>, which would militate against lengthening based on mora preservation.
2. Wells (1990) describes [a:, t œ II se]· The RP form is [a:]. Pronunciations marked with the
symbol t are widespread in English among educated speakers, but nevertheless judged
to fall outside RP. In England, [ae] is a localized northern form (though it is standard in
AmE) (see Wells 1990: xii). [a:] in sp and sk also has the same variety.
3. Examples are collected from Fergusson (1985) and Gunji (1991), and I cross-referenced
their pronunciation with Wells (1990). Rare examples are omitted.
4. [oi] is an exception in the GVS. It traces back to ME /i:/, which was supposed to result in
[ai] through GVS.
5. The first MEOSL applies to tale in the following way.
a. Syllabification b. Schwa Loss c. Syllable Loss Compensatory

Lengthening
t a : 1
What compensates the stray mora depends on the weight of the stressed syllable. Com-
pensation by a vowel occurs when the preceding syllable is light as above. But when it is
already heavy as in (6c), the stray mora is compensated by a final consonant as in (6d).
Notice that the word-medial sC in (6a) is heterosyllabic. But if the word-medial sp is
also taken as a complex segment, two ways of syllabification are logically possible. One is
ga. sps and the other is gasp.3. The former would make the initial syllable light, then the
stray mora delinked from schwa should be compensated by a vowel, but it is contrary to the
fact. The latter syllabification would create a heavy syllable, so the compensation by a vow-
el is not expected. But the uncompensated stray mora here needs to disappear without rea-
son, which would weaken the principle of structure preservation. Thus both syllabifications
are problematic, and we do not regard word-medial sC clusters as complex segments.
6. Among normal consonant clusters, obstruent plus liquid clusters such as able, bible, fable
underwent lengthening (see Jordan 1974: sec. 25, Remark 2 and Nakao 1985:132-3). This
is because the initial syllable is formed to be open by Sonority Contour Principle, (cf. Ya-
mane 1998a). Thanks to Jeroen van de Weijer for addressing this question.
7. It may be ascribed to the OCP effect, as an anonymous reviewer indicates, but for now
the statement, "the more similar the segments, the stronger the interaction" (Itô &
Mester 1995:205)) best describes my intuition. An empirical evidence is also found in OE
Homorganic Lengthening of Id (81.1%) (Minkova & Stockwell 1992). We assume that
the strong interaction oí Id, sharing [cor] and [voice], triggers lengthening (Yamane 1997).
8. St words which underwent PVFL are limited to the exeptions to the second MEOSL.
So st in these words did not acquire monomoraicity until PVFL began. Thanks to an
anonymous reviewer for addressing this point.
9. PVFL can be considered to be a postlexical change. Yamane (1996a, b) demonstrates
that a constraint-based system can treat this kind of change as a constraint-reranking.
10. Takeru Honma (p.c.) points out that if we assume that the OCP targets only specified
features, with the underspecification of [cor], we will gain an account for Cor/Lab asym-
metry in the sense that tVst type is found more often than *pVsp and *kVsk (cf. Davis
1991, Cairns 1988). But the OCP effect on the continuant part of coronals as in (12b) (i.e.
the sCVs type is less frequent than the sCVf type of words) cannot be captured, unless
[cor] is specified. A fuller study of the underspecification problem lies outside the scope
of this paper, but is an important topic. For the details, see McCarthy & Taub (1992).
11. * COMPLEX is also sustained by the implicational universale described by Prince & Smo-
lensky (1993:186): "If the segment inventory of a language includes a complex segment
with primary place π and secondary place ψ, it has a simple segment with place π and a
simple segment with place ψ."
12. Each system of synchronic phonology outlined agrees with the generalization "banning
the worst of the worst" (Prince & Smolensky 1993:180). From a diachronic viewpoint,
the way of constraint reranking here conforms to the principle of family invalidation;
"it is always the lowest constraint in the family of ranked constraints that is overridden
first when that family is subject to change" (Yamane 1996b). The earlier realization of
st complex follows from this.
13. sC in onset is an unmarked sequence, from a typological viewpoint (Morelli 1998). But
whether this holds true of sC in coda remains to be answered.
14. Then does the gemination bring good results only to the speakers? The analysis devel-
oped here states that gemination is desirable for both speakers and hearers. As for the
speakers, the articulatory effort is minimized, since the number of contrasts is reduced.
As for the hearers, the acoustic salience is enhanced, since Q is prolonged. Featural lev-
el or C level, gemination turns out to be in favor of both speakers and hearers. It is no
wonder that phonological phenomena is caused by the balance between speakers and
hearers (cf. Yamada 1994, 1997). Both Quebec French and Japanese seem to achieve
such a balance, although each function is realized at different levels as below.
Quebec French Japanese

Deletion Gemination
(i) Minimization of articulatory effort C-slot level Feature level

(ii) Maximization of acoustic salience Feature level C-slot level
15. It is logically possible to regard CODACOND as a derived constraint, or NOCODA&*PLACE

through LC (Smolensky 1993). However, Fukazawa and Lombardi (2000) are against
this view, proposing that the two constraints to be conjoined must belong to the same
constraint family.
16. In order to capture the variation of epenthesis between Modern Standard Hindi (or Sin-
halese) and Japanese, DEP-EDGE (Alderete 1995, Benua 1997: 127) may play a crucial
384 Noriko Yamane
role. Since this constraint prohibits an epenthetic material at edges of morphological or

prosodie domains, it must be ranked low enough to be overridden in Modern Standard
Hindi, unlike Japanese. The whole ranking in each language needs further investiga-
tion. Thanks to Jeroen van de Weijer for letting me addressing this matter.
17. According to Maekawa (1988:138), a high vowel is deleted after a voiceless fricative or
affricate, but when it is preceded by a voiceless stop, it is not deleted but rather de-
voiced.
18. Allowing multiple winners is not at all unfamiliar in OT. See Hammond (2000).
References
Alderete, John
1995 Faithfulness to Prosodie Heads. Ms., University of Massachusetts, Amherst.
Benua, Laura
1997 Transderivational Identity: Phonological Relations Between Words. Doctoral
dissertation, University of Massachusetts, Amherst.
Broselow, Ellen
1987 Non-obvious Transfer: On predicting Epenthesis Errors, Interlanguage pho-
nology, ed. by G. Ioup and S. Weinberger, 261-278, Newbury House, Cam-
bridge, MA.
1991 The Structure of Fricative-Stop Onsets, Handout presented at the workshop
Properties of Phonological Feature Organization, University of California,
Santa Cruz.
Cairns, Charles E.
1988 Phonotactics, Markedness and Lexical Representation. Phonology 5, 209-36.
Côté, Marie-Hélène
1997 Phonetic Salience and the O C P in Coda Cluster Reduction. CLS 33: The Main
Session, 57-71, Chicago Linguistic Society.
Davis, Stuart
1991 Coronals and the Phonotactics of Nonadjacent Consonants in English, Pho-
netics and Phonology 2: The Special Status of Coronals, ed. by C. Paradis and
J-F. Prunet, 49-60, Academic Press, San Diego and London.
Dobson, Eric J.
1968 English Pronunciation 1500—1700. vol. II. Clarendon Press, Oxford.
Fergusson, Rosalind
1985 Rhyming Dictionary.Market House Books, England.
Flemming, Edward S.
1995 Auditory Representations in Phonology. Doctoral dissertation, University of
California, Los Angeles.
Fudge, Erik C.
1969 Syllables. Journal of Linguistics 5, 253-86.
Fujimura, Osamu
1997 Syllable Features and the Temporal Structure of Speech. Proceedings of LP'
96, ed. by B. Palek, 91-111, Charles University Press.
Fujiwara, Yasuaki
1990 Koeishi Inritsu Kenkyuu (A Metrical Study of Old English Poems). Keisuisha,
Hiroshima.
Fukazawa, Haruka and Linda Lombardi

2000 To Be Simple or Not to Be: Constraints in Optimality Theory. The Proceed-
ings of the First Tokyo Conference on Psycholinguistics, ed. by Otsu, 35-54,
Hituzi Shobo, Tokyo.
Goldsmith, John A.
1976 Autosegmental Phonology. Doctoral dissertation, MIT. [pub. by Garland,
New York, 1979],
Gunji, Toshio
1991 Eigo Gyaku Biki Jiten. Kaitakusha, Tokyo.
Hammond, Michael
2000 The Logic of Optimality Theory, to appear in the proceedings of W E C O L .
[ROA-390-04100].
Hayes, Bruce
1980 A Metrical Theory of Stress Rules. Garland, New York.
1989 Compensatory Lengthening in Moraic Phonology. Linguistic Inquiry 20, 253-
306.
Itô, Junko
1986 Syllable Theory in Prosodie Phonology. Doctoral dissertation, University of
Massachusetts, Amherst. [Pub. by Garland, New York, 1988.]
1995 The Core-Periphery Structure of the Lexicon and Constraints on Reranking.
University of Massachusetts Occasional Papers in Linguistics 18 (Papers in
Optimality Theory), 181-209, GLSA.
Jordan, Richard
1974 Handbook of Middle English Grammar: Phonology, tr. and rev. by E.J. Crook.
Mouton, The Hague.
Katayama, Motoko
1998 Optimality Theory and Japanese Loanword Phonology. Doctoral dissertation,
University of California, Santa Cruz.
Kawagoe, Itsue
1995 Consonant Gemination in Loanword Phonology and Rhythm Clash. Gengo
Kenkyu (Journal of the Linguistic Society of Japan) 108, 46-72.
Kim,Suksan
1986 A Nonlinear Analysis of Reduplicating Preterites in Germanic. Linguistic
Journal of Korea 11,141-55.
Kirchner, Robert
1998 An Effort-Based Approach to Consonant Lenition. Doctoral dissertation,
University of California, Los Angeles.
Kitahara, Mafuyu
1996 Consonant Gemination and Accent in Japanese Loanwords. Proceedings of
Formal Approaches to Japanese Linguistics 2. (MITWPL 29), 61-79, MIT
Press, Massachusetts.
Kohler, Klaus J.
1967 Modern English Phonology. Lingua 19,145-76.
Lamontagne, Gregory A.
1993 Syllabification and Consonant Coocurrence Conditions. Doctoral dissertation,
University of Massachusetts, Amherst.
Maekawa, Kikuo
1988 Boin no Museika [Vowel Devoicing]. Nihongo to nihongokyoiku 2 (Japanese
and Japanese Education), ed. by M. Sugito, 135-53, Meijishoin, Tokyo.
386 Noriko Yamane

1995 Faithfulness and Reduplicative Identity. University of Massachusetts Occa-
sional Papers in Linguistics 18 (Papers in Optimality Theory), 249-384, GL-
SA.
McCarthy, John and Alison Taub
1992 Review Article: The Special Status of Coronals: Internal and External Evi-
dence, ed. by C. Paradis and J.-F. Prunet. Phonology 9, 363-70.
Minkova, Donka
1982 The Environment for Open Syllable Lengthening in Middle English. Folia
Linguistica Histórica III/l, 29-58.
1985 Of Rhyme and Reason: Some Foot-Governed Quantity Changes in English.
Current Issues in Linguistic Theory 41, 163-78.
Minkova, Donka and Robert Stockwell
1992 Homorganic Clusters as Morie Busters in the History of English: the Case of
-Id, -nd, -mb. History of Englishes: New Methods and Interpretations in Histor-
ical Linguistics, ed. by M. Rissanen et al., 191-206, Mouton de Gruyter, Berlin
and New York.
Morelli, Frida
1998 Markedness Relations and Implicational Universals in the Typology of Onset
Obstruent Clusters. NELS 28,107-20.
Nakao, Toshio
1985 On-Inshi (English Historical Phonology). Taishukan, Tokyo.
1993 Optimality Theory. Constraint Interaction in Generative Grammar. Ms., Rut-
gers University, New Brunswick, and University of Colorado, Boulder.
Sagey, Elizabeth C.
1986 The Representation of Features and Relations in Non-Linear Phonology. Doc-
toral dissertation, MIT.
Selkirk, Elisabeth O.
1982 The Syllable. The Structure of Phonological Representations II, H.G. van der
Hulst and N.S.H. Smith, 337-83, Foris Publications, Dordrecht.
Smolensky, Paul
1993 Harmony, Markedness, and Phonological Activity. Handout from ROW-1.
[Revised, ROA-87,1995.]
1995 On the Internal Structure of the Constraint Component Con of UG. Handout
of a talk at UCLA, 4/7/95. [ROA-86.]
Weijer, Jeroen M. van de
1995 Continuancy in Obstruents and in Liquids. Lingua 96, 45- 61.
1996 Segmental Structure and Complex Segments. Niemeyer, Tübingen.
Wells, John C.
1990 Pronunciation Dictionary. Longman, England.
Yamada, Norio
1994 Phonological Changes in English: A Functional Explanation in Historical
Phonology. Doctoral dissertation, Tsuda College, Tokyo.
1997 Optimality Theory and English Historical Phonology. Studies in English Lin-
guistics: A Festschrift for Akira Ota on the Occasion of His Eightieth Birthday,
ed. by M. Ukaji, T. Nakao, M. Kajita and S. Chiba, 1133-53, Taishukan, Tokyo.
Yamane, Noriko
1995 An Integrated Account of Middle English Open Syllable Lengthening. Κ 15,
36-46, Kansai Linguistic Society.
1996a Rekishiteki On-in Henka to Saitekisei Riron: Schwa Shoushitsu no Baai [Di-
achronie Phonological Change and Optimality Theory: The Case of Schwa
Loss]. On-in Kenkyuu: Riron to Jissen [Phonological Studies: Theory and
Practice], ed. by On-in Ron Kenkyuu Kai, 167-72, Kaitakusha, Tokyo.
1996b Diachrony in Optimality Theory: The Directionality of Change & The
Change of Optionality. Talk given at the symposium 'Phonology and its Inter-
faces' on the 14th National Conference of the English Linguistic Society of Ja-
pan.
1997 More on Middle English Open Syllable Lengthening: Is/ + Stop Clusters. Talk
given on the 15th National Conference of the English Linguistic Society of Ja-
pan.
1998a Chuu-Eigo ni okeru Kai-onsetsu Chou-onka Saikou: Isl to Heisaon no Ren-
ketsu [More on Middle English Open Syllable Lengthening: Is/ + Stop Clus-
ters]. JELS 15, 261-270, English Linguistic Society of Japan.
1998b Great Vowel Shift, Chain Shift, and Local Conjunction. Talk given at the
workshop 'Great Vowel Shift and the Phonological Theory' on the 16th Na-
tional Conference of the English Linguistic Society of Japan.
1999a On-in Riron ni okeru Tsuuji-Henka no Yosoku-Kanousei : sC Rensa no Baai
[The Predictability of Diachronic Sound Change in Phonological Theory: The
Case of sC Clusters]. Journal of the Phonetic Society of Japan 3:1,29-40, Toyo
Shuppan, Tokyo.
1999b Local Conjunction in the Great Vowel Shift: An Answer to the Structural Co-
herence Problem from Optimality Theory. Ms., Nagoya Bunri University.
Yamane, Noriko and Shin-ichi Tanaka
2000 English Word-Final sC Clusters in Japanese Loanword Phonology. Phonolog-
ical Studies 3, ed. by the Phonological Society of Japan, 37-44, Kaitakusha, To-
kyo.
Author Index
Abercrombie, D. 327f, 331,333-35,338,352- Cairns, C.E. 383fnl0

53, 353fnl Catford, J.C. 195, 202, 353fn3
Ackema, P. 246, 274 Chalfont, C : see Ingleby, Chalfont and
Aitken, A.J. 338 Brockhaus
Akasaka, Y. 3, 7, 38, 43fn37 Charette, M. and A. Göksel 197,210,212fn7
Akasaka, Y. and K. Tateishi 3-46,43fn37, Cho, Y.Y. 200
295fn2, 295fn3 Chomsky, N.A. 3, 5-7,17, 24, 26, 34, 37, 39,
Akinaga, K. 132,137fn7 43fn33
Alderete, J. 83fn3, 85-86,136fn3,189fn6, Chomsky, N.A. and M. Halle 14-15, 36,
383fnl6 40fnl, 195
Aigeo, J.T. 320fnl7 Cinque, G. 295fn4
Allen, M. 271,290 Cobb, M. 197
Aoyagi, S. 247 Côté, M.-H. 371, 373
Avery, J.P. and K.D. Rice 200 Crowley, T. 319fn2
Culicover, P.W.: see Rochemont and Culi-
Backley, P. 200 cover
Baker, M. 245 Cutler, Α.: see Hawkins and Cutler
Bates, D. 270 Cyran, E. 196-97, 200, 210
Bauer, L.J. 292
Becker, T. 299, 305 Dauer, R.M. 327-28
Beckman, J.N. 87, 91, 93 Davis, S.M. 383fnl0
Beckman, M.E. and A. Shoji 137fn4 Di Sciullo, A.-M. and E.S. Williams 245,
Benua, L.H. 87, 91, 94, 98,106fn5-106fn7, 249, 278
309ff, 315f, 383fnl6 Dobson, E.J. 363
Beurden, L. van 320fn5
Bhat, D.N.S. 196 Evers, V.: see Lahiri and Evers
Bisetto, A. and S. Scalise 272f
Bloch, Β. 233 Fabb, N.A.J. 292
Bloch-Rozmej, A. 200, 210 Fergusson, R. 382fn3
Blumstein, S.E.: see Lahiri and Blumstein Fiengo, R. 24
Bolinger, D.L. 36 Flemming, E.S. 373, 375
Booij, G.E. 310 Fletcher, J. 329ff
Borer, H. 245 Flikeid, K. 195
BoSkovic, Z. and D. Takahashi 20, 33-34 Fudge, E.C. 270, 357
Bresnan, J.W. and S. Mchombo 245 Fujihara, C. 292
Broadbent, J.M. 200 Fujimura, O. 357
Brockhaus, W. 202 Fujiwara, Y. 195, 358
Brockhaus, W.: see also Ingleby, Chalfont Fukazawa, H. 83fn3, 87, 91-92, 94, 97,100,
and Brockhaus; Williams and Brockhaus 106fn9
Fukazawa, H. and L. Lombardi 383fnl5
390 Author Index
Fukazawa, H. and M. Kitahara 41fnl9, Itô, J. and R.A. Mester 9,41fnl9, 67-69, 71-
83fn3, 83fn4, 85-109 77,79-81, 83fn3, 83fn4, 85-86, 88f, 91, 94,
Fukazawa, H. and V. Miglio 105fnl 97-98,105fn2,106fn3,106fn8,137fn4,
Fukazawa, H., M. Kitahara and M. Ota 91, 186f, 241nll, 241fnl2, 310, 314
94, 97,105fn2 Itô, J., R.A. Mester and J.E. Padgett 97,
Fukui, M. 120,122f 105fn2
Fukui, N. 19 Itô, J., Y. Kitagawa and R.A. Mester 163
Fukui, N. and Y. Takano 19, 25ff, 31, 38, Itô, J.: see also Kubozono, Itô and Mester
43fn33
Fukui, N.: see also Saito and Fukui Jaeggli, O.A. 320fn4
Johnstone, T.M. 195, 208
Göksel, Α.: see Charette and Göksel Jones, D. 353fnl
Goldsmith, J.A. 85, 369 Jordan, R. 382fn6
Greenberg, J.H. 304
Gunji, T. 382fn3 Kager, R. 71, 77
Kageyama, T. 245-76, 278,283-84,286-89,
Hale, K.L. 18 303, 307f
Halle, M. and A.P. Marantz 245 Kageyama, T.: see also Shibatani and
Halle, M. and J.-R. Vergnaud 62 Kageyama
Halle, M.: see also Chomsky and Halle
Kahn, D. 40fn4
Halliday, M.A.K. 336
Katamba, F.X. 303
Hamans, C. 319fnl
Katayama, M. 189fn6, 378-79
Hammond, M.T. 384fnl8
Kawagoe, I. 370
Hankamer, J.: see Sag and Hankamer
Kayano, S. 145ff, 148ff
Haraguchi, S. 3, 47-65, 215, 218, 232, 233f
Kaye, J.D. 195, 220, 225
Harris, J.M. 193,196f, 198ff, 200
Kaye, J.D., J. Lowenstamm and J.-R. Verg-
Harris, J.M. and G.A. Lindsey 193,196f,
naud 193,196-97, 200, 212fn8
198ff, 200
Kayne, R.S. 24ff, 31,38, 42fn32
Haspelmath, M. 320fnl0
Kibe, N. 105fn2
Hattori, N. 184,190fn21
Kim, S. 358
Hawkins, J.A. and A. Cutler 306
Hayes, B.P. 3, 5, 8,118,142,180, 360 Kindaichi, H. 49
Heuven, V.J. van: see van Zanten and van Kindaichi, K. 142,144
Heuven Kiparsky, R.P.V. 195, 273
Hirayama, T. 221, 240fn2 Kirchner, R.M. 375
Hock, H.H. 319fn3 Kitagawa, Y.: see Itô, Kitagawa and Mester
Hoji, H.: see Saito and Hoji Kitahara, M. 378
Honma, T. 67-84,105fn2, 383fnl0 Kitahara, M.: see Fukazawa and Kitahara;
Huang, S. 320fn6 Fukazawa, Kitahara and Ota
Kohler, K.J. 357
Ingleby, M., C. Chalfont and W. Brockhaus Kubozono, H. 15-16, 86,111-42,136fn3,
197 137fn4,137fn7,159ff, 164ff, 168,171,176f,
Inkelas, S. 83fn5, 261 180,188fnl, 189fn7,189fn9,246,261,270,
Inkelas, S. and D. Zee 141 294fn2, 295fn2, 299-302, 308-309, 320fn8
Ishiwata, T. 301 Kubozono, H. and S. Ohta 117,314,320fnl4
Itô, J. 13, 67-69,71, 300, 301, 302, 313-14, Kubozono, H., J. Itô and R.A. Mester
320fnl2, 320fnl5, 320fnl6 189fn3
Autor Index 391
Kubozono, H.: see also Tanaka and Kubozo- Michelson, K. 136fn3

no Miller, D.G. 269
Kuno, S. 20-21, 41fn22 Minkova, D. 360
Kurisu, K. 77 Minkova, D. and R. Stockwell 383fn7
Kuwamoto, Y. 314 Miyagawa, Sh. 19-20
Morelli, F. 383fnl3
Ladd, D.R. 270 Morita, J. 292
Lahiri, A. and S.E. Blumstein 204 Motwani, P. 308
Lahiri, A. and V. Evers 204
Lamontagne, G.A. 357, 364 Nakao, T. 363, 382fn6
Lass, R.G. 319fn3 Nakatani, L.H., K.D. O'Connor and C.H.
Leben, W.R. 85 Aston 336
Lee, D.-Y. 200, 212fn6 Namiki, T. 272, 277-97, 294fnl, 299, 304f
Lee, D.-Y. and S. Yoshida 210, 212fn8 Nanjo, K. see Nishihara, van de Weijer and
Lehiste, I. 327 Nanjo
Levin, B.C.: see Rappaport Hovav and Nasukawa, K. 211fn2
Levin Nespor, M. and I.B. Vogel 141-42
Levin, J. 113 NHK 118,120,184
Liberman, M.Y. and A.S. Prince 14, 40fnl Ni-ikura, S. et al. 319fnl
Lieber, R. 245, 254, 266, 274, 278, 291-93, Nishihara, T., J.M. van de Weijer and K.
299, 303, 320fn4 Nanjo 12, 278, 299-324
Lindsey, G.A.: see Harris and Lindsey Nitta, T. 49, 54,60
Lombardi, L. 93,106fn4
Lombardi, L.: see also Fukazawa and Lom- Ohso, M.: see Quackenbush and Ohso
bardi Ohta, S.: see Kubozono and Ohta
Lowenstamm, J.: see Kaye, Lowenstamm Okuda, 0.151ff
and Vergnaud Ota, M.: see Fukazawa, Kitahara and Ota
Lubowicz, A. 87 Ota, M. 333-34
Otsu, Y. 41fnl9,105fn2, 295fn3
MacMahon, M.K.C. 336
Maddieson, I. 200 Padgett, J.E. 105fn2,155fn5
Maekawa, K. 384fnl7 Padgett, J.E.: see also Itô, Mester and
Marantz, A.P.: see Halle and Marantz Padgett
Masuya, Y. 327-56, 353fn2, 354fnl5 Paradis, C. and J.-F. Prunet 200, 207
McCarthy, J.J. 67, 74, 82, 85, 99 Pesetsky, D.M. 269
McCarthy, J.J. and A.B. Taub 383fnl0 Pilsudski, B. 142
McCarthy, J.J. and A.S. Prince 11, 40fnll, Poser, W.J. 11,132,163,194, 233, 246-47,
40fnl2, 67, 91,310, 314, 316, 369 249, 261, 265f
McCawley, J.D. 9,15,41fnl6,41fnl8,47,52, Postal, P.M. 255
116,160,166,233 Potter, B. 136fn3
Mchombo, S.: see Bresnan and Mchombo Prince, A.S. and P. Smolensky 67,75,78,91,
McKoon, G.: see Ward, Sproat and McKoon 111, 118,136, 309, 366, 383fnll, 383fnl2
Mester, R.A. 67-69, 71, 80-82, 83fn6, Prince, A.S.: see also Liberman and Prince;
190fnl5 McCarthy and Prince
Mester, R.A.: see also Itô and Mester; Itô, Prunet, J.-F.: see Paradis and Prunet
Mester and Padgett; Itô, Kitagawa and
Mester; Kubozono, Itô and Mester Quackenbush, H.C. 295fn5
392 Author Index
Quackenbush, H.C. and M. Ohso 137fn7 Takahashi, D.: see Boskovic and Takahashi
Takano, Y. 19, 27, 29, 32, 42fn32
Ralli, A. 303 Takano, Y.: see also Fukui and Takano
Rappaport Hovav, M. and B.C. Levin 255 Tamura, S. 144,145ff, 148ff
Refsing, K. 155fnl5 Tanaka, S. 118,120,136fn2,159-92,189fn4,
Rennison, J.R. 196 190fnl9
Rice, K.D. 105fn2 Tanaka, S. and H. Kubozono 129,137fn5
Rice, K.D.: see also Avery and Rice Tanaka, S. and N. Yamane 126,185,189fn9,
Roach, P.J. 331 190fnl9
Rochemont, M.S. and P.W. Culicover 27 Tanomura, T. 137fn7
Roeper, T. and M.E.A. Siegel 249 Tateishi, K. 3,7,11,38,40fn7,40fn9,40fnl0,
Ross, J.R. 18, 24, 28 137fn4,163,190fnl7, 240fn7
Tateishi, K.: see also Akasaka and Tateishi;
Sag, I.A. and J. Hankamer 257 Selkirk and Tateishi
Sagey, E.C. 357, 364 Taub, A.B.: see McCarthy and Taub
Saito, M. 18-19 Taylor, J. 268
Saito, M. and H. Hoji 18 Trommelen, M. and W. Zonneveld 278, 303
Saito, M. and Ν. Fukui 19
Trubetzkoy, N.S. 83fn2,116
Sato, H. 105fn2,189fn5
Scalise, S. 278, 280,293, 295fn6, 299, 304
Uldall, E.T. 327, 334, 338,353fn2
Scalise, S.: see also Bisetto and Scalise
Umegaki, M. 302
Scheler, M. 319fn3
Urbanczyk, S.C. 87, 91, 94, 97
Schultink, H. 303
Uwano, Z. 136fn3,189fn5
Selkirk, E.O. 29, 36, 39, 41fn20, 42fn30,
136fn3,141-42,147,155fnl4, 246, 263,
Vance, T.J. 105fn2
278,290-91,304,310, 357f
Selkirk, E.O. and K. Tateishi 7, 22, 41fn20 Vergnaud, J.-R.: see Halle and Vergnaud;
Shibatani, M. 205, 209, 300f Kaye, Lowenstamm and Vergnaud
Shibatani, M. and T. Kageyama 245,248-49 Verhijde, M.M. and J.M. van de Weijer
Shimamura, R. 292 42fn29
Shin, S.-H. 319fn2 Viola, M.: see Fukazawa and Viola
Shiraishi, H. 155fn5 Vogel, I.B. 299, 302, 305-308, 318
Shoji, Α.: see Beckman and Shoji Vogel, I.B.: see also Nespor and Vogel
Sibata, T. 49,119
Siegel, M.E.A.: see Roeper and Siegel Walker, R. 166
Slobin, D.I. 306 Ward, G„ R.W. Sproat and G. McKoon 255f,
Smolensky, P. 85,105fnl, 365, 366, 383fnl5 259
Smolensky, P.: see also Prince and Smolen- Weijer, J.M. van de 357, 369
sky Weijer, J.M. van de: see also Nishihara, van
Spaelti, P. 77 de Weijer and Nanjo; Verhijde and van de
Spencer, A.J. 290 Weijer
Sproat, R.W. 255 Wells, J.C. 382fn2, 382fn3
Sproat, R.W.: see also Ward, Sproat and Wenk, B.J. and F. Wioland 329
McKoon Wiese, R. 319fnl
Stockwell, R: see Minkova and Stockwell Williams, B. and S.M. Hiller 332-33, 335
Subandi 293 Williams, E.S. 270,277-78,289, 294fnl, 299,
Szymanek, B. 319fnl 302-303
Autor Index 393
Williams, E.S.: see also Di Sciullo and Will- Yoshida, S.: see also Lee and Yoshida
iams Yoshida, Y. 133,189fn5,219-22,226,228-29,
Williams, G. 196ff, 200 233,240fn5
Williams, G. and W. Brockhaus 197 Yoshida, Y. and H. Zamma 215-41
Yamada, E. 40fn5
Zamma, H.: see Yoshida and Zamma
Yamada, N. 383fnl4
Zanten, E. van and V.J. van Heuven 334
Yamane, N. 357-87, 382fn6, 383fn9, 383fnl2
Zee, D. and S. Inkelas 29
Yamane, N. and S. Tanaka 371, 381
Zee, D.: see also Inkelas and Zec
Yamane, N.: see also Tanaka and Yamane
Yip, M.J.W. 85, 99,106fn9 Zonneveld, W.: see Trommelen and Zonne-
Yokotani, T. 240fn7 veld
Yonekawa, A. 301,320fnll Zubizarreta, M.L. 36, 39
Yoshida, S. 193-214 Zwicky, A.M. 278, 304-305
Language Index
Ainu Indonesian 293, 334

- Saru dialect 136fn3,141-57 Italian 272f, 293,295fn2, 295fn6,299, 302,
- Shizunai dialect 151,155fnl5 304, 306ff, 320fn9, 327
Arabic 331
- Cairene Arabic 195 Japanese
- eastern Arabic dialects 195, 208 - common Japanese 58, 63
- Fukuoka Japanese 48
Catalan 74 - Ibaragi dialects 58
Caucasian languages 353fn3 - Hokuriku dialects 205
Chinese 111, 262, 270, 295fn2, 301, 320fn6 - Izumi Japanese 48
- Kagoshima Japanese 47-48,136fn3
Dakota 62, 92f, 104 - Kansai-style Japanese 47-48, 218
Dutch 93f, 104, 293, 300, 303-304, 319fnl - Kanto dialects 206
- Kumi Japanese 48
English 3-43, 62, 96f, 179,194, 246,255-57, - Kyoto Japanese 215-41, 240fn3
259, 269-72, 274, 277f, 280, 290ff, 294fnl, - Kyushu dialects 205
294fn2, 299-300, 302ff, 308, 313, 318, - Miyakonojo Japanese 48
320fnl2, 320fnl7, 327-87 - Nakamura Japanese 48
- American English 336, 382fn2 - Old Japanese 113
- Old English 358ff, 383fn7 - Old Kyoto Japanese 48
- Middle English 360ff - older Japanese 295fn7
- Late Middle English 360 - Osaka Japanese 47-48, 218, 240fn3
- early Modern English 362ff - Shikoku dialects 205
- Modern English 364ff - Tohoku dialects 49, 58,195, 205, 206
- Received Pronunciation (RP) 382fn2 - Tokyo Japanese 40fnl0, 47-48, 50,52, 54-
- Scottish English 336-37, 351 55,57-58,61,70,76, 86, l l l f f
- Papua New Guinean English 319fn2 - Tsuruoka Japanese 47ff
European languages 300
Korean 195, 208,293, 319fn2
French 141,199, 319fnl, 320fn4, 329ff
- Acadian French 195 Latin 270, 304
- Quebec French 195, 208,371ff, 383fnl4 Lushootseed 94ff
German 82,293, 307, 319fnl Makassarese 136fn3

Germanic languages 304 Mohawk 136fn3
Greek 270,303, 327
Polish 272
Hebrew 272 Portuguese: Brazilian Portuguese 195
- Tiberian Hebrew 74
Hindi 369, 373, 383fnl6 Romance languages 299, 304
Language Index
Russian 195, 320fn4, 331 Telugu 331

Thai 327
Selayarese 136fn3 Turkish 272
Sinhalese 369,373,383fnl6
Shona 93 Vietnamese 293,303
Slavic languages 195
Somali 304 West-African languages 200
Spanish 199, 327
Swahili 303 Yoruba 331
Subject Index
accent bimoraicity 240fn4

- accent deletion 53, 55 borrowings, see loanwords
- accent language 218, 221, 232 branching
- accent loss 179-80 - in morphology 10-16
- accent shift 47, 50ff, 57ff, 63, 64 - in phonology 7-10
- accentedness 47, 56ff, 61, 233f - in syntax 5,16-18
- accentless nouns 119, 217, 231 - left-branching structures 15, 28
- accentual systems 47 - right-branching structures 15-16,28
- Accentual Phrase 34 see also syllable weight, headedness,
- accentuation 165,260,273 heaviness
- acute accent 49
- adjectives 47, 60ff, 131-34 category projections (in syntax) 274
- antepenultimate accent 116-19,219,221-23 clash avoidance 229f
- avoidance of final accent 162 clipping, see truncation
- change in progress 184-85 coalescence 57, 59, 61
- de-accentuation 132,176 coda condition 314, 317
- default 165 complex segments 358ff
- grammar 186 - affricates 201
- in foreign nouns 116f - contour segments 201
- lexical accent 9-10, 216-19, 221ff, 225-27, - complex segment formation 361
230, 232, 256, 267, 273 compounds 62,104,160-62,247,300ff, 303ff,
- nouns 47, 51, 53, 57,216f, 238f 318
- parsing of 162 - unaccented compounds 176-77
- phrasal accent 247,256 - borrowed compounds 306
- pitch accent 47, 216 - compound accent rule 14-16,123-31
- pre-accenting morphemes 57, 236 - compound-specific submeaning 283
- pre-no deaccenting rule 132-34 - head vs. non-head nomináis 160,172-76
- preservation of 164-65,171 - noun-noun compound accent rule 123-27
- pseudo accent 225,227-30 - possessive compounds 268
- unaccentedness 47f, 56, 59,119-23 - postsyntactic compounds 248
- verbs 47, 56ff, 131-32 - quadrimoraic compounds 177
see also compounds, domains, faithful- - recursive compounds 279, 290
ness, loanwords - reversible compounds 280, 290
acquisition 306 - short heads vs. long heads 161,188fnl
acronyms 183 - Sino-Japanese compounds 125,167, 301
affixes 62 - X-iti compounds 129-31
affricates, see complex segments - X-taroo compounds 127-29
affrication 193,194-96 connected speech 224
anaphora 255-56 consonant clusters 114, 357
- anaphoric island constraint 255, 258 consonant elision 112,115f, 134-36
Subject Index 397
constraints focus movement 19

- constraint conflict 161 foreign stratum, see lexical strata
- free ranking 175
- gravitation 186f gemination 313f, 370-72
- local conjunction 85,105fnl, 105fn2, 365 government phonology 196
- maintain contrast 373 - c-command relations 202f, 209
- (minimal) reranking 126,173 - empty heads 197, 200f, 203-205,207, 209
- morpheme realization 88 - head vs. operator 196f
- No Phrase constraint 249, 266 - licensing 197, 220
- non-finality 171,178-79 see also segmental structure: elements
- ranking invariance 186 grammar
- reranking 126,171,178,181 - Binding Theory 38
- self-conjoined constraints 83fn3 - modularity assumption 3, 6-7, 43fn35
- universal constraints 300 - phonology-syntax interface 7
see also coda condition, contiguity, culmi- see also interfaces
nativity, edgemostness, nonfinality, faith-
fulness constraints
headedness
contiguity 314, 317
- morphological headedness 165,188fn2,
contraction, see truncation
299ff, 302ff, 308
coronal/labial asymmetry 358ff
- morphological headedness: Righthand
culminativity 180
head rule 277ff, 294, 299ff, 302ff, 319
- phonological headedness 320fn8
derivedness 315
heaviness 3-43
diachronic change 185
see also syllable weight
diminutives 320fn4
historical phonology 357ff
dispersion theory (of contrast) 373-78
hypocoristics 315ff
dissimilation 92
domains
inaccessibility (of position in domain) 219,
- accent domain 217-20
23If, 237-39
- phonological domains 216-17, 219f, 225
- headship 219, 224f, 229 incorporation 269
downstep 225, 227 inflection 305
- comparative, superlative (in English) 271
edge (of phonological domain) 223ff - plural (in English) 271, 305
edgemostness 124,126 interfaces 18-34
elision, see consonant elision invisibility (of epenthetic vowels) 11 Iff, 134
epenthesis, see vowel epenthesis iso-chestabicity, iso-syllabicity 353
excrescent vowels 113 isochrony, isochronism 327, 352-53
see also vowel epenthesis
expandability 282, 293 lenition 202
lengthening, see vowel lengthening
faithfulness level-ordering (in morphology) 96f, 271
- accentual faithfulness 164 lexical integrity 245
- faithfulness constraints 124,136fn3 lexical strata 165
- faithfulness conversion 186 - alien stratum 97
- positional faithfulness 136fn3 - foreign stratum 97,125,168,173,310, 315
- relativized faithfulness constraints 87 - mimetics 97,105fn2,165,183,190fn20
398 Subject Index
- native/Yamato stratum 9,97,105fn2,125, Obligatory Contour Principle (OCP) 85-109

166,173,217,221,252 - typology of effects 100
- Sino-Japanese stratum 97,105fn2,125, onbin 115, 240fn9
167,173,252, 257, 262-63, 301 Optimality Theory 67,136,159,300
lexicalist hypothesis 245 - Correspondence Theory 90, 91ff, 98ff,
licensing, see Government Phonology 105fn2,369, 375
loanword phonology 111, 369-81 - emergence of the unmarked (TETU) 91-
loanwords 111, 116-23,136fnl, 270, 299ff, 92,107fnl0,124
307ff - lexicon optimization 67, 78ff, 81ff
- loanword accent rule 116f - richness of the base 77ff, 81ff
- unaccentedness in loanwords 119ff - sympathy theory 74ff
output-output conditions 138fnll
markedness
- context-free markedness 71 palatalization 193-94,198f, 207
- contextual markedness 71 particles, case marking 216ff, 224,230
mimetics, see lexical strata pitch 105fn2
minimalism 24 - pitch sharing 219f, 223,231f
- asymmetric c-command 25 place of articulation
- case checking 19, 33,42fn32, 43fn34 - coronals 193,200
- Interpret α 3, 36 - dental-alveolars 193,199
- Merge 5,17, 36 - labials 49,199
- Move 5 - palatals 199
- Move a 24, 31 - palato-alveolars 193ff, 199
- reconstruction 33 - velars 200
minor phrase 41fnl6, 247, 261 prefixation (in English) 270
modularity assumption, see sub grammar - Aoyagi prefixes 247
mora language vs. syllable language 116 prenasalization 56
mora prepositions 272
- mora theory prosodie rules 262
- moraic nasal 56,115,137fn7, 200, 211fn3 - P-structure rules 141,147
- moraic obstruent 115 - pure phonological rules 141,147
- mora vs. syllable 116 prosodie structure 29, 39, 302
- quadrimoraic heads in compounds 169, prosodie units 23, 29
175 - filled foot 338
morphological markers - foot 163,327
- accusative 51 - iambs 175
- genetive 50, 52-53 - silent foot 338
- negative 57f, 62 - trochees 175
- negative imperative 57f - mora 302
movement, syntactic 19 - prosodie category formation 163
- prosodie word 6, 300, 314
native morphemes, see lexical strata - single-PrWd hypothesis 179
nicknames, see hypocoristics - syllable 328-29
nonfinality 124,126 - as chest pulse 328f
non-locality 171 - chestable 328
no-skipping 314 - unfilled foot 338
- utterance 22,39
Subject Index 399
reduplication 93ff - front truncation 300, 308-11

Rendaku (sequential voicing) 67ff, 86ff, 254 - templatic truncation 309
Rendaku: Lyman's Law 68ff, 86ff typology 225ff
rhythm 22, 328f, 350ff
- leader-timing 329 Universal Grammar 309
- pauses 338
- stress-timing 331,352 variation 169,175,181
- syllable-timing 331, 352 velar nasal 49
- trailer-timing 329 verb forms
- conditional 58f
s + stop clusters 357ff - past tense 58ff
sandhi 142-43 - present tense 47, 56-57
scrambling 18-24 verb paradigms 233ff
segmental structure versification 151ff
- elements 106fn4,196,198f voiced velar nasalization 67,72ff
self-embedding (in syntax) 20 voicing rules 59, 93f, 97ff, 105fn2
stress 8-9,269, 307, 334 vowels
- silent stress 327 - high vowels 60
structural ambiguity (of syntactic structures) - vowel devoicing 114,121
21 - vowel epenthesis 111, 113-17,134ff,
subcategorization (of affixes) 254 136fn3, 378ff
syllable structure 111, 114-16,120, 312 - vowel lengthening 216-17, 240fn2,
syllable weight 115ff 241fnll
syntactic atomicity 245 - vowel shortening 61
- vowel system 198
tone
- contour tones 54 weight, see syllable weight, see also heaviness
- tone language (Japanese as a) 218 word order: SOV languages 304
- tone melodies 218, 232 - SVO languages 304
- tone shift 47 word formation 245, 252
truncation 12-14,299ff, 306ff, 320fnl7 - root 263ff, 273
- back truncation 300, 308f, 310, 320fn9 - stem 263ff, 273
- double truncation 300f, 308-11, 319fn2, - word 263ff, 273
320fnl2 - Word Plus (W + ) 246, 263ff, 273
List of Contributors
Yukiko Akasaka Taro Kageyama

Graduate School of Language and School of Humanities
Culture Kwansei Gakuin University
Osaka University 1-1-155 Uegahara
1-8 Machikaneyama-Machi Nishinomiya, 662-8501
Toyonaka Japan
Osaka 560-0043 <tkage@sannet.ne.jp>
Japan
<akasaka@ff.iij4u.or.jp> Mafuyu Kitahara
Haruka Fukazawa Department of Linguistics
Department of Human Science Indiana University
Faculty of Engineering Bloomington, IN 47405
Kyushu Institute of Technology USA
1-1 Sensui-cho, Tobata-ku <mkitahar@yo.rim.or.jp>
Kitakyushu, 804-8550
Japan Haruo Kubozono
<fukazawa@dhs.kyutech.ac.jp> Dept. of Linguistics
Faculty of Letters
Shosuke Haraguchi Kobe University
Faculty of Languages and Cultures Rokkodai, Nada-ku
Meikai University Kobe, 657-8501
8 Akemi Japan
Urayasu, 279-8550
Japan <kubozono@lit.kobe-u.ac.jp>
<shoharaguchi@dream.com>
Yosihiro Masuya
Takeru Honma
Faculty of Social Sciences and Hu- Graduate School of Humanities &
manities Institute for Language and Culture
Tokyo Metropolitan University Konan University
1-1 Minami-Ohsawa, Hachioji-shi 9-1 Okamoto 8 Chôme
Tokyo, 192-0397 Higashinada-ku
Japan Kobe, 658-0072
<honmat@st.rim.or.jp> Japan
<ymasuya@konan-u.ac.jp>
402 List of Contributors
Takayasu Namiki Nagoya University

Department of English Furo-cho, Chikusa-ku
Faculty of Education Nagoya-shi, 464-8601
Ibaraki University Japan
2-1-1 Bunkyo <tanaka@lang.nagoya-u.ac.jp>
Mito-shi
Koichi Tateishi
Ibaraki-ken, 310-8512
Department of English and
Japan
American Studies
<namiki@mito.ipc.ibaraki.ac.jp>
Kyoto University of Foreign Studies
Kensuke Nanjo 6 Saiin-Kasame-Chô
Department of English Ukyô-Ku
Faculty of Letters Kyoto, 615-8558
St. Andrew's University Japan
1-1 Manabino, Izumi <tatekoo@nn.iij4u.or.jp>
Osaka 594-1198 Jeroen van de Weijer
Japan Department of Linguistics
<knanjo@aol.com> Holland Institute of Generative
Linguistics
Tetsuo Nishihara
Leiden University
Department of Education,
P.O. Box 9515
English-Teaching Course
2300 RA Leiden
Miyagi University of Education
The Netherlands
Aza-Aoba Aramaki Aoba-ku
<vdweijer@rullet.leidenuniv.nl>
Sendai-city, 980-0845
Japan Noriko Yamane
<nisihara@staff.miyakyo-u.ac.jp> Department of Information Culture
Nagoya Bunri University
Hidetoshi Shiraishi 365 Maeda
Sakhalinskij Oblastnoj Inazawa-cho, Inazawa City
Kraevedcheskij Muzej Aichi, 492-8520
Kommunisticheskij prospekt dom. 29 Japan
Yuzhno-Sakhalinsk <nrkyamane@aol.com>
693010 Russia
<toshi_shiraishi@hotmail.com> Shohei Yoshida
International Student Center
Shin-ichi Tanaka Yokohama National University
Department of Linguistics and 79-1 Tokiwadai, Hodagaya-ku
Informatics Yokohama, 240-8501
Graduate School of Human Japan
Informatics <shohei2@ynu. ac.j p>
List of Contributors 403
Yuko Yoshida Hideki Zamma

Institute for Language and Kobe City University of Foreign
Culture Studies
Doshisha University 9-1 Gakuenhigashi-machi
Kyo-Tanabe City Nishi-ku,
Kyoto, 610-0394 Kobe City 651-2187
Japan Japan
<yuyoshid@mail.doshisha.ac.jp> <zammah@nifty.com>

(Studies in Generative Grammar 51) Jeroen Maarten Van de Weijer, Tetsuo Nishihara - Issues in Japanese Phonology and Morphology (2001, de Gruyter) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Studies in Generative Grammar 51) Jeroen Maarten Van de Weijer, Tetsuo Nishihara - Issues in Japanese Phonology and Morphology (2001, de Gruyter) PDF

Uploaded by

Copyright:

Available Formats

Issues in Japanese Phonology and Morphology

The series Studies in Generative Grammar was formerly published by

© Printed on acid-free paper which falls within the guidelines

Library of Congress Cataloging-in-Publication Data

Issues in Japanese phonology and morphology / edited be Jeroen van

Die Deutsche Bibliothek — Cataloging-in-Publication Data

Issues in Japanese phonology and morphology / ed. by Jeroen van de

© Copyright 2001 by Walter de Gruyter G m b H & Co. KG, D-10785 Berlin.

We are pleased to thank a number of institutions and individuals for their

Leiden, The Netherlands

I. Studies in Japanese Phonology

Yukiko Akasaka and Koichi Tateishi

Haruka Fukazawa and Mafuyu Kitahara

Domain-Relative Faithfulness and the OCP: Rendaku Revisited.. 85

Epenthetic Vowels and Accent in Japanese: Facts and Paradoxes 111

Yuko Z. Yoshida and Hideki Zamma

II. Studies in Japanese Morphology

Tetsuo Nishihara, Jeroen van de Weijer, and Kensuke Nanjo

III. Studies in Contrastive Japanese-English Phonetics

Author Index 389

List of Contributors 401

Yukiko Akasaka and Koichi Tateishi

Weight distinctions in linguistics have been considered to belong in the

What is branching, which we identify as "heaviness"? Branching is a node

(1) Branching: a node A branches iff it immediately dominates Β and C

Syl Syl Syl

The phonological and morphological trees above are mostly left-branch-

(2) Heavy NP Shift

We will come back to such examples later in section 4 and conclude

As phonology is basically concerned with linear strings of sounds, the rel-

(3) English Main Stress (Nouns 4 )

If we follow Hayes (1995) and analyze the English case as right-to-left

(4) a. The final syllable is ignored.

(5) Moraic Trochee in English

(6) a. (4a) ba[ F t na]na b. (4b) A[ F t men]ca

We can generalize such cases in the following way:

(7) The main stress foot must be branching into moras.

This is a case of branching condition on Structural Change (and Structur-

(8) Yamato nouns

(9) Foreign nouns6·7

However, if the antepenultimate mora happens to be the second mora of

(10) The preantepenultimate patterns with heavy syllables

( 1 2 ) ALIGN(ACCENT, LEFT, *BR(SYL)) 1 0 :

(12) is higher-ranked than (11), so that the antepenultimate regularity

Heaviness effects are observed in morphology, too. In this section, we will

3.1. Minimal word in English

In English, a person's name can be contracted to form a hypocoristic form

(13) English hypocoristics

These hypocoristics all consist of two moras, as pointed out by McCarthy

(13')[b[ Mora e]t[ Mora i]]n

A nickname with one mora cannot exist in English:

(14) Betty > *Be [be]

nor can a lexical word with one mora exist.

(16) Minimal Word: [Foot Mora Mora]

In other words, we can say the following:

(17) A word must be minimally complex/heavy (i.e. branching). 12

Here, we see that branching into moras is an output condition on word

3.2. Japanese hypocoristics

Japanese also has hypocoristic word formation patterns (Poser 1990,

(18) name nickname

3.3. Word contraction in Japanese

Itô (1990) observes interesting patterns of word contraction forms in Jap-

(19) Japanese contracted words