LIBRARY

MiChiQan State

University

 

 

PLACE IN RETURN Box to remove this checkout from your record.
TO AVOID FINES return on or before date due. '
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

0°03 09

 

UXWJQEESU

 

MAY 0 2 2009

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DateDuo.p65‘p.1 5

VOWEL

 

 

 

Depanm

VOWEL DEVOICING IN TOKYO JAPANESE: A VARIATIONIST APPROACH
By

Terumi lmai

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Department of Linguistics and German, Slavic, Asian, and African Languages

2004

VOWEL

This
variationist
aiew inves
differences
investigata
extremely n
social and I

The l

mnsonants

 

deVOiCing 9,
Data fmm 4‘1
instrumental

Effeci
bOUHdary. Cc
Claim that de
SUpponeCj, a!
and Maekaw,

aCcented Vov

ABSTRACT
VOWEL DEVOICING IN TOKYO JAPANESE: A VARIATIONIST APPROACH
By

Terumi lmai

This study investigates variation in Japanese vowel devoicing within a
variationist framework. There are numerous studies Of this phenomenon, but only
a few investigate its social aspects; Yuen (1997) reports style and gender
differences, and age difference are reported by Maekawa (1988); no study has
investigated the effect of social class. Previous studies have also been based on
extremely limited data, but this study uses a large corpus, incorporating both
social and linguistic factors.

The linguistic factors are vowel identity, preceding and following
consonants, morpheme boundary type, pitch accent pattern, and consecutive
devoicing environments. The social factors are age, sex, social Class, and style.
Data from 42 respondents” tape —recorded interviews were analyzed
instrumentally for voicing and treated to a logistic regression statistical analysis.

Effects Of the preceding and following consonant, accent, morpheme
boundary, consecutive devoicing, and style confirm previous studies, but the
claim that devoicing is prohibited between two fricatives (Tsuchida 1997) was not
supported, and vowel identity was statistically significant, contrary to Han (1962)
and Maekawa (1983). Previous claims about the effect of age in the devoicing Of

accented vowels, interaction between accent and the following consonant, and

   

inleraclieii
completel-
devoicing I.
effect of s
inleractior

sex differe
N8\

differently'
word acce
significam
devoicing
the previo
SOI
agefsex p
nonstanct
Prescriptii
leed Da‘
phonetic
Standard,
mamm'
(3) a Dun

pattern

interaction between morpheme boundary and the following consonant were not
completely supported, and the effect of morpheme boundary in a consecutive
devoicing environment was also not supported. Finally, the earlier study of the
effect Of sex was not entirely supported because sex and age show an
interaction; younger males devoice most, and younger females devoice least, but

sex differences are weak or nonexistent in middle-age and Older respondents.

New findings in this study include the following: [i] and [p] behave

differently from other Obstruents; pitch associated with the vowel, rather than the
word accent, was found to be important; morpheme boundary was found to be a
significant factor, but only a word level boundary demotes devoicing, and
devoicing is also demoted in a consecutive devoicing environment even though
the previous devoiceable vowel was not actually devoiced.

Social factors show a mixed pattern regarding standardness. Style,
age/sex patterns, and articulatory data suggest that devoicing may be
nonstandard, but the social class pattern (for /u/ only), speaker perception, and
prescriptive authority suggest it is a standard feature. In order to account for this
mixed pattern, several possible accounts are suggested: (1) phonological vs.
phonetic devoicing (Tsuchida 1997) may be related to the perception of
standardness. (2) language change may be occurring among younger speakers,
or a combination of these two may be able to account for this mixed pattern, or
(3) a purely variationist approach in which various segments of the society
associate themselves with certain levels Of performance may account for the

pattern.

Copyright by
TERUMI IMAI
2004

For my parents

’l

 

 

 

 

iv.
who has I
was alwa:
Hbflmeh
dissertatic
Carol, wh
weekends
Dr. Yen-H(
Ph0n0i09y
I WC
their feedb
Anne Viol;
and Dr, St
and helper
I th;
introcjuCed
WillingneS:
My
encourage
HirokatSu I

Understanc

ACKNOWLEDGMENTS

I would like to express my deepest appreciation to Dr. Dennis Preston,
who has supported me throughout the process of completing my dissertation. He
was always available for discussion and remained enthusiastic about my study.
His timely responses kept going, and I would not have been able to finish this
dissertation without his constant encouragement. I am also grateful tO his wife
Carol, who always welcomed me warmly when I visited their house to work on
weekends, and also encouraged me tO carry on. In addition, I would like to thank
Dr. Yen-Hwei Lin for training me in phonology. I developed an interest in
phonology and phonological/phonetic variation after taking her courses.

I would like to extend my appreciation to my other committee members for
their feedback that proved to be very helpful: Dr. MutsukO Endo Hudson and Dr.
Anne Violin-Wigent. I am also grateful tO Dr. Timothy Vance, Dr. Richard Wright,
and Dr. Shigekc Okamoto for their comments and suggestions that inspired me
and helped me tO understand certain concepts/issues better.

I thank the respondents who participated in this study, and those who
introduced me tO the potential respondents. Without their assistance and
willingness to cooperate, this study could not have been carried out.

My special thanks go to my friends and colleagues for their constant
encouragement: Erica Benson, Betsy Evans, Chunhua Ma, Jane Ozanich, and
Hirokatsu AsanO. Finally, I would like to thank my family in Japan for their

understanding and continuous support.

vi

 

 

_T I

LIST OF
LIST OF

CHAPTI
INTROU
1.1

1.

1.

1.
1.2
1.3
1.4

CHAPTI

BACKC
2.1

2.2

NNMMNN

2.2

TABLE OF CONTENTS

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

LIST OF TABLES ............................................................................................................... x

LIST OF FIGURES ........................................................................................................... xii
CHAPTER 1

INTRODUCTION- _ __ ....... J

1-1 Japanese PIIOIIO'OQY -- .............................. 3

1.1.1 Syllable vs. mora ____________ 3

1. 1.2 Japanese phonemes ______ 4

1.1.3 Japanese accentuation system _ 6

1-2 Brief discussion of Japanese morpheme types ..................................... 8

1.3 Research questions and hypotheses .......................... 10

1-4 Organization- - -__.-- - - ....................... 11
CHAPTER 2

BACKGROUND _" ........... 13

2.1 Phonological and phonetic accounts of Japanese vowel

devoicing .................................................................................... .13

2.2 Linguistic factors 16

2.2.1 Preceding and following consonants 17

2.2.2 Vowel identity __ 21

2-2-3 Accent- .................. 22

2.2.4 Morpheme boundaries 24

22-5 Consecutive devoicing .......................... 25

22-6 Summao'.-- .......................... 28

2.2 Social Factors 29

2-3-1 Region ............................................................... 30

2.3.2 Age _ _ 33

2-3-3 Sex and speech style ............................. 34

2-3-4 Social class ............................. 36

2-3-5 Summary ................................................................................................... 39

vii

CHAPTE
METHOD
3.1 \I

3.2

3.3

CHAPT
RESUL
4.1

CHAPTER 3

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

METHODOLOGY. ............................................................................................................. 4 1

3.1 Variables _ ................................................................... 4 1

3-1-1 Linguistic variables . .................................................... 4 2

3. 1.1.1 Vowel identity, .................................................................. 42

3. 1.1.2 Consonant identity ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 42

3.1.1.3 Morpheme boundary type ............................................. 44

3.1.1.4 Pitch accent pattern ........................................................ 45

3.1.1.5 Consecutive devoicing environment ,,,,,,,,,,,,,,,,,,,,,,,,,, 46

3.1.1 Social variables - .......................... 4 3

3.1.2.1 Age ..................................................................................... 50

3.1.2.2 Sex ..................................................................................... 51

3.1.2.3 Social class ...................................................................... 52

3.1.2.4 Speech style ..................................................................... 55

3.2 Data Collection -- ................................................................ 56

3.2.1 Respondents ................ . _ _- _ ....................................... 56

3-2-2 Interview procedures ............................. __ 57

3.3 Data Handling . ........................................ 57

3.3.1 Acoustic analyses of the data __ ,,,,,,,,,,,,,,,,,,,,,,,, 57

3-3-2 Statistical analyses of the data ........................................ .62
CHAPTER 4

RESULTS ....................................................................................... 64

4-1 Linguistic factors--- __ __ _ ................................. 64

4.1-1 Preceding and following consonant. ............................................... 65

4.1.2 Vowel identity ................................................ 74

4.1.3 Accent ..................................................................... 75

4.1.4 Morphemeboundary. 84

4.1.5 Consecutive devoicing. ........................................................................ 89

4.2 Social factors ................................................... 9 2

4-2-1 Speech style- .............................................................. 93

4-2-2 Age and sex- ..................................................................... 94

4.2.3 Social class ............................................................................................. 99

4-3 Summary ............................................................................ 10"

 

viii

CHAF

DISCI

 

CHA
CON

APP

REF

CHAPTER 5

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

DISCUSSION ................................................................................................... 104

5 1 Linguistic factorsm ....................................... 104

5-1-1 Preceding and following consonants .......................................... 104

5.1.1.1 Status of [I] and [p] ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 105

5.1.1.2 Devoicing before a pause ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 108

5.1.1.3 Status of afin’cates ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 1 10

511' Vowel identity. _ __ ........................................................... 112

5-1-3 Accent ...................................... 113

5.1.4 Morpheme boundary __________ 116

5-1-5 Consecutive devoicing.-............... ......... 118

5-2 Social factors.-- ............................................ 120

5-3 Summary ......................................................................................................... 124

CHAPTER 6

CONCLUSION ____________ 127

6.1 Phonological vs. phonetic devoicing and standardness _______________ 128

6.2 Language change among younger speakers __________ _ _129

6.3 Language change and phonological/phonetic devoicing ,,,,,,,,,,,,,, 131

6.4 Variationistic quantitative approach ..................... 132

6-5 Summary and further study suggestions ............................................ 133

APPENDICES ................................................................................. 136

APPENDIX A: Word list .............................................. 137

APPENDIX B: Reading passage (Japanese) ................................................. 144

APPENDIX C: Neighborhood scores ............................................................... 146

APPENDIX D: Socioeconomic scores ............................................................. 147
APPENDIX E: Sample transcription of an interview conversation

(excluding interviewer’s part)____ _____ 149

REFERENCES 150

ix

 

 

Table 3.2

Table 3.3:
Table 3.4;
Table 4.1:

Table 4.2;

 

Table 4.3;

Table 4.4:

Table 45;

Table 4.5;
Table 4.7.
Table 4.8;
Table 4.9:
Table 4.10
Table 4.11
Table 4.12
Table 4.13
Table 4.14

Table 4.15

LIST OF TABLES

 

 

Table 1.1: Phonetic Inventory of Japanese Consonants____ ,,,,,,,,,,,,,,,,,,, 4
Table 2.1: Effect of preceding and following consonants (Yuen 1997) ,,,,,,,,,,,,,,,, 19
Table 3.1: Age Groups ..................................................................................................... 51
Table 3.2: Social Classes ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 55
Table 3.3: All Social Factors ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 56
Table 3.4: Coding of /kohitsuji/ __ ........................................................ 61
Table 4.1: Preceding consonant for IV - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 66
Table 4.2: Preceding consonant for lu/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 67
Table 4.3: Following consonant for /i/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 68
Table 4.4: Following consonant for /u/ - Goldvarb" ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 70

Table 4.5: Devoicing between preceding and following fricatives: [s] and [I] ..... 72

Table 4.6: Devoicing between a preceding affricate and following fricative ........ 72
Table 4.7: Devoicing between a preceding fricative and following affricate ,,,,,,,, 72
Table 4.8: Devoicing between preceding /h/ and following non-lh/ fricative 73

Table 4.9: Devoicing between preceding non-lh/ fricative and following lh/ ,,,,,,, 73

Table 4.10: Devoicing between allophones of /hl ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 74
Table 4-113 Vowel identity- Goldvarb ........................................................................... 75
Table 4.12: Pitch accent pattern for /i/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 76
Table 4.13: Pitch accent pattern for /u/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 77
Table 4.14: Accented vowels and the following consonant ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 79
Table 4-151 Accented vowels and age ......................................................................... 80

 

Table 4.

 

Table 4.
Table 4.
Table 4.
Table 4.
Table 4.2'
Table 4.2I

Table 4.2

 

Table 42
Table 4.2:
Table 4.21

Table 42

Table 4.2
Table 4.2

TabIe 4.3

Table 4.16: Age/Sex and Pitch Accent Pattern ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 82

Table 4.17: Morpheme boundary types for /i/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 85
Table 4.18: Morpheme boundary types for /u/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 86
Table 4.19: Morpheme boundary and the following consonant _____________________________ 87

Table 4.20: Morpheme boundary and consecutive devoicing environment ________ 88

Table 4.21: Consecutive devoicing environment for /i/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,, 89
Table 4.22: Consecutive devoicing environment for /u/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,, 90
Table 4.23: Consecutive devoicing environment and Style ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 92
Table 4.24: Speech style for /i/ - Goldvarb __________________________________________________________________ 93
Table 4251 Speech style for W - Goldvarb ................................................................ 94
Table 4.26: Age and gender combined for IV - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 94

Table 4.27: Age and gender for /i/ - Older Male, Female, and Middle Female

combined ....................................................................................................... 96
Table 4.28: Age and gender combined for /u/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 98
Table 4.29: Social class for M - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 99
Table 4.30: Social class — Working and Upper combined — for /ul _____________________ 100

xi

 

Table 4.1

I
Table 4.1

 

Table 4.1

Table 4.1

 

 

Table 42’
Table 4.2}

Table 4.2

Table 4.2:

Table 4.2'

Table 42
6

Table 42

Table 4.27

Table 4.28
Table 4.29

Table 4.30

Table 4.16: Age/Sex and Pitch Accent Pattern ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 82

Table 4.17: Morpheme boundary types for /i/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 85
Table 4.18: Morpheme boundary types for /u/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 86
Table 4.19: Morpheme boundary and the following consonant ,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 87

Table 4.20: Morpheme boundary and consecutive devoicing environment ________ 88

Table 4.21: Consecutive devoicing environment for /i/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,, 89
Table 4.22: Consecutive devoicing environment for /u/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,, 90
Table 4.23: Consecutive devoicing environment and Style ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 92
Table 424: Speech style for IV - Goldvarb .................................................................. 93
Table 4251 Speech style for NI - Goldvarb ................................................................ 94
Table 4.26: Age and gender combined for IV - Goldvarb ________ _ ....94

 

Table 4.27: Age and gender for /i/ - Older Male, Female, and Middle Female

combined ....................................................................................................... 96
Table 4.28: Age and gender combined for /u/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 98
Table 4.29: Social class for /u/ - Goldvarb ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 99
Table 4.30: Social class — Working and Upper combined - for lul ,,,,,,,,,,,,,,,,,,,,, 100

xi

n
I . -.—._.—__ ""

 

 

Figure 3
Figure 3
Figure 4

Figure 4.

Figure 4.

LIST OF FIGURES

Figure 3.1: Isee+katsu#hi/ - Voiced .............................................................................. 53
Figure 3.2: lsee+katsu#hi/ - Devoiced .......................................................................... 59
Figure 4.1: Age and Gender for li/ ,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, 95

Figure 4.2: Age and Gender for /i/ - Older Male, Female, and Middle Female
combined 97

ooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo

Figure 43: Age and Gender for NI ............................................................................... 98

xii

 

 

T
sociolinc
linguistic
M are u
between
1962, Mc
Vance IS
Madsen
OOVOICITII
°I Phone
BeCkmar
1995‘ N;
SDeleting
A
casItal, .
and Dho
Such as
Studies 1
Yuen (1

frequent

CHAPTER 1
INTRODUCTION

This study investigates the variation in Japanese vowel devoicing within a
sociolinguistic (variationist) framework, looking at social factors as well as
linguistic factors. Earlier research suggests that the Japanese high vowels /i/ and
lu/ are usually devoiced when they occur between voiceless consonants or
between a voiceless consonant and a pause (Sakuma 1929, Martin 1952, Han
1962, McCawley 1968, Kawakami 1977, Nihon Hoosoo Kyookai (NHK) 1985,
Vance 1987, Maekawa 1983 and 1988, Sugito 1988, Kondo 1995, Nagano—
Madsen 1995, Tsuchida 1997, Yuen 1997, Varden 1999). Japanese vowel
devoicing has attracted many researchers and has been well studied in the fields
of phonetics and phonology (Han 1962, McCawley 1968, Maekawa 1983,
Beckman and Shoji 1984, Vance 1987, Sugito and Hirose 1988, Kondo 1994 and
1995, Nagano—Madsen 1995, Tsuchida 1997, Varden 1999), but no thorough
sociolinguistic investigation has been done.

Although it is believed that vowel devoicing is more likely to occur in
casual, rapid speech than in slower, more formal speech, most of the phonetic
and phonological studies on Japanese vowel devoicing used controlled data,
such as reading framed sentences or isolated words. There are few quantitative
studies that have looked at the differences in devoicing in different speech styles.
Yuen (1997) shows that the more casual the speech style becomes, the more

frequently devoicing occurs and that men devoice more than women. In

 

particul
speech
devolci.
fiequen:

vowel d

I
l

Pronunc
be devo-
speakers

one can

 

irequentl
Women, i

Tl
399.8ex,
In Vanatic
answer It

All
Social 0m
derived fr
This WOrlt
made ab c
Identity, tr
boundary

which are

particular, his results show that women resist devoicing in the most formal
speech style; formal reading style. With these results, one might think that vowel
devoicing in Japanese is a nonstandard feature because it occurs more
frequently in more casual styles and men use it more than women. However,
vowel devoicing is also considered a standard feature in Japanese; the Japanese
Pronunciation and Accent Dictionary (NHK 1985) indicates where vowels should
be devoiced in standard Japanese, and the perception of the Tokyo dialect
speakers suggests that it is a standard Tokyo feature. Considering these facts,
one can ask questions such as “why does a standard feature happen more
frequently in casual conversation?” and “why do men use the feature more than
women, if it is a standard feature?”

The present study investigates these apparent contradictions by looking at
age, sex, social status, and speech style, which are common social factors used
in variationist sociolinguistics studies (see Chapter 3) and tries to suggest an
answer to this apparently contradictory pattern.

Although linguistic factors have been much more thoroughly treated than
social ones in previous studies of vowel devoicing, they have all too often been
derived from casual Observation or very limited studies Of actual speech samples.
This work focuses, therefore, secondarily, on the Claims of previous research
made about such factors. The linguistic factors used in this study are vowel
identity, the identity of the preceding and following consonants, morpheme
boundary type, pitch accent pattern, and consecutive devoicing environments,

which are commonly discussed in phonetic/phonological studies of Japanese

 

vowel c

 

quanbt

suppleI

Japane

morpho

1.1 Ja

When iii

affricate

II-I-I).t

accentu;

phonGIGS

DSrCGiVe

vowel devoicing (see Chapter 2). It is hoped that, by investigating these factors
quantitatively in different speech styles, natural language usage will provide
supplementary information to previous studies.

In the following subsections, I will provide some background on the
Japanese language; I will first introduce Japanese phonology (1 .1) and

morphology (1.2); and then discuss my research questions and hypotheses (1.3).

1.1 Japanese phonology

The basic Japanese syllable structure is (C)V. A coda is allowed only
when it is a part of a geminate, when it is a nasal followed by a stop, fricative, or
affricate that shares the place feature with itself, or when it is a moraic nasal.

In the following, I will first briefly summarize the role of mora in Japanese
(1.1.1), then introduce Japanese phonemes (1.1.2), and outline the Japanese

accentuation system (1.1.3).

1.1.1 Syllable vs. mora

Japanese is a mora-timed language, which means that the basic
phonological unit that native speakers perceive is a mora. For example, in the
English word London, there are two syllables (London), but Japanese speakers

perceive it as having four moras (Lo.n.do.n) (T sujimura 1996). This is also

 

reflect
distinct

accent

1.1.2 .1

 

 

 

J

transcril

Phonetl

and he

baSEd r

47

Tie:

‘71
3
8
z"
r

L/

>

3:

3.
3

77

reflected in their writing system, in which one symbol represents one mora‘. This
distinction between syllable and mora is also important in considering the pitch

accent system in Japanese (cf. 1.1.3 below).

1.1.2 Japanese Phonemes

Japanese has five vowels: la, i, u, e, 0/. la, i, e, O/ are phonetically
transcribed as [a, i, e, 0], but M is usually an unrounded high back vowel,
phonetically transcribed as [w] (T sujimura 1996).

Japanese consonants consist of stops, fricatives, affricates, approximants,

and nasals. Table 1.1 shows the inventory of the (major) Japanese consonantsz

based on Tujimura (1996) and the IPA.

Table 1.1: Phonetic Inventory of Japanese Consonants

 

 

 

 

 

 

 

 

 

 

 

bilabial alveolar aleveo- palatal velar uvular glottal
palatal

Stops voiced b d g

voiceless p t k
Fricatives voiced z

voiceless q, s 5 9 h
Affricates voiced dz d3

voiceless ts ll
Approximants
liquids voiced f
glides voiced i w
Nasals voiced m n ,3 N

 

 

 

 

 

 

 

 

 

 

 

 

1 Small ‘ya', ‘yu’, and ‘yo’, which form a part of a palatalized sound in combination with the
greceding letter, are exceptions to this; e.g, Us: (‘hi’ + small ‘ya'=’hya').

There are other consonants, such as alveO-palatal and palatal nasals. which are caused by
assimilation to the following consonant, but they are excluded here for sake of simplicity.

 

 

 

ahmai

lullet
Oamrb
phoner
behwe
andlh‘
Mghv.
belt)”

0141

those

kaku
reCer
Seun
Dhon

1996

All the (oral) stops (lp, b, t, d, k, g/) are phonemes in Japanese. The
phonemic status of the fricatives and affricates, however, is not so simple. For
example, [3] occurs before the vowels lal, lul, /e/, and /o/, but not before /i/. [1‘]
occurs before /i/ instead. However, [I] can also occur before lal, /e/ (in recent
loan words), lul, and IOI; therefore, it is considered a phoneme. A similar
alternation occurs with its voiced counterpart /z/. [2] occurs before the vowels /a/,
lul, lel, and lol, but not before lil. [d3] occurs before /i/ instead, but since [d3] can
occur before lal, le/ (in recent loan words), /u/, and /0/ also, it is considered a
phoneme. Similarly, [t] occurs before non-high vowels lal, /e/, and /o/, but not

before the high vowels Ill and /u/. [tj‘] occurs before /i/ and [ts] before /u/. [c], [<11],

and [h] are other examples that show similar alternation. [h] occurs before non-

high vowels lal, /e/, and /O/, but not before the high vowels [i] and /u/. [9] occurs
before /i/ and [<11] occurs before /u/. Thus, it seems that [tj] and [ts] are allophones
of /t/, and [9] and [<11] are allophones of /h/. However, there are cases where

those sounds occur before other vowels, as in [tja] ‘tea', [méotsarluto] ‘Mozart’,

[caktu] ‘hundred’, and [geckm] ‘fork,’ respectively. Some Of these words are

recent loan words, but others are not. Therefore, the standard treatment of these
sounds is that they are allophones when they occur before /i/ or lul, but are
phonemes when they occur before other vowels (Shibatani 1990, Tsujimura

1996).

Japanese has one liquid M, which is similar to flapped [d] in English. It

may be phonetically transcribed as [1]. Other approximants are the glides /j/ and
/w/.

Japanese nasals include [n], [m], [13], and [N]. [n] and [m] occur in the
onset position, but [N] occurs in the coda position. [1]] is considered as an
allophone of /g/ or /n/ because it only occurs intervocalically, as in /kagami/
[kagami] “mirror” (at least among Older speakers), or before a palatal stop as in
/kangaeru/ [kaygaeru] ’to think’; it never occurs in the onset position. On the other

hand, Shibatani (1990) posits the phoneme /N/ as a moraic nasal. When it is

followed by a pause, it is a uvular nasal [N], but when it is followed by a

consonant, it assimilates to the following consonant and changes its place of

articulation. However, it is possible to consider it as an allophone of /n/, and n is

the standard spelling (in romanization) for the coda nasal. Tsujimura (1996)
seems to take this position; /n/ is realized as [n] in the onset position, as [N]
before a pause, and as [n], [m], or [13] before another consonant due to a place

assimilation.

1.1.3 Japanese Accentuation System

Japanese is a pitch-accent language, in which each mora is associated
with a specific pitch, and “the pitch or tonal pattern Of the entire word is

predictable given the location Of the accent of a word” (T sujimura 1996: 74). In

Japanese, each more is associated with either high (H) or low (L) pitch, and pitch
accent pattern is phonemic; a different pitch accent pattern with the same

segmental sequence may result in different words. Shibatani (1990) gives an

example of three-way distinction of the same segmental sequence lhali/ in the

Kyoto dialect, as in (1 ).

(1 ) lhaj'I/ a. HH ‘edge’

b. LH ‘chopsticks’

c. HL ‘bridge’

In Japanese, accent is on the mora with a high-pitch followed by a low-
pitch, as shown in (2), taken from the Tokyo dialect, where ‘+’ indicates a bound

morpheme boundary, ‘#’ a compound word boundary, and ‘##’ a word boundary.

(2) a. kumo HL ‘clouds’
b. kawa(##ga) LH(L) ‘river-Subj.’

c. kan+kaku#kl+kan LHHHHLL “sense organ’

In a word with an accent on the final mora, as in (2b), we can tell that the
accent is on the final mora by attaching a particle such as ga. Once we know the
accent location, the pitch accent pattern of the rest of the word is predictable in
the Tokyo dialect. All the moras preceding the accented mora receive a high-

pitch, and the initial mora receives a low-pitch by the Initial Lowering Rule

 

 

(Hera:

syllab!
a mor

 

128

l
influenr
decus:

ITIOfphr

ahech~
ending
shown

iflfiecu‘

(Haraguchi 1977), as in (20). Thus, pitch accent can be different within a single
syllable (kan in 2c), which shows that the pitch accent-bearing unit in Japanese is

a mora.

1.2 Brief discussion of Japanese morpheme types

In the present study, morpheme boundary is included in order to see the
influence of different morpheme types on vowel devoicing (see Chapter 2). I will
discuss only the relevant parts of Japanese morphology here, namely,
morpheme types.

Japanese, like other languages, has free morphemes, such as nouns,
adjectives, and verbs, and bound morphemes, such as verbal conjugation
endings (inflectional morphemes) and prefixes (derivational morphemes), as
shown in (3), taken from Tsujimura (1996: 142-143). (3a-d) show verbal and

adjectival conjugation endings, and (3e) and (3d) show the prefix ‘su-‘.

(3) a. tabe-ru ‘eat-Non-past tense (verb)=eat, will eat’
b. tabe-ta ‘eat-Past tense (verb)=ate’
c. ooki—i ‘big- Non-past tense (adjective)=is big’

d. ooki-kat-ta “big-Past tense (adjective)=was big’
e. su-asi ‘bare leg’

f. su-hada ‘bare skin’

Japanese is also rich in compounding. Shibatani (1990) gives ample
examples Of different types of compounding, such as native compounds, Sinc-
Japanese compounds, and hybrid compounds (237-238), some of which are

reproduced in (4) below.

(4) a. aki-zora Native, Noun-Noun ‘autumn sky’

b. tatj‘r-yomi Native, Verb-Verb ‘reading while standing”

C. ki-soku SinO-Japanese rule

d. ita-tjoko Hybrid, native-foreign ‘chocolate bar’

(43) is an example Of a native compound where two nouns are combined;
aki ‘autumn’ and sore ‘sky’. (4b) is another example Of a native compound, but
two verbs are combined; tatsu ‘stand’ and yomu ‘read’. (4c) is an example of a
SinO-Japanese compound, but both ki and soku are bound morphemes (and as a
unit, means ‘rule’), unlike other examples in (4), in which both members of the

compounds are free morphemes. Finally, (4d) is an example of a hybrid

compound, in which a native word ita ‘board’ and a foreign loan word tfoko

‘chocolate’ are combined.

These are the distinctions used in the present study: free vs. bound
morphemes, and within free morphemes, compound boundary vs. word
boundary (across different grammatical categories, such as verbs, nouns,

adjectives).

Of course there are many other features of Japanese; this study is not a
grammatical survey of the language. I have attempted here only to provide
background information on the language features that will prove important to the
consideration of the distribution and occurrence of vowel devoicing as it may be

influenced by these factors.

1.3 Research Questions and Hypothesis

The specific questions to be answered in the present study are the

following:

1. Which preceding and following consonants promote devoicing?

2. Is there an interaction between the preceding consonant and the
following consonant?

3. Does vowel identity have any influence on Japanese vowel
devoicing?

4. Does morpheme boundary have an effect on devoicing, and if so, in
all styles? Is there a difference among different morpheme types?

5. Does pitch accent have any effect on devoicing? If so, does it have
the same effect for different age groups?

6. Is vowel devoicing demoted in a consecutive devoicing
environment? What is the actual behavior in a speaker's natural
usage?

7. Does the devoicing rate change in different styles?

8. Is there any difference between men and women in terms Of
devoicing?

9. Do different age groups behave differently in terms of devoicing?

10. DO different social classes behave differently in terms Of devoicing?

10

I will provide a detailed hypothesis to each question related to linguistic
factors (1) through (6) in Chapter 3. As for (7), according to Yuen (1997), there
are stylistic differences, and the more casual the style becomes, the more
devoicing occurs. As for (8), Yuen (1997) shows that there is a difference and
that men devoice more than women. As for (9), there are some reports that the
younger generation behaves differently from Older generations in term of the
devoicing of accented vowels (refer to Chapter 2 for details). As for (10), if vowel
devoicing has a social meaning, i.e. degree of standardness, then we should
expect some different behaviors in different social classes if it is indeed a
nonstandard feature; lower social class speakers with a higher percentage of
devoicing, and higher social class speakers with a lower percentage because the
pattern in style and sex differences suggest this may be a nonstandard feature.
We would also expect that the age group that is most sensitive to social prestige
will be least likely to devoice vowels in a more formal style. In all these cases,
however, it will be important to recall that devoicing is considered standard (in
some environments) in Tokyo Japanese, according to both dictionaries and

speaker perception.

1.4 Organization

In the next chapter, I will provide an overview of the issues commonly

touched on in discussions of Japanese vowel devoicing. Chapter 3 provides

11

detailed information on the variables used in the present study, both linguistic

and social, coding of those variables, and data collection and handling
procedures. Chapter 4 discusses the results of the statistical analyses Of the data,
and Chapter 5 provides the detailed discussion of the results. Finally, Chapter 6
concludes with a summary, proposed interpretations of the results, and further

research suggestions.

12

CHAPTER 2
BACKGROUND

In this chapter, I will first review different accounts Of vowel devoicing in
Japanese, then discuss the linguistic and social factors that have been argued or

assumed to affect Japanese vowel devoicing.

2.1 Phonological and phonetic accounts of Japanese vowel devoicing
Japanese vowel devoicing has been traditionally described as a
phonological process, in which the feature [-voice] spreads from neighboring
voiceless Obstruents to a high vowel (/i/ or /u/) in an unaccented syllable
(McCawley 1968, Vance 1987). However, as summarized in Tsuchida (1997),
this is controversial. Some studies show that there is a difference in the effect of
preceding and following consonants on vowel devoicing between fricatives and
stops. A preceding fricative seems to promote devoicing more than a preceding
stop (Han 1962, Maekawa 1983 and 1988, and Tsuchida 1997) and a following
stop more than a following fricative (Takeda and Kuwabara 1987 and Yoshida
and Sagisaka 1990 [cited by Nagano—Madsen 1995], Nagano—Madsen 1995). If
vowel devoicing is due to the [-voice] spreading, it should not show different
degrees Of effect between fricatives and stops since both are specified for [-
voice]. It has also been pointed out that devoicing between two fricatives is less
likely to occur (Sakuma 1929 [cited in Tsuchida 1997], Tsuchida 1997). The

phonological [-voice] spreading approach cannot explain this tendency either.

13

Another problem with the [-voice] spreading account is that more recently the
feature [voice] has been argued to be a privative feature, not a binary one (e.g.
Lombardi 1991 ). Therefore, the argument that vowel devoicing is a phonological
(categorical) rule Of [-voice] feature spreading cannot be entirely supported.

Some researchers argue that vowel devoicing is a phonetic vowel
reduction process due to a glottal gestural overlap (Jun and Beckman 1993,
Kondo 1994). However, if Japanese vowel devoicing is due to a phonetic vowel
reduction process, it is difficult to account for the fact that high vowels in different
consonant environments are more or less likely to devoice (Sakuma 1929,
Tsuchida 1997). A vowel reduction approach also cannot account for the fact that
in a consecutive devoicing environment, in which there is more than one
devoiceable vowel in a row, devoicing Of both vowels in the two successive
syllables is avoided (Han 1962, Maekawa 1988, Tsuchida 1997 among others,
see 2.2.4. below for more details).

Tsuchida (1997) argues that there are two types Of vowel devoicing in
Japanese: phonological and phonetic. She argues that voiceless fricatives and
devoiced vowels are both specified for [spread glottis] ([s.g.] hereafter) based on
the articulatory data, and that when a high vowel occurs next to a voiceless
fricative and devoiced, it receives [s.g.] from the neighboring fricative in a
phonological operation. When a high vowel comes between two stops, the
feature [s.g.] is inserted by 3 Gen operation. This is articulatorily supported
because the devoiced vowel and the neighboring voiceless consonants are

produced with a single glottal opening. Tsuchida further argues, based on the

14

facts that a high vowel between two voiceless fricatives is less likely to devoice,
(NHK 1985, Tsuchida 1997), that vowel devoicing is prohibited in phonology
when a high vowel occurs between two fricatives. She also argues that, when the
vowel is devoiced in that environment, two glottal openings were Observed,
instead of one observed in other environments of phonological devoicing. She
accounts for this restriction by an OCP violation, which is caused by having two
[s.g.] features in a row when the vowel is devoiced, where the feature [s.g.] is
shared by the preceding voiceless fricative and the devoiced high vowel. This is

schematically shown in (5) below.

(5) voiceless fricative hi h vowel voicelessl fricative

[s.g.] [s.g.]

Tsuchida also argues, based on the fact that the devoicing before

allophones of /h/ ([h], [(1)], and [9]) is less likely to occur (NHK 1985, Nagano-

Madsen 1995, Tsuchida 1997), that the feature [s.g] cannot be shared across
syllables (i.e. by a high vowel and the following /h/) because /h/ does not have a
supralaryngeal node.

However, it is well known that devoicing sometimes occurs in the following
environments: (1) when a high vowel occurs between two voiceless fricatives
(Sakuma 1929, NHK 1985, Tsuchida 1997), (2) when a high vowel occurs before
an allophone of IN (Sakurai 1985, Tsuchida 1997), (3) when a high vowel occurs

after a voiceless consonant and before a voiced consonant (Han 1962, Maekawa

15

1988, Tsuchida 1997), and (4) when a non-high vowel occurs between voiceless
consonants (Maekawa 1988, Tsuchida 1997). Tsuchida (1997) claims that these
are all cases of phonetic devoicing caused by undershooting a target. If the goal
is total voicing, sometimes speakers fail to reach the target, resulting in phonetic
devoicing.

Since the scope of this study is primarily to determine social factors in
Japanese vowel devoicing, I will not seek to determine if vowel devoicing is a
phonetic process, phonological rule, or a combination. I use the phonetic identity
of the preceding and following consonants rather than their features in my
characterization Of units and environments, although I will make some reference
to those features in the discussion of my findings.

Nevertheless, a careful acoustic analysis suggests that devoicing in some
of the environments which Tsuchida calls “phonetic” are as completely devoiced
as others which she suggests are sanctioned, and they are acoustically

indistinguishable.

2.2 Linguistic factors
In this section, I will discuss the major linguistic factors that affect
Japanese vowel devoicing; i.e. preceding and following consonants, vowel

identity, accent, consecutive devoicing environments, and morpheme boundary

types.

16

2. 2. 1 Preceding and following consonants

The effect Of preceding and following consonants is the most discussed
factor of vowel devoicing in Japanese. As mentioned in the previous subsection,
it has been pointed out that fricatives and stops (and affricates) have a different
degree of effect on the frequency of vowel devoicing in Japanese. A preceding
fricative promotes devoicing more than a stop or an affricate (Han 1962,
Maekawa 1983, Kondo 1997, Yuen 1997) and a following stop, more than a
fricative (Takeda and Kuwabara 1987, Yoshida and Sagisaka 1990 [cited by
Nagano-Madsen 1995], NaganO-Madsen 1995, Yuen 1997). However,
researchers disagree on which part Of the context has a stronger effect on vowel
devoicing, namely, the preceding consonant or the following consonant.

Traditionally, it was assumed that the preceding consonant has the most
significant effect on vowel devoicing in Japanese. Han (1962) says that a vowel
following an affricate is "more unvoiced" (89) than one following a stop, which is
generally not unvoiced, and that fricatives ”show greater effect on the unvoicing
of the vowel” (89) than affricates. However, information about her respondents
and the details of the experiments are not provided. She also limited the following
consonant to [k]. Maekawa (1983) also found, in his acoustic analysis Of 10
minutes of speech of a female TV broadcaster, that a preceding fricative
promotes devoicing more than a stop or an affricate. However, he does not
provide the following consonant identity. Neither of these studies investigated the

importance of the following consonant.

17

On the other hand, most of the newer phonetic studies of Japanese vowel
devoicing show that it is actually a following stop that has a stronger effect on
devoicing (Kawakami 1977, Takeda and Kuwabara 1987, Yoshida and Sagisaka
1990 [all cited in Nagano—Madsen 1995], NaganO-Madsen 1995, Yuen 1997).

Nagano-Madsen (1995) found, in her acoustic study of a female Standard

Japanese speaker’s speech using frame sentences, that vowels followed by [l],

[c], and [(1)] are less likely to devoice, and that the preceding consonant did not

show any influence. Kawakami (1977) says, however, that a preceding fricative
or an affricate does not promote devoicing even though a preceding stop
generally promotes it, contradicting other studies. Yuen (1997) found that a
preceding fricative promotes devoicing more than a stop or an affricate, but a
following stop promotes devoicing more than a fricative, an affricate, or a pause.
He found that the following stop has a statistically stronger effect on vowel
devoicing than a preceding fricative. However, lmai's (1997) research, using
casual conversational data from four young native Tokyo dialect speakers, found
that a preceding consonant identity has a statistically more significant effect on
vowel devoicing than a following consonant identity, contrary to Yuen's results. In
addition, lmai's (1997) results show that vowel identity is also important. A
preceding fricative very strongly promotes the devoicing of /i/, but a preceding
stop or affricate, both Of which share a [+stop] feature, strongly promotes
devoicing for lu/.

Tsuchida (1997) argues, as described in the previous subsection, that the

devoicing of a high vowel between two fricatives and before /h/ is prohibited. This

18

 

 

seem
stop p
prohib
comm
has a
the po.
(1997)
conson
Table 2

COTTSOT’IE‘

 

 

 

The l0ll0\
preceding
COTTSOnar.|
COnSonan

S
4I8IUS

(199
.‘if lite
YUen (199'

seems consistent with the claim, made by most phonetic studies, that a following
stop promotes devoicing because these two environments, in which devoicing is
prohibited, have a following fricative (lh/ is usually categorized as a fricative) in
common. Thus, Tsuchida seems to assume that the following consonant identity
has a strong effect on vowel devoicing. However, she also seems to agree with
the position that a preceding fricative promotes devoicing more than a stop. Yuen
(1997) reports that there is an interaction between the preceding and following
consonants, e.g., preceding fricatives and following stops promote devoicing.
Table 2.1 is Yuen's results Of a factorial test for preceding and following

consonants, reproduced from Yuen (1997: 31).

Table 2.1: Effect of preceding and following consonants (Yuen 1997)

 

 

 

 

 

 

 

 

 

DF F-value P-value

Preceding F (2, 1783) 4.255 0.0143
Consonant

Following F (3, 1783) 38.177 < 0.0001
Consonant

Preceding * F (6, 1783) 28.201 < 0.0001
Following

 

The following consonant has a stronger effect on the voicing mean3 than the
preceding consonant because its F-value is higher than that Of the preceding
consonant. The results also indicate that the combination Of the two types Of
consonant is significant“. It is possible that this preceding and following

consonant interaction may be the reason for the controversy of which consonant

 

3 Yuen (1997) used the measure of "voicing mean” which is calculated based on their voicing
status; if the vowel is deleted, it was assigned 0, if devoiced, 1, and if fully voiced, 2.
4 Yuen (1997) does not provide further details of the interaction.

19

context has stronger effect on vowel devoicing. However, lmai (1997) did not find
a statistically significant interaction between the preceding and the following
consonants.

There is also some debate about the effect of a pause. The traditional
description of Japanese vowel devoicing is that a high vowel is devoiced when it
occurs between voiceless consonants or after a voiceless consonant and before
a pause. However, Maekawa (1988) says that an utterance-final pause (including
word-final position in words in isolation) has the same influence as a voiceless
consonant, i.e., it promotes devoicing, but a sentence-intemal pause seems to
demote devoicing. This tendency was Observed in his study conducted in a town
in the Tottori prefecture (in the western part Of Japan), in which the respondents
were asked to read a passage (he only reports the results of 29 male
respondents). However, he seems to attribute the negative influence Of a
sentence-intemal pause to the raised pitch right before a pause, as in watashi ga
jussai no toki## (the accent mark indicates raised pitch, and '##' marks a word
boundary) 'when l was 10 years old,‘ where it is Obligatory that the vowel is
voiced (Kawakami 1977, cited by Maekawa). lmai (1997) also found that the
following pause demotes devoicing, but, in this study, a mora with a raised pitch
before a pause was excluded. Therefore, we cannot account for the negative
influence Of a sentence-intemal pause simply by the pre-pausal raised pitch.

This negative influence of a sentence-intemal pause seems to be
compatible with at least a part of Sugito's (1988) findings in her study of regional

differences of Japanese vowel devoicing, specifically the results from the Tokyo

20

dialect, because devoicing was demoted before a sentence-intemal pause in
Tokyo but not in Nagoya (see 2.3.1. below).

To summarize, the relative importance Of the preceding and the following
consonant is still controversial, but most studies seem to agree that a preceding
fricative and/or a following stop promotes devoicing. The effect Of a pause seems
to differ depending on the position: A sentence-final pause seems to promote

devoicing, but a sentence-intemal pause demotes it.

2. 2.2 Vowel identity

The effect of the vowel identity on vowel devoicing is also controversial.
Han (1962) argues that only the high vowels (/i/ and /u/) undergo devoicing
because the high vowels are inherently shorter than the non-high vowels (lal, lel,
and lo/). Among the high vowels, she argues that “/u/ is more readily unvoiced

than /i/" based on the results of her experiments, in which lu/ was devoiced at a

slower rate than /i/ in the same environment (e.g., /kiji/ ‘shore’ and /ku]'r/ ‘comb’).

However, Yuen (1997) reports otherwise: he found that /i/ is more likely to be
deleted/devoiced than lu/5. On the other hand, Maekawa (1983) found no
difference in the devoicing rate of /i/ and /u/ in his analysis of a female TV
broadcaster’s speech. lmai (1997) also found no difference in the devoicing rate

between the high vowels.

 

5 Yuen (1997), and other phoneticians, uses a phonetic transcription ([i] and [03]). but I will use a
phonemic transcription for the sake of simplicity.

21

2.2.3 Accent

Accent is another well-studied factor of Japanese vowel devoicing. It has
been argued that a high vowel between voiceless consonants either does not
undergo devoicing when it bears an accent, or that the accent is shifted to either
the preceding or the following vowel to allow devoicing (Sakuma 1929, Han
1962, Sakurai 1985, Maekawa 1988, Sugito and Hirose 1988, Nagano-Madsen
1995). However, more recent studies show that even an accented vowel can
undergo devoicing, particularly among younger speakers (Sugito 1982, Sakurai
1985, Maekawa 1988, Sugito and Hirose 1988, Kondo 1993, Nagano-Madsen
1995, lmai 1997, Tsuchida 1997).

Han (1962) claims that a high vowel between voiceless consonants is
generally not devoiced when it has a high pitch but is devoiced when it has a low
pitch. However, Maekawa (1988) found no effect of pitch height (i.e., high or low)
on vowel devoicing in his analysis of a female TV broadcaster’s speech. Han
studied minimal pairs with a pitch height difference (e.g., /krkukotol 'to listen'
LHHL and /kikukoto/ 'with Kikuko' HLLL, where L indicates a low pitch, H
indicates a high pitch, and the italicized letter indicates a devoiced vowel). She
claims that /i/ in the former and lu/ in the latter are devoiced because they have a
low pitch, which is not entirely supported because it is possible that other factors
may be playing a role in the former (e.g., the word-initial position or the first
syllable in a sequence Of devoiceable vowels, which is discussed in 2.2.4 below).

Although these two studies used the pitch height of the mora, a more

common description of the interaction between pitch accent and vowel devoicing

22

 

btha
andi

Made

shfie
no ac
never
They
devoi

aCOUS'

deVoi
that u

a”Tor

is that vowel devoicing is blocked in an accented mora, which has a high pitch
and is followed by a low pitched mora (Sakurai 1985, Sugito 1988, Nagano-
Madsen 1994).

Sugito and Hirose (1988) state that, in the Tokyo dialect, accent is usually
shifted to the following vowel if the vowel is devoiced, but in some cases there is
no accent shift and the high vowel is still devoiced. They claim that accent shift
never happens in the Kinki dialect, thus producing accented, devoiced vowels.
They found that ”it is the following descending tones that make preceding
devoiced vowels heard as accented” (21 ). NaganO-Madsen (1995) found, in her

acoustic study Of a female speaker of Standard Japanese using frame

sentences, that devoicing was strongly demoted before [1'], [9], and [(1)] when the

vowel was accented. This agrees with the description in the Japanese
Pronunciation and Accent Dictionary (NHK 1985) that devoicing is inhibited
before /h/ when the vowel is accented. However, Nagano—Madsen (1995) states
that ”recent works using longer words and a sentence database found no
overriding influence of accent in preventing devoicing" (566) and "due to the
restricted and uncontrolled corpus, these studies [Kondo 1993 and Nagano-
Madsen 1994] did not reveal much clarity as to the relationship between
accentuation and segmental contexts” (566).

However, Sakurai (1985) states that the effect Of accent on vowel
devoicing is weaker among younger generations. Tsuchida (1997) also argues
that there is no effect Of accent on vowel devoicing in native Japanese words

among younger speakers, but Older speakers still maintain this restriction.

23

 

How
speai
of Bel
acce
a VOV.

vowel

more I

not ve

devoi
there
seen

DENJS

you

Dre

However, in loan words, accent does have a strong effect, even among younger
speakers, and she attributes this difference to the difference in the representation
of accent in native words and loan words: native words have an underlying
accent linked to a vowel, but loan words have a floating accent, which is linked to
a vowel by a phonological rule, which therefore avoids accented, devoiced
vowels. (Refer to Tsuchida 1997 for more details.)

On the other hand, Yuen (1997) still found that unaccented vowels are
more likely to be devoiced than accented ones. Therefore, the effect Of accent is
not very clear.

However, a high vowel before a pause and after a voiceless consonant is
devoiced only when the vowel has a low pitch (Martin 1952, NHK 1985), and
there is no study that has found contradictory results. Therefore, accent would
seem to have a strong effect in this environment; Le, a high vowel before a
pause.

In summary, it seems that the effect Of accent is weak particulariy among
younger generations. However, having a high pitch in the word-final mora

prevents devoicing, regardless Of age.

2. 2.4 Morpheme boundaries

It is also argued that morpheme boundaries have a blocking effect on
vowel devoicing when combined with other factors (Sakurai 1985, Vance 1987,

1992, Kondo 1997, Tsuchida 1997). Sakurai (1985) says that when a

devoiceable vowel is followed by Isl, lhl, or ljl, and there is a morpheme

24

boundary after it, it is less likely to devoice (e.g. /booekj#suijun/ 'a trade standard'
[underlining indicates a non-devoiced vowel and '#' marks a compound word
boundaryj). Other studies argue that, if it is the only devoiceable vowel, a high
vowel may be devoiced at a word boundary, but in a consecutive devoicing
environment, the vowel at a morpheme (or word) boundary remains voiced

(Vance 1987, 1992, Kondo 1997, Tsuchida 1997).

2.2.5 Consecutive devoicing

When there is more than one devoiceable vowel in a row, previous
research suggests that not all of the vowels are likely to be devoiced (Sakuma

1929, Martin 1952, Han 1962, Kawakami 1977, Maekawa 1988, Tsuchida 1997),

avoiding consecutive devoicing, as in lkjj'rtsu/ 'temperament' (LHH pitch accent

pattern). The first high front vowel /i/ is devoiced, but the second one is not.

Han (1962) says that “the unvoicing of vowels in two successive syllables
is rare, and the unvoicing Of three successive vowels does not occur" (91 ). On
the other hand, Kawakami (1977) says that when there are many devoiceable

vowels, sometimes those vowels are voiced. His example is as follows.

(6) lj'lj‘utsu/ ‘expenses’ -> [fij‘tqtsm], Ujugtsm], [jjj‘urtsw], [[jurtsm], and
mruesmr

Kawakami (1977) Claims there are five possible realizations of ljijutsul;

one in which the first two vowels are devoiced ([jjj‘ugtsmj), one in which the first

25

 

VOWE
TITSI ‘1
the fi

whicf'

loo rr-

 

claim

simila
cock')

realiz;

OCCUT

SIOw

8We
in if
Ciisl

1811

vowel is deleted and the second vowel is devoiced ([l‘jugtsmj), one in which the
first vowel is devoiced and the second vowel is voiced ([ljjmtst. one in which
the first vowel is deleted and the second vowel is voiced ([[j‘urtsw]). and one in

which the first vowel is voiced and the second vowel is devoiced ([j‘rjugtsml). He

claims that all the vowels are not devoiced because, if they were, there would be

too many voiceless consonants in a row, which might cause confusion between

similar words, such as Ij'rjamo/ -> [jjamo] (a kind of fish) and [jamo] (‘a fighting

cock’). This seems to indicate that there is considerable variation in the
realization Of a word with consecutive devoicing possibilities.
Sakuma (1929) states that devoicing in two consecutive syllables does

occur, as in (7).

(7) [k(u;)tjjsaki] ‘muzzle’
[kjkugtji] person’s last name

However, he also says that when a speaker is pronouncing the word
slowly, it is usually only the first vowel that is devoiced ([kugtjisaki] and [kjkurtji]).

More recent studies seem to agree that the devoicing of consecutive
syllables is avoided. Maekawa (1988) reports the results Of his study conducted

in the Tottori prefecture and says that, for the word lsukikirai/ (‘likes and

dislikes’), there were three different realizations; three speakers pronounced it as

[smkikirafl, nineteen as [sugkikirafl and four as [sugkjkirai]. He says that there are

26

 

hvoir
segu]

incat

I
many
sever

envhc

 

 

hstol'

envuo

~Tsu.
Pact
Vth

IChk

two interpretations of this; one is that the first syllable in the devoiceable
sequence is more likely to devoice, and the other is that the vowel following a
fricative is more likely to devoice.

Avoidance of consecutive devoicing has been mentioned and studied by
many researchers, but the system of avoidance is still not clear. There are
several factors that control the devoicing pattern in consecutive devoicing
environments. (Refer to Tsuchida [1997] for a more thorough review.) Below is a
list of the restrictions in the patterning of devoicing in a consecutive devoicing

environment derived from previous studies.

1. An accented vowel resists devoicing (Martin 1952, Han 1962), e.g.,
[kujjkumO] 'strangely.‘

2. The vowel in a word-initial mora is more likely to devoice (Martin 1952,
Yuen 1997), e.g., [fikitsuméru] 'spread completely.‘

3. A preceding fricative promotes devoicing (Han 1962), e.g., [sekijjtsu]
'room made of rock.’

4. A high vowel preceding a word boundary resists devoicing (Vance 1987,
1992, Kondo 1997), e.g., [kankaku#kjkan] 'sense organ.’

Tsuchida (1997) provides a phonological analysis of the interaction of these
factors in the framework of Optimality Theory. She posits a set of constraints,
which are responsible for the devoicing pattern. However, Tsuchida concludes as

follows.

27

[W]hen there are more than two devoiceable vowels in consecutive
syllables, the locus of devoicing is determined in two steps: first, either the
initial or second vowel in the sequence is selected based on the ranked
constraints, then the voicing of the following vowels is determined in an
alternating pattern. (272)

Even though she can account for the devoicing pattern of shorter ,
consecutive devoicing environments (up to two) using her ranked constraints,
she cannot account for the patterns in a longer environment (with three

devoiceable vowels) with only her constraints.

2.2.6 Summary

In this subsection, various linguistic factors that have been argued to have
an effect on vowel devoicing were discussed. Most studies agree that a
preceding fricative and/or a following stop promotes devoicing. However, Yuen's
(1997) claim that there is an interaction between the preceding and the following
consonants is interesting because researchers disagree on which side of the
vowel context has a stronger effect on devoicing. The effect of a pause is
different depending on the position: a sentence-final pause promotes devoicing,
but a sentence-intemal pause demotes it. Vowel identity does not seem to be a
factor, but this is still controversial. Accent does not have an effect among
younger generations but has an effect among Older generations. In consecutive
devoicing environments, there seems to be a set Of constraints involved: an
accented vowel and a vowel at a word boundary resist devoicing, but the first

vowel in a sequence Of devoiceable vowels and a vowel with a preceding

28

fricative are more likely to devoice. A high vowel in a devoicing environment at a
morpheme boundary also resists devoicing, particularly in a consecutive

devoicing environment and before lh/.

2.3 Social factors

Most studies of vowel devoicing in Japanese are phonetic ones, which try
to discover the articulatory mechanisms of the phenomenon, and there are only a
few studies that investigate the social aspects of this phenomenon. Maekawa
(1988) mentions the "standardness" of vowel devoicing in Japanese. It is
considered “standard” in both production and perception: 1) there are some fixed
patterns for devoicing where more than two vowels are devoiceable, and 2) there
are cases in which, if a vowel is not devoiced, it sounds strange (at least to
speakers Of standard or devoicing varieties). Maekawa (1988) also mentions the
possible influence of age, sex and speech style but says there is almost no
research in the field that has focused on these social factors; he suggests that a
relatively well-studied social factor concerning vowel devoicing is regional
differences, which have been studied by Japanese dialectologists (language
geographers and comparative dialectologists).

In this section, I will discuss the social factors that are argued to affect
vowel devoicing in Japanese, namely, region, age, sex, speech style, and social

class (socioeconomic status).

29

2. 3. 1 Region

A relatively well-studied social factor Of Japanese vowel devoicing is
region. Since Kindaichi 1954 (cited in Maekawa 1988), Japanese dialectologists
have agreed that there is regional variation in vowel devoicing in Japanese.
Mase 1977 (cited in Maekawa 1988) divides the dialects in Japan into eastern
and western dialects and considers vowel devoicing as an important factor in
describing the phonological characteristics of them, saying that eastern
Japanese dialects are “consonant-prominent,” and western dialects are “vowel-
prominent.” NHK (1985) provides a map of Japan and divides it into two regions
in terms of devoicing: ”more noticeable" and "less noticeable." Roughly, eastern
and southern Japan are marked as dialects of more noticeable devoicing and
western and northern Japan as dialects of less noticeable devoicing. However,
these are all based on impressionistic judgments.

Sugito (1988) conducted an acoustic study of the speech Of 10 males in
their 30s to 50s from eight different cities in Japan, namely, Tokyo, Sendai,
Nagoya, Osaka, Kochi, Okayama, Kumamoto, and Naha, using a reading
passage. She found that there is a variation in the frequency of vowel devoicing
among the different cities. Her results confirmed that cities in the eastern and
southern parts of Japan (Tokyo, Sendai, Kumamoto, and Naha) have higher
percentages of devoicing (over 50%), and western cities (Osaka, Okayama, and
Kochi) have lower percentages (less than 35%), except Nagoya, which had the
highest percentage (almost 70%) of all. She attributes this to the Nagoya area's

diversity and suggests that we need to investigate that area in more detail. Also,

30

 

 

Nag

and
rebt

mou

seen
eash

Japa

(hfien
basec
invest
Tahar

"the;

 

 

 

 

 

afiihw
c0nsor
COHSei
SGVenl
ThlSis
Study.
Memo

devote

Nagoya has a significantly higher devoicing rate than that of Tokyo in vowels
before a sentence-intemal pause (e.g., before a comma) as in ooku ’(it is) a lot,
and' (the connective form of ooi '(it is) a lot'). She suggests that this might be
related to the Tokyo speakers' awareness of ”standard" and that that factor
motivates them to speak clearly, i.e., to avoid devoicing in certain environments.

Although Nagoya seems to be an exceptional case, Sugito's (1988) results
seem to confirm the general description of the regional differences, which is that
eastern and southern Japan have more devoicing and western and northern
Japan less devoicing.

Except for Sugito (1988), there is no study that statistically shows regional
differences in the frequency Of vowel devoicing, because the earlier studies are
based on impressionistic judgments. There are, however, a few studies that
investigated the frequency Of vowel devoicing in other regions, such as Osaka.
Tahara, et.al. (1998, cited in Morris 2003) asked twenty people from Osaka, both
male and female, and in different age groups, to read a set of sentences. There
are two sentences in which the high vowels /i/ and /u/ occur between voiceless
consonants in an unaccented syllable, which are not at a word boundary or in a
consecutive devoicing environment. Out of twenty respondents, sixteen or
seventeen devoiced those vowels, which gives us a devoicing rate Of 82.5%.
This is comparable to the devoicing rate by Tokyo speakers from Sugito's (1988)
study. Based mostly on this high percentage of devoicing by Osaka speakers in
the most favored environments for devoicing, Morris (2003) suggests that the

devoicing rate in Tokyo and Osaka (or Kinki) dialects may not be very different in

31

 

 

anei

enhn

omy

dfier
vanoc
devou
efiect

Nkues

 

dfhuel

Japanr
be ext]
Wficek
Olllere
raliter
Congo,
cOOSOI
VOICed
asjn/
reQIOn

10 ITIVe

an environment in which no other factors interact. However, the devoicing
environments of the vowels taken from Tahara et.al.'s study were very limited;
only two vowels were used.

Thus, there is no study that consistently investigated the regional
differences in the frequency Of vowel devoicing taking into consideration the
various factors that affect devoicing. Sugito (1988) compared the overall
devoicing rate from difference dialects, but she did not take into account the
effectof accents, consecutive devoicing environments, and word boundaries.
More study is necessary in this area before we can find a pattern of regional
differences in devoicing.

However, Maekawa (1988) says that although regional difference in
Japanese vowel devoicing is evident, the details Of the differences cannot simply
be explained by the devoicing rate in /...CVC.../ environments (C indicates a
voiceless consonant and V indicates a vowel). He suggests that regional
difference in vowel devoicing should be understood as forming a continuum
rather than being categorical. Moving from east to west, one finds different
consonant articulation, such as affrication of [k], and devoicing before a voiced
consonant, which ultimately reaches the point where a high vowel between
voiced consonants is deleted, resulting in the gemination of a voiced consonant,
as in /kaji##da/ —> /kad##da/ ’there is a fire.‘ Thus, in order to understand the
regional differences and draw dialect lines of more or less devoicing, one needs

to investigate various consonant environments and the distribution of the overall

32

devoicing rate from different regions, considering age difference and errors

caused by different methods.

2.3.2 Age

Age differences in vowel devoicing have been assumed, particulariy in
association with accent; namely, Older generations do not devoice accented
vowels, but younger generations do (NHK 1985, Sakurai 1985, Tsuchida 1997).
However, no sources are cited in those statements, and there is no systematic
and extensive study that focuses on age differences, so I must regard these
comments as impressionistic or based on anecdotal or very limited data.

Mineta (1988, cited in Maekawa 1988) reports an age difference in vowel
devoicing in her study conducted in a village in the Aichi prefecture. According to
her, younger generations devoiced 82.3% of the high vowels between voiceless
consonants and older generations devoiced 61.9%. However, this difference is
likely due to the difference in the devoicing rate of accented vowels: the Older
generation's devoicing rate (about 30%) was only half that of the younger
generation's in that context. Sugito (1988) also reports an age difference in the
devoicing rate in Osaka with more devoicing among younger generations, which
she interprets as a change. However, she does not mention the effect of accent
in the age difference, implying that younger generations devoice more in general.

On the other hand, Maekawa (1988) says that, in his study conducted in a
town in the Tottori prefecture, he could not see any strong age effect in vowel

devoicing. It was true that younger generations devoiced more than older

33

generations, but the difference was slights, and his conclusion was not based on
any statistical analysis.

In general, there seems to be an age difference in the devoicing Of
accented vowels. However, the evidence for age difference in general is weaker
because we have different results from different dialects, i.e. age difference is
reported in the Osaka and Aichi dialects, but not in the Tottori dialect. This seems
to show that there may be an interaction between age difference and region. It
should be noted that there has been no study investigating age differences in

vowel devoicing among Tokyo dialect speakers.

2.3.3 Sex and speech style

Sex and speech style are two of the most commonly studied factors Of
language variation. However, I am aware of only one study that investigated the
effect of these factors on Japanese vowel devoicing (Yuen 1997)’.

Yuen (1997) hypothesized that there would be a sex difference in vowel
devoicing because female speech and male speech are very different in
Japanese. He further hypothesized that male speakers would devoice vowels
more than female speakers do because female speech is more polite than male
speech; therefore, he claims, female speech is slower “for producing clearer and
more audible speech” (Yuen 1997: 11). He suggests that this leads to a longer

duration of vowels in female speech than male speech and that vowel devoicing

 

6 According to the chart Maekawa provides, [IT] was constantly devoiced more frequently than [ki]
in all generations, indicating the effect of the preceding consonant.
7 lmai (1997) included sex difference but not style difference.

34

 

WOL

neg

sen

 

que
uiT

beh-

incn
devc
nae:
Iengl
Spe

HOW1
Ache
Ohio
hacto
Spec
Prec

Ah“)

GEVO
Was I

OnWj

would be less likely to occur in female speech than in male speech, assuming a
negative correlation between vowel duration and devoicing.

Yuen used three different speech styles: reading sentences using a frame
sentence, reading a passage from Japanese fiction, and conversation (answering
questions). His subjects were 9 speakers Of the Tokyo dialect, who were raised
in Tokyo and lived there at least 15 years; there were 4 males and 5 females, all
between 20 and 30 years of age. He found that, as the degree of formality
increases, the speech rate decreases and the vowel length increases; therefore,
devoicing decreases. This correlation of voicing (or non-devoicing), speech
rates, and vowel length was significant (p=0.05). He also found that the vowel
length Of males is indeed shorter than that of females, that males have faster
speech rates than females, and that males devoice more than females.
However, speech style alone statistically has a stronger influence than sex.
Actually, speech style was the most significant of all the linguistic and social
factors he investigated, and sex was the second. The complete ranking of his
factors, both linguistic and social, in terms of the relative importance is as follows:
Speech Style > Sex > Word Position > Accent > Following Consonant >
Preceding Consonant > Vowel Quality. Yuen used an ANOVA post-hoc test, an
ANOVA factorial test and a correlation 2 test for his statistical analysis.

lmai (1997) included sex as one Of the factors expected to affect vowel
devoicing and found that males devoiced more than females, but this difference
was not statistically significant because of the small number of respondents; i.e.

only four in this study.

35

Yuen (1997) attributed the sex difference in Japanese vowel devoicing to
females' greater politeness and tendency to speak slower, especially in formal
style. However, I would interpret this as an indication of standardness or
appropriateness because, if we look at the interaction chart of voicing means and
speech style for males and females (Yuen 1997: 28, Figure 11), the difference
between males and females is about twice as large in the formal reading style
than in the conversation style and the casual reading style. If women do
something a lot more in more formal style, in which they pay most attention to
their speech, they may be doing something they consider to be standard, or
prestigious. Most sociolinguistic research suggests that women are more

prestige-conscious (Wolfram and Schilling-Estes 1998: 188-192).

2.3.4 Social Class

Although social class is one of the major social factors Of language
variation, there has been no study that investigating social class differences in
Japanese vowel devoicing. Actually, no language variation studies in Japanese
have investigated this factor, although some researchers have used the
yamanote/shitamachi distinction. Yamanote (‘uptown’) is associated with an
image Of white-collar workers and greater prestige, while shitamachi (‘downtown’)
is usually associated with an image Of blue-collar workers, merchants, and lower
prestige. Hibiya (1995) used social class as a variable in her study of the
variation of Inl, but she used this distinction between yamanote and shitamachi.

However, this distinction may not be valid; it is descended from the historical

36

regional separation Of two social classes and not based on the actual social
status of the people who currently live there. Besides, people frequently move
within Tokyo and also from other areas of Japan to Tokyo.

Social class is used extensively in many sociolinguistic studies; in the
US, for example, occupation, education, income, type of housing, and
neighborhood are used to calculate the socioeconomic status of respondents and
classify them into such different social groups as Lower, Working (Lower/Upper),
Middle (Lower/Upper), and Upper classes, based on scores assigned to these
various subcategories. However, this method has not been used in the language
variation studies in Japan.

There are several social classifications systems in Japan but none of them
is agreed upon by Japanese sociologists (Kosaka 1994a). The Social
Stratification and Social Mobility (SSM) survey, which started in 1955 and is
conducted every 10 years, uses the categories below (from Table 3-8; Kosaka

1994a: 47).

Professional
Managerial
Clerical
Sales
Skilled
Semi-skilled
Unskilled
Farmer

PPNP’P‘PSPN.‘

37

However, the above classification does not take into account the size Of
firms and job status, which obviously influence socioeconomic status.

Seiyama (1994) uses the terms “non-manual" for 1 to 4 in SSM categories
and “manual” for 5 to 7, and further divides the people within these new
categories into “upper" and “lower" non-manual and manual in order to compare
the Japanese data with US. data. Farmers, including people engaged in fishing
and forestry, form a category by themselves, and almost all of them are self-
employed. Seiyama also uses the terms “white-collar" and “blue-collar,” which
correspond to “non-manual” and “manual,” respectively. He also points out that
the salaried vs. self-employed distinction is important in Japan because of a large
proportion of self-employed workers. Seiyama’s new classification, partly based

on social mobility, is as follows.

1. Salaried white-collar

2. Self-employed white-collar
3. Salaried blue-collar

4. Self-employed blue-collar

5. Farmer

Kosaka (1994a) introduces another (“neO-Marxist”) scheme, but it does

not provide a continuous scale.

1. Capitalist Class (managers, self-employed and family workers with
more than four employees)

38

 

 

Ho
pro
col
inte

sen

23.

(life
east
Japg
lack
an a
dch
mon

deVO

2. New Middle Class (full-time or part-time managerial, professional,
and full-time clerical)

3. Working Class (Others)

4. Old Middle Class (self-employed and family workers with fewer
than four employees)

Thus, sociologists use different classifications for different purposes.
However, they all seem to draw a line between white-collar, non-manual,
professional, managerial, clerical and sales people on the one hand, and blue-
collar, manual, skilled, semi-skilled, and unskilled workers on the other. I
interpret this, roughly, as a distinction between Middle and Working Class. In this

sense, their classification is not so different from the ones used in the US.

2.3.5 Summary

Dialectologists (i.e. Kindaichi and Mase) agree that there is regional
difference in the frequency of vowel devoicing, and a general description is that
eastern and southern Japan have more devoicing and western and northern
Japan less devoicing. However, we need more study in this area because of the
lack of extensive studies that focus on regional differences. There seems to be
an age difference in the devoicing Of accented vowels; younger generations
devoice more and older generations less, but age difference in general needs
more extensive study. As for sex and speech style differences, females tend to

Clevoice less than males, and vowel devoicing is more likely to occur in casual

39

style. Finally, there are no studies of the effect of social class difference on
Japanese vowel devoicing. In general, research on the effects Of social factors

on vowel devoicing is needed.

40

CHAPTER 3

METHODOLOGY

In this chapter, I will discuss the variables (factors) used in the present
study, both linguistic and non-linguistic (3.1 ), coding of those variables (3.1), data

collection (3.2), and the data handling procedures (3.3).

3.1 Variables

Two kinds Of variables are used in the present study: linguistic and non-

linguistic (social) ones:

<Linguistic variables>

Vowel identity

Preceding consonant identity
Following consonant identity
Morpheme boundary type
Pitch accent pattern

outsource

Consecutive devoicing environment

<Social variables>

1 . Age
2. Sex
3. Social class

4. Speech style

41

These linguistic variables and stylistic varieties were guaranteed to occur
by being incorporated into the word list and a reading passage in the data
collection phase of the project; the other social variables were realized through
respondent selection.

Each variable was coded in an Excel file row for each occurrence Of the
dependent variable — a voiced or devoiced vowel. I will discuss variable

selection, coding, and expected results in the following subsections.

3.1. 1. Linguistic variables

3. 1.1. 1 Vowel identity

Vowel identity is included to see if there is any difference in the rate Of
vowel devoicing among the five vowels, particularly between the two high vowels,
/i/ and /u/. Even though some researchers argue there is a difference between
the two (e.g., Han 1962), other studies, particularly sociolinguistic ones, did not
find any (e.g,, Maekawa 1983). I expect to find no difference in the frequency Of
vowel devoicing for IV and lul, and, although it is clear that the non-high vowels
(lel, lo] and la/) will be much less frequently devoiced, l have no prediction about

any possible difference in the rate among them.

3. 1.1.2 Consonant identity

The type of consonant is important in vowel devoicing in Japanese;

therefore, I identify both preceding and following consonants as variables. I

42

 

 

 

Vo

 

coded the exact identity of the preceding and following consonants instead Of
combining them according to place and/or manner of articulation. By coding for
exact identity, it will be easy to study groups of consonants based on place and
manner of articulation or other factors later, but initial coding by groups would
presuppose similarities.

The numeric values for the preceding and following consonants used in

my coding scheme are as follows.

1=[ll

2=ISI
3=Itll
4=[ts]

5 = [9] (palatal fricative)
6 = [111] (bilabial fricative)

7=[h]
8=[I]
9=Ikl
10=[b]

11 = pause (Only for the following environment.)

Voiced consonants were coded as themselves (e.g., b, d, 9), not as numbers.
I coded all the vowels between voiceless consonants, and also coded the
high vowels after a voiceless consonant and before a voiced consonant. In

Japanese the preceding consonant inventory is limited before IV and /u/ because

of the allophonic alternation of ltl, Isl, and /h/ (see Chapter 1). /t/ becomes [tj']

43

 

 

i
”la—

before Ill, and [ts] before /u/. Similarly, /h/ becomes [9] (palatal fricative) before /i/,

and [11)] (bilabial fricative) before /u/. Finally, ls/ becomes [1] before /i/, but remains

[s] before /u/. Thus, [t] never occurs before /i/ or /u/ except in loanwords, [s] never
occurs before /i/, and [h] never occurs before a high vowel.

My hypothesis is that a preceding fricative will have a greater influence on
the application Of vowel devoicing than a stop or an affricate, and a following stop
more than a fricative or an affricate. It is also expected that devoicing is demoted
before any allophone of lhl, and between two fricatives.

Another hypothesis is that these preferences, and perhaps all the others
listed here, will show a gradient effect according to speech style. Even if
devoicing before an allophone of /h/ and between two fricatives is prohibited in
phonology, phonetic devoicing occurs, particularly in conversation, which is
usually a more allegro speech style than that found in word lists and reading
passages. My prediction is that devoicing occurs more frequently in more casual

styles (i.e. word list < reading passage < conversation).

3.1.1.3 Morpheme boundary type

Researchers (e.g. Sakurai 1985, Vance 1987) agree that a morpheme
boundary has an effect on vowel devoicing in Japanese: vowel devoicing is
avoided at a morpheme boundary, particularly in a consecutive devoicing
environment (Sakurai 1985, Vance 1987, 1992, Kondo 1997, Tsuchida 1997).

However, we do not know if the avoidance of devoicing depends on the type of

44

 

de
re;

olc

Ber

BCC

 

morpheme boundaries. Therefore, morpheme boundary type is included in the
present study, and different types of boundaries were included in the word list
and reading passage. Morpheme boundaries are coded depending on their

status as a bound or free morpheme. The actual coding used is as follows.

1 = no boundary (morpheme internal)

2 = pause (e.g. word-final position)

3 = bound morpheme boundary (inflectional and derivational morphemes,
e.g., oishi+katta ‘it was tasty”; and bound content morphemes, as in
Sino-Japanese compounds: eg. shitsu+ren 'lost love')

4 = compound word boundary: e.g., booeki#suijun 'a trade standard'

5 = word boundary

3.1.1.4 Pitch accent pattern

Pitch accent pattern is included to see if there is any effect on vowel
devoicing, and if there is a change among different age groups since it has been
reported that younger generations tend to devoice even accented vowels, which
older generations tend not to devoice (Sugito 1982, Sakurai 1985, Maekawa
1988, Sugito and Hirose 1988, Kondo 1993 [cited in Tsuchida 1997], Nagano-
Madsen 1995, lmai 1997, Tsuchida 1997).

Most of the research on devoicing has used the distinction between
accented and unaccented vowels, but I use pitch accent pattern because the

accented/unaccented distinction is contained in this representation, and, if there

45

is even a greater sensitivity than to just the accented/unaccented distinction, this
coding may be able to capture it.

Pitch accent pattern is coded to reflect all possible patterns for the
preceding and following mora combinations as shown below. Each letter

represents a tone-bearing unit (=mora) or a pause (#), and the middle letter is the

target vowel.
1=LLL 7=HHL 13=#L#
2=LLH 8=HHH 14=#H#
3=HLL 9=#LL 15=LL#
4=HLH 10=#LH 16=LH#
5=LHL 11=#HL 17=HL#
6=LHH 12=#HH 18=HH#

It is expected that older generations devoice less in pitch accent patterns
5, 7, and 11 than younger generations because these are all cases of accented
vowels. It is also expected that devoicing is less likely to occur in pitch accent
patterns 14, 16, and 18 because the vowel is before a pause and high-pitched

(refer to 2.2.3).

3.1.1.5 Consecutive devoicing environment

Consecutive devoicing is usually avoided, as discussed in 2.2.5; however,
consecutive devoicing may occur (Maekawa 1988, lmai 1997). All instances of

the high vowels /i/ and lu/ and any other vowels that were actually devoiced were

46

coded to see whether or not they were in a typical devoicing environment
(between voiceless consonants or after a voiceless consonant and before a
pause), whether or not the preceding vowel was in a typical devoicing
environment, and whether they (or their preceding vowels) were or were not

actually devoiced. The coding is as follows.

0 = the vowel is not in a consecutive devoicing environment (i.e., the
previous vowel is not in a typical devoicing environment.)

1 = the previous vowel is in a typical devoicing environment, but it is not
devoiced.

2 = the previous vowel is in a typical devoicing environment, and it is
devoiced.

x = the vowel is not in a typical devoicing environment

For example, in a phrase /i+shiki#fu+mee##no##kyoo+aku+han/ ('a brutal
criminal who is unconscious', where '+‘ indicates a bound morpheme boundary,
'#' a compound word boundary, and '#lt' a word boundary), the second III is
coded as '0' because the previous vowel is not in a devoicing environment (‘#i-
sh’). The third /i/, however, would be coded as either '1' or '2’ depending on the
voicing of the previous vowel, which is in a devoicing environment — ‘shik’. If the
previous vowel was devoiced, the third ‘i’ would be coded as '2'; if the previous
vowel was not devoiced, it would be coded as '1.’ The first /ul is coded as 'x'
because, although it is preceded by [f], a voiceless consonant, it is followed by
[m], a voiced consonant; therefore it is not in a typical devoicing environment.

Finally, the second /u/ is coded as '0' because it is preceded by [k] and followed

47

by [h], therefore in a devoicing environment, but the previous vowel is not in a
devoicing environment — ‘O-ak’.

My hypothesis is that consecutive devoicing is more likely to occur as the
speech style becomes more casual because phonetic devoicing is more likely to

occur in more casual, faster speech.

3. 1.2 Social variables

Vowel devoicing in Japanese, between voiceless consonants and in a low-
pitched syllable, is considered standard in Tokyo dialect (cf. Japanese
Pronunciation and Accent Dictionary). Maekawa (1988) states that it is
considered standard because there are some fixed patterns in consecutive
devoicing environments and that if someone does not devoice the final vowel in a
sentence-final polite copula, as in /soo##desu/ (‘It is so.’), it sounds strange to
the ears Of Tokyo speakers. The first reason appeals to prescriptive authority,
and the second presupposes that Tokyo speakers are speakers of the standard
variety and that anything which “sounds strange” to them is not.

The findings in Yuen (1997) suggest that vowel devoicing in Japanese has
social meaning, since he found sex and speech style differences. However, it is
male speakers who devoiced more, not female speakers, in his study, which
might suggest that this feature is nonstandard, because women are reported, in
many sociolinguistic studies, to use more standard features than men.
Furthermore, female speakers' voicing mean was much higher than that of male

speakers’ in the most formal style, formal reading style, in Yuen (1997). This also

48

seems to suggest that devoicing is a nonstandard feature because females
devoice much less in the most formal style than males do, and females are
known to use more prestigious, or standard forms, in more formal styles.

The style differences also suggest that devoicing behaves like a
nonstandard feature because more devoicing was found in more casual styles in
Yuen (1997). He attributed these differences to women’s being more polite than
men by speaking slower and more clearly, therefore resulting in less devoicing.

If we had only Yuen’s (1997) results, we would have thought that
Japanese vowel devoicing is just a fast speech phenomenon. However,
considering its status as a standard Tokyo feature, according to the general
perception of Japanese speakers (cf. Maekawa 1988) and to prescriptive
authority (i.e. NHK 1985), we must conclude that it has deeper social meaning. If
it does have such meaning, it is an interesting phenomenon since articulatory
studies show that it is an unconsciously monitored process, one in which
speakers make an (unconscious) effort to produce a devoiced vowel and the
neighboring voiceless consonant with a single glottal Opening (Sawashima 1971
[cited in Maekawa 1988], Tsuchida 1997). .

Another piece of support for the position that devoicing is nonstandard
exists in the report that younger speakers tend to devoice even accented vowels,
whereas older speakers tend not to. This, however, might also suggest that there
may be language change going on among younger speakers, or it might even

reflect an age grading differentiation regarding vowel devoicing in Japanese.

49

Neither of these hypotheses, however, has been sufficiently studied to allow us
to suggest with any certainty that the feature is standard or nonstandard.

From these previous observations and studies, it is possible to suspect
that there are some social meanings attached to Japanese vowel devoicing, but
there has been no study that looked at this phenomenon in a language variation
framework.

In order to investigate the social meaning of this phenomenon more
thoroughly in the framework of variationist sociolinguistics, I included four
common social factors: age, sex, social class, and speech style, although, as l
have noted above, speech style will also be an important consideration in

determining the influence of some linguistic factors.

3.1.2.1 Age

Three age groups are used in this study: Younger, Middle, and Older. At
first, real age was coded, and the devoicing percentage was obtained for each
respondent. However, no grouping tendency could be derived from these
percentages, and the respondents were grouped together using thirty and sixty
years of age as the boundaries. Respondents younger than 31 are coded as 'Y,’
those over 30 and younger than 61 as 'M,’ and those over 60 as 'O.’ Table 3.1

shows the age groups and number Of respondents for each sex.

50

Table 3.1: Age Groups

 

 

 

 

 

Age Group Male Female Total
Younger (15-30) 8 7 15
Middle (31-60) 7 7 14
Older (61-90) 6 7 13
Total 21 21 42

 

 

 

 

 

 

If devoicing is a feature changing in Tokyo (i.e. increasing), one would
expect Younger respondents to generally devoice more than Older respondents,
and Middle respondents to fall somewhere in between. If devoicing is a stable
variable, however, one would expect a typically age-graded pattern in which the
Younger and Older groups would devoice most, if it is a nonstandard form

(Chambers and Trudgill 1980).

3.1.2.2 Sex

Sex is included because the difference between men and women in vowel
devoicing in Japanese has been reported (Yuen 1997), and sex has been shown
to be an extremely important variable in numerous Japanese sociolinguistic
research (Haig 1990, lde and McGlOin 1990, Okamoto 1995, lde 1997, lde and
Yoshida 1999). In the present study there are 21 respondents in each sex group.

It is difficult to predict whether males or females will devoice most. Since
devoicing is considered a standard (NHK 1985, Maekawa 1988), one might
expect more frequent female devoicing, in keeping with widely-observed sex-

related practice (Wolfram 1969; Labov 1972; Wolfram and Fasold 1974; Trudgill

51

1983), but if devoicing is associated with rapid, casual speech, then males might

dominate.

3.1.2.3 Social class

Social class is included in the present study because it has been shown to
be an important factor in sociolinguistic studies (e.g. Labov 1966, Trudgill 1974,
Wolfram and Fasold 1974), although there is no study of Japanese vowel
devoicing that uses social class as a variable.

I divided my respondents into three groups: Upper Middle, Lower Middle,
and Working. Since no previous sociolinguistic studies on Japanese use social
class (or socioeconomic status) as a variable, I basically adopted the criteria
commonly used in sociolinguistic studies in the US. (Labov 1966; Shuy, Wolfram,
and Riley 1968; among others, e.g. Warner Index of Social Status 1960), using
socioeconomic scores adapted to the Japanese urban Tokyo environment (cf.
Kosaka 1994b)

Each respondent was assigned a socioeconomic score according to their

occupation, education level, and neighborhood.

Occupation: There are five scores for occupation as follows:

1 = blue-collar, heavy physical labor; such as plumbers

2 = blue-collar, light-physical labor; such as daycare workers and noodle
shop workers

52

3 = white-collar, lower prestige; such as employees of a mid-size to big
company, presidents of a small company, and owners of a small store

4 = white-collar, higher prestige; such as managerial at a bigger company,
engineers, and university instructors/professors

5 = most prestigious; such as doctors, lawyers, and presidents of big
corporations

If the respondent is a student, their parents' occupation was used tO

determine their score.

Education: There are also five scores for education level as follows:

1 = high school education or below

2 = vocational college/technical school
3 = junior college

4 = university

5 = graduate school

If the respondent is a middle school student, I assigned the score that they
are most likely to get in the future. For example, if the respondent is a middle
school student and their parent is a doctor or an architect, l assigned ’4' to them
because it is very likely that they will receive university education.

These two scores of occupation and education were multiplied by two and
combined with the score of neighborhood, based on Warner (1960)’s estimates
of their relative importance in determining status, although I gave weight to
occupation and education, due to my assumption that education may have

greater prestige in Japanese society.

53

Neighborhood: The scores for neighborhoods were derived from informal

discussions I had with friends who live in Tokyo. I used the following 5

categories:
1 = 'not good'
2 = 'so—so'

3 = 'adequate'
4 = ’good'
5 = 'very good'

As a result of this scale, two of my respondent neighborhoods are classified as 1,

eight as 2, seven as 3, ten as 4, and three as 5 (see Appendix C).

Categorization: These three scores (occupation, education, and neighborhood)
were combined resulting in a range Of 5 to 25 (see Appendix D), and the
respondents’ scores ranged from 5 to 23. First, these raw numbers were entered
into the Excel coding sheets together with the coding for other variables, and
devoicing percentages were obtained for each respondent. As it was with age, no
natural grouping of the respondents or correlation with devoicing was found.
Therefore, groups were made so that there are about the same number Of
respondents in each group. As a result, those respondents whose score falls
between 6 and 11 were grouped as Working, those between 12 and 17 as Lower
Middle, and those between 18 and 23 as Upper Middle. An investigation of the

occupation of those who fall into these groups makes this grouping appear to be

54

a reasonable one. Table 3.2 shows social class and number Of respondents for

each sex.

Table 3.2: Social Classes

 

 

 

 

 

Social Class Male Female Total
Working (6-1 1) 7 7 14
Lower Middle (12-17) 7 9 16
Upper Middle (18-23) 7 5 12
Total 21 21 42

 

 

 

 

 

 

3.1.2.4 Speech style

I used three speech styles in this study: Word list (see Appendix A),
Reading passage (See Appendix B), and Conversation (see Appendix E for a
sample conversational transcription). These styles form a continuum Of formality,
conversational style being the least formal and word list style being the most
formal. The more formal the style is, the more attention speakers pay to their
speech (Labov 1966) and such attention usually results in a greater incidence of
formal or prestigious forms. Yuen (1997) found that speech style is the most
significant factor in vowel devoicing in Japanese. I expect to find a similar result,
namely, that devoicing is promoted as the speech style becomes more casual,
although the status Of devoicing in some environments will require an even closer

look at the distribution of style and other variables.

55

3.2 Data collection

3. 2. 1 Respondents

Most researchers agree that regional differences affect the application of

vowel devoicing (see 2.3.1 above). Because this is the first attempt to see the

differences according to social groups and age groups, inclusion of region would
make this project unmanageable. Since all the respondents in this study are from
the Tokyo area, this factor is controlled. All of them grew up in the Tokyo area,

and most of them were also born there and have at least one parent who grew

up there.

The data from 42 respondents was used for the final statistical analyses.

The following table shows the composition of respondents by age, sex, and

social class.

Table 3.3: All Social Factors

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Age Group Sex Social Class Total
Working Lower Middle Upper
Middle
Younger Male 3 2 3 8
(1 5-30) Female 2 3 2 7
Middle Male 2 3 2 7
(31 -60) Female 1 3 3 7
Older Male 2 2 2 6
(51-90) Female 4 3 0 7
Total 14 16 12 42

 

56

 

3.2.2 Interview procedures

I interviewed each respondent in three stages. First, I asked them some
questions to warm them up to the task and make them feel comfortable about
being taped. The early questions I asked are demographic: where they were
born and raised, where their parents were born and raised, their occupation, their
parents' occupation, where they live now, what they like to do for leisure, etc.,
since such information is important in determining their social status. Then I
asked them to read a word list and a reading passage. I prepared a word list of
90 words, phrases or short sentences, containing all the possible combinations of
the preceding and following voiceless consonants, including a pause, for both
high vowels (see Appendix A). I also prepared a reading passage, which
contained as many words/phrases from the word list as possible within a
reasonable length (see Appendix B). The whole interview session was recorded
on tape. I used a Sony Walkman to record the conversation using regular
cassette tapes. Each interview session varied in length, ranging from

approximately 20 minutes to 2.5 hours.

3.3 Data handling

3.3.1 Acoustic analyses of the data

The recorded data were transferred and digitized using the sound analysis
program Praat. Voicing of the vowels was determined by investigating the 1)

waveforms, 2) spectrograms, 3) intensity, and 4) pitch track, as well as by ear to

57

confirm the instrumental procedures. If the vowel is completely devoiced, its
sound wave does not show a periodic wave, nor does it show clear forrnants in
the spectrogram. Its intensity drops, and the pitch track is lost.

Figure 3.1 shows a fully voiced token of the word /see+katsu#hi/ ‘living
expenses,’ and Figure 3.2 shows a clearly devoiced vowel /u/, between /ts/ and

/h/, in the same word.

 

hi

 

 

 

 

Figure 3.1: /see+katsu#hi/—Voiced

The first big chunk of periodic wave is the first syllable, with a long vowel
lseel, and the pitch track (thick black dotted line in the middle of the spectrogram)
remains rather stable throughout the syllable. There is also a relatively high
intensity line (the thin white line above the pitch track). The second big periodic
wave chunk is the second syllable /ka/. Again, there are clear forrnants, the pitch
track stays about the same, and intensity rises in the vocalic portion of the signal.

The third periodic wave peak is rather small, but one can still see the clear

58

formants for the vowel /u/ in the third syllable /tsu/. There is a clear pitch track,
which remains stable until it reaches the final syllable. Intensity also rises where
the vowel /u/ is produced. Finally, the last peak is the final syllable /hi/. There are
clear formants, a pitch track, and an intensity measure. The pitch track shows a

decline, which means the previous syllable Itsu/ was accented (cf. Sugito 1982).

 

I" 1.1 .‘
l'io‘lrllv"

 

Figure 3.2: Isee+katsu#hil-Devoiced

In Figure 3.2, the first big chunk Of periodic wave is the first syllable, /see/,
and there are the clear formants in the spectrogram and a well-defined pitch
track, which remains very stable throughout the syllable. The intensity level is
also high. The small wave peak following is the second syllable lkal. There are
clear formants, the pitch track remains at about the same height, and intensity
rises as the vowel is produced. However, there are no formants, pitch track, nor
intensity movement where the third syllable /tsu/ should be. There is a small

intensity raise where /ts/ is released, but the next pitch track and formants

59

 

 

31)

represent the vowel III in the final syllable /hi/. The pitch track for lhi/ shows a
sharp drop within the syllable, which indicates the devoiced Itsu/ was accented
and the accent was manifested by the steep drop Of the pitch height in the
following syllable (cf. Sugito and Hirose 1988).

There are cases, however, in which it is hard to determine voicing. In such
cases, if three of the above criteria were met, I coded the vowel as devoiced. For
example, if there is no periodic wave, no clear formants, no intensity, but there is
a pitch track where the vowel is supposed to be in the spectrogram, I
nevertheless regarded the vowel as devoiced.

All five vowels were analyzed for voicing: high vowels after a voiceless
consonant and non-high vowels between voiceless consonants. I chose these
options because high vowels can be devoiced even before a voiced consonant
(Maekawa 1983 and 1988, Sugito 1988), and even non-high vowels can be
devoiced between voiceless consonants (Sakuma 1929, Maekawa 1988, Jun
and Beckman 1994, Tsuchida 1997), which is the best environment for devoicing
in general.

The pitch accent pattern is also confirmed by using the pitch tracker in
Praat, discussed more thoroughly below.

After the acoustic analysis was done, all the tokens were coded in an
Excel sheet for each respondent according to the coding criteria explained in 3.1.

For each respondent, there are approximately 270 tokens in the Word list

style, 180 tokens in the Reading passage style, and between 150 to 350 tokens

6O

in the Conversation style. There are 42 respondents in total, and the total number
of tokens used for the statistical analyses is over 30,000.
Table 3.4 shows a sample line of data coding for the word /kohitsuji/

‘lamb.’

Table 3.4: Coding of Ikohitsujil

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Vowel Voi. Pre.C FOI.C P.A.P. M.B. C.D.E. S.S. Gnd. Age Res. Style
0 + 9 5 10 4 x M F 62 R W
i - 5 4 5 1 0 M F 62 R W
u + 4 j 3 1 x M F 62 R W

 

 

In the first row, from the leftmost column, are vowel identity, voicing,
preceding consonant, following consonant, pitch accent pattern, morpheme
boundary, consecutive devoicing environment, social status, sex, (actual) age,

respondent identity, and speech style. For example, the vowel /i/ is devoiced, so

it is coded as ‘-‘ in the voicing column. It is preceded by [c], and coded as 5 for

the preceding consonant; it is followed by [ts] and coded as 4 in the following
consonant column. The accent pattern of Ikohitsuji/ is LHLL; therefore the code is
5 (LHL). Since lhitsuji/ is a single morpheme that means ‘sheep,’ it is coded as 1
for the morpheme boundary type. /i/ is in a typical devoicing environment, but the
previous vowel is not, therefore it is coded as 0 for consecutive devoicing
environment. For the social factors, this respondent is Lower Middle Class,
coded as M; sex is Female, her age is 62, her ID is ‘R,’ and the tokens were

taken from the Word List ('W’).

61

3.3.2 Statistical analyses of the data

The coded data were processed first in tables to see the general pattern Of
devoicing according to each variable, than in GoldVarb (a logistic regression
program) to find out significant factors and the relative significance of the values
within each factor. GoldVarb is capable of dealing with the very small numbers in
some cells that may arise in the study Of conversational data.

In the first GoldVarb run, high vowels and non-high vowels were analyzed
separately to see the difference in the vowel devoicing rate. In the non-high
vowel run, there were many ‘knockouts,’ where no vowels were devoiced, and
the overall devoicing rate was 5%, which is very low“. Therefore, I eliminated the
non-high vowels from the following detailed statistical analyses.

In the statistical run for high vowels, there was only one knockout: the
Pitch Accent Pattern #14 (#H#), which occurred only three times. The overall
devoicing rate was 45%: /u/ was devoiced at 45%, and /i/ at 46%, suggesting no
statistically significant difference in devoicing rate between the two.

The knockouts (many for non-high vowels and one (#H#) for high vowels)
are not unexpected. Non-high vowels are not likely to devoice compared to high
vowels, and a single mora utterance with a high pitch accent is doubly unlikely to
devoice because it is a single mora utterance; therefore, it is more natural to
voice the vowel so that it can be heard.

The knockout environment and a few other very low devoicing rate

environments were eliminated from the subsequent statistical runs to Obtain

 

8 524 non-high vowels were devoiced out of 9,560 occurrences (5%), of which lo/ and le/ were
devoiced at 7%, and la/ at 4%.

62

better results. It should be noted that these eliminated environments are all
linguistic ones: no social factors were eliminated. Therefore, all the statistical
results reported in the next chapter include only the high vowels /i/ and /u/
between voiceless consonants or after a voiceless consonant and before a
pause. Low devoicing rate pitch accent patterns were eliminated. For the
analyses Of linguistic environments, the vowels were analyzed separately

because of the allophonic alternation in the preceding consonant.

63

CHAPTER 4

RESULTS

This chapter will discuss the results of the statistical analyses of the data.
First, I will report on the results for linguistic factors in 4.1, focusing particularly on
the issues that have been discussed in the literature, namely, preceding and
following consonants, vowel identity, accent, morpheme boundary, and
consecutive devoicing. Then, I will discuss the results for social factors, namely,

speech style, age and sex, and social class, in 4.2.

4.1 Linguistic factors

The following discussion is based on the results Of statistical analyses
using Goldvarb, a logistic regression program. This program enables one to
compare cells with a relatively small number of tokens and choose significant
factors and factor groups. Goldvarb gives a weight for each factor, which
indicates the likelihood of the application of the rule; in this case, devoicing. For
example, if the weight associated with a factor is above 0.5, it promotes

devoicing, but if the weight is below 0.5, it demotes devoicing. The higher the

64

weight, the more the factor promotes the rule’s application. In this Goldvarb
analysis, only high vowels, IV and /u/, in a typical devoicing environment
(between voiceless consonants or after a voiceless consonant and before a
pause) are included. Devoicing in other environments or of non-high vowels is so
infrequent that it would be difficult to build a statistical model around a data set
that included such items. For linguistic environments, the statistical analysis was
done for the high vowels Ill and /u/ separately because Of the different inventories

of the preceding consonants due to Japanese phonotactics.

4. 1 . 1 Preceding and following consonant

As discussed in 2.2.1, most linguists seem to agree that a preceding
fricative and a following stop separately promote devoicing. This is confirmed in
the present study. Table 4.1 shows the results of the Goldvarb run for the

preceding consonants for the vowel /i/.

65

Table 4.1: Preceding consonant for /i/ — Goldvarb

 

 

 

 

 

 

 

 

 

 

Preceding Weight Number
Consonant (devoiced/total)
1 [I] 0.642 1986/2507
2 m] 0.522 816/1645
3 [9] 0.513 609/944
4 [k] 0.434 752/1456
5 [p] 0.142 318/705

 

The table shows that the fricatives and affricates promote devoicing; their

weights are all above 0.5. Stops, however, Clearly demote devoicing with weights

all below 0.5.

Preceding [p] most strongly demotes devoicing for lil, but it should also be
noted that affricates and fricatives are mixed. The alveopalatal affricate [tj] is
ranked between the fricatives, [I] and [C]. In other words, the alveopalatal fricative
[1'] strongly promotes devoicing, but the palatal fricative [g] and the alveopalatal

affricate [if] somewhat promote it, and there is only a 0.01 difference between the

latter two.

As for the stops, even though both stops demote devoicing, the labial stop

[p] strongly demotes it with the weight Of only 0.142.

I now turn to the results for the vowel lul. Table 4.2 shows the results of

the Goldvarb run for preceding consonants.

66

 

Table 4.2: Preceding consonant for /u/ - Goldvarb

 

 

 

 

 

 

 

 

 

 

 

Preceding Weight Number
Consonant (devoiced/total)

1 [1],] 0.706 700/1107

2 [5] 0.642 1 998/2660

3 [ts] 0.577 699/1 326

4 [k] 0.341 965/21 1 1

5 [I] 0.276 278/705

6 [p] 0.1 51 289/506

 

The preceding bilabial fricative [(1)] is the best promoter for /u/ devoicing.
The next best promoter is [s], followed by the affricate [ts]. Stops also demote /u/
devoicing, and, again, [p] most dramatically. However, there is a striking
difference from the results for lil; the alveopalatal fricative [j], which is the best
promoter for IV devoicing, is a strong demoter for lul. Since [[1 is an allophone of
/s/ before lil, but it is a phoneme before lul, perhaps this striking difference can
be accounted for phonologically. I will return to this question in the more detailed
interpretation of results to follow in Chapter 5.

Now I turn to the results for the following consonants. Table 4.3 shows the

results of the Goldvarb run for consonants following /i/.

67

 

Table 4.3: Following consonant for /i/ - Goldvarb

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Following Weight Number
Consonant (devoiced/total)

1 [t] 0.847 1829/201 7
2 [k] 0.826 1036/1 145
3 [3] 0.756 309/388
4 [ts] 0.731 321/379
5 [tj] 0.697 315/497
6 [I] 0.432 254/397
7 [p] 0.185 244/585
8 pause 0.1 19 102/882
9 [h] 0.039 54/575
10 [9] 0.014 14/184
1 1 [(1,] 0.005 3/207

 

 

The top two promoting consonants are both stops ([t] and [k]), with a very
high weight (above 0.8), but the alveolar fricative [3] comes next. The next two
best promoters are both affricates ([ts] and [tj']). Even though [I] somewhat
demotes devoicing (its weight is slightly below 0.5), there is a large gap between
[[1 and [p], and everything below [I] strongly demotes devoicing; [p], pause, and
the allophones of /h/ ([h], [c], and [111]). The general tendency is, then, that
following stops promotes devoicing most, followed by affricates and fricatives,
although [3] promotes devoicing more than the affricates do, and the split

between [s] as a promoter and [j] as a demoter is not accounted for.

68

There are two exceptions to this general tendency, however. First, the
behavior of [p] is strange. [p] ranks considerably lower than other stops, below
affricates and even below the fricatives ([s] and [j]); it has a very low weight of
0.185. While other following stops strongly promote devoicing, [p] strongly
demotes it.

Second, while alveolar and alveopalatal fricatives ([s] and [1], respectively)
are ranked higher, the allophones of /h/, namely [h], [c], and [(1)], are ranked at the
bottom of the list, with very low weights (below 0.04). This seems to support
Tsuchida’s (1997) claim that devoicing before /hl is phonologically prohibited.
Otherwise, we cannot explain why the following [3] and [1'] promote devoicing
(strongly or relatively), but the allophones of /h/ strongly demote it.

Table 4.4 shows the results of the Goldvarb run for the following

consonants for the vowel lul.

69

Table 4.4: Following consonant for /u/ -— Goldvarb

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Following Weight Number
Consonant (devoiced/total)

1 [ts] 0.861 418/534

2 [t] 0.840 785/919
3 [k] 0.818 1481/1830

4 [1]] 0.575 353/486
5 [s] 0.436 462/755

6 pause 0.352 974/1957
7 [I] 0.264 253/741
8 [p] 0.143 73/223
9 [h] 0.065 46/340
1 0 [9] 0.048 44/284
1 1 [4,] 0.041 40/348

 

Here, the general tendency is similar to that for /i/: following stops and
affricates promote devoicing and fricatives demote it. This tendency is actually
even clearer here. All the stops and affricates except [p] are promoters, and all
the fricatives are demoters. Following pause is also a demoter, as it is for the
vowel IV. A difference from the ranking for /i/ is that the following alveodental
affricate [ts] is the best promoter for /u/.

If we remove [p] from this ranking, we get a clear rank by feature; for the
preceding consonant, fricative > affricate > stop, and for the following consonant,
stop > affricate > fricative. The ordering for preceding and following is Obviously a

complete reversal.

7O

It is interesting that [p] behaves more like fricatives than stops in the
following environment. [p] historically changed to [(1)], a bilabial fricative, which
changed to [h] eventually, and its distribution in modern Japanese is very limited.
It seems that [p] demotes devoicing in general, both in the preceding and the
following environments. In the next chapter, I will comment further on this fact.

So far, predictions from previous research are confirmed; a preceding
fricative and a following stop promote devoicing, and devoicing is demoted
before allophones of /h/. Another hypothesis to be investigated concerns the
interaction between the preceding and following consonants; that is, devoicing is
expected to be demoted between two fricatives (Tsuchida 1997).

The results of a cross-tabulation of the preceding and following
consonants show that [s__s] and [Ls] environments for the vowel lu/ strongly
demote devoicing (24% and 31%, respectively), but the [Ls] environment for /i/,
on the contrary, promotes it (84%). Similarly, [s__l] and [L1] environments for /u/
very strongly demote devoicing (19% and 0%, respectively), but the [L1]
environment for /i/ does not (53%). These results are shown in Table 4.5 below.
The percentages are given first, and the numbers in parentheses indicate the

actual number Of devoiced tokens and the total number of occurrences.

71

Table 4.5: Devoicing between preceding and following fricatives: [s] and [j]

 

 

 

 

 

 

 

lu/ Preceding: is} Preceding: m
Following: [5] 24% (23/95) 31% (16/51)
Following; [I] 19% (21/109) 0% (0/52)

/i/

Following: [3] N/A 84% (58/69)
Following: [j] N/A 53% (57/108)

 

 

 

 

An affricate + [j] combination for both vowels demotes devoicing: [tj‘_j'j for
/i/ at 11% and [ts __l'] for /u/ at 20%. However, an affricate + [s] combination does

not: [tj_s] for /i/ at 98% and [ts_s] for /u/ at 50%. This is shown in Table 4.6 below.

Table 4.6: Devoicing between a preceding affricate and following fricative

 

 

 

 

 

 

 

Preceding: Its} Preceding: [1)]
(lull (/i/)
Following: [8] 50% (66/131) 98% (133/136)
Following; [11 20% (20/100) 11% (5/47)

 

On the other hand, devoicing between a preceding fricative and a

following affricate seems to strongly promote devoicing, as shown in Table 4.7.

Table 4.7: Devoicing between a preceding fricative and following affricate

 

 

 

 

 

 

 

Preceding: [S] Preceding: [j] Preceding: [I]
W) (/u/) (lil)
Foilowigq: [ts] 79% (66/64) 84% (37/44) 98% (136/139)
Following; [(5] 99% (78/79) 92% (86/93) 96% (68/71)

 

 

__l] for Iu/ at 72% and 49%, respectively.

Devoicing before a (non-lhl) fricative and after an allophone of /h/ is

promoted; [c_s] and [LIT for /i/ at 94% and 80%, respectively, and [([l_s] and [ill

Table 4.8: Devoicing between preceding /h/ and following non-lh/ fricative

 

 

 

 

Preceding: [ill] Preceding: [g]

(/u/) gm
Following: [5] 72% (112/156) 94% (46/49)
Following; [IT 49% (23/47) 80% (73/91)

 

 

 

 

Devoicing before an allophone Of lh/ and after a non-lh/ fricative is strongly

demoted, as shown in Table 4.9.

Table 4.9: Devoicing between preceding non-lh/ fricative and following lh/

 

 

 

 

 

 

 

 

Preceding: [S] Preceding: [j] Preceding: [j]
(lull (/u/) (/i/)
Following: [h] 22% (10/45) 11% (8/73) 10% (7/69)
Following; [4}] 5% (2/44) 5% (2/42) 6% (4/62)
Following; [4,] 37% (31/83) 3% (3/99) 2% (1/46)

 

Devoicing between allophones of /h/ is also strongly demoted, as shown in

Table 4.10.

 

Table 4.10: Devoicing between allophones of /h/

 

 

 

 

 

 

 

 

Preceding: [ill] Preceding: [g]
(lull (/i/)
Following: [111 10% (8/84) 15% (17/115)
Following: [9] 4% (3/70) 13% (5/40)
Following: [1)] 2% (1/41) 0% (0/43)

 

Thus, it is not conclusive that devoicing between two fricatives is demoted.
Devoicing after a fricative and before allophones of /h/ is strongly demoted, but
this is probably because of the prohibition against devoicing before an allophone
of /h/, as argued by Tsuchida (1997). The present results show that, except for
some cases, such as the devoicing of /u/ between non-lh/ fricatives and before [0
and after an affricate, devoicing between two fricatives is, in fact, promoted.

In summary, a preceding fricative and a following stop promote devoicing,
devoicing is demoted before allophones of /h/, and the prohibition Of devoicing
between two fricatives claimed in Tsuchida (1997) was not confirmed in the

present study.

4.1.2 Vowel identity
The difference in devoicing by the identity of the vowel is still controversial.

In the present study /i/ was devoiced at 45.61% and /u/ at 44.9%. From this, /i/

74

appears to be more frequently devoiced than lul, but the percentage difference is
small. However, the results of a Goldvarb run show that this difference is

statistically significant, as shown in Table 4.11.

Table 4.11: Vowel identity — Goldvarb

 

 

 

 

 

 

Vowel Weight Number
Identity (devoiced/total)
1 /i/ 0.605 4484/7261
2 /u/ 0.409 4929/8419

 

 

 

The vowel /i/ promotes devoicing, but /u/ demotes it. It is worth noting that
these data include tokens of sentence final /u/ in a copula (e.g., /desu#/), which is
almost categorically devoiced for many speakers of Tokyo Japanese, so the

greater frequency for III-devoicing is perhaps even more surprising.

4.1.3 Accent
Table 4.12 shows the results of a Goldvarb run for pitch accent pattern,
discussed in 3.1.1.4 above, for /i/. The potentially devoiced vowel is in the middle

position.

75

Table 4.12: Pitch accent pattern for /i/ - Goldvarb9

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Pitch Accent Weight Number
Pattem (devoiced/total)

1 #L# 0.757 1/3
2 HL# 0.740 48/188
3 LL# 0.705 37/215
4 HLL 0.608 825/907
5 #LH 0.593 568/875
6 LLL 0.573 440/61 0
7 LLH 0.551 553/795
8 HHH 0.492 831/1223
9 LHL 0.483 145/264
10 #LL 0.465 5/6
1 1 HHL 0.460 371/629
12 LHH 0.460 454/620
13 HLH 0.356 94/153
14 #HH 0.315 4/7
15 HH# 0.254 1 5/349
16 #HL 0.218 89/282
17 LH# 0.058 1/131

 

 

 

 

 

All the patterns that promote devoicing (above 0.5) are low-pitched.
Although the #L# pattern, which is a single mora utterance, is the best promoter
statistically, we can safely dismiss it because Of the small number Of tokens.
Excluding #L#, the top two promoters are both pre-pausal low-pitched vowels.

All the accented vowels (9, 11, and 16) demote devoicing (below 0.5),

although the LHL pattern (9) is ranked rather high with a weight of 0.483. An

 

° The #H# pattern is excluded in this Goldvarb run because it was a “knockout,” which means
there was no variation. In this case, there were three occurrences of this pattern, and none of

them were devoiced.

76

initial accented vowel (16) is a strong demoter with a weight of 0.218. However,
the strongest demoters are pre-pausal high-pitched vowels.

Among the patterns that demote devoicing (below 0.5), there are two Iow-
pitched vowels, namely, 10 and 13. 10 has a relatively small number of
occurrences, but 13 has a large number. When a low-pitched vowel is located
between two high-pitched vowels, it seems to resist devoicing.

Table 4.13 shows the results Of a Goldvarb run for the pitch accent pattern

for the vowel lul.

Table 4.13: Pitch accent pattern for /u/ - Goldvarb

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Pitch Accent Weight Number
Pattern (cl evoiced/total)

1 LLL 0.762 591/738
2 HLL 0.713 422/738
3 HHH 0.656 827/1493
4 LLl-l 0.608 701/863
5 LL# 0.600 528/878
6 HL# 0.599 437/715
7 #LL 0.590 3/4

8 HLH 0.504 135/224
9 LHH 0.478 229/366
10 HHL 0.408 97/341
1 1 #HL 0.400 64/96
12 #L# 0.299 1/3
13 #LH 0.273 774/1185
14 LHL 0.130 110/398
15 #HH 0.095 2/7
16 HH# 0.089 7/203
17 LH# 0.004 1/165

 

 

 

 

 

 

77

Similar to the results for /i/, all the patterns that promote devoicing (above
0.5) are low-pitched except for the HHH pattern (pattern 3), which is a good
promoter for lul. Accented vowels (10, 11 and 14) all demote devoicing (below
0.5), although the HHL and #HL patterns (10 and 11, respectively) are ranked
rather high compared to Table 4.12. Although accented vowels are demoters,
they are not the strongest ones. The strongest demoters are the pre-pausal
vowels that are high pitched (16 and 17), and this is clearer in the case of /u/
than of lil (weights of 0.089 and 0.004 for /u/ vs. 0.254 and 0.058 for /i/,
respectively).

There are two patterns with a low-pitched vowel that demote devoicing
(below 0.5), namely, 12 and 13. 12 is the case of a single syllable utterance with
a small number of occurrence, but 13 has a relatively large number of occurrence.
Besides, it is the low-pitched vowel in an initial mora followed by a high-pitched
vowel, where one would expect a promotion Of devoicing based on earlier studies
(Martin 1952 [cited in Vance 1987], Yuen 1997). It is true that low-pitched /u/ in
an initial mora followed by another low-pitched vowel promotes devoicing, and
low-pitched /i/ in an initial mora followed by a high-pitched vowel also promotes it,

but any other vowel in an initial mora resists it.

78

It is also interesting that the pre-pausal low-pitched vowels (5 and 6) are
among the promoters but not the strongest for /u/. The tokens include the
occurrences of the copula (“desu#”), where the final vowel is almost categorically
devoiced for most speakers Of the Tokyo dialect. One would have expected that
such environments would be the best promoters for devoicing.

Nagano—Madsen (1995) found that devoicing was strongly demoted before
[I], [9], and [ill] when accented, and NHK’s Japanese Pronunciation and Accent
Dictionary (1985) also states that devoicing is inhibited before /h/ when the vowel

is accented. Table 4.14 shows the devoicing of these environments.

Table 4.14: Accented vowels and the following consonant

 

 

 

 

 

 

 

 

 

 

Following LHL HHL #HL
Consonant
[IT 41% (31/75) 59% (94/158) 13% (6/48)
[9] N/A 28% (1 1/40) 0% (0/1)
[lb] 0% (0/1 ) 3% (3/96) 2% (2/128)
[h] 0% (0/7) 12% (26/223) 0% (0/7)

 

From the table, it seems that a following lh/ does demote devoicing when
the vowel is accented, but it is not Clear if the following [1'] also demotes it.
However, this tendency may be a reflection Of the prohibition of devoicing before

/h/ in general.

79

It has been argued that there is a generational difference in vowel
devoicing in an accented syllable; namely, younger speakers devoice accented
vowels more than older speakers do (Sakurai 1985, Tsuchida 1997). Table 4.15
shows the results of a cross tabulation Of pitch accent pattern (only accented

vowels) and age.

Table 4.15: Accented vowels and age

 

 

 

 

 

 

 

 

 

Age LHL HHL #HL
Younger 37% (66/236) 50% (173/343) 44% (60/135)
Middle 44% (97/219) 50% (166/336) 34% (41/120)
Older 34% (70/208) 43% (127/294) 42% (52/123)

 

For the LHL pattern, Middle Age respondents are the best devoicers, then
Younger, and Older are the least, although the percentage differences are small.
For the HHL pattern, Younger and Middle Age respondents devoice at the same
percentage (50%), and Older respondents devoice less (43%). For the #HL
pattern, Younger respondents devoice most at 44%, then Older at 42%, and
Middle come last at 34%. The percentage differences are very small in any
pattern. It seems that Older respondents devoice least in general (or at least in
two out Of three patterns), but Younger and Middle Age respondents devoice at a

similar rate. It may be true that, if we take an average percentage, the order is

80

Younger (43.7%), Middle (42.7%), Older (39.3%), in terms Of more devoicing, but
the differences would be very subtle. This is, in fact, similar to the general results
for age differences (see 4.2.2 below). I found an interaction between age and sex
and concluded that I should not look at the age groups alone, apart from sex.
From this table, it is not clear if age has any effect on the devoicing of accented
vowels, independent of the effect of age on devoicing in general.

A cross tabulation of age/sex group and pitch accent pattern shows an

interesting result, as shown in Table 4.16 below.

81

Table 4.16: Age/Sex and Pitch Accent Pattem1°

1

174/221

163/221

84 126/1
1

 

The Age/Sex group is ordered so that it reflects the overall devoicing rate
from left to right (high to low devoicing rate). The top three pitch accent patterns
are the ones with an accented vowel.

It is always the case that Younger Males devoice most in each pitch

accent pattern, and it is almost always true that Younger Females devoice least

 

1° The patterns with a very low frequency (#LH, #HH, and #L#) and those with a very low
devoicing rate (LH# and HH#) are excluded from this table.

82

of all, except for three pitch accent patterns, namely #HL, LL#, and HL# (the
shaded cells). Among the accented vowel patterns, it is true that Younger Males
devoice more than Older respondents, but that is also true with non-accented
vowel patterns. It is only the #HL pattern that shows a different pattern from the
general picture. In this pattern, Younger Females (38%) devoice more than
Middle Age Males (34%), Middle Age Females (34%), and Older Males (33%),
although the difference is small. Another unusual fact about this pattern is that
Older Females (50%) devoice almost as much as Younger Males (52%) do (the
#HL shaded cell). The accented vowel in the word-initial mora somehow
promotes devoicing among younger speakers and Older females.

The other two patterns where Younger Females devoice more than Middle
Age and Older respondents are both pre-pausal low-pitched vowels. In both
cases Younger Males still devoice the most, but Younger Females are the
second most frequent devoicers. In this environment, we can say that younger
speakers devoice more than Older generations.

In summary, the most demoting pitch accent patterns are those with a
high-pitched vowel before a pause; there seems to be a tendency for low-pitched

vowels to be more likely to be devoiced than high-pitched vowels, and accent

83

seems to have some influence on devoicing, with accented vowels tending to
resist devoicing, contrary to claims in the recent literature (Sugito 1982, Sakurai
1985, Maekawa 1988, Sugito and Hirose 1988, Kondo 1993, Nagano—Madsen
1995, lmai 1997, Tsuchida 1997).

The proposed interaction between accent and the following consonant (i.e.
devoicing is demoted before [[1 or lhl) was not supported in the present study.

As for the interaction between accent and age, it is true that Younger
Males devoice more accented vowels than Older and Middle Age speakers, but
this does not extend to Younger Females in general. Only the accented vowel in
the initial mora promotes devoicing among Younger speakers, both males and
females, and extends even to Older Females. In terms of age differences,
Younger respondents devoice more in the environment of pre-pausal, low-

pitched vowels.

4. 1.4 Morpheme boundary

Table 4.17 shows the results of a Goldvarb run for morpheme boundaries,

discussed in 3.1.1.3 above, for the vowel /i/.

84

Table 4.17: Morpheme boundary types for IV - Goldvarb

 

 

 

 

 

 

 

 

 

 

Morpheme Weight Number
Boundary (devoiced/totgl)
1 3 0.620 21 14/2900
(Bound morpheme)
2 1 0.600 1901/2493
(No boundary)
3 2 0.273 102/878
(Pause)
4 5 0.1 75 256/540
(Word boundary)
5 4 0.163 108/446
(Compound)

 

 

It has been argued that a morpheme boundary demotes devoicing (Vance
1987, 1992, Kondo 1997, Tsuchida 1997). From the table, we can see that
devoicing is promoted when the vowel is morpheme internal (type 1), and when it
is at a bound morpheme boundary (type 3). When the vowel is before a pause
(type 2), devoicing is demoted, and a word boundary and compound word
boundary strongly demote devoicing. While it is true that devoicing is demoted at
a word boundary, a bound morpheme boundary does not demote devoicing but
promotes it.

Table 4.18 shows the results of a Goldvarb run for morpheme boundaries

for the vowel lul.

85

Table 4.18: Morpheme boundary types for /u/ - Goldvarb

 

 

 

 

 

 

 

 

 

 

Morpheme Weight Number
Boundary (devoiced/total)
1 1 (No boundary) 0.645 2132/2870
2 3 (Bound 0.571 870/1798
morpheme)
3 2 (Pause) 0.467 976/1955
4 5 (Word boundary) 0.302 687/1004
5 4 (Compound) 0.195 264/790

 

 

The general tendency is the same here. When the vowel is morpheme
internal or followed by a bound morpheme, devoicing is promoted. When the
vowel is followed by a pause, it is demoted. When the vowel is at a word
boundary, particularly at a compound boundary, it is strongly demoted. Pause
has a higher weight for /u/, which may be because of the copula “desu#”, which
is almost categorically devoiced for many speakers of the Tokyo dialect.

As for Sakurai’s (1985) claim that devoicing is demoted at a morpheme
boundary and before a fricative, Table 4.19 shows the results of a cross
tabulation of morpheme boundary and the following consonant for both /i/ and

/u/". The shaded cells are the environments that fit Sakurai’s description.

 

1’ These vowels were combined because of similar results in individual analyses.

86

Table 4.19: Morpheme boundary and the following consonant

1 3
(NO bound.) (Bound (Compound) (Word

103/191

11

1

112/1

1437/151

1442/1

 

It seems that the vowels are less likely to devoice at a word level
boundary and before some fricatives, but the results are not consistent, and
Sakurai’s claim cannot be supported.

It has been also argued that morpheme boundaries have a blocking effect
on vowel devoicing in consecutive devoicing environments (Sakurai 1985, Vance

1987 and 1992, Kondo 1997, Tsuchida 1997). Table 4.20 shows the results of a

87

cross tabulation Of morpheme boundary and the consecutive devoicing
environment for both IV and lul. The shaded cells are the target environments:

before a word level boundary and in a consecutive devoicing environment.

Table 4.20: Morpheme boundary and consecutive devoicing environment

 

 

1 3 4 5
(No bound.) (Bound (Compound) (Word
morph.) boundary)
0 76% 66% 35% 71%

(Not in a cons.

devoicirlgenv.) (3422/4363) (2531/3851) "(339/977) (840/1186)

 

 

 

 

 

 

 

 

. . 68% 57% : 6% 29%
$121933ngng (510/746) (249/439) ' » (5/65) (59/207)...
. . 40% 50% a ~ 14% 29%
(P’ev'°“sv's 1 2/2 6 206/411 . e , . . .‘ ll
devoiced) ( 0 5 ) ( I (28,194) {:13 (444.5“

 

The table shows that even when it is not in a consecutive devoicing
environment, a compound word boundary demotes devoicing. As for the vowels
at a word level boundary in a consecutive devoicing environment, it seems that
devoicing is demoted more than the rest of the morpheme boundary types (i.e.
no boundary and bound morpheme boundary), but that confirms two independent
tendencies, not the interactive deterrent to devoicing suggested in earlier work.

Consecutive devoicing is discussed in greater detail below.

88

In summary, devoicing is promoted when the vowel is morpheme-internal
or when it is at a bound morpheme boundary. It is demoted at any other
boundary type; pause, word boundary, and compound boundary, confirming

previous studies.

4. 1.5 Consecutive devoicing
Table 4.21 shows the results of a Goldvarb run for the consecutive

devoicing environment for the vowel IV.

Table 4.21: Consecutive devoicing environment for /i/ - Goldvarb

 

 

 

 

 

 

 

 

 

Consecutive Devoicing Weight Number
(devoiced/total)
1 0 (Not in a consecutive 0.538 3862/5650
devoicing environment)
2 1 (Previous vowel is not 0.423 453/993
devoiced)
3 2 (Previous vowel is 0.286 166/614
devoiced)

 

The only promoting factor here is the environment in which the previous
vowel is not in a devoicing environment (1 in Table 4.21). Devoicing is demoted
when the vowel is in a consecutive devoicing environment (2 and 3), and,

unexpectedly, even when the previous vowel is not devoiced (2). However, when

89

the vowel is in a consecutive devoicing environment, the vowel is more likely to
be devoiced when the previous vowel is not devoiced than when the previous
vowel is devoiced; 2 has a much higher weight than 3. In other words, there is a
strong tendency for the avoidance Of consecutive devoicing.

It is interesting, however, that when the previous vowel is also in a
devoicing environment, the devoicing of the vowel is demoted even when the
previous vowel is actually not devoiced. It seems that as soon as the speakers
realize that there are potentially devoiced vowels in a sequence, they try to avoid
devoicing, and end up voicing both more than half the time. I will comment further
on this in the discussion chapter below.

Table 4.22 shows the results of a Goldvarb run for the consecutive

devoicing environment for the vowel /u/.

Table 4.22: Consecutive devoicing environment for lul - Goldvarb

 

 

 

 

 

 

 

 

 

Consecutive Devoicing Weight Number
(devoiced/total)
1 0 (Not in a consecutive 0.642 4299/6684
devoicing environment)
2 2 (Previous vowel is 0.108 238/886
devoiced)
3 1 (Previous vowel is not 0.082 392/847
devoiced)

 

9O

Here, as with III, the only promoting factor is the one in which the previous
vowel is not in a consecutive devoicing environment. However, the difference is
that when the vowel is in a consecutive devoicing environment, devoicing is
much more strongly demoted. Also, the order is reversed; it is demoted even
more strongly when the previous vowel is not devoiced than when the previous
vowel is devoiced. Since the weight difference is very small, I will not focus on
the reversed order, but it is still true that in both cases of consecutive devoicing
environments, devoicing is very strongly demoted. In other words, it does not
matter if the previous vowel is devoiced or not. The speakers appear to be aware,
subconsciously of course, of the danger Of consecutive devoicing and try to avoid
it, resulting in not devoicing any vowels at all.

It is natural to hypothesize that consecutive devoicing would be more
common in more casual speech. Table 4.23 shows the results of a cross

tabulation Of consecutive devoicing environment and style.

91

Table 4.23: Consecutive devoicing environment and Style

 

 

 

 

 

 

 

 

0 1 2
(Not in a consecutive (Previous vowel (Previous vowel
devoicinLenvironment is not devoiced) is devoiced)
C 86% 87% 51 %
(2986/3469) (62/7 1 ) (75/147)
RP 71 % 49% 36%
(2424/341 0) (231/475) (1 05/292)
WL 50% 43% 21%
(2756/5465) (552/1295) (224/1061)

 

 

The shaded cells are the cases of consecutive devoicing. It is confirmed
that consecutive devoicing is more common in the Conversational style, at 51%,
less likely in the Reading Passage style, at 36%, and most unlikely in the Word
List style, at 21%. However, even in the most formal style, the Word List style,
consecutive devoicing does occur rather more frequently than previous research

or prescriptive rules would suggest.

4.2 Social Factors

This section will discuss the results of the statistical analyses using
Goldvarb for the social factors, namely style, age, sex, and social class. The
statistical analyses were done for both vowels separately since it was found that

there are cases in which social factors have a very different effect on Ill and /u/.

92

4. 2. 1 Speech style
As noted in 3.1.2.4 above, speech style is the most significant factor in
vowel devoicing in Japanese (Yuen 1997). Table 4.24 shows the results of a

Goldvarb run for speech style for the vowel /i/.

Table 4.24: Speech style for /i/ - Goldvarb

 

 

 

 

Speech Style Weight Number
(devoiced/total)
1 Conversation 0.816 1 585/1 785
2 Reading passage 0.486 1097/1732
3 Word list 0.335 1799/3740

 

 

 

 

 

 

The results confirm Yuen’s findings; the more formal the speech style is,
the less frequently devoicing occurs. In fact, only conversational style promotes
devoicing, and the word list and reading passage styles demote it, even though
devoicing is more likely to occur in reading passages than in word lists.

Table 4.25 shows the results of a Goldvarb run for speech style for the

vowel lul.

93

Table 4.25: Speech style for NI - Goldvarb

 

 

 

 

 

 

 

 

Speech Style Weight Number
(devoiced/total
1 Conversation 0.742 1 585/ 1 785
2 Reading passage 0.590 1097/1732
3 Word list 0.330 1 799/3740

 

Here, both the conversational and reading passage styles are promoters

for devoicing, and only the word list style demotes it. However, the general order

is the same; the more casual the style, the more likely devoicing occurs.

4. 2.2 Age and sex

The individual results for age and sex revealed an interaction between

them. Therefore, these factors are combined. Table 4.26 shows the results for

age and sex combined for the vowel IV.

Table 4.26: Age and sex combined for /i/ - Goldvarb

 

 

 

 

 

 

 

 

 

 

 

Age & Sex Weight Number
(d evoiced/totaj

1 Young Male 0.614 911/1303
2 Middle Male 0.517 795/1248
3 Middle Female 0.486 766/1243
4 Older Male 0.477 620/1030
5 Older Female 0.473 461/1258
6 Younger Female 0.418 628/1175

 

94

 

 

Yuen (1997) argued that men devoice more than women, but this is not

true across age groups. Since Middle Females devoice more than Older Males,

for example, we cannot say that men always devoice more than women. Figure

4.1 shows this visually.

 

 

l Male
El Female

 

 

 

 

Older Middle Young

 

 

 

Figure 4.1: Age and Sex for /i/

The sex difference is very subtle for Older and Middle respondents. It is

obviously the Younger Males who contribute to the high devoicing rate for males

and the Younger Females who contribute to the low devoicing rate for females. In

fact, the differences between Older Males, Older Females, and Middle Females

95

are so small that they were combined to form one group, and the results were not
significantly different. Table 4.27 and Figure 2 show the results of the

combination.

Table 4.27: Age and sex for li/ - Older Male, Female, and Middle Female

 

 

 

 

 

 

combined
Age 8: Sex Weight Number
(devoiced/total)
1 Younger Male 0.614 911/1303
2 Middle Male 0.517 795/1248
3 Older Male, Female 0.479 2147/3531
& Middle Female
4 Younger Female 0.418 628/1 175

 

 

 

 

 

96

 

Devoicing by Gender and Age

II

 

Younger Middle Male Older Male, Younger
Male Female & Female
Middle
Female

 

 

 

Figure 4.2: Age and Sex for /i/ - Older Male, Female, and Middle Female
combined

It is clear that Younger Males are the greatest promoters of devoicing, and
Younger Females are the biggest demoters. Middle males only somewhat
promote devoicing and the rest (Older Males, Older Females, and Middle
Females) are all demoters.

Table 4.28 shows the results for age and sex combined for the vowel /u/.

97

Table 4.28: Age and sex combined for lu/ - Goldvarb

 

Here, the order is exactly the same, and the weights are very similar to

those of /i/. Figure 4.3 below shows this visually.

 

I Male
[3 Female

 

Older Middle Young

 

 

 

Figure 4.3: Age and Sex for /u/

98

The differences between males and females are larger for /u/ for all age
groups, but the sex difference for Young respondents is still much larger than
that for Older and Middle Age respondents. Because the sex differences in each

age group are larger than those for /i/, I could not combine any groups.

4.2.3 Social class

There are always problems in social class classification; in this case, the
basis for neighborhood classification is weak, and income is not included
because it is a hard question to ask during an interview. However, assuming it
might suggest some directions for future study, I will discuss the results of the
statistical analyses for social class in the following.

Although social class was selected by Goldvarb as a significant factor for
devoicing of /u/, it was thrown out as insignificant for IV. Table 4.29 shows the

results for social class differences for /u/.

Table 4.29: Social class for /u/ - Goldvarb

 

 

 

 

 

 

 

 

Social class Weight Number
(devoiced/total)
1 Middle 0.520 1859/3149
2 Working 0.490 1550/2731
3 Upper 0.486 1 520/2537

 

 

99

Although significant, the differences among social classes are very small.
The Middle Class somewhat promotes devoicing, while the Working and Upper
Middle Classes only slightly demote it, and the difference between them was not

significant. Table 4.30 shows the results Of this combination.

Table 4.30: Social class — Working and Upper combined - for Iu/

 

 

 

 

 

Social class Weight Number
(devoiced/total)
1 Middle 0.519 1859/3149
2 Working& Upper 0.488 3070/5268

 

 

 

 

Since Younger Males devoice most, it is reasonable to hypothesize that
this feature is nonstandard because younger males are prone to the covert
prestige Often associated with nonstandard features. One would, therefore,
expect to find the typical pattern of nonstandard features associated with social
class, in which the lower the social class, the more nonstandard features one
should find. However, that is not the case here. The pattern is that Middle class
people devoice most, and Working and Upper classes devoice least. This pattern
is more typical for a standard feature, particularly if the greater use by the Middle

Class is representative of the linguistic insecurity often exhibited by Lower Middle

100

Class respondents in much sociolinguistic work. I will provide a more detailed

discussion on this distribution in the interpretation section below.

4.3 Summary

The results for the more sophisticated statistical treatment Of this large
sample show both correspondences and a lack of correspondence with previous
work. I’ll summarize each correspondence and non-correspondence below. Note
that within a single factor, we have several different results and some Of them
may show correspondence, and others may show non-correspondences. Those

factors that confirmed previous studies are as follows.

(1) Effect of the preceding and following consonant: a preceding fricative
and a following stop promote devoicing most, and devoicing is
demoted before /h/.

(2) Effect of accent: accented vowels resist devoicing, and pre-pausal
high-pitched vowels very strongly demote devoicing. This confirms the
traditional description Of Japanese vowel devoicing as opposed to
more recent studies (Sugito and Hirose 1988, Kondo 1993, Nagano-
Madsen 1995, Tsuchida 1997).

(3) Effect Of morpheme boundary: word boundary, including compound

boundary, demotes devoicing, whereas no boundary and bound
morpheme boundary promotes it.

101

(4) Avoidance of consecutive devoicing: devoicing is demoted in a
consecutive devoicing environment.

(5) Effect Of speech style: the more casual the speech style is, the more
frequently devoicing occurs.

Those factors that did not support the previous studies are as follows.

(1) Interaction between the preceding and following consonants: the claim
that devoicing is prohibited between two fricatives (T suchida 1997) is
not supported by the present study.

(2) Effect of vowel identity: the vowel /i/ promotes devoicing, but /u/
demotes it. This contradicts the claim made by Han (1962), but
supports Yuen’s (1997) results.

(3) Effect of age in the devoicing Of accented vowels: the present study did
not completely support the claim that younger speakers devoice
accented vowels more than Older speakers do. It is only young males
who devoice more than Older speakers, and younger females tend to
devoice less than older speakers.

(4) Interaction between accent and the following consonant: the claim that
devoicing is demoted before [I] or /h/ when accented (NHK 1985,
Nagano-Madsen 1995) is not supported by the present study.

(5) Interaction between morpheme boundary and the following consonant:
the claim that devoicing is demoted before a fricative at a morpheme
boundary (Sakurai 1985) is not supported by the present study.

(6) Effect of morpheme boundary in a consecutive devoicing environment:
the claim that devoicing is demoted at a morpheme boundary in a
consecutive devoicing environment is not supported in this study; the
two demoting factors are independent.

102

(7) Effect of sex: the claim that men devoice more than women do (Yuen
1997) was not entirely supported in the present study because sex and
age show an interaction; younger males devoice most, and younger
females devoice least, but these differences are weak or nonexistent in

middle-age and older respondents.

Both correspondences and a lack of correspondences, as well as new

findings from the present study, will be discussed further in the next, interpretive,

Chapter.

103

Chapter 5

Discussion

This chapter will interpret the detailed results reported in the previous
chapter. I will discuss linguistic factors in 5.1, and social factors in 5.2, and

summarize the major findings in 5.3.

5.1 Linguistic Factors

The present study has generally confirmed the results Of previous studies
in terms Of the promoters and demoters of devoicing by the preceding and
following environments. It was also found that accented vowels resist devoicing,
contrary to the claims made in recent studies (Sugito 1982, Sakurai 1985,
Maekawa 1988, Sugito and Hirose 1988, Kondo 1993, Nagano-Madsen 1995,
lmai 1997, Tsuchida 1997). In terms of pitch accent patterns, a pre-pausal low-
pitched vowel was the best promoter for /i/, but not for lul, and a pre-pausal high-
pitched vowel was the strongest demoter for both vowels. The effect Of
morpheme boundary demotion was generally confirmed, but the results for
consecutive devoicing environments showed an unexpected pattern. I will

discuss these issues in more detail in the following subsections.

5. 1.1 Preceding and following consonants

The general picture Of promoting and demoting factors in terms Of

preceding and following consonants Obtained from the present study is

104

consistent with the one repeatedly reported in the literature; preceding fricatives
promote devoicing most, then affricates, and stops demote it; following stops

promote devoicing most, then affricates, and fricatives demote it. However, the

behavior 0le] and [p] do not conform to the general picture, and a following

pause demotes devoicing, contrary to the standard description Of vowel
devoicing in Japanese (e.g. McCawley 1968, Vance 1987).

The present study also supports Tsuchida’s (1997) Claim that devoicing
before an allophone of /h/ is phonologically prohibited; I found that it is strongly
demoted. However, it did not clearly support another claim of hers that devoicing

between two fricatives is also phonologically prohibited, because devoicing

before [8] or U] and after [s], [I], or an allophone Of /h/ is rather promoted, except

for some cases for the vowel /u/ (refer to Tables 4.5 and 4.8).

5.1.1.1 Status of [I] and [p]

As seen in the previous chapter, the behavior Of [j] and [p] is unusual.
First, a preceding [1] is the best promoter for the devoicing of /i/ but is a strong
demoter for /u/ (refer to Tables 4.1 and 4.2). When [s] and [1‘] occur in the
following environment, [1'] tends to fall behind [s]; [s] is a good promoter for /i/ with

a weight of 0.756, but [[1 is a demoter, with a weight Of 0.432, and both [6] and [j]

demote the devoicing of /u/, but [3] demotes less, as shown by its higher weight,

(0.436), compared tO [1] (0.264) (refer to Tables 4.3 and 4.4).

105

According to Shibatani (1990), the “palatalized consonants ([1] and [tj'j)

have a skewed pattern of distribution” (163). He argues that the combinations [SH

and [[e] do not occur”; therefore, they are in complementary distribution in the
environment of front vowels, and the distribution Of [l'] is limited before non-front
vowels (163). He goes on to argue that [l] and [1]] before /i/ are allophones of /s/

and /1/, respectively, but before other vowels, as in [la] and [tja], they are derived

from lsja/ and ltja/ (164). I understand that this suggests that they are equivalent

to palatalized phonemes lsyl and ltyl, respectively, but Shibatani does not give an

explanation for this skewed distribution of the palatalized phonemes. Thus, [I] is

an allophone of ISI before lil, but it is a phoneme before lul. Since preceding Is/ is

a promoter for devoicing but III is not, the different behavior of those two

fricatives can be accounted for to some degree by this distinction. Even this
apparent underlying difference, however, does not account for the difference
between the two if we accept the generalization that preceding fricatives are
expected to promote devoicing. The results here suggest a variable phonological
interpretation.

A preceding [p] is a very strong demoter for the devoicing of both high
vowels (refer to the Tables 4.1 and 4.2). This is expected because [p] is a stop,
and stops are expected to demote devoicing when they occur in the preceding

environment. However, in the following environment, where stops are expected

 

'2 [is] only occurs in foreign words like ‘Shakespeare’ (Ijeekusupial) and in interjections. Vance
(1987) gives an example of [jerii] ‘sherry’ (21). He also states that He] is easy to pronounce for
native speakers of Japanese, but [Si] is very difficult, and is usually pronounced as [[1].

106

to promote devoicing, [p], and only [p] among stops, actually functions as a
demoter. For both /i/ and /u/, [p] is a very strong demoter with a weight below 0.2
(cf. Tables 4.3 and 4.4).

According to Shibatani (1990), “[p] does not occur in initial position in
native or Sino—Japanese words” (163), and “the distribution of [p] is extremely
limited in the native and Sino-Japanese vocabularies. [it] is limited to the

environment of consonant clusters, e.g. yappan' ‘as expected’, simpai ‘worry”’
(166). Historically, [h] was derived from [111] which was itself derived from [p],
which explains the alternation between [h] and [p] (hoo “law" vs. ken-poo
“constitution"), and [h] and [b] (hito “person” vs. hito-bito “people”) (Shibatani
1990; 166-7). [ill] remains only before /u/ in modemJapanese, and the
distribution Of [p] is also limited in the native and SinO-Japanese vocabularies,
although it is a phoneme.

Considering the status of both Ill and /p/, which have very limited

distributions in modern Japanese, I am tempted to suggest that vowel devoicing

is sensitive to the status of the phoneme. It seems that /j/ and /p/ do not have full

status as phonemes, and this is reflected in the process of vowel devoicing.
However, more study is needed to determine the status of these two phonemes
and the relationship of this status to vowel devoicing. As with the sibilant
distinctions discussed above, the data here point to a variable phonological

rather than phonetic influence.

107

5.1.1.2 Devoicing before a pause

The standard description of vowel devoicing in Japanese is that the high
vowels (/i/ and /u/) are devoiced between voiceless consonants or between a
voiceless consonant and a pause (Sakuma 1929, McCawley 1968, Kawakami
1977, Vance 1987, Maekawa 1988, Sugito 1988, Kondo 1994, Nagano-Madsen
1995). However, the present study found that a following pause demotes
devoicing of both vowels, and it demotes li/ devoicing more strongly (0.119), than
lu/ (0.352) (cf. Tables 4.3 and 4.4). This contradicts the standard description of
vowel devoicing in Japanese.

If one also takes pitch accent patterns into consideration, however, a pre-
pausal low-pitched vowel was the best promoter for /i/, and the same
environment also promotes the devoicing of lul, although it is not the best
promoter. On the other hand, pre-pausal high-pitched vowels demote devoicing
very strongly (cf. Tables 4.12 and 4.13).

It is not only in this study that the status of the following pause has been
controversial. Other recent studies show that devoicing of word-final vowels is
not common, or less likely (Maekawa 1988, Takeda and Kuwabara 1987,
Nagano—Madsen 1994, Kondo 1997 [all cited in Tsuchida 1997]). Maekawa
(1988) states that an utterance-final pause has the same effect as a following
voiceless consonant, i.e. it promotes devoicing, but that a sentence-intemal
pause demotes devoicing. His study was conducted in the Tottori prefecture (in

the western part of Japan) and showed that devoicing of vowels before a

108

sentence-intemal pause is less likely, perhaps due to the fact that most of those
vowels have a raised pitch accent.

Kawakami (1977) reports that a high vowel before a pause is devoiced
only when the vowel is low-pitched, and this is supported by the present study;
the strongest demoter among pitch accent patterns is a pre-pausal high-pitched

vowel (cf. Tables 4.12 and 4.13). However, Takeda and Kuwabara (1987) show

that word-initial and word-medial [11] was devoiced at 31% and 24%, respectively,

but word-final [[1] was devoiced at only 1% (T suchida 1997), and I assume these

are all low-pitched vowels. Kondo (1997) even suggests that devoicing before a
pause may be limited to utterance final ldesu/ (polite copula) and lmasu/ (polite
verb suffix) (Tsuchida 1997; 299). However, the present study shows different
results. The devoicing of both the low-pitched high vowels /i/ and /u/ is promoted
before a pause.

The present study does not distinguish sentence-final vowels from
sentence-intemal ones, partly because it is difficult to tell if the utterance is a
sentence or a part of one in spontaneous conversations. However, the fact that
the pre-pausal low-pitched vowel is the best promoting environment for /i/, but
the same environment is not the best promoter for /u/ seems to make this
irrelevant. If a high rate of devoicing Of a low-pitched vowel before a pause is due
to the sentence final copula and verb suffix, /u/ should have as its best promoter
the pre-pausal low-pitched vowel.

None of the studies mentioned above discuss the influence of vowel

identity in general. Therefore, I cannot compare my results with theirs. Besides,

109

Kondo (1997) used isolated words, while the present study used three different

speech styles, which might have caused the difference (see 5.2 below).

5.1.1.3 Status ofAffn'cates

The internal organization of an affricate is still controversial. A traditional
view is that it is a combination of two features; a stop and a fricative, in that order.
However, more recent studies suggest that an affricate is actually a stop
(Steriade 1993, Rubach 1994), or that the features are not ordered (Lombardi
1990)

A previous articulatory study shows that affricates pattern with stops,
rather than fricatives, in terms of the glottal gesture in word-initial and word-
medial positions (Sawashima and Niimi 1974, Cited in Tsuchida 1997). [8] shows
a large glottal opening in both positions, but both stops and [ts] show larger
glottal openings in word-initial position. This seems to be in line with the claim
that affricates are a kind Of stop (Steriade 1993, Rubach 1994). In Yoshida and
Sagisaka (1990, cited in Tsuchida’s 1997 study), this is also supported because
the following affricates had a devoicing rate similar to that Of the following stops
(both stops and affricates have over a 70% devoicing rate), but not to that of
fricatives (between 50% and 60%). Tsuchida also stipulates that affricates do not
pattern with fricatives based on the fact that a high vowel following an affricate
and followed by a fricative is expected tO devoice according to NHK (1985), as in
(8) below, whereas the devoicing of a high vowel between two fricatives is

prohibited.

110

(8) [tlll‘lo] ‘blood’
[tjjsud3i] ‘Iineage’
[atsqsa] ‘thickness’

(Data taken from Tshichida 1997; 303)

In the present study, the tendency of affricates to pattern with stops in the
following environment, reported in Yoshida and Sagisaka (1990), is supported (of.
Tables 4.3 and 4.4). However, affricates do pattern with fricatives in the
preceding environment (Cf. Tables 4.1 and 4.2). Both fricatives and affricates are
promoters for both vowels in this environment. If an affricate is actually a
combination of a stop and a fricative, then it is the stop that is adjacent to the
vowel in the following environment, and the fricative in the preceding
environment. Therefore, for the preceding and following environment patterns,
the present study supports the traditional view, in which affricates are a
combination of a stop and a fricative.

The present study also shows that devoicing after an affricate and before
a fricative is strongly demoted when the following fricative is [j], which is
consistent with the traditional view of an affricate if we assume Tsuchida’s claim

that devoicing between two fricatives are prohibited. However, devoicing was not

demoted when the following fricative is [3] (Cf. Table 4.6). Moreover, devoicing

after [tj] and before [5] for /i/ is strongly promoted (98% devoicing rate).

111

It is difficult to account for this difference, although it could be related to

the status of [l] as a phoneme, as discussed above. In addition, however, the

present study did not support Tsuchida’s claim that devoicing is prohibited
between two fricatives; therefore, this argument may not be valid in general. In

that case, the present study seems to support the traditional view of affricates.

5.1.2 Vowel identity

In the present study, it was found that vowel identity is one of the
significant factors for vowel devoicing; the vowel /i/ is more likely to devoice than
the vowel lul. Previous studies disagree on this issue; Han (1962) argues that lul
is more readily devoiced than /i/; Yuen (1997) reports that /i/ is more likely to
devoice than /u/, and Maekawa (1983) and lmai (1997) did not find any
difference between the two. However, the present study analyzed a much larger
number of tokens than any of these studies; therefore, the slight difference in the
devoicing percentages between the two vowels turned out to be significant.
These computations include tokens of sentence final polite copula and polite verb
suffix (ldesu/ and lmasul, respectively), which are almost categorically devoiced
for many speakers of Tokyo Japanese (Maekawa 1988). This implies that the
devoicing of /i/ is even more likely than that of /u/ if we disregard these instances.

I can only speculate on why /i/ is more likely to devoice than /u/. In

Japanese, consonants before /i/ are highly palatalized. This is Obviously realized

in the allophonic alternation for /t/ and /s/. For example, /s/ becomes [1] before /i/

but remains [s] before all other vowels. Because of this palatalization, speakers

112

are capable of hearing the difference between devoiced [IT] and [[u] (Beckman

and Shoji 1984). If the vowel is recoverable from the co-articulation of the vowel
and the preceding consonant, devoicing may be promoted even more since
possible ambiguity is avoided.

This is just a speculation, and we obviously need more study in the

difference between /i/ and /u/ in terms of devoicing rate.

5.1.3 Accent

As discussed in 5.1.1.2 above, the strongest demoter of all the pitch
accent patterns is that of pre-pausal high-pitched vowels, which is consistent with
previous studies (Martin 1952, NHK 1985).

It was also found that low-pitched vowels are more likely to devoice than
high-pitched vowels in general, and accent seems to have some influence on
devoicing because accented vowels demote devoicing, contrary to claims made
in recent studies (Sugito 1982, Sakurai 1985, Maekawa 1988, Sugito and Hirose
1988, Kondo 1993, Nagano-Madsen 1995, lmai 1997, Tsuchida 1997). Although
it is true that accented vowels are frequently devoiced (cf. Tables 4.12 and 4.13),
the more sophisticated statistical model used in this study shows that accent
demotes devoicing in all pitch accent patterns.

It is interesting that the major factor turned out to be the pitch associated
with the vowel, not the word accent, because most previous studies focus on the
accented vowel and its influence on vowel devoicing. Han (1962) and Maekawa

(1983) are the only studies, as far as I know, that looked at L and H pitches. Han

113

claims that a high vowel between voiceless consonants is generally not devoiced
when it has a high pitch but is devoiced when it has a low pitch. However, it is not
clear if she meant “accented vowels” by “high pitch.” On the other hand,
Maekawa concluded that pitch accent does not have an influence on the rate of
vowel devoicing. He found that low-pitched high vowels were devoiced 83% of
the time, whereas high-pitched vowels were devoiced at a 57% rate (from Table

7, 75), but he attributed this difference to the frequent occurrences of the low-

pitched [Sin] and [[1] in ldesu/ (polite copula), lmasu/ (polite verb suffix), /dejita/

(past tense Of ldesul), and lmaj'lta/ (past tense of lmasul). When he excluded

those cases, the difference between high-pitched and low-pitched vowels was
very small (50% and 57%, respectively).

However, Maekawa had a very small number Of tokens (453), and he only
reports percentages, while the present study has a large number Of tokens (over
15,000) and uses a sophisticated statistical program engineered to handle cells
with a relatively small number.

In trying to account for the peculiarity Of ldesu/ and lmasul, Maekawa

(1983) points to (1) the intonation associated with the sentence ending and (2)

the post-accent position of those devoiced vowels, as in [de‘jjta], [ma‘jjta],

[de‘Sug], and [ma’sulj (where indicates an accent shift from high to low).

Fujimura (1967) also states that vowel devoicing is more likely after the accented
syllable (cited in Maekawa 1983), but it is not clear if he meant the vowel right
after the accented vowel, or just any low-pitched vowel after the accented vowel

(Maekawa 1983, Note 13).

114

With the vowels immediately after the accented one (Of. Tables 4.12 and
4.13), those patterns with the HLX sequence (X indicates high or low pitch
segment; 2, 4, and 13 in Table 4.12 and 2, 6, and 8 in Table 4.13) are mostly
promoters for both vowels. The only exception is the HLH pattern for the vowel /i/,
which demotes devoicing. A low-pitched vowel placed between two high-pitched
vowels resists devoicing. This may be because of the effort to preserve the pitch
shift which otherwise may be cancelled if devoicing occurred. It is also possible
that, since this pitch accent pattern is only possible at a word boundary, it
demotes devoicing because of the word boundary (cf. 4.1.4).

Another finding Of the present study is that accented vowels resist
devoicing, contrary to recent studies (such as Sugito 1982, Sakurai 1985,
Maekawa 1988, Sugito and Hirose 1988, Kondo 1993, NaganO-Madsen 1995,
lmai 1997, Tsuchida 1997). In both Table 4.12 and Table 4.13, all the patterns
with XHL have a weight below 0.5. An accented vowel is a demoting environment,
but devoicing still occurs. In Table 4.16, it is clearly shown that Younger Males
devoice accented vowels more than 50% of the time, which may have caught the
attention of older speakers who say that younger speakers devoice accented
vowels. However, one cannot conclude that younger speakers devoice more
accented vowels than Older speakers as claimed by recent studies because
younger females tend to devoice less than older speakers. This is true in two of
the three patterns (LHL and HHL), and younger females devoice the vowel in the
#HL pattern only slightly more than Older Males, Middle Age Males, and Middle

Age Females. However, Older Females devoice the vowel in the same pattern

115

more than Younger Females do, and almost as much as Younger Males do. This
peculiar behavior of Older Females needs to be accounted for, since this is the
only pattern in which Older Females show a similar devoicing rate to that of
Younger Males, who are the biggest promoters of devoicing.

Thus, the claim that younger speakers devoice more accented vowels
than older speakers is not supported. It is true that younger males devoice more
accented vowels than older speakers, but it may only be a reflection of the
general tendency for younger males to devoice more than older speakers. On the
other hand, there is one environment where younger speakers, both males and
females, devoice more than older speakers; pre-pausal low-pitched vowels.
However, the percentage differences among Older, Middle Age and Younger
Female respondents are rather small for the LL# pattern. More extensive study is

needed to determine if there is any age difference in this particular environment.

5.1.4 Morpheme boundary

It has been argued that morpheme boundaries have a blocking effect on
vowel devoicing, particularly in consecutive devoicing environments (Sakurai
1985, Vance 1987 and 1992, Kondo 1997, Tsuchida 1997). For example,
Tsuchida (1997) states that “devoicing may occur across a word boundary when

there is only one devoiceable vowel” (256) and gives the following examples.

(2) [kjakul#seN] “passenger boat’ ([kjaku] ‘guest’ + [seN] ‘boat’)

[okj#kaéru] ‘rearrange’ ([Oki] ‘put’ + [kaeru] ‘change’)

116

[limpjilljl'lgij ‘mysticism’ ([[impfl ‘mystery’ + [[091] ‘principle')

However, the last example is a case of a compound word, where both
elements can be used as an independent word. In the second example, on the
other hand, [oki] is a verb stem and cannot stand alone”, i.e., it requires some
other morpheme, in which sense it is a bound morpheme. Finally, in the first
example, [kjaku] is an independent word, but [seN] is a bound morpheme. Thus,
it is not clear what the researchers (e.g. Vance 1987, 1992, Kondo 1997,
Tsuchida 1997) mean by ‘a word boundary‘ or ‘a morpheme boundary.’

The present study found that a word level boundary has a blocking effect,
but not a bound morpheme boundary (cf. Tables 4.17 and 4.18). As expected,
the most promoting morpheme boundary type is the morpheme-intemal vowel,
but the next promoting boundary type is a bound morpheme boundary. All other
boundary types, i.e. pause, word boundary, and compound word boundary,
demote devoicing. It is interesting that a compound word boundary demotes
devoicing more strongly than a word boundary, which may not even form a
grammatical unit, such as a subject and a verb, as demonstrated in Vance (1992).

The reason why a morpheme-intemal vowel promotes devoicing is
probably straightforward: there is less risk of being misunderstood. In English,
word-final t/d deletion (Guy 1980) shows similar patterning. Within a single
morpheme, the word-final t/d is more likely to be deleted than that in a multiple

morpheme word; e.g. “mist” vs “missed.” That is, the ‘t’ in “mist” is more likely to

 

'3 It can stand alone when it is used as a conjunctive (“put (it) and ~”). However, in this usage,
extra meaning is attached to the verb, namely “and”, so I do not consider this as the same verb
standing alone.

117

be deleted than the ‘t’ in “missed” [mist]. The reason seems to be that if you
delete the ‘t’ in “missed,” you lose morpheme information and there is a risk of
confusing “miss” and “missed.” Similarly, if you devoice the morpheme-internal
vowel, you may be signaling it is one word, and if you don’t devoice, you may be
signaling that there is a word level boundary. If you devoice the vowel at a word
level boundary, you may lose the information and the cluster of words may sound
like a single word. Again, this is only a speculation, but is it true that both
phenomena (vowel devoicing in Japanese and consonant cluster simplification in
English) make the distinction between morpheme-intemal and morpheme-

external.

5. 1.5 Consecutive devoicing

Devoicing of consecutive syllables has been reported (e.g. Sakuma 1929,
Kawakami 1977, Maekawa 1988) but other studies claim it is avoided,
particularly in a slower, more formal speech (e.g. Han 1962, Tsuchida 1997).

Tsuchida (1997) argues that there is a regular pattern for the possible
devoicing sites; the initial two vowels are evaluated with the ranked constraints,
and, after it is decided which one of the initial two is devoiced, the rest of the
sequence follows an alternating pattern (cf. 2.2.5).

The present study shows, however, that devoicing is demoted in a
consecutive devoicing environment even when the previous vowel is not
devoiced (Cf. Tables 4.21 and 4.22). The weight is higher for /i/, which is 0.423,

but it is very low for /u/, which is 0.082. In the case of lul, devoicing is more

118

strongly demoted when the previous vowel is not devoiced than when the
previous vowel is devoiced, although the difference was very small. This seems
to indicate that the speakers are subconsciously aware of consecutive devoicing
before they utter the word, and, in trying to avoid it, end up not devoicing any
vowels at all.

This is cognitively interesting. Morris (2003) found in her perception study
of Japanese vowel devoicing that speakers of Tokyo Japanese judged a person
who did not devoice the vowel in the most favorable environment for devoicing as
a Tokyo speaker because they did not pay attention to the actual devoicing, but
to the environment in which the vowel occurred. In other words, they did not care
if the person actually devoiced the vowel or not, but, since it was in such a likely
environment for devoicing, they just assumed that it was devoiced. This is

counterintuitive, however, and we often hear comments which note that someone

who does not devoice the high vowels in lkusa/ (‘grass’) and /tj'lkara/ (‘power’)

sounds like they are from Osaka or Kochi (places in the western part of Japan)
(Sakuma 1929). In Tokyo Japanese, the final vowel in the utterance lhai,
soo##desu/ (‘Yes, it is so.’) is almost categorically devoiced, and if someone.
does not devoice it, it gives a strange impression (Maekawa 1988). These
comments indicate that peOple appear to recognize vowels that should be
devoiced but are not. However, according to Morris (2003), this is not the case.
People appear not to pay attention to the actual devoicing in the most likely
environments. This is consistent with one finding of the present study. People do

not pay attention to the actual devoicing Of the previous vowel in a consecutive

119

devoicing environment, but they are aware (obviously subconsciously) of the
environments and that there is a danger of consecutive devoicing. In response to
that, which might result in lexical ambiguity, people tend to try to avoid the
situation. This is another piece of evidence that shows the difference between
prescriptive grammar and descriptive grammar. What a dictionary says and what
people think they should do (or actually do) is often very different from what they

do in their unmonitored, natural language use (e.g. Trudgill 1972).

5.2 Social Factors

As seen in 4.2 above, the results of the present study show variation in
devoicing in terms of speech style, age and sex, and, for the vowel lul, social
class. It was confirmed that the more casual the speech style becomes, the more
devoicing occurs. There is an interaction between age and sex, and younger
males devoice most, younger females devoice least, and the rest fall in between.

From these patterns, it is reasonable to hypothesize that vowel devoicing
in Japanese is a nonstandard feature, because younger males are often prone to
be attracted to the covert prestige associated with nonstandard features, and
nonstandard features usually occur in more casual speech style. Moreover,
younger females are often prone to be attracted to the overt prestige associated
with standard features. This may explain why younger males devoice most and
younger females devoice least.

However, social class difference was also significant, at least for the vowel

lul. Middle class respondents are more likely to devoice and Working and Upper

120

Middle classes are less likely to devoice. This social class patterning is more like
that of a standard feature, because the typical pattern of a nonstandard feature is
that the lower the social Class, the more frequently the nonstandard feature is
used. The pattern in which Lower Middle class speakers use the feature most,
and Working and Upper Class speakers don’t use it as much, may indicate the
linguistic insecurity Of the Lower Middle class speakers, which means the feature
is considered prestigious and associated with standardness. Overuse of such
features by Lower Middle Classes is known as “hypercorrection” (Labov 1966).
Kroch (1978) argues that “ordinary unconscious phonological changes are
phonetically motivated processes” (21); therefore, “non-prestige dialects tend
tO be articulatorily more economical than the prestige dialect“ (20). The socially
prestigious dialects resist these natural processes because they want tO “mark
themselves Off as distinct from the common people” (30). If this reasoning is
applied to the present case, and if vowel devoicing is a nonstandard feature, it
should be more economical than non-devoicing. This seems to be articulatorily
supported because phoneticians found that vowel devoicing occurs because of
the glottal gestural overlap, an effort to minimize the glottal gesture, which is
manifested in the sequence of a vowel and the neighboring voiceless consonants
produced with a single glottal opening (Yoshioka 1981; Yoshioka, Lquvist, and
Hirose 1982; Jun and Beckman 1993; Tsuchida 1997; Fujimoto, et. al. 2002). By
devoicing the vowels between voiceless consonants, the glottal gesture becomes

more economical.

121

However, it needs to be pointed out that the phonetic outcome of the
extreme case of vowel devoicing is vowel deletion, and this would be
considerably less natural for Japanese because it produces consonant clusters.
CV syllables are considered to be the most unmarked syllable structure, but as a
result of vowel deletion, lkusa/ becomes /ksa/. Also, even when the vowel is not
deleted, a devoiced vowel is more marked than voiced vowels, as indicated by
the constraint SONVOI (sonorants are voiced; Ito, Mester, and Padgett 1995).

If it is actually the case that vowel devoicing in Japanese in a nonstandard
feature, why is it considered standard to devoice certain vowels in Tokyo
Japanese? Why do people make comments that the western dialects give an
impression Of being smooth and soft, and Tokyo dialect gives an impression of
being crisp and clear, (at least partly) because vowel devoicing occurs frequently
in Tokyo dialect, but very infrequently in the western dialects (Sakuma 1929)?

Maekawa (1988) also acknowledges that there is a certain social measure
(probably standardness) in vowel devoicing in Japanese, because 1) there are
certain vowels, such as the final vowel in ldesu/ (polite copula) in sentence-final
position, which must be devoiced, and when they are voiced, it gives a strange
impression, 2) in a consecutive devoicing environment, only certain patterns are
actually used out of possible combinations Of devoiced and voiced vowels, and if
the speaker does not use one of the possible patterns, that also gives a strange
impression, and 3) Japanese vowel devoicing is actively (but still unconsciously)
controlled by the speakers because of the “positive effort Of widening of the

glottis for the devoiced vowel segment” (Sawashima 1971, Cited in Maekawa

122

1988). This “strange impression“ gives rise to a social standard because it does
not interfere with communication or semantic meaning of the utterance, but it
reveals such speaker information as region, age, sex, and even emotional state
(Maekawa 1988). Maekawa argues that this is not the case with the vocalization
of lh/ intervocalically, which is a feature that also shows variation. If the speaker
does not vocalize lhl, it does not give a strange impression. Therefore, Maekawa
(1988) argues that vowel devoicing has a social standard, which the vocalization
Of /h/ does not have.

It is not clear, however, that the “strange impression” is related to
“standardness” because it could be only a regional difference. In short, we don’t
know if it has social meaning other than region, although region is not completely
isolated from standardness in Japanese. However, Sakuma (1929)’s comments
that, if someone does not devoice the high vowels in a devoicing environment,
they sound like they are from somewhere in the western part of Japan, gives an
impression that he considers those people in the western part to be nonstandard
speakers or “hillbillies”, people who do not know how to speak “properly.” SO far,
then, vowel devoicing shows linguistic characteristics of a nonstandard feature,
but, according to people’s perception, it is a standard feature.

Thus, speech style and sex/age patterns suggest that vowel devoicing in
Japanese may be a nonstandard feature, and articulatory data seems to support
this: a glottal gesture becomes more economical by devoicing. However,

Japanese speakers’ general perception and prescriptive authority suggest

123

othen/vise: it is a standard Tokyo feature, and the social class pattern (at least for

the vowel lul) seems to support this.

5.3 Summary
The present study has confirmed previous studies regarding the influence
of preceding and following consonants: preceding fricatives promote devoicing

most, then affricates, and stops demote it; following stops promote devoicing

most, then affricates, and fricatives demote it. However, the behavior of [j] and

[p] does not conform to the general picture; they both demote devoicing. It was
suggested that this might be because they do not have full status as phonemes,
and vowel devoicing is demoted next to a weak status phoneme. It was also
found that a low-pitched vowel before a pause is a robust promoting environment
for devoicing, more so for the vowel li/ than lul. The behavior of affricates in
devoicing environments was also examined, and it was argued that the present
study supports the traditional view of affricates, which is that affricates are a
combination Of a 310p and a fricative.

Vowel identity was also found to be a significant factor in vowel devoicing
in Japanese. The vowel /i/ is more likely to devoice than the vowel /u/, an
important finding since the forms ldesu/ and lmasul, which are nearly
categorically devoiced in Tokyo speech, did not provide enough data to
overcome the higher probability for /i/ devoicing.

As for accent, it was found that the pitch associated with the vowel, rather

than the accent itself, is more important. Low-pitched vowels are more likely to

124

devoice than high-pitched vowels. Moreover, it was found that accent has an
influence on vowel devoicing in Japanese, and accented vowels do demote
devoicing, although it is true that accented vowels do undergo devoicing quite
frequently. Furthermore, there may be an age difference in the environment of
pre-pausal low-pitched vowel, where younger speakers uniformly devoice more
than Older speakers. However, age difference in the devoicing of accented
vowels was not supported in the present study.

Morpheme boundary was also found to be a significant factor; a word level
boundary demotes devoicing, particularly, a compound word boundary. However,
a bound morpheme boundary does not demote devoicing, it actually promotes it.

One peculiar finding of this study is that in a consecutive devoicing
environment, devoicing is demoted even though the previous devoiceable vowel
is not actually devoiced. As a result, both vowels tend to be voiced, and this
suggests that speakers have considerable sensitivity to the weight of devoicing
environments.

It was argued that social factors show a mixed pattern regarding
standardness. Speech style, age/sex patterns, and articulatory data suggest that
devoicing may be nonstandard, but the social class pattern (for lul), speakers’
perception, and prescriptive authority (e.g. dictionaries) suggest it is a standard
feature.

In the first chapter, I mentioned that the focus of this study is not to
determine whether vowel devoicing is a phonological or phonetic process, but I

made several phonological references in my discussion above. In the following

125

chapter, I will try to account for the mixed pattern of social factors, by making
reference to the phonology/phonetics distinction made by Tsuchida (1997),
introduced in Chapter 2, although I will take other approaches to this problem as

well and make some final suggestions for further research.

126

CHAPTER 6
CONCLUSION

The present study has investigated vowel devoicing in the Tokyo dialect
quantitatively by collecting a substantial amount of data, looking at various
linguistic and social factors, and using a sophisticated statistical program.
Although its primary purpose was to uncover social meaning, it was also hoped
that an investigation of such a large data set would provide supplementary
information on previously studied linguistics factors. By using this method, I have

in fact reached some findings that are not reported in previous studies and have

added more evidence concerning some controversial issues; [1‘] and [p] behave

differently from other Obstruents; the vowel /i/ promotes devoicing, whereas /u/
demotes it; accent demotes devoicing; a word boundary also demotes, and
devoicing is demoted in a consecutive devoicing environment, regardless Of the
voicing of the previous vowel. Since the amount of data collected is very large, it
is more reliable than most earlier studies, which used only limited data. In
addition, analyses using the sophisticated statistical program, Goldvarb, make
this study even more reliable in determining the significant factors influencing
vowel devoicing in Japanese.

As outlined in the previous chapter, the analysis of social factors faces a
problem; there is a mixed pattern in terms of standardness. Style and age/sex
patterns suggest that devoicing is a nonstandard feature, and social class

patterns and speakers’ perceptions, as well as prescriptive authority, suggest the

127

contrary. I will propose several possible accounts for this mismatch in the

following subsections.

6.1 Phonological vs. phonetic devoicing and standardness

One possible answer to the question “How can a feature behave like a
nonstandard feature, but be considered standard?” may be that native Japanese
speakers consider vowel devoicing in Japanese as a standard, but there may be
a subtle quantitative requirement, one which interacts with the various weighted
environments. One of my respondents told me during the interview that younger
people these days don’t pronounce the end of sentences clearly, and it annoys
him. He doesn’t consider it appropriate. He gave an example of a pastor at his
Church, who is in his 40’s and from Kyushu (the southern island). He speaks fast
and devoices a lot Of vowels at the end of the sentences.

This perception might be related to the phonological and phonetic
devoicing that Tsuchida (1997) suggests. Phonological devoicing may be
considered standard, but phonetic devoicing may be nonstandard, perhaps even
causing ambiguity in speech as a result of consonant cluster homophone
formation and less clarity overall. Thus, young male speakers devoice most
because of the covert prestige attached to the nonstandardness of phonetic
devoicing, and younger females devoice least by adopting only phonological
devoicing because of the overt prestige associated with it, and avoiding phonetic
devoicing. (Younger females are more prestige-oriented because they have not

established their social status yet, as I discuss in more detail below.)

128

If this is the case, maintaining a certain level of devoicing is a highly
monitored albeit unconscious process. In Kroch’s terms, it may be a feature

selected by the prestige group of speakers to mark themselves Off.

6.2 Language change among younger speakers

Another possible explanation for this uncommon patterning of social
factors may be that it reflects gender differences among younger speakers, at
least for the sex and age interaction pattern. Japanese has considerable gender
differences at all levels of language (Shibamoto 1985, Haig 1990, lde and
McGloin 1990, Okamoto 1995, lde 1997, lde and Yoshida 1999) and is
particularly well-known for sentence final particle differences (e.g. McGloin 1990,
Okamoto 1995). However, it has been reported that gender differences are
lessening among younger speakers. Okamoto (1995) shows that younger female
speakers actually use more masculine sentence-final forms than feminine ones.
If this is the case and if younger male and female speech are becoming similar in
terms Of some linguistic features, such as sentence-final particles, it is possible to
hypothesize that younger people may try to express gender differences in
another form. This could be manifested in their use of vowel devoicing; less
devoicing is feminine, and more devoicing is masculine. This differentiation does
not extend to older age groups because they still maintain the traditional gender
differences in other areas. This might explain the large difference between men
and women among Younger speakers, but rather small differences among

Middle Age and Older speakers.

129

On the other hand, the speech style pattern may be accounted for on
purely phonetic grounds. Since vowel devoicing is caused by glottal gestural
overlaps, the faster the speech becomes, the more overlaps occur. However, this
cannot exclusively account for social class patterning, for Lower Middle class
speakers devoice more than other classes.

The social class classification used in this study followed the criteria used
in sociolinguistic studies in the US. (e.g., Labov 1966; Shuy, Wolfram, and Riley
1968; among others) using socioeconomic scores, as outlined in Wamer’s Index
of Social Status (1960). In Japan there are no previous sociolinguistic studies
that use social class as a factor, except for those that used the yamanote
(uptown)/shitamachi (downtown) distinction (e.g. Hibiya 1995). Although it’ seems
reasonable to assume that the same criteria apply to the socioeconomic situation
in Japan as in the US, the treatment is not established in the field. Furthermore, it
may be the case that different factors, which were not used in the present study,
play a significant role in determining socioeconomic status in Japan.

The fact that I did not obtain any clear social status differences in the
present study may mean that the social class criteria I used are not entirely
accurate for measuring social status in Japan. It may also mean that vowel
devoicing in Japanese is not strongly related to social class differences. In any
case, we need more extensive studies of social status in Japan as it relates to

variable linguistic phenomena.

130

6.3 Language change and phonological/phonetic devoicing

By looking at the patterning of social factors more closely, it is possible to
combine the above two possibilities. The gender differences among younger
speakers might suggest language change or an age-grading differentiation. If
phonetic vowel devoicing is considered nonstandard, Older speakers may not use
that feature as much because they tend to embrace conservative prescriptive
values and to use more standard features. Middle Age speakers may be more
lax in this restriction, since they are younger and perhaps slightly less
conservative or do not rely on this linguistic feature to show their standardness;
therefore, they use more of the feature in question, i.e. phonetic devoicing. Since
Younger speakers have not established their status yet, female speakers devoice
less so that they sound more standard, or prestigious, by avoiding phonetic
devoicing, but male speakers opt for the covert prestige of the feature: young
people's (or male) speech. Thus, the mixed pattern Observed in the present study
reflects a combination of language change in the gender differences among
Younger speakers and the standard/nonstandard distinction based on the status
of the devoicing: phonological or phonetic.

The mixed pattern, which seems difficult to account for, may be in part
accounted for by adopting the notion that there is both phonological and phonetic
devoicing, as Tsuchida (1997) argues. Previously, phonetic devoicing may have
been regarded as nonstandard; therefore, males devoice more than females in
general, but the gender differences among Middle Age and Older speakers are

relatively small, because gender is not marked by this feature. However, younger

131

males may have developed this distinction and started to devoice more as a sign
Of covert prestige attached to young people’s speech. Then young females may
have interpreted that as masculinity and started devoicing less than their male
counterparts to indicate femininity and to compensate for the features that have

lost gender differences, such as sentence-final particles.

6.4 Variationlstic quantitative approach

It is also possible to look at this patterning from a purely variationist point
of view. On this view, vowel devoicing in Japanese carries a dual meaning —
standard and nonstandard. Standard vowel devoicing is signaled by the right
amount of devoicing, not too much, not too little. According to Labov, nasality

shows a similar distinction.

Frequently, if you ask somebody what he thinks of this
style of speech (nasalized), he’ll say it’s very “nasal”;
and if you produce a speech of this sort (denasalized),
he’ll say that’s very ”nasal” too. In other words, the
denasalized speech characteristics of some urban
areas and extremely nasalized speech are treated in
the same way. (Labov, discussion of Hoenigswald
1966:23-4)

In order to be considered standard, one has to have the right amount of nasality.
If there is too much nasality, they are called ‘nasal’; if there is too little nasality,
they are also called ‘nasal’, even though the right term may be ‘denasalized’.
Another view Of Labov’s Observation is that excessive nasality or lack of nasality

both result in nonstandardness. Similarly, excessive devoicing or lack Of

132

devoicing results in nonstandardness in the case of Japanese vowel devoicing. If
someone does not devoice enough, they are considered a speaker of a different
dialect (probably from the western part of Japan), and if someone devoices
considerably more, they are considered nonstandard within the Tokyo dialect and
associated with young or masculine speakers because Of the new identity
attached to the feature among youth. This may be why young male speakers
devoice most, young females devoice least, and the rest fall in between, with

males devoicing slightly more than females in the same age group.

6.5 Summary and further study suggestions

I have proposed two possible accounts for the social patterns found in this
study, in which vowel devoicing seems to show both standard and nonstandard
behavior. In both accounts, standardness is related to the right amount of
devoicing and nonstandardness is related to excessive devoicing (in the Tokyo
dialect).

In one account, the right amount Of devoicing is based on the
phonetic/phonological distinction proposed by Tsuchida (1997). Phonological
devoicing is standard and phonetic devoicing is nonstandard. In the other
account, the right amount of devoicing may be somewhere in between the young
males and young females, closer to that of Middle Age and Older speakers. In
any case, the maintenance of the right amount of devoicing may be the standard
feature that is picked up by the prestige group of speakers, according to Kroch

(1976).

133

In order to show that the phonological vs. phonetic devoicing is the reason
for this mixed patterning, we need to investigate what exactly constitutes
phonological and phonetic devoicing. The present study supported some of
Tsuchida’s (1997) claims (e.g. devoicing before allophones of /h/ is
pohonologically prohibited) but not others (e.g. devoicing between two fricatives
is phonologically prohibited). This may be a difficult task because we would have
to develop stringent tests to show the line between phonology and phonetics.

In order to show that it is the amount of devoicing that determines the
standard value of devoicing, we need to know what exactly the right amount of
devoicing is. Is it the amount used by younger females that is considered most
standard, because they are certainly not considered nonstandard, although their
usage shows very low frequency? Is it the amount used by Middle Age speakers,
because they seem to be most sensitive to the standard feature in many
sociolinguistic studies? A carefully planned, and manipulated, perceptional study
of vowel devoicing might be able to tease out some Of these Classifications.

Finally, as mentioned in 6.2, more sensitive and locally applicable study of
social status in Japan is definitely needed. Is the yamanote (uptown)/shitamachi
(downtown) distinction still valid, and is it the only valid distinction? Are there
other factors that play a significant role in determining social status in Japan? Is it
at all possible to determine a uniform scale to cover all age groups because of
the rapidly changing society Of younger speakers? These questions should be
pursued in order to better understand the social meaning Of Japanese vowel

devoicing and the social structure of the Japanese language as a whole.

134

The present study is the first attempt to investigate Japanese vowel
devoicing in a variationist framework, using the common social factors used in
many variationist studies, including socioeconomic status. I hope this work will
help provide a basis for further sociolinguistic studies of vowel devoicing and the

Japanese language in general.

135

APPENDICES

136

APPENDIX A

Table A-1: Word List

rase . C foll. C
ressha n

7
4
r
11
10

kusatsu#e#iku#kisha

CD-b-bfiN-hajagppmfi“

kakashi

natsu
toohoku#chihoo

susuki

nishi

 

i
i
a
u
a
i
u
i
u
a
u
i
a
u
a
u
u
i
u
u
i
u
u
o
a
a
i
u
0
u
i
u
u
i
i
u
i
i

—sAOJ-SCONNOOCONAACDCDm-h-h-CD-h-lSCDN-ba-tAOONCDCDSCO—SJ-‘s‘l—sco

137

Table A-1 (cont’d)

W
m
a
m
m
w
m

Jaru#no#suchuwaadesu

1959870821

1

1191987039

1

0U.Iueoe.la.

sushimeshi

 

138

Table A-1 (cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

No. word/phrase VI pre. C foll. C
26 shoohisha#no#shuchoo i 5 1
u 1 3
27 ishiki#choosa i 1 9
i 9 3
a 2 1 1
28 fusoku u 6 2
o 2 9
u 9 1 1
29 futsuu#no#shufu u 6 4
u 1 6
u 6 1 1
30 pishari#to#shimeru i 10 1
o 8 1
i 1 m
31 seekatsuhi#o#kasu a 9 4
u 4 5
a 9 2
u 2 1 1
32 fukoo#na#futari u 6 9
u 6 8
33 gn#no#supuun u 2 1 0
34 himitsu#shugi i 5 m
u 4 1
u 1 J
35 fuhoo#nyuukoku u 6 7
o 9 9
u 9 1 1
36 gakeppuchi e 9 10
u 1 0 3
i 3 1 1
37 rekishiteki#haikei i 9 1
i 1 8
e 8 9
i 9 7
38 seefu#hihan u 6 5
i 5 7
39 fufuku#o#mooshitateru u 6 6
u 6 9
i 1 8
a 8 8
139

 

l
-. 4 .
r

Table A-1 (cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

No. word/phrase VI pre. C foll. C
40 chishiki i 3 1
i 1 9
i 9 1 1
41 shifuku#no#kookoo i 1 6
u 6 9
u 9 n
42 samui#shichookakushitsu i 1 3
a 9 9
u 9 1
i 1 4
u 4 1 1
43 asa#hayaku#okita a 2 7
i 9 8
a 8 1 1
44 tokee#ga#p_ipi#tto#naru O 8 9
i 1 0 1 0
i 10 8
45 tsukue#no#ue#o#yoku#fuku u 4 9
u 9 6
u 6 9
u 9 1 1
46 shutaiteki u 1 8
e 8 9
i 9 1 1
47 kachitsuzukeru#ureshisa a 9 3
i 3 4
u 4 z
i 1 2
a 2 1 1
48 enpitsu i 1 0 4
u 4 1 1
49 sansuu#no#kuku u 9 9
u 9 1 1
50 chichi#to#haha i 3 3
i 3 8
O 8 7
a 7 7
a 7 1 1
51 gufuttnoltuchippanashi u 6 n
i 3 1 0
i 1 1 1
52 ichi#kara#hajimeru i 3 9

 

140

 

Table A-1 (cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

No. word/phrase VI pre. C foll. C
53 omoikitte#hipparu i 9 8
e 8 5
i 5 1 0
54 haha#no#hi a 7 7
i 5 1 1
55 oshite#kudasai i 1 8
e 8 9
u 9 d
56 kasu#hoo§g§#ii a 9 2
u 2 7
57 kashitsuki a 9 1
i 1 4
u 4 9
i 9 1 1
58 kachi a 9 3
i 3 1 1
59 hifu i 5 6
u 6 1 1
60 fushin u 6 1
61 shutsubotsu#suru u 1 4
u 4 b
u 4 2
u 2 r
62 kojinteki#higai e 8 9
i 9 5
i 5 g_
63 ishiki#fumee#no#kyooaku#han i 1 9
i 9 6
u 6 m
u 9 7
64 hokori#o#suteru o 7 9
u 2 8
65 kyohi#hannoo i 5 7
66 putsuri#to#kireru u 10 4
u 4 r
o 8 9
i 9 r
67 jippi#seekyuu i 10 2

 

141

 

Table A—1 (cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

No. word/phrase VI pre. C fOII. C
68 sakana#ga#ippiki#pichipichi#haneru a 2 9
i 10 9
i 9 10
i 10 3
i 3 10
i 10 3
i 3 7
69 pita#tto#tomaru i 10 8
a 8 8
o 8 8
70 ibento#no#shusaichi#o#utsusu u 1 2
u 4 2
u 2 11
71 hanasu#hito u 2 5
i 5 8
o 8 11
72 paathii#no#shuhin u 1 5
73 shushoo u 1 1
74 akushu u 9 1
u 1 11
75 hihyoo i 5 5
76 matsu#hoo#wa#taihen#da u 4 7
77 sanpi l 10 11
78 hikakuteki#takai#chiteki#suijun i 5 9
a 9 9
u 9 8
e 8 9
i 9 8
a 8 9
i 3 8
e 8 9
i 9 2
79 otaku#ppoi#fuchidori#megane a 8 9
u 9 10
u 6 3
i 3 d
80 mezurashii#shuhoo#no#hinshu#kairyoo u 1 7
u 1 9
81 tsuchi u 4 3
i 3 11
82 kaihatsu#fukanoo a 7 4
u 4 6
u 6 9

 

 

 

 

 

142

 

Table A-1 (cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

No. word/phrase Vl pre. C foll. C
83 shippai#o#kakusu#tsumori i 1 1 0
a 9 9
u 9 2
u 2 4
u 4 m
84 hinshitsu#kanri i 1 4
u 4 9
85 tsutaeru u 4 8
86 OOkikute#udeppushifiga#tsuyoi i 9 9
u 9 8
u 10 1
I 1 J!—
87 pukupuku#shita#te u 1 0 9
u 9 10
u 10 9
u 9 1
i 1 8
a 8 8
e 8 11
88 pikapika#no#kutsu i 10 9
a 9 10
i 10 9
u 9 4
u 4 11
89 chiisana#kutsushita u 9 4
u 4 1
i 1 8
a 8 11
90 kokoro o 9 9

 

 

 

 

 

 

143

APPENDIX B

Reading Passage (Japanese)

as: act/:6 3531113 is was amt utut c
sears/viz. EB . Ill EHEEid’. Bxbflefinit’B—otllewlzt

L: as b‘L‘L’tb l.\ O<i 57': 63:51. f:<
655'6‘4'. 33. £11: lzfi<t Ill. Oxaauaaae. .tEl 111/sea

SEE 171-15: LXS/v Lolafln 75\<
Ifl.‘ 1m LJ (1)/311357567“: f '03“. £§¥$Mi 851 0) kW ’8 ll! 3'05

aaaac;5c Lamb use
waist/$1971.61. Bar: :3 0) iii 16 B61 wtttlzahéwbimt’acd.

we :63. 5 accents. in so)
Berti/wwfleolafevcd. not. 4611 15 91:31AM aces.

t—ra— LLJ:<U/v rim-9‘ 1: so) ‘4'wa ‘lb
x—zt—a) 131.11. 751 7: flats. tie/3:5 1b 11%—lift eta/ester.

at): 1:815) Bffi 11W) I595 [r1913
fig¥$hbi¥fiwi 6 Té’fiél13Hi‘C‘Ll‘m 31115550215le

email/:4. Luz—BUM» Loser} the? let.
act-c7. 131111116111 t 51413 a) $15 Id: 9w] fitted-cuaaae.

imam—53 J: 3:51).“ 105‘s; Liars—3 U/ousmwixi
semi Izlia< 137:1 use. :9 Lu $52 0) 511155 lat

091125 3515 303531941115 IA $1.1"? L‘olfié’ USU?»
11?}; flint £5. oth‘i‘d’. 8155 fig Izt’rosz. fl 7'31 -E 65:157-

1: sitar/u lpébfit: .S~f< rat. 1»
Hia‘CIAZ-ELE. £53 Altai/“714751 TE 1.11366: 53 ofzw't‘. IOTEE

1M soc-a a 16.: ecu six—a 1615
1.7:. lb urmxnnwmx E 7:101:51; L1... Iii OWN—Ha: 7:

5‘6 5'1. 9.1: ammo/o as: out“
50c m 5 ill 19193.1 {—579 an manned". i§¥elula 91- r:

L‘ 91115751< LHL/11357.6(
fictitfiEUifiho E? LkbokwaN. fli¥$ Mtft

cu sis/om le‘3LJ:< use trauma 6
fiflewc$$fiebte are LiLr:.I~‘/‘7’I~d>;<74w7x=&fied>

144

:at w) 9w» thb—I‘t ewes
b<¥9thaflea>§ ‘G'd'o JALUDJUI-JU—‘T‘ZCD as: muémra 8

‘\‘3T 3'35 ‘C XLISit (A 35%
5!:75‘L‘C i L‘ {915} E¥IZA1LT 1‘27bl2fiéfzht E o‘CL‘i'fi'o

55¢ :15 :tL a5e<sua5 95:5 Jrcu
5715115 (tuifiho‘éh éMf-Iat mum:- E 71%? 5'6 5931‘5'0

its: EAL‘fl'A’ t5ti<sti5 we
:97 HDioiwa $331115 mwmna Kittie)? wxxfir—JBMLVEL

$50

145

APPENDIX C

Table A-2: Neighborhood Scores

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Neighborhood Score
1 Adachi-ku 1
2 Arakawa-ku 1
3 Edogawa-ku 2
4 Taitoo-ku 2
5 lkebukuro (T oshima-ku) 2
6 Sugamo (Toshima-ku) 2
7 Kootoo-ku 2
8 Takasago (Katsushika—ku) 2
9 ltabashi—ku 2
10 Takadanobaba (Shinjuku-ku) 2
1 1 Nishi-Tookyoo-shi 3
12 Komagome (Toshima-ku) 3
13 Chihaya (Toshima-ku) 3
14 Shibamata (Katsushika-ku) 3
1 5 Kawasa ki-shi (Kanagawa) 3
16 Shinbashi (Minato-ku) 3
17 Nerima-ku 3
18 Asakusa (Chuuoo—ku) 4
19 Minato-ku 4
20 Nakano—ku 4
21 Ishikawadai (Oota-ku) 4
22 Kitami (Setagaya-ku) 4
23 Bunkyoo-ku 4
24 Suginami-ku 4
25 Chiyoda-ku 4
26 Mitaka-shi 4
27 Yokohama-shi (Kanagawa) 4
28 Shibuya-ku 5
29 Mejiro (Toshima-ku) 5
30 Meguro (Oota-ku) 5

 

146

 

Table A-3: Socioeconomic Scores

APPENDIX D

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ALeGroup Sex Age Occupation x2 Education x2 Neighborhood Total
Y Male 22 1 2 1 2 1 5
M Female 61 1 2 1 2 2 6
Y Male 24 1 2 1 2 2 6
Y Male 28 1 2 1 2 3 7
Y Male 27 1.5 3 1 2 2 7
M Male 60 2 4 1 2 2 8
0 Male 62 1 2 1 2 4 8
Y Male 28 1.5 3 1 2 3 8
Y Male 25 1 2 2 4 2 8
M Female 35 2 4 2 4 1 9
0 Female 87 2 4 1 2 3 9
0 Female 76 2 4 1 2 3 9
0 Male 65 2 4 1 2 3 9
Y Male 24 1 2 1 2 5 9
Y Male 26 1.5 3 2 4 2 9
M Female 60 3 6 1 2 2 10
M Male 54 2 4 2 4 2 10
0 Male 60 2 4 1 2 4 10
0 Male 78 3 6 1 2 2 10
Y Female 27 2 4 2 4 2 10
0 Female 70 3 6 1 2 3 11
Y Female 29 3 6 1 2 3 11
M Male 36 3 6 1 2 3 11
M Female 35 3 6 2 4 4 14
Y Male 20 3 6 2 4 4 14
Y Male 19 2 4 4 8 2 14
M Male 33 1.5 3 4 8 4 15
Y Female 25 3 6 3 6 3 15
Y Male 26 2 4 4 8 3 15
M Male 61 3 6 4 8 2 16
0 Female 84 3 6 3 6 4 16

 

147

 

Table A-3 (cont'd)

 

 

 

 

 

 

 

 

 

 

667777777888999900000111333
111111111111111122222222222
245333533224353524444335555
8868886888888888W888W40u88888
443444344444444454445544444
6466666668868686888868W8WNw
323333333443434344443454555
013M01929829054837945505265
224 46761561312233122381411
b bun-o Ebbb bbbb ugly Mb
mmmmmmmmmmmmmmmmmmmmmmmmmmm
eMeeMMeeeeMMMeeeeMeMMMMMeeM
F FF FFFF FFFF F FF
YYMMMMOOYMOYMYYYMMYYYMOYMYY

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

148

APPENDIX E

Sample transcription of an interview conversation (excluding interviewer's part)

Counter #

23—25
29-33

34-35
36-37
38-39
39-40
42-43
44-49

49-51
51 -52
52-54
54-55
57-62

64-65
67-83

83-84
84-85
85-87

87-88

Tokens

ima de ieba, kankoku no, .. souru desu,

eeto issai .. n:: ni-juu-..nen desu yo ne: shuusen, shuusen de:
modotte kimashita no de: .. maa, i .. ni-sai desu ne, ni-sai
chikaku desu ne, hai

hai soo desg.

soo da to omoimasg, hai.

(so)datta basho wa, tookyoo desg

e:: watakgshi wa suginamiz-ku desg

tookyoo no itabashi-ku desg.

itabashi-kg to itte mo honto ni mina, minami-choo juusho ga
minami—choo nandesu gaz, honto ni .. itabashj-ku no hazure ..
moo sugu .. dooro ip-pon de toshima-ku .. to yuu tokoro nanodez,
n:: itabashi-kg to ittemo .. n: ikebukuro, kara sugu desg shi,
moyori no eki wa: kanamechoozz desu ne: ..... soo desu ne:
ichi-ban chikai no wa kanamechoo

aruite .. juu .. juu-go-hun .. kakaranai .. gurai kanazz

chjkai desg, dakara itsumo: jitensha

ikebukuro-eki made demo ni-jup-pun tarazu desg kara: .. ne::
dakarazz, hai, benri-na toko ni, itabashi-kg tte yuuto minna
itabashi no hazure no hoo to yuu ka, un: saitama ni chikai hoo 0
500200 sareru kamo shirenai ndesu ga: honto ni moo toshima-
ku .. sugu tonari ga toshima-kg to yuu

go-juu-has-sai dezsg.

soo desu ne:, shgfu, de, ez, desg keredomoz, ima ano ..
boranthia, katsudoo to iimasg kaz, sore no, ano guruupu ni haitte
mashjtez, de, ano, jiritsg-shien saabisg to iu koto dez, e:: .. roojin-
kaigo tokaz, sorekara .. e:: .. okaasan, ikuji: no, okaasan no: soyu,
sapooto to yuu yoona, shigoto shitemashite enupiioo hoojin ni,
narimashjta nodez, ano jibun—tachi de, honto ni, shigoto ..... ano
jimu-teki na (koto kara nani kara) zenbu jibun-tachi: de:
yattemasu node, (hh) ano, boranthia no honto no, soo yuu
shigoto to, sorekara jimu-tekl na shigoto tte yuu kana? hoojin to
shite no shigoto o, ryoohoo yattemasu node kekkoo isogashiku,
hanbun, shigoto mitai ni nattemasg.

soo desu ne:, un

e::to mada ni-nen tarazu desg kedo:

chotto maa, kodomo-tachi mo .. nee fgtari kekko(hh)n
shimashita shi:

eriko to mariko a2: masahiro ga kokkon shjta nde:

149

REFERENCES

Beckman, Mary and Atsuko Shoji. 1984. Spectral and Perceptual Evidence for
CV Coarticulation in Devoiced lsi/ and lsyul in Japanese. Phonetica 41. 61-71.

Chambers, J. K. and Peter Trudgill. 1980. Dialectology. Cambridge, UK:
Cambridge University Press.

Cho, Young-mee Yu. 1993. The Phonology and Phonetics of ‘Voiceless’ Vowels.
BLS 19. 64-75.

Fujimoto, Masako, Emi Murano, Seiji Niimi and Shigeru Kiritani. 2002.
Differences in Glottal Opening Pattern between Tokyo and Osaka Dialect
Speakers: Factors Contributing to Vowel Devoicing. Folia Phoniatn’ca et
Logopaedica 54.133-143.

Fujimura, Yasushi. 1967. Nihongo-no onsei — Gengo Keishiki-no Onkei Kijutsu-
kara Onpa made-no Dootei. NHK Soogoo Hoosoo Bunka Kenkyuujo.

Haig, John H. 1990. A phonological Differences in Male-Female Speech among
Teenagers in Nagoya. In Aspects of Japanese women '3 language, ed. by
Sachiko lde and Naomi Hanaoka McGloin. 5-22. Tokyo: Kuroshio.

Han, Mieko Shimizu. 1962. Unvoicing of Vowels in Japanese, Onsei no Kenkyu
10, Nihon Onsei Gakkai.

Haraguchi, Shosuke. 1977. The Tone Pattem of Japanese: An Autosegmental
Theory of Tonology. Tokyo: Kaitakusha.

Hibiya, Junko. 1995. The velar nasal in Tokyo Japanese: A case of diffusion from
above. Language Variation and Change 7. 139-152.

Hoenigswald, Henry. 1966. “A proposal for the study of folk-linguistics,” in:
William Bright (ed.), Sociolinguistics. The Hague: Mouton, 16-26.

lde, Sachiko. (ed.) 1997. Joseigo no Sekai. Tokyo: Meiji Shoin.

lde, Sachiko and Naomi Hanaoka McGloin. (eds.) 1990. Aspects of Japanese
women's language. Tokyo: Kuroshio.

lde, Sachiko and Megumi Yoshida. 1999. Sociolinguistics: Honorifics and Gender

Differences. The Handbook of Japanese Linguistics, ed. by Natsu ko
Tsujimura, 444-480. Malden, Mass: Blackwell Publishers.

150

lmai, Terumi. 1997. Vowel Devoicing in Japanese. MA thesis. Michigan State
University, East Lansing, MI.

Ito, Junko, Armin Mester, and Jaye Padgett. 1995. Licensing and
Underspecification in Optimality Theory. Linguistic Inquiry 26. 571 -61 3.

Jun, Sun-Ah and Mary E. Beckman. 1993. A Gestural-Overlap Analysis of Vowel
Devoicing in Japanese and Korean. Paper presented at the Annual Meeting
of the Linguistic Society of America, Los Angeles.

Kawakami, Shin. 1977. Nihongo Onsei Gaisetsu. Tokyo: Oofuusha.

Kindaichi, Haruhiko. 1954. On’in [Phonology]. Nihon Hoogengaku [Japanese
Dialectology], ed. by Misao Tojo. Tokyo: Yoshikawa Koobunkan.

Kondo, Mariko. 1993. The Effect of Blocking Factors and Constraints on
Consecutive Vowel Devoicing in Standard Japanese. Poster presented at
LabPhone 4, Oxford, 11-14 August, 1993.

Kondo, Mariko. 1994. Mechanisms of Vowel Devoicing in Japanese. Proceeding
of the 1994 lntemational Conference on Spoken Language Processing, vol.2.
479-482.

Kondo, Mariko. 1995. Temporal Adjustment of Devoiced Morae in Japanese.
Proceedings of the 13‘” lntemational Congress of Phonetic Sciences, vol.3.
61 -64.

Kondo, Mariko. 1997. Mechanisms of Vowel Devoicing in Japanese. A doctoral
dissertation, University of Edinburgh.

Kosaka, Kenji. 1994. Aspects of Social Inequality and Difference. In Social
Stratification in Contemporary Japan, ed. by Kenji Kosaka. 34-53.

Kosaka, Kenji. (ed.) 1994. Social Stratification in Contemporary Japan. London:
Kegan Paul lntemational.

Kroch, Anthony. 1978. Toward a Theory of Social Dialect Variation. Language in
Society 7. 17-36.

Labov, William. 1966. The Social Stratification of English in New York City.
Washington DC: Center for Applied Linguistics.

Labov, William. 1972. Sociolinguistic Patterns. Philadelphia: University of
Pennsylvania Press.

151

Lombardi, Linda. 1990. The nonlinear organization of the affricate. Natural
Language and Linguistic Theory 8: 375-425.

Lombardi, Linda. 1991. Laryngeal features and laryngeal neutralization. Doctoral
dissertation, University of Massachusetts, Amherst.

Maekawa, Kikuo. 1983. Kyotsugo ni okeru Boin no Museika no Kakuritsu ni
tsuite. [Probabilities for the occurrence of vowel devoicing in Common
Japanese]. Gengo no Sekai 1-2. 69-81.

Maekawa, Kikuo. 1988. Boin no Museika. [Vowel Devoicing] Koza Nihongo to
Nihongo Kyoiku, ed. by Miyoko Sugito, Vol.2. 135-153. Tokyo: Meiji Shoin.

Martin, Samuel. 1952. Morphophonemics of standard colloquial Japanese.
Supplement to Language: Journal of the Linguistic Society of America, v. 28,
no. 8, pt. 2. Language dissertations, no. 47. Baltimore : Linguistic Society of
America

Mase, Yoshio. 1977. Toozai hoogen no tairitsu [The opposition of the eastern
and western dialects]. In lwamani Kooza Nihongo 11: Hoogen [Dialect].
Tokyo: lwanami Shoten.

McCawley, James. 1968. The Phonological Component of a Grammar of
Japanese. The Netherlands: Mouton

McGloin, Naomi Hanaoka. 1990. Sex Differences and Sentence-Final Particles.
In Aspects of Japanese women '8 language, ed. by Sachiko lde and Naomi
Hanaoka McGloin. 23-41. Tokyo: Kuroshio.

Mineta, Akemi. 1988. Aichi-ken Minami-shitaragun Sakute-mura Hougen-no Boin
no Museika-ni tuite [On vowel devoicing in the dialect of Sakute village,
Minami-shitara county in Aichi prefecture]. Proceedings of the Phonetic
Society of Japan Annual Meeting.

Morris, Midori Yonezawa. 2003. Perception of devoicing variation and the
judgment of speakers' region in Japanese. A doctoral dissertation, Michigan
State University, East Lansing, MI.

Nagano—Madsen, Y. 1994. Vowel Devoicing Rates in Japanese from a Sentence
Corpus. Lund University, Department of Linguistics. Working Papers 42. 117-
127.

Nagano-Madsen, Y. 1995. Effect of Accent and Segmental Contexts on the

Realization of Vowel Devoicing in Japanese. Proceesings of the 13‘”
lntemational Congress of Phonetic Sciences, Vol.3. 564-567.

152

Nihon Housou Kyoukai. (ed.) 1985. Nihongo Hatsuon Akusento Jiten [Japanese
Pronunciation and Accent Dictionary]. Tokyo: Nihon Hoosoo Shuppan
Kyookai.

Okamoto, Shigekc. 1995. “T asteless” Japanese: Less “Feminine” Speech Among
Young Japanese Women. Gender Articulated : Language and the Socially
Constructed Self, ed. by Kira Hall and Mary Bucholtz. 297-325. New York:
Routledge

Rubach, Jerzy. 1994. Affricates as Strident Stops in Polish. Linguistic Inquiry 25.
119-143.

Sakuma, Kanae. 1929. Nihon Onseigaku. [Japanese Phonology]
Tokyozkyoubunsha.

Sakurai, S. 1985. Kyoutsuugo-no Hatsuon-do Chuui Subeki Kotogara [On
Pronunciation of Standard Japanese]. Nihongo Hatsuon Akusento Jiten
[Japanese Pronunciation and Accent Dictionary], ed. by Nihon Hoosoo
Kyookai. 128-134. Tokyo: Nihon Hoosoo Shuppan Kyookai

Sawashima, Masayuki. 1971. Devoicing of vowels. Annual Bulletins 5, Research
Institute of Logopedics and Phoniatrics, University of Tokyo.

Sawashima, Masayuki and S. Niimi. 1974. Laryngeal Conditions in Articulations
of Japanese Voiceless Consonants. Annual Bulletin of the Research Institute
of Logopedics and Phoniatrics 8. 13-17.

Seiyama, Kazuo. 1994. lntergenerational Occupational Mobility. In Social
Stratification in Contemporary Japan, ed. by Kenji Kosaka. 54-77.

Shibamoto, Janet S. 1985. Japanese Women's Language. Orlando: Academic
Press.

Shibatani, Masayoshi. 1990. The Languages of Japan. New York, NY:
Cambridge University Press.

Shuy, Roger, Walt Wolfram, and William Riley. 1968. Field Techniques in an
Urban Language Study. Washington, DC: Center for Applied Lingusitics.

Steriade, Donca. 1993. Closure, Release, and Nasal Contours. Nasality, ed. by
M. Huffman and R. Krakow. San Diego: Academic Press.

Sugito, Miyoko. 1982. Nihongo Akusento-no Kenkyuu [Study on Japanese
Accents]. Tokyo: Sanseidou.

153

Sugito, Miyoko. 1988. Nihon no 8 Toshi ni okeru Boin no Museika. [Vowel
Devoicing in Eight Cities in Japan] Osaka Shoin Women ’s College Collected
Essays 25. 1-10.

Sugito, Miyoko and Hajime Hirose. 1988. Production and Perception of Accented
Devoiced Vowels in Japanese. Annual Bulletin of the Research Institute of
Logopedics and Phoniatrics 22. 21 -39.

Tahara, Hiroshi, et al. 1998. Jinmonkon Database Vol. 2 (CD-ROM). Kanagawa,
Japan. Sougou Kenkyuu Daigakuin Daigaku.

Takeda, K. and H. Kuwabara. 1987. Analysis and Prediction of vowel devoicing
(in Japanese). Proceedings of Acoustic Society of Japan. 105-106.

Trudgill, Peter. 1972. Sex, Covert Prestige and Linguistic Change in the Urban
British English of Norwich. Language in Society 1, 2. 179-195.

Trudgill, Peter. 1974. The Social Differentiation of English in Norwich. Cambridge,
UK: Cambridge University Press.

Trudgill, Peter.1983. On Dialect: Social and Geographic Factors. Oxford: Basil
Blackwell.

Tsuchida, Ayako. 1997. Phonetics and Phonology of Japanese Vowel Devoicing.
A doctoral dissertation, Cornell University, Ithaca, NY.

Tsujimura, Natsu ko. 1996. An Introduction to Japanese Linguistics. Oxford:
Blackwell.

Vance, Timothy. 1987. An Introduction to Japanese Phonology. Albany, NY:
State University of New York Press.

Vance, Timothy. 1992. Lexical Phonology and Japanese Vowel Devoicing. The
Joy of Grammar, ed. by Diane Brentari, Gary N. Larson, and Lynn A.
MacLeod. 337-350. Philadelphia: John Benjamins Publishing Company.

Varden, Kevin J. 1999. On High Vowel Devoicing in Standard Modern Japanese:
Implications for Cunent Phonological Theory. A doctoral dissertation,
University of Washington.

Warner, W. Lloyd. 1960. Social class in America. New York: Harper.

Wolfram, Walt. 1969. A Sociolinguistic Description of Detroit Negro Speech.
Washington, DC: Center for Applied Linguistics.

154

Wolfram, Walt and Ralph W. Fasold 1974. The Study of Social Dialects in
American English. Englewood Cliffs, NJ: Prentice-Hall.

Wolfram, Walt and Natalie Schilling-Estes. 1998. American English. Oxford:
Blackwell.

Yoshida, N. and Y. Sagisaka. 1990. Factor Analysis of Vowel Devoicing in
Japanese (in Japanese). ATR Technical Report TR-l-0159, ATR Interpreting
Telephony Research Laboratories, Kyoto.

Yoshioka, Hirohide. 1981. Laryngeal Adjustments in the Production of the
Fricative Consonants and Devoiced Vowels in Japanese. Phonetica 38. 236-
251.

Yoshioka, Hirohide, Anders Lquvist, and Hajime Hirose. 1982. Laryngeal
Adjustments in Japanese Voiceless Sound Production. Journal of Phonetics
10. 1-10.

Yuen, Chris L. 1997. Vowel Devoicing and Gender in Japanese. MA thesis.
University of California, San Diego, CA.

155

.qr-r

 

IES

Ill

HSITY LIBRAR

MICHIGAN STAIE UNIVE

III

Ill“),

Illlllli

Ill

3

I

7
3
0
4
4
0
5
2
o
3
9
2

4|