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

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DatoDuopss-sz

 

PERCEPTION OF DEVOICING VARIATION AND

THE JUDGMENT OF SPEAKERS' REGION IN JAPANESE

By

Midori Yonezawa Morris

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Linguistics and
Germanic, Slavic, Asian, and African Languages

2003

ABSTRACT

PERCEPTION OF DEVOICING VARIATION AND
THE JUDGMENT OF SPEAKERS' REGION IN JAPANESE

BY

Midori Yonezawa Morris

Vowel devoicing occurs in the Tokyo dialect, but many people are not
aware of it because it is allophonic. It is generally believed that it does not
occur in the Kinki (Kyoto-Osaka area) dialect, although it could occur as
frequently as in Tokyo in the most general devoicing environments. It has
been reported that respondents can identify speakers' regions upon hearing
others' pronunciation but that such judgments are sometimes affected by
social information or stereotypes about pronunciation itself (Preston 1996,
and others). There are some perceptual studies of Japanese using phonemic
features, but few using allophones. I conducted an experiment to examine
how vowel devoicing affects judgments Tokyo and Kinki respondents make
on whether or not others were from the same regions as their own after
hearing each word with devoicing variation in a recorded word list on a tape.

Responses were analyzed to examine how phonological and social
factors affected the judgment. The overall results show that both Tokyo and
Kinki respondents tended to determine speakers' regions based on the

general belief (devoicing in Tokyo and nondevoicing in Kinki), but the Kinki

results are much less distinct, supporting the assumption of devoicing in
Kinki.

Generally, [-continuant] of the consonant that precedes devoiceable
vowels contributes to judgements based on the general belief, while 'palatal
[+continuantl' tends to promote the judgment 'from Tokyo / non-Kinki' with
both devoiced and nondevoiced tokens. This perception phenomenon
coincides with the optimal environment for vowel devoicing in production
— a palatal fricative and /i/ (Imai 1997). The respondents apparently 'know‘
the optimal environment and make judgments, assuming that the vowel is
devoiced because of its environment. These results also suggest that Kinki
respondents also have the same knowledge, which is consistent with the
assumption of devoicing in Kinki. Of the following consonant, [+continuant]
tends to promote the judgment 'from Tokyo/non-Kinki' regardless of
devoicing status. These results suggest that the respondents assume that the
vowel is devoiced when they cannot retrieve the devoicing status after
hearing devoicing environments and hear [+continuant], as well as the
optimal environment for vowel devoicing in production. These results
nicely illustrate making use of linguistic information along the time line,
using knowledge about one's own phonological system in production, and
'hearing' what is not in the actual information, thus reflecting results in

other studies as in Beddor, et a1. (2002), for example.

Copyright by
Midori Yonezawa Morris
2003

ACKNOWLEDGMENTS

I am deeply indebted to Professor Dennis R. Preston, the chairperson of
my advisory committee, and Professors David Dwyer, Mutsuko Endo
Hudson, and Yen-Hwei Lin for their insightful comments and suggestions.
My "outside" reader, Professor Anne Violin-Wigent took particular care in
making detailed comments on the manuscript. Special thanks go to Dennis
Preston for his patience and constant encouragement of and confidence in
me.

I also would like to express my appreciation to Dr. Yukihiko Nakai, for
his advice on my experiment and suggestions for further information on
studies related to my project.

I also want to thank all the speakers and respondents in my experiment
and all those who introduced me to other participants for their interest and
encouragement. It seems to me that I kept writing so that I could answer
their questions.

I recognize my parents and friends in my hometown as well for my
Kinki dialect, which first motivated me to start this study. Finally, I would
deeply like to thank my husband, Dick, who understands the Kinki dialect
and its culture, and gave generous support during the final stages of this

dissertation.

TABLE OF CONTNETS

List of tables ................................................................................................................... x
1. Introduction ............................................................................................................. 1
2. Previous studies ...................................................................................................... 6
2.1. Some differences between the Tokyo and Kinki dialects ................... 6
2.2. Vowel devoicing in the Tokyo dialect ................................................... 8
2.3. Vowel devoicing in the Kinki dialect .................................................. 15
2.4. Perception of dialects and attitudes toward them .............................. 22
3. Methods .................................................................................................................. 33
3.1. Preparing a test tape .................................................................................. 33
3.2. Response collection .................................................................................. 45
3.3. Preparation for analysis ........................................................................... 46
3.4. Analysis ....................................................................................................... 50
4. Results ..................................................................................................................... 53
4.1. Overall results by token type .................................................................. 53
4.1.1. Overall results by the five token types ..................................... 53

4.1.2. Overall results by devoicing variation and regional accents
...................................................................................................... 62
4.2. Insignificant factors for overall results ................................................ 63
4.2.1. Insignificant factor groups .......................................................... 63

4.2.2. Devoicing and nondevoicing at the end of the stimulus 65

vi

4.2.3. Words of different length and accent patterns ....................... 66
4.3. Overall results by phonological and social factors ............................. 67
4.3.1. Position of devoiceable vowels in terms of mora count ...... 67
4.3.2. Position of devoiceable vowels in terms of boundary types
...................................................................................................... 68
4.3.3. Pitch accent pattern ....................................................................... 69
4.3.4. Preceding consonant .................................................................... 70
4.3.5. Following consonant ................................................................... 73
4.3.6. Identity of devoiceable vowels ................................................... 75
4.3.7. Voicing of devoiceable vowels .................................................. 75
4.3.8. Gender of the speaker .................................................................. 76
4.3.9. Gender of the respondent ........................................................... 77
4.3.10. Age of the respondent ................................................................ 77
4.4. Results for Tokyo devoiced data and Tokyo nondevoiced data ..... 79
4.4.1. Position of devoiceable vowels in terms of mora count ....... 80
4.4.2. Position of devoiceable vowels in terms of boundary types
...................................................................................................... 81
4.4.3. Pitch accent pattern ....................................................................... 82
4.4.4. Preceding consonant .................................................................... 82
4.4.5. Following consonant ................................................................... 83
4.4.6. Identity of devoiceable vowels ................................................... 86
4.4.7. Gender of the speaker .................................................................. 87
4.4.8. Gender of the respondent ........................................................... 88

vii

4.4.9. Age of the respondent .................................................................. 88

4.5. Results for Kinki devoiced data and Kinki nondevoiced data ....... 90
4.5.1. Position of devoiceable vowels in terms of mora count ...... 90

4.5.2. Position of devoiceable vowels in terms of boundary types

...................................................................................................... 91
4.5.3. Pitch accent pattern ....................................................................... 91
4.5.4. Preceding consonant .................................................................... 91
4.5.5. Following consonant ................................................................... 94
4.5.6. Identity of devoiceable vowels ................................................... 96
4.5.7. Gender of the speaker .................................................................. 96
4.5.8. Gender of the respondent ........................................................... 96
4.5.9. Age of the respondent .................................................................. 97
4.6. Summary of the results ............................................................................ 98
5. Discussion ............................................................................................................. 104
5.1. Comparison of saliency among token types ...................................... 105
5.2. Significant and insignificant factors in overall results .................... 113
5.3. Phonological factors ................................................................................ 114
5.3.1. Voicing of devoiceable vowels ................................................. 114
5.3.2. Preceding consonant ................................................................... 115
5.3.3. Following consonant .................................................................. 117
5.3.4. Position of devoiceable vowels in terms of boundary types
.................................................................................................... 121

viii

5.3.5. Pitch accent pattern and position of devoiceable vowels in

terms of mora count .................................................................. 122

5.3.6. Identity of devoiceable vowels .................................................. 124

5.4. Social factors .............................................................................................. 125
5.4.1. Gender of the speaker and gender of the respondent .......... 125

5.4.2. Age of the respondent ................................................................ 126

5.5 Speaker's identity ...................................................................................... 127

6. Conclusion ........................................................................................................... 133
Appendix I ................................................................................................................. 140
Appendix II ................................................................................................................ 141
References .................................................................................................................. 144

ix

Table 1

Table 2

Table 3

Table 4

Table 5

Table 6

Table 7

Table 8

Table 9

Table 10

Table 11

Table 12

Table 13

Table 14

Table 15

Table 16

Table 17

Table 18

Table 19

Table 20

LIST OF TABLES

Devoicing by people from Tokyo and Osaka in Sugito (1988) ......... 18
Variations of devoicing .......................................................................... 19
Devoicing variation in different positions (Tokyo subjects) ........... 20
Example words containing environment of devoiceable / i / ......... 37
Example words containing environment of devoiceable / u / ........ 38
List of the words containing no devoiceable vowel .......................... 41
Coding for each word (examples) .......................................................... 44
Tabulation .................................................................................................. 54
Overall results by token types (Tokyo) ................................................. 57
Overall results by token types (Kinki) .................................................. 57
Overall two-way results by token types (Tokyo) ................................ 61
Overall two-way results by token types (Kinki) ................................. 61
Results by regional accents and devoicing variation (Tokyo) ......... 62
Results by regional accents and devoicing variation (Kinki) .......... 63
Final results by mora count (Kinki) ...................................................... 68
First results by boundary type (Tokyo) ................................................. 68
First results by boundary type (Kinki) .................................................. 68
Final results by boundary type (Tokyo) ................................................ 69
Final results by accent pattern (Kinki) .................................................. 70
First results by preceding consonant (Tokyo) ..................................... 70

Table 21

Table 22

Table 23

Table 24

Table 25

Table 26

Table 27

Table 28

Table 29

Table 30

Table 31

Table 32

Table 33

Table 34

Table 35

Table 36

Table 37

Table 38

Table 39

Table 40

Table 41

Table 42

Final results by preceding consonant (Tokyo) .................................... 72

First results by preceding consonant (Kinki) ...................................... 72
Final results by preceding consonant (Kinki) ..................................... 73
Final results by following consonant (Tokyo) .................................... 74
Final results by following consonant (Kinki) ..................................... 74
Final results by voicing of the vowel (Tokyo) .................................... 76
Final results by voicing of the vowel (Kinki) ..................................... 76
Final results by gender of the speaker (Kinki) .................................... 77
Final results by gender of the respondent (Tokyo) ............................ 77
First results by age of the respondent (Tokyo) .................................... 78
First results by age of the respondent (Kinki) ..................................... 78
Final results by age of the respondent (Tokyo) ................................... 79
Final results by age of the respondent (Kinki) .................................... 79
Final results by mora count (Tokyo, devoiced) .................................. 80
Final results by mora count (Tokyo, nondevoiced) .......................... 80
Final results by boundary type (Tokyo, devoiced) ............................. 81
Final results by boundary type (Tokyo, nondevoiced) ...................... 81
Final results by accent patterns (Tokyo, nondevoiced) ..................... 82
Final results by preceding consonant (Tokyo, devoiced) ................. 83
Final results by preceding consonant (Tokyo, nondevoiced) .......... 83
First results by following consonant (Tokyo, devoiced) ................... 84
First results by following consonant (Tokyo, nondevoiced) ........... 85

xi

Table 43

Table 44

Table 45

Table 46

Table 47

Table 48

Table 49

Table 50

Table 51

Table 52

Table 53

Table 54

Table 55

Table 56

Table 57

Table 58

Table 59

Table 60

Table 61

Table 62

Table 63

Table 64

Final results by following consonant (Tokyo, devoiced) ................. 86

Final results by following consonant (Tokyo, nondevoiced) .......... 86
Final results by vowel identity (Tokyo, devoiced) ............................. 87
Final results by gender of the speaker (Tokyo, devoiced) ................ 87
Final results by gender of the speaker (Tokyo, nondevoiced) ......... 87

Final results by gender of the respondent (Tokyo, nondevoiced) .. 88

First results by age of the respondent (Tokyo, devoiced) ................. 88
First results by age of the respondent (Tokyo, nondevoiced) .......... 89
Final results by age of the respondent (Tokyo, devoiced) ................ 89
Final results by age of the respondent (Tokyo, nondevoiced) ........ 90
Final results by boundary type (Kinki, devoiced) .............................. 91
First results by preceding consonant (Kinki, devoiced) .................... 92
Final results by preceding consonant (Kinki, devoiced) .................. 92
First results by preceding consonant (Kinki, nondevoiced) ............ 93
Final results by preceding consonant (Kinki, nondevoiced) ........... 93
First results by following consonant (Kinki, devoiced) .................... 94
First results by following consonant (Kinki, nondevoiced) ............ 95
Final results by following consonant (Kinki, nondevoiced) ........... 95
Final results by gender of the respondent (Kinki, devoiced) .......... 96

Final results by gender of the respondent (Kinki, nondevoiced) 96
First results by age of the respondent (Kinki, devoiced) .................. 97

First results by age of the respondent (Kinki, nondevoiced) ........... 97

xii

Table 65 Final results by age of the respondent (Kinki, devoiced) ................. 98

Table 66 Final results by age of the respondent (Kinki, nondevoiced) ......... 98
Table 67 Results by speaker (Tokyo, all types of tokens) ................................. 128
Table 68 Results by speaker (Kinki, all types of tokens) .................................. 129

xiii

1. Introduction

Vowel devoicing in the Tokyo dialect is a common topic in Japanese
phonology. The most general description of vowel devoicing is the one
found in, e.g., Vance (1987), Tsujimura (1996), and Kondo (1997): The high
vowels /i/ and / u / are devoiced between voiceless consonants and between a
voiceless consonant and a pause. For example, / i / and /u/ in /kzkan/
’duration of time,’ /kukan/ 'distance between two places,‘ and / hon desu/
'(It) is a book' are devoiced. (Italics indicate devoiced vowels.) Studies of
different aspects of vowel devoicing in the Tokyo dialect have been
extensively reported. They include its physiological characteristics, what
factors interact with it, its environmental conditioning, whether some trace
of the vowel remains after devoicing, and its variability (e.g., Kindaichi 1958,
Han 1962, Sugito 1969, Yoshioka 1981, Vance 1987, Jun and Beckman 1993,
Kondo 1994, Imai 1997, Kondo 1997, Kondo 1999). Vowel devoicing in non-
Tokyo dialects has also been studied, but not yet as fully as for the Tokyo
dialect, and studies on perception of devoicing are even fewer.

Although it is generally believed that vowel devoicing does not occur
in the Kinki dialect (Horii 1972, Peng 1993), which is spoken in the Kyoto-
Osaka area, there are studies which show that it occurs there and in other
dialects outside Tokyo (e.g., Sugito 1962, Sugito 1969, Sugito 1988), and a

database of the Kinki dialect is available (Tahara et a1. 1998). One of the

problems with the investigation of devoicing in Kinki is that the data are
scarce and limited due to coverage of only a few phonological environments.
Another problem is that, when devoicing in different dialects is compared,
the target vowel in one dialect is not compared with a vowel in the same
phonological environment in another, which could lead to an inaccurate
evaluation of overall devoicing rates if high- or low-frequency environments
are over- or under-represented in one data set or another. When the
devoicing of vowels in the same phonological environments is compared,
using the limited data available from some of the previous studies of the
Kinki dialect, the ratio of devoicing turns out to be almost the same in Tokyo
and Kinki. Before comparing vowel devoicing in non-Tokyo dialects with
that in the Tokyo speech, it is necessary to collect more data and to analyze
them taking environmental effects into account to determine the factors
influencing its distribution and variation in non-Tokyo dialects.

The studies mentioned above focus on vowel devoicing in production.
Perception of it is another important and interesting issue. Previous studies
of variety perception show not simply that hearers can discriminate different
dialects, but also that hearers' judgments are based on low-level phonetic and
other variables. Some studies also show that the respondent judgments (e.g.,
evaluations of dialects) match their knowledge about dialect distinctiveness,
and furthermore such evaluations are affected by social information. On the

other hand, social information has also been shown to interfere with both

evaluation and identification of dialects (Graff, Labov, and Harris 1983,
Kerswill 1985, Preston 1993, Niedzielski 1999, Purnell, Idsardi, and Baugh
1999, Strand 1999). It should also be noted that in many of these studies the
respondents cannot describe the linguistic forms they used in making their
identifications and/ or evaluations, even though some of these studies
controlled for the manipulation of only linguistic factors.

Warner (1997) conducted an experiment about the perception of the
Kinki accent pattern, and showed that people from outside the region can
acquire knowledge of such suprasegmental differences in another dialect.
How people perceive and judge vowel devoicing, however, is a perhaps even
more valuable investigation because devoicing is not phonemic and not
obvious for every speaker of Japanese, as accent patterns are.

It is difficult to predict the results of any work on vowel devoicing
perception. Although devoicing is described in dictionaries as a 'standard'
language phenomenon and Tokyo is taken to be the source of the standard
variety, devoicing or nondevoicing as a clue for Tokyo people to judge a
speaker as a Tokyo person or not is complicated by the factors outlined above,
namely, that devoicing and nondevoicing are allophonic and that Japanese
people are not usually aware of them. In addition, since the devoicing rate
could be similar in the Kinki and Tokyo dialects, people might,
unconsciously, have some knowledge of that. The results of judgments by

non-Tokyo, Kinki respondents are at least unpredictable.

In this dissertation, I will present the results of an experiment that
investigates how people from Tokyo and Kinki make judgments of whether
or not a speaker is from the same region as their own upon hearing him or
her speaking. All features except for vowel devoicing will be held constant in
the primary data presented to respondents.

I will show, as a result of this experiment, that devoicing and
nondevoicing contribute to judgments by both Tokyo and Kinki respondents
just as the differences of the Tokyo and Kinki pitch accent patterns do. The
results are consistent with an assumption that devoicing is a Tokyo feature
and nondevoicing is a Kinki feature. That is, the respondents tend to use
devoicing and nondevoicing as clues when making judgments. Perhaps
more importantly, the respondents tend to use devoicing/nondevoicing as a
clue even though it may not reflect the actual distribution of
devoicing/nondevoicing in different dialects. In other words, they appear to
use an unconscious stereotype of devoicing in making judgements. The
results for Kinki respondents are not strong, and I will suggest that their
responses are more finely tuned to an interaction between voicing/devoicing
and other phonological features such as devoicing in accented mora. I will
also show that the respondents could not describe the reason why they judged
certain speakers as non-Tokyo or Kinki, just as indicated for other features in

previous perception studies.

Examination of phonological factors and social factors that significantly
affect judgments are also investigated and show distinct differences between
the Tokyo and Kinki results. Both Tokyo and Kinki respondents tend to
make use of the most salient factors influencing vowel devoicing in
production (Imai 1997), letting them override actual devoicing and
nondevoicing as a clue. For example, when Tokyo respondents hear
preceding consonants that most strongly promote devoicing in production,
they tend to judge the speaker as a Tokyo or non-Kinki person regardless of
actual devoicing or nondevoicing of the following devoiceable vowels. It
seems that the respondents also know, again, of course, unconsciously, the
best environments for devoicing in production and use that knowledge to
assume that the vowel should be devoiced. I will also show that the identity
of the following consonant also has an influence on devoicing/nondevoicing
perception. In short, the Tokyo respondents appear to assume that
devoiceable vowels are devoiced after having heard a phonological
environment which promotes vowel devoicing, and judge the speaker as a
Tokyo person. Again, this tendency was weaker for the Kinki respondents.

In the following section, I will review previous studies of differences in
the Tokyo and Kinki dialects, of vowel devoicing in them, and of perception
of dialects in general. I will then outline how I controlled factors, collected
data, and analyzed them, and then show the results, discuss them, and give

the conclusions of my study.

2. Previous studies

2.1. Some differences between the Tokyo and Kinki dialects

There are many dialects in Japan, but the hyéjungo (vowels with carets
indicate long vowels), 'the standard language,’ is based on a dialect spoken in
Tokyo (Sanada 1991). One of the major dialects other than the Tokyo dialect
is the Kinki dialect, which is spoken in the Kyoto-Osaka area (Hirayama et. a1.
1982). The Tokyo and Kinki dialects differ at various linguistic levels. An
example of difference in vocabulary is toriniku 'chicken meat,’ which is used
in Tokyo, while kashiwa is used in Kinki. An example of morphological
difference is the use of clause-final forms such as -to, -ba, and -tara 'if; when.‘
These forms are generally called conditional forms, and these three have a
complex set of usage restrictions in the Tokyo dialect. In the Kinki dialect, on
the other hand, -tara is almost always used.

The accent system in Japanese involves low and high pitch. In the
Tokyo dialect, some words are called "accentless," although they actually have
a low pitch in the first mora and high pitches in the rest of the word, and the
pitch stays high even when a morpheme is added (e.g., LHH+H, where "+"
indicates a morpheme boundary and ""H and "L" indicate high and low
pitch, respectively). All other words are "accented," and may have a low pitch
or a high pitch in the first mora. If the first mora of a word is high-pitched,

the rest of the moras are all low-pitched (e. g., HLL). If the first mora is low-

pitched, there are one or more successive high pitched moras after the first
mora, and the moras following the last high pitched one are all low (e.g.,
LHHLL). The accent pattern of a word can be, then, described as either
"accentless" or "accented" according to the position of the last high pitched
mora. Accent is phonemic, and may create a minimal pair such as dme 'rain'
(HL) and ame (accentless) 'hard candy' (LI-I). The Kinki dialect has a pitch-
accent system as well, and accent patterns are the same as Tokyo patterns for
some words, as in (1a) below, but some are different, as in (1b). In addition to
the patterns that exist in Tokyo, the Kinki dialect has some patterns of its
own, namely, more than one high pitched mora at the beginning of the word,
more than one low pitched mora at the beginning of the word, and
descending or ascending pitch within a mora (Makimura 1979, Warner 1997)
as in (1c), (1d), (1e), and (1f), respectively. 'D' and 'A' indicate descending and
ascending pitch, respectively. The Kinki patterns in (1c), (1d), (1e) and (1f) are

not found, at least phonemically, in Tokyo.

(1a) 'seat' Tokyo: seki (HL) Kinki: seki (HL)

(1b) 'scary; scared' Tokyo: kowai (LHL) Kinki: kowai (HLL)
(1c) 'box' Tokyo: hako (LH) Kinki: hako (HH)
(1d) 'egg white' Tokyo: shiromi (HLL) Kinki: shiromi (LLH)
(1e) 'rain' Tokyo: ame (HL) Kinki: ame (LD)

(1f) 'hand' Tokyo: te (H) Kinki: te (A)

Sugito (1962) describes differences of pronunciation between the Tokyo

dialect and the Kinki dialect. According to her, in the Kinki dialect the larynx

is narrow in vocalization, vowels are clear and longer, and there is no
nasalized / g/, while in Tokyo the larynx is wide open in vocalization, many
vowels are voiceless, and /g/ is nasalized in the second or later mora and
particles. Additionally, the diphthong /ei/ in some areas of Kinki is
pronounced / e: / in Tokyo. She also points out other phonetic differences
among consonants.

For this study, I controlled those features that are different in Tokyo
and Kinki so that I could focus on the effects on the respondents' judgments
by devoicing or nondevoicing only and compare these effects with that of

other features.

2.2. Vowel devoicing in the Tokyo dialect

Jun and Beckman (1993) state that vowel devoicing relates to a glottal
gestural overlap of the vowel and neighboring consonants. Yoshioka (1981)
used electrode techniques and showed what occurs in the laryngeal muscles
in vowel devoicing. He asked a subject to pronounce a set of words and
showed that vowel devoicing is accompanied by reciprocal movements of
two groups of laryngeal muscles, namely abductor and adductor, that cause
the glottis to open. Han (1962) treats four factors as important in devoicing
vowels: (i) shorter vowels, i.e., high vowels, (ii) faster speech tempo, (iii)
unaccented syllables, and (iv) neighboring sounds which makes vowels

shorter, i.e., voiceless consonants.

The most general description of vowel devoicing in the Tokyo dialect
is that unaccented high vowels / i / and /u/ are devoiced between voiceless
consonants and between a voiceless consonant and a pause (Vance 1987, p.48).

Examples are shown in (2) below. Italicized vowels indicate devoiced vowels.

(2) kikan "duration of time"
kukan "distance between two places"
Hon desu. "(It) is a book."

Vowels in the devoicing environments stated above, however, are not
devoiced in certain conditions. For example, devoicing interacts with
accentuation. Vance (1987) explains how accentuation on a devoiced vowel is

avoided, using the examples in (3).

(3a) taka-i 'is expensive' taka-katta 'was expensive'
(3b) fuka-i 'is deep' fuka-katta 'was deep'

When the accent is on the second mora in the present form as in (3a), it shifts
regularly to the first mora in the past form of an adjective. However, as seen
in (3b) it does not shift to avoid accentuation on the devoiceable /u/, which is
between two voiceless consonants. Two words which have the same
segments in the same order but different accent patterns have a devoiced
vowel in one and a nondevoiced vowel in the other, as seen in the data from

Yoshioka (1981). In his data, /sisee/ (accentless, i.e., LHH) 'posture' is

generally pronounced with devoiced /i/ while /sisee/ (HLL) 'death and life'
is generally pronounced with nondevoiced /i/. Han (1962) says, furthermore,
that vowel devoicing does not occur in a high pitched syllable, not simply in
an accented syllable.

Devoicing interacts with intonation, too. Vance (1987) gives an
example of a question without an interrogative particle as in Nani ka
arimasu? ‘Is there something?’ to show that the sentence final /u/ is not
devoiced since it carries rising intonation.

When a word contains two or more consecutive moras with vowels in
a devoicing environment, the accented vowel is not devoiced and the rest
take alternate devoicing (Kindaichi 1958, Han 1962), as in kutsushita
/kutusita/ 'a sock' and shikifuku /srkihuku/ 'a formal suit.‘ The latter
example, pronounced in the pitch pattern LHHH, contradicts Han's (1962)»
statement because the second devoiced mora /hu/ is pronounced in a high
pitch. Kondo (1999) conducted a production experiment using words
containing consecutive devoicing environments, and argues that the
devoicing site is determined by the largest syllable structure. She suggests that
the structure may be more flexible than previously considered, allowing a
heavy syllable with a non-nasal, non-geminate coda as seen in the first
syllable in shdhichi /loohiti/ 'consumer belt.’

It is also pointed out that devoicing tends not to occur at a word

boundary (Sugito 1969, Tsuchida 1993). Tsuchida does not give data, but

10

Sugito gives such examples as hayaku#haku (# indicates a word boundary)
'put on quickly,‘ hayaku#hiku 'pull quickly,’ and hayaku#fuku 'blow
quickly,’ in which /u/ at the word boundary is not devoiced. Vance (1987)
points out that devoicing morpheme-medially is preferred to devoicing
morpheme-finally as in oshi—tsukeru /osi-tukéru/ 'push against' rather than
/ osi—tukéru / .

The studies cited above are qualitative, and even prescriptive, as in
Kindaichi (1958). There are reports of data that shows variation of vowel
devoicing in both unaccented and accented moras. Vance (1987) states that
another way to avoid accentuation on a devoiced vowel other than not
shifting an accent in the adjective conjugation, as in the example (3) above, is
to simply nondevoice an accented vowel in a devoicing environment. He
does not give any examples, but another adjective /hzk1’1-i/ 'is low' might be
one because its past form is pronounced /hiki’1-katta/ (Kindaichi 1958) 'was
low' with /u/ nondevoiced.

Sugito (1969) asked nine people from Tokyo and Osaka to read
sentences containing words with devoiceable vowels and compared how they
devoice and nondevoice them. Her results show that 3 Tokyo speakers did
not devoice 54 to 59 vowels out of 163 in devoicing environments. She
concludes that devoiceable vowels tend to be nondevoiced at word

boundaries or in consonant combinations which are difficult to pronounce

such as [kuh] and [INN- Lovins (1976) also asked five men and four women

11

from Tokyo or near Tokyo to read sentences containing devoiceable vowels.
She reports that accented vowels in her data were devoiced 24% of the time.
Yoshioka's (1981) data also include examples of variation, both in
nondevoicing of devoiceable vowels and devoicing of accented high vowels
between voiceless consonants (e.g., 5 cases out of 28 of /sisee/ 'posture' and 8
cases out of 28 of /sisee/ 'death and life,’ respectively). Sugito (1969) further
reports that when a mora follows a mora with a devoiced vowel which was
supposed to be accented, the following mora starts with a high pitch and then
the pitch descends rapidly. For example, in /hfisu/ 'lie on the stomach' the
frequency of / u / in the second mora drops rapidly from the beginning toward
the end, while the frequencies in the mora following a devoiced vowel stay
almost the same in accentless words. She concludes that this pitch contour
helps listeners determine the position of accent in the preceding mora in spite
of vowel devoicing. Kitahara (1998) basically confirms Sugito's results. He
reports that the pitch elevation occurs at the beginning of the mora following
one with a devoiced accented vowel, but it occurs only when the mora
preceding it is in a low pitch, and that there is no clue to locate the accented
mora for listeners when the preceding mora is in a high pitch.

Devoicing before a pause shows a different distribution of variation for
/i/ and / u/ according to Sugito (1988). She gives data for vowel devoicing in
a reading passage, and shows remarkable differences in frequencies of

devoicing sentence-finally, namely /u/ in /desu/ (copula), and clause-finally,

12

namely /i/ in idd-shi /idoo-si/ 'moves and' and /u/ in 6ku /ooku/ 'many
and' and in waruku /waruku/ 'bad and.’ In her data, all the Tokyo people
devoice one of the two sentence-final / u / s and 89% of them devoice the
other, while none of them devoices the clause-final /i/ and none of them
devoices the clause-final /u/ in /ooku/, while 22% of them devoice the other
clause-final /u/ (See Table 1).

Devoicing in an unaccented high-pitched mora occurs frequently,
contradicting Han's (1962) statement that it does not occur. Lovins (1976)
reports that fewer than 2% of the devoiceable vowels in her data are
nondevoiced in such moras.

Devoicing also has variation in two or more unaccented moras in a
consecutive devoicing environment, in spite of Han's (1962) general rule
which calls for undevoicing accented vowel and alternate devoicing for the
rest. In Sugito's (1988) data, kakuchi-tomo /kakuti-tomo/ 'in all regions' has
a consecutive environment, and 67% of the Tokyo people devoice / u/ and
78% devoice /i/ (See Table 1). This means at least 45% (67-(100-78)) of them
devoice both /u/ and / i / . Imai (1997) also reports that consecutive devoicing
occurs frequently, even in three or four consecutive environments. The data
I collected also contain examples of consecutive devoicing. I asked fourteen
people from Tokyo to read the same passage and thirteen of them read a

phrase with consecutive devoicing as in /hanasu-kfl<ai/ 'chance of talking.‘

l3

Devoicing at a word boundary and at a morpheme boundary also
occurs quite frequently. The example /hanasu#kfl<ai/ in my data shows a
very high rate of devoicing at a word boundary. /u/ is word-final and / i / in
/ki/ is morpheme-final, and both are devoiced by 13 people out of the 14.
Many other examples are found in data in previous studies, such as /si-see/
(23 people out of 28) in Yoshioka (1981), kaku-chi-tomo /kaku-ti-tomo/ 'in all
regions' (67%) and /kaku-ti-tomo/ (78%) in Sugito (1988), and /hiraku#to/
'when someone opens it' (100%) in Sugito (1969).

Devoicing of non-high vowels and devoicing in the environment that
is not a general devoicing environment are also reported. Imai (1997) reports
that /a/, /e/, and /o/ besides /i/ and /u/ are sometimes devoiced. She also
reports that devoicing sometimes occurs after nasals among other voiced
consonants.

There are also studies about the probability of devoicing depending on
different consonants in the devoicing environment. Han (1962) states that
effects on devoicing are greater with fricatives than affricates, and with
affricates than stops. She also states that the effect of /l/ is greater than the
effect of pitch accent.

Imai (1997) collected data from natural casual conversation and
examined approximately 1,600 tokens of vowel devoicing. Based on her
analysis, the most significant promoting factor in vowel devoicing is the

following voiceless segment and the second one is the preceding voiceless

14

segment. She determined a best promoter of devoicing in terms of
articulation of the preceding segment for each of 5 vowels: the fricative for
/ i / , the feature [+stop] for /u/ and / 0/, and the stop consonant for /e/ and
/ a / . She further states that "the vowel /i/ with a preceding fricative most
strongly promotes devoicing." She also points out that a shared feature
between the preceding consonant and the vowel promotes devoicing; the
feature palatal for / i / and the feature labial for /u/ (pp. 52-53).

In this study, I use the most general description of the devoicing
environment: unaccented high vowels between voiceless consonants. I
eliminated non-high vowels, accented vowels, and unaccented high vowels
in consecutive devoicing environments. This made it possible for me to
examine devoicing variation without the influence of other factors such as

accentuation, intonation, vowel height, and syllable structure.

2.3. Vowel devoicing in the Kinki dialect

It is generally believed that, in the Kinki dialect, vowels are nondevoiced. In
particular, it is often suggested that Kinki speakers nondevoice /u/ at the end
of a sentence as in /desu/ (copula) or /masu/ (polite present affirmative
ending). Maeda (1977) describes vowels in the Osaka dialect (one of the
dialects within the Kinki area) as very resonate and pronounced more
deliberately than consonants, as opposed to those in the Tokyo dialect, in

which vowels are pronounced less deliberately. Horii (1972) and Peng (1993)

15

state that vowel devoicing does not occur in the Kinki dialect. Tsujimura
(1996) states that devoicing is "not common (p. 25)," and Shibatani (1990)
states that it is "less noticeable in Kyoto (p. 161)," rather than not present at
all. None of them gives any data.

In spite of those widespread ideas, devoicing in the Kinki dialect has
been reported. Nakai (1991) reports that vowel devoicing at the end of the
sentence tends to occur in some regions within Kinki. He collected
pronunciations of Itadakimasu 'I humbly receive' (a greeting phrase before
having a meal) by children from more than 200 elementary schools in Kinki
and examined the accent pattern and devoicing of /u/ . He reports that a
more standardized accent is used and devoicing tends to occur more in inner
Kinki, which is Osaka city and its suburban area, while the Kinki accent is
used and nondevoicing tends to occur in the peripheral areas, but the Kinki
accent is used and devoicing tends to occur in the very center of Osaka city.

Devoicing occurs within a word as well in Kinki. Tahara, et a1. (1998)
created a database of the Kinki dialect on a CD-ROM, by asking twenty male
and female subjects of different age groups, who were from Osaka, to read a
set of sentences. In the sample data I checked, 2 sentences include vowels in
the most general devoicing environments: (i) Ka tsukamaeta de 'I caught a
mosquito,‘ and (ii) Ha ochiteru wa 'Leaves have fallen and are on the
ground.‘ Sixteen people out of 20 devoiced /u/ in the former sentence and 17

people devoiced / i / in the latter.

16

There are more studies which show devoicing in the Kinki dialect.
Sugito (1969) shows that 3 Kinki dialect speakers devoice 28 to 50 vowels out
of 163 in devoicing environments. She also reports on devoicing by an 83
year-old male subject from central Osaka city. Nakai (1991), cited above,
similarly reports on devoicing in the very center of Osaka city based on an
extensive data set, thus showing that devoicing by Sugito's particular subject
is not a mere idiosyncratic example. Sugito shows that one of the subjects
devoices accented vowels as well, as in /kzisa/ 'grass.‘

Sugito (1988) also gives data to show that Osaka people devoice vowels
less frequently than Tokyo people do. She chose 8 cities in Japan (Sendai,
Tokyo, Nagoya, Osaka, Kochi, Okayama, Kumamoto, and Naha) and asked 9
or 10 males from each city to read a weather report in kyétsfigo 'common
language; language for mutual understanding.’ She examined devoicing
tendencies in these data in terms of phonological environments and of the
regions where the subjects live. An excerpt from her table is shown in Table 1
below.

In her data, the subjects from Tokyo sometimes fail to devoice vowels,
but the frequencies for devoicing by the Osaka subjects are not higher in any
word than those by the Tokyo subjects, confirming the statements by
Tsujimura, Shibatani, and Maeda. At the same time, the studies reviewed

above show how frequently Osaka people devoice vowels in certain

17

environments, and the results clearly deny the general belief that Osaka

people do not devoice vowels.

Table 1 Devoicing by people from Tokyo and Osaka in Sugito (1988)

 

 

 

 

 

 

 

 

 

 

 

 

 

Tokyo Osaka
Sentence-finally desu 1 89% 56%
desu 2 100% 78%
Clause-finally idé-shi 0% 0%
6ku 0% 0%
waruku 22% 0%
Within words itsuka-wa 100% 67%
kakuchi-tomo 67% 33%
kakuchi-tomo 78% 67%
kaifuku-suru 78% 11%
kaifuku-suru 22% 11%
Total 55.6% 32.3%

 

 

 

 

 

However, devoicing in the Kinki dialect is still not fully investigated.
In data from Tahara et a1. (1998), 82.5% of devoiceable vowels (33 out of 40) are
devoiced by Osaka people. This is higher than the frequency of any word
pronounced by Osaka people, and it would not even be exceptional as a
frequency in Tokyo, in Sugito's (1988) data. This does not seem impossible,
knowing that there is a tendency for devoicing before a pause, which is the
most common environment of nondevoicing in Kinki described by non-

linguist Japanese speakers, in inner regions of Kinki, although it is only one

18

phrase pronounced by only one age group in Nakai (1991). Nevertheless, it is
tempting to assume that the devoicing rate in Osaka (Kinki) is not all that
different from that in Tokyo. Devoicing by an 83 year-old male subject from
the very center of Osaka in Sugito (1969) also supports this tempting
assumption. Table 2 shows the percentages of devoicing and nondevoicing in
some of the previous studies (calculation mine, based on numbers of the

tokens in each study).

 

 

 

 

 

 

Table 2 Variations of devoicing
Tokyo subjects Osaka subjects
Devoicing Nondev. Devoicing Nondev.
Sugito (1969) 65.8% 34.2% 29.4% 70.6%
Sugito (1988) 55.6% 44.4% 32.2% 67.8%
Yoshioka (1981) 56.5% 43.5% N/ A N/A
Tahara et a1. (1988) N/ A N/ A 82.5% 17.5%

 

 

 

 

 

 

 

The data in Tahara et al. are different from those in the others, in that all
tokens have high vowels in the most general devoicing environments, i.e.,
high vowels in unaccented mora between two voiceless consonants and not
in consecutive devoicing environments. The others, on the other hand,
have devoiceable vowels that may interact with other factors and may
therefore remain nondevoiced, such as in accented moras, in consecutive

devoicing environments, and before pauses. The comparison, then, is not for

19

the same environments, and the conclusion that Kinki speakers devoice
vowels less frequently is not a reasonable one. For example, if one compares
/kusa/ in Tokyo and /kusa/ in Kinki (both mean 'grass') as Sugito (1969)
does, a lower rate of devoicing in Kinki does not mean simply a difference
due to dialectal difference. Kinki speakers may well avoid devoicing due to
the accentuation, just as Tokyo speakers do.

Table 3 shows the percentages of devoicing and nondevoicing in

unaccented and accented moras separately from two sets of data.

Table 3 Devoicing variation in different positions (Tokyo subjects)

 

Unaccented Accented Consecutive

 

Devoicing Nondev. Devoicing N ondev. Devoicirg
Yoshioka (1981) 76.8% 23.2% 16.1% 83.9% N/ A
Morris 80.6% 19.4% N / A N/ A 92.9%

 

 

 

 

 

 

 

 

 

 

The frequencies of devoicing unaccented vowels (also without any
interaction with other factors) are approximately the same as that in Tahara et
al. (1998). Sugito's data (1969, 1988) could not be separated by unaccented and
accented tokens because the accent pattern used for each token is not
indicated. The low frequencies of devoicing in Sugito (1969) might be
explained, therefore, by the mixture of unaccented and accented tokens. The

frequencies in unseparated data in Yoshioka (1981) are similar to those in

20

both of Sugito's, so it is reasonable to suppose that the devoicing rate in
unaccented moras might increase in Sugito's data when counted separately,
but the factor that lowers the devoicing frequency in Sugito (1988) is not
certain. It should be pointed out, however, that the passage she asked her
subjects to read contains only one devoiceable vowel with no possible
interaction with other factors out of ten devoiceable vowels. Of the other 9
devoiceable vowels, 5 are before a pause, and 4 are in two words containing
two consecutive devoiceable vowels. It is natural, then, to suspect those
interactions are connected with low devoicing frequency.

The data I reviewed above are not, however, sufficient to determine
the devoicing in Kinki in detail, due to a lack of variety of phonological
environments, the limited number of the subjects, and the lack of gender and
age variety. It would not be unreasonable, however, to make a tentative
assumption that devoicing of a vowel in an environment in which no other
factor suppresses devoicing occurs with about the same frequency in Tokyo
and Kinki. Different frequencies in other environments could emerge
because of different influences of other factors, which may be the same in
Tokyo and Kinki.

To avoid some of the gaps in previous studies, I collected data from
both male and female speakers, and I selected the words I used in the
experiment by the following criteria, so that I could compare the words in the

same phonological environment: (i) pronounced in the same pitch accent in

21

Tokyo and Kinki, (ii) containing only one high vowel in a general devoicing
environment, (iii) containing a devoiceable vowel that does not interact with
other factors. That is, I excluded words containing a devoiceable vowel in
such positions as in accented moras, in consecutive devoiceable moras, and
before a pause. I did, however investigate both morpheme and non-

morpheme boundary environments.

2.4. Perception of dialects and attitudes toward them

Labov (1972) discusses the benefits of sociolinguistic investigation and
states that "The speech community is not defined by any marked agreement
in the use of language elements, so much as by participation in a set of shared
norms; these norms may be observed in overt types of evaluative behavior,
and by the uniformity of abstract patterns of variation which are invariant in
respect to particular levels of usage" (pp. 120-121) and "Once the social
significance of a given linguistic variant has been determined, , this
variable may then serve as an index to measure other forms of social
behavior" (p. 120).

There are many studies on dialect perception within Lobov's general
framework. Kerswill (1985) conducted a perception experiment using
standard Norwegian (Bergen) and a rural dialect (Stril). He presented 9
samples spoken by different people to subjects and asked them to judge the

speakers as Bergen or Stril. He concludes that the subjects made judgments by

22

perceiving degrees of mixing of the two dialects and rated the samples
relatively along a linguistic continuum from purely Bergen to purely Stril,
and that morpho-lexical cues are utilized most. In his methods, however,
cues at the morpho-lexical and phonetic levels are not controlled; therefore it
is not possible to identify precisely the salience of each type of cue. He also
mentions that judges' comments do not help because their descriptions of the
samples they heard did not match the actual features. This implies those
features are unconscious.

Preston (1993) conducted 4 experiments in the United States, namely,
asking respondents to draw maps of areas where people speak differently and
labeling them, to rate the 50 states for 'correctness' and 'pleasantness,‘ to rate
the 50 states for the degree of difference they perceive between them and the
respondent's own dialect, and to identify the regions from which 9 different
speech samples came. In this last experiment of area identification,
respondents used the numbers 1 to 9 which were assigned to the cities along a
north-south line from Saginaw, Michigan to Dothan, Alabama, as the places
the speakers are from and the numbers 1 through 9 for the voices,
representing the order in which they were placed. The respondents not only
placed the voices correctly by region (rnisplacing only one), but also indicated
larger distances (as determined by mean scores differences) between some

adjacent areas, which can be interpreted as major boundaries of

23

distinctiveness, and these, interestingly, often corresponded to the major
boundaries which resulted from the map-drawing and rating tasks.

Purnell, Idsardi, and Baugh (1999) conducted 4 perception experiments
using African American Vernacular English, Chicano English, and Standard
American English. They show that the subjects can significantly discriminate
the callers' ethnicity over the phone. They also show that such dialect
discrimination is possible by hearing only one word 'Hello,‘ and, more
specifically, that the hearers' ability can be explained on the basis of the
frequency of the second formant of the first syllable alone.

Niedzielski (1999) conducted perception experiments confounded with
social information, namely indicating to the respondents whether the speaker
was from Michigan or Canada, using words that contain vowels involved in
Canadian Raising (CR) and the Northern Cities Chain Shift (NCCS). She
shows that the respondents, who are all Detroiters, tend to choose ’standard'
vowels as matches to the ones they actually heard when they are told the
speaker is from Michigan and choose raised vowels (CR) when they are told
the speaker is from Canada, although the speaker was the same person, one
from Michigan, pronouncing only raised vowels. She also shows that they
tend to choose 'standard' or 'hyperstandard' vowels when told the speaker is
from Michigan, while more people tend to choose shifted vowels (NCCS)

when told the speaker is from Canada. She states that Detroit listeners use

24

social information for perception and judgment and hear the sounds to fit the
phonetic features of those socially stereotyped categories.

Strand (1999) reports on perception experiments (Strand and Johnson
1996) on the effects of gender information using the words sod and shod,
which form a minimal pair with /s/ and /l/ . There is an acoustic continuum
between /s/ and /I/, and the boundary between them is at higher frequency in
female speech than in male speech. The hearers drew boundaries between
the two sounds at lower frequencies when they heard the sounds looking at a
male face, in accordance with the lower frequencies of male speakers'
pronunciation of /s/ and /l/, than when they heard the same sounds without
looking at a face. These results suggest that gender stereotypes alter the
hearers' perception on the basis of the interaction of phonetic information at
a low level and socially constructed information at a higher level.

These studies support the following ideas: (i) respondents may identify
someone's ethnicity or dialect region based on his or her speech, (ii) they can
do this based on forms they are not aware of and cannot describe accurately,
(iii) such judgments are affected by social as well as linguistic information.

Speakers of Japanese also usually can tell whether someone is from the
same region as their own upon hearing him or her talking, sometimes even
when he or she tries to imitate a non-native dialect. This is a consequence of
judging such speech as having or not having shared features with their own

dialect. It is quite obvious that people can easily recognize differences in

25

vocabulary and differences in structures to some extent. It is also expected that
it is easy to perceive such differences and make judgments about whether
someone is from the same region as one's own based on pitch accent patterns
because they are phonemic.

Warner (1997) conducted an experiment on the recognition of accent
patterns to examine some factors from the viewpoint of dialect acquisition.
In earlier studies using segmental differences, reviewed by her, it was found
that forms in lexical opposition and forms which do not exist in one's own
dialect make acquisition harder. In her experiment, Tokyo and Kansai
(Kinki) people listened to a word list in the Kansai and Tokyo dialects,
respectively, and responded by choosing the meanings that they thought were
meant. The word lists consisted of pairs of words that have the same
segments but of different types because of accent patterns: (i) words in lexical
opposition, in which a word is in the listener's dialect but with a different
meaning, as in hfishi, meaning 'chopsticks' in Tokyo and 'bridge' in Kansai,
versus hashi, meaning 'bridge' in Tokyo and 'chopsticks' in Kansai, (ii)
homophones in the listener's dialect but different words in the speaker's
dialect, as in kdmi, meaning 'god' or 'paper' in Kansai, versus kdmi, meaning
'god' and kamf, meaning 'paper' in Tokyo, and (iii) words with accent
patterns that do not exist in Tokyo, such as 3, meaning 'picture' in an
ascending pitch in Kansai, versus é, meaning 'picture' and e (accentless),

meaning 'handle' in Tokyo. Her results show that the first type and the third

26

type of words explained above make acquisition of pitch accent patterns
harder, and she concludes that forms in lexical opposition and forms which
do not exist in one's own dialect make it harder to acquire segmental
differences as well as suprasegmental differences such as accent patterns.

It does not seem, on the other hand, that some features of
pronunciation, in particular allophonic features such as vowel devoicing,
could be consciously activated or perceived in natural conversations,
although it does surface in such cases as, for example, the training of
announcers. Han (1962) states that "native speakers of Japanese are not
usually aware of the fact that some of the Japanese vowels are not heard by
foreigners." Some Japanese people cannot recognize or tell the difference
upon hearing two pronunciations of the same words, one with a devoiced
vowel and another with a nondevoiced vowel. This is quite reasonable
because devoicing and nondevoicing are allophonic and are not features they
would be aware of. That is why the respondents in Kerswill (1985) and Strand
(1999) did not describe what they heard exactly as they heard it and those in
Niedzielski (1999) did not realize their own shifts.

Some careful speakers of Japanese describe Tokyo devoicing and
believe that Kinki speakers do not devoice, but they seldom suggest variation
in these behaviors, believing it to be categorical. Most speakers, however, do
not comment on this phenomenon. Some people apparently have conscious

knowledge that devoicing occurs in the Tokyo dialect, particularly at the end

27

of the sentence. However, they may well not know that it occurs in the Kinki
dialect as well or that nondevoicing occurs occasionally in the Tokyo dialect.
This gap may create a stereotype about dialects. The Kinki dialect and the
Tokyo dialect are sometimes contrasted in frequency of vowel devoicing, but
speakers of other dialects are not mentioned as often. According to Sugito
(1988), in fact, Tokyo is not the place where people devoice vowels most, and
Osaka is not the place where people devoice them least. Among 8 cities she
investigated, Nagoya and Kochi are the places where people devoice vowels
most and least, respectively.

Maekawa (1983) states that vowel devoicing may be required "as a
norm in the society." He has an impression that a word with nondevoicing,
even pronounced in the standard pitch accent, sounds as if it has "local
characteristics," which probably means non-Tokyo or non-standard. He says
it is problematic to bring such a personal judgment into a discussion of the
standard language, but also admits to having idiosyncratic views about the
lexicon (pp. 78-79). His impression, without quantitative investigation nor
reasonable explanations, suggests the possibility of similar responses from
others to devoicing variation and may indicate the possibility for a number of
personal judgments which may form meaningful patterns.

It seems quite reasonable then to ask the question, "Is vowel devoicing
one of the shared norms among speakers of the Tokyo dialect." and to collect

quantitative data from Tokyo people to answer this question. It would be a

28

straightforward or even simple question if devoicing had no variation,
displayed considerable dialectal differences, and was a conscious feature.
However, the answer may not be so simple because of variation, dialectal
differences and similarities, and the unconscious nature of devoicing.
Perhaps the Tokyo judgment norm will be that someone who devoices
vowels is from Tokyo and that someone who does not is not. That would be
phonologically simple, but does not reflect the reality of distribution of
devoicing, at least, in the Kinki dialect as shown in Tables 2 and 3, since the
non-Tokyo (i.e., Kinki) data Tokyo speakers hear contains devoiced vowels in
the most general environments. In other words, the criteria speakers of
Japanese use in identifying the regional background of speakers of devoiced
and nondevoiced tokens may be based on linguistic stereotypes, entailing an
oversimplified knowledge of the distribution of devoicing. Furthermore, if
they use not only phonological information but also social information when
they hear such tokens, what information would affect their judgments and
how?

Since devoicing is a "standard" feature, will Kinki respondents use it
differently as a clue to regional identity? In other words, is it one of the
shared norms among them, when the data they hear contain devoiceable
vowels in the most general environment? Their judgments could be the
mirror image of the Tokyo people's, based on phonological environments

and stereotypes; that is, their basis for judgment could be that a nondevoicer

29

is from Kinki and a devoicer is not. Or, because their devoicing rates are as
high as about 80%, their basis could be that a devoicer is from Kinki and
nondevoicer is not.

Since I have mentioned social factors, it is important to consider
attitudes toward the standard language and dialects. People, including those
from Tokyo, tend to believe that those who are from the Tokyo metropolitan
area speak the standard language. Actually, as investigated by Sanada (1991),
no single dialect in Tokyo is identical with the standard language. Instead,
one of the dialects spoken in Tokyo was selected to establish the standard (pp.
92-95). The standard language is an abstract ideal form to aim for, as opposed
to kydtsfigo 'common language; language for mutual understanding,’ which
is used in real life (pp. 202-205).

Sanada (1991) describes the process of promoting use of the standard
language by the government after the Meiji Restoration (1868) and states that
dialects were "devalued as disturbing or useless for promoting use of the
standard language (p. 105, translation mine)" Mogami (1994) points out that
the standard language may well have the image of nationalism, coersiveness,
and authority because of the process of its establishment in the history of the
country. Sibata, cited by Sanada (pp. 106-107), states that teaching the standard
language has divided people on the basis of availability of education,
economic status, and power relations in spite of the original purpose of

unification. Some people may think that no one speaks the standard

30

language entirely accurately, while others, Yoshioka (1995) states, may find
prestige in the Tokyo dialect and call it the standard simply because Tokyo is
the capital.

Warner (1997) noted that the Kinki dialect "has some prestige as the
descendant of the language of the ancient court (p. 365)." Umesao (1995) states
that Kyoto people take a pride in being the center of culture in Japan, based on
the fact that Kyoto, in Kinki, was an ancient capital of Japan. Shibuya (1995)
conducted a study of Kyoto people's attitudes toward their own dialect, Osaka
dialect, and the language for mutual understanding, and of their
codeswitching between their dialect and the language for mutual
understanding. They gave positive descriptions about their own dialect, such
as 'deliberate,’ 'beautiful,’ and 'gentle,’ the highest among 14 dialects, and
they gave negative descriptions of their dialect the lowest ratings out of the
14. In 7 different situations, the percentages of approval for speaking the
language for mutual understanding were the lowest among the 14 dialect
regions. The results suggest that the Kyoto speakers tend to speak their own
dialect in any situation. The results on their attitude and codeswitching
clearly show that they consider their own speech to have strong prestige.

For this study, I conducted an experiment using vowel devoicing to
determine its significance in judgments about whether or not the speaker is
from the same region as one's own. I chose words with devoiceable vowels

in the most general devoicing environment and neutralized other possible

31

features that could be a clue for making judgments as much as possible, and
collected data from Tokyo people and Kinki people. I eliminated the
responses to speakers who were known to the respondents personally to
avoid any judgment based on social information. In the experiment, I asked
my subjects "Is the speaker from your own region?" rather than "Does the
speaker use the standard language (the Kinki dialect)?" This is to avoid the
strict reaction "No one speaks the standard language" and reactions
influenced by 'value,’ added during the history of the language policy
although those ways of thinking might be changing nowadays.

My hypothesis is that a speaker is more likely to be judged as a Tokyo
person by Tokyo people and as a non-Kinki person by Kinki people if he or
she devoices a devoiceable vowel, and as a non-Tokyo person by Tokyo
people and as a Kinki person by Kinki people if he or she does not. In other
words, I predict that both Tokyo people and Kinki people have stereotypes
that Tokyo (or non-Kinki) people devoice vowels and Kinki (non-Tokyo)
people do not, and I believe that judgments are based on the stereotypical

perceptions of standard language behavior which I have outlined above.

32

3. Methods

To find out how devoicing and nondevoicing would affect a listener's
judgment on where the speaker is from, I collected data by presenting the
subjects with a word list and asking them to make a judgment for each word.
Then, I analyzed the responses from Tokyo people and Kinki people to find
out which factors determined their judgments and compared the results of

the subjects from the two areas.

3.1. Preparing a test tape

First, I prepared a test tape with a list of words. Target words in the list were
selected so that vowel devoicing was the only factor which might sound
'foreign' to the respondents. I selected lexical items that are used commonly
in Tokyo and Kinki and pronounced in the same pitch accent patterns in both
areas. I eliminated two other factors which would be used as clues to make
judgments about the speaker. One is non-word-initial /g/, which is nasalized
in some regions, and the other is /ee/ (the long / e / ), which is pronounced as
[ei] rather than [e:] in certain regions (Kindaichi 1958). These exclusions were
important because very often those factors are conspicuous and allow people
to tell where the speaker is from even by hearing only one word in a long

passage. Loan words were excluded from this study.

33

Among the selected lexical items, I further selected words that had only
one high vowel between voiceless consonants in an unaccented mora with a
low pitch. I eliminated words that had high vowels in positions where
devoicing is often avoided in the Tokyo dialect, such as in high—pitched
moras (both accented and unaccented), before pauses, and in consecutive
devoicing environments. I eliminated all these because responses based on
them may not have resulted from the simple criterion sought here, that is,
that devoicing represents Tokyo and nondevoicing represents non-Tokyo,
and comparison of responses to such data would have deviated from my
assumption that rate of vowel devoicing in the most general devoicing
environment is the same in Tokyo and Kinki.

The pitch accent pattern of the words selected for this study is either
HLL or LHL at the beginning of the word, common in Tokyo and Kinki, and
such words contained a devoiceable vowel in a low-pitched mora. For
example, atafuta 'hurriedly' is pronounced as HLLL and /u/ is devoiceable,
and nadeshiko 'a type of pink flower' is LHLL and /i/ is devoiceable. Using
these two distinct pitch accent patterns made it possible also to compare how
the effects of accent pattern may differ in respondent judgments about
whether or not a speaker is from the same region as the respondent.

Words pronounced in the other two patterns, LHH with accent on the
last mora and LHH without any accented mora, are not included. The

reasons are the following. First, devoicing in the high-pitched mora tends to

34

be avoided in Tokyo and may therefore affect the responses. Second, words
which contain a devoiceable vowel in the first mora do not seem to have
parallels in the two areas investigated here. It seems that words pronounced

in LHH in the Tokyo dialect are pronounced as HLL, LHL, or HHH in Kinki as

shown in (4):

(4a) (a place name) Tokyo: kyobashi (LHHH) Kinki: kyobashi (HLLL)
(4b) 'rice planting' Tokyo: taue (LHH) Kinki: taue (LHL)

(4c) 'replacement' Tokyo: kawari (LHH) Kinki: kawari (HHH)
In such cases, the respondents may judge whether the speaker is or is not
from the same region as their own on the basis of differences of the pitch
accent pattern.

There are 110 possible environments of the high vowels between
voiceless consonants. For /i/, there are 5 possible preceding consonants [k, J,
tJ, 9, p] and 10 possible following ones [k, s, J, t, tJ, ts, h, 9, «b, p]. For /u/, there
are 6 possible preceding consonants [k, s, J, ts, (j), p] and the same 10 possible
following ones. [tJ] preceding /u/ is not included here although it is
considered to be a voiceless phoneme, because there seems to be no word
which contains [tJu] followed by another voiceless consonant or a pause
except for very few loanwords, as Kondo (1999) observes.

Ideally, each environment of vowel devoicing should have a sample

word, but not all environments can be filled by actual words in order to

35

examine which consonant and which position (preceding or following
devoiceable vowels) promotes or demotes judgment based on my hypothesis.
Among the total 110 possible environments, words were found for 58
environments. The default verb, sum 'to do,‘ a particle such as de 'at; by,‘
and mo 'also' or suffix such as -san 'Mr./Ms.' is added to the word in some
cases, to avoid word final devoicing or to create a devoicing environment.
For example, I could not find a word that has [JitJ] in the second mora and [tJ]
in the third mora's onset, so I chose Toshi [toJi], which is a two-mora
truncation of a man's name, and added —chan [tJan], a suffix added to names
when addressing a child or very close family member or friend. Adding this
suffix creates the environment [JitJ] and avoids a word final devoiceable
vowel / i / . It is not ideal to combine morphemes or words, but I sometimes
did simply because it was difficult to find‘ sample words containing
devoiceable vowels within a morpheme in the various phonological
environments I wanted to study. However, the effects of a boundary on the
production and perception of devoicing are separate issues, and by also
focusing on such boundaries, it is possible to investigate their effects on
perception and judgment. Furthermore, devoicing frequently occurs at
morpheme boundaries and sometimes consecutive devoicing occurs even at
word boundaries. The list of words and phrases containing / i / in the possible
environments of devoicing is shown in Table 4, and the one containing /u/

is shown in Table 5.

36

835838 wfi3o=£ SECS 3538 3385.8: “35:8:8 wficmumum 53:8 8638 ~munum>

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

agave. .3883 8.
“53363 Eefgemfim. m
.98 m3oE U53 ~98.
33363 c
.83 32 9a awe .3.
uwlfifims u
.8: :63 can.
age ;
.mumnmmhwmfiu
368$ 82“. tom 0:. .850 «5 «m. .83»
as auigum uuimuias 9-3::ng .Smfium my
.28... 6:5. .mufim 3058 m an.
Satémoh 63:63»: 2
.95— EEH wfi .Efimuomfiwu
ban was Emum .8“. Ho .8250qu owe—U. 3:8“. EB 3533.
Eifiéms 8-39 «.QO 38-383 33....33 :63 3-3:3 a
23:3. .98 coax—Ema. .: 3% $2. @2633 063 how mm.
as. 365.2% in 23.. Esteem 43.33: .
.mnfiwoomu.
gfiufi:
Bang ._8 mm. .396. .3 36$. .Emmgx £232. .25 35m.
§-=§§ SPENSN 3.3.38: §t3m§¥ 33.38: m
I .533“
EA Mo wmb m.
.932 69c .336 8383: .588
M: m .8983 _. .3? Row m. .93. .Emohm >02 23:. m m: m .5953 r
gukufizu: cEfiETS figfis 25352.5 u=§63§~ x
a u 2 _ x
€5.8on urn—088m.—

 

 

) \ 3983036 mo «5:58.338 wring—Bu 8.83 395%.“—

¢ £an

37

98:99:00 wE30=£ 53:00 86200 83:08.8: “38:98:00 wfivmofim 53:00 25200 8038>

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.8 .3 :3. 33958.
88:88: 3853: :
.8>.8m 85g.
8-3.388
385526»
.282...
85 :8 8.
8.3.3 a
.83 888: .3.
8.38.: 8
8:8: 38:85 .wfivmm: 88¢.
338» 33% A
.28 0%. .98 33333.
885:3 afifi QE$3§ 3
.808 80.
8.33%& 2
.33
m 98 8:83 3885:;
“8&8: 8. 8:88 .88.: .9886
:88»? .85 um 9:38:05. “:5 08 tab. 538.38. .83 08 8.508. .083. 08 2:9.
2-38.8 33% 2-33% awn ovxmfifi mauvafizmm 353 3.8.: $52 3-03.9 .
.888 .32. .238» .32. 8:2 8.22... .3.
oguémamz 888338» 8-3853 _
.3088. .8080fi8. 8838880. 38:00. .8830: 23 a.
3.5388“ 338%.: 35383.3 3.8.38.8 8338: m
8:3qu 0.8 338*
88:8 85 .3. .8 88w. .880 82w. .8 .883. .8458. m m: m 80:03 _.
8-5383 8786? 0.3.8383 fiéfifim 8E8: u:§-§o~§ x
a m 3 _ m 0—
888800 “3:885

 

 

 

\: \ 8380888 «0 38:80:38 98:83:00 883 29:88:

m 3an

38

The position of a devoiceable vowel relative to a morpheme boundary
could affect judgments about the speaker because of possible effects of
devoicing variation in production. A devoiceable vowel in this study is
either at a word boundary, at a morpheme boundary, or within a morpheme,

and the effects will be compared among these three. Examples are given in (5)

below.

(5a) At a word boundary katsu#tsumetai 'also cold'
(5b) At a morpheme boundary Toshi-chan 'little Toshi'
(5c) Within a morpheme atafuta 'hurriedly'

Another aspect of the position of a devoiceable vowel focused on in
this study is the fact that it may occur in either the second or third mora from
the beginning of the word. In either case, I chose the words with unaccented
devoiceable vowels only. In my data, a devoiceable vowel is in the second
mora of a word with the initial HLL pattern or in the third mora in HLL or
LHL. For example, Toshi'-Chan 'Toshi (a truncated name)-(suffix to address a
child)‘ has a devoiceable vowel in the second mora in the HLLL pattern, and
Hiroshi-kun 'Hiroshi-(suffix to address a boy)’ in the third mora in HLLLL. It
is also possible to compare this with such a word as nadeshiko, 'a pink,‘
which has a devoiceable vowel in the third mora in LHLL.

Two types of words that do not contain any environment for vowel

devoicing are selected for the purpose of comparison: (i) words that are

39

common in both the Kinki and Tokyo areas but pronounced with different
pitch accent patterns, and (ii) words that are common and pronounced with
the same pitch accent in both areas. In other words, they are words that might
be judged as 'foreign' by pitch accent alone and completely neutral words
which give no clue to region, neither from accent nor devoicing. Using these
words, it is possible to compare the extent of the effects on the respondent
judgments between accent patterns and devoicing variability. It is also
possible to compare the extent of effects between the different Tokyo and
Kinki accent patterns. The words containing no devoiceable vowel are listed
in Table 6.

The subjects who pronounced these words and phrases were 14 male
and female speakers of Japanese. Their native dialects were not controlled,
because variation did not show up frequently enough in the data I collected
from Tokyo people prior to this study. An examination of the responses
showed that the hometown origin of the speakers did not influence
judgments in this study. The list of the subjects with their gender and
hometown is shown in Appendix I. They were asked to read the list in a
'natural' way.

Both versions of devoiced and nondevoiced vowels of the same word
were needed to see if the respondents judged them differently. I examined
whether or not the vowels of the recorded pronunciations were devoiced by

using the Signalyze program. It converts the recorded sounds into wave

40

 

 

 

 

154: A15: .38. “.3235:
15: 153 Essa—ohm. £55
4524 15434 .Emvfim 89A 988m. 325$:
gum..— Ag .88 a mo Moan. 3335mm mono?»
EA #5 68:38. 838 3.582
mmm dm 668 823 has“? $53
15E..— .SE .20 Son. .2853
AI..— .EE .3 £88 “cacao so». 333:
15:5 .5234 932?? poem a 53» 93.5%. tugunmss
a; E :33. 583mg
5; H24 mam. 533 “swoon
.533 .534 .9855. 333:8 “Gamma“.
15: 4H,: .Uoumum 233m. 383 fits
a: EU .coam. £333: £983
.3? EEE .Emfiam a. 33353
E E5 .wcucma out. ~35
E E Amfimc 833 .mudxmfimM. 335:3
E E3 .Emfimomamu. {35$
2:9 5 28% 6.5 038. s 283.. 6.5 225

 

 

 

 

 

$32, mEmoumogu o: wfifimucoU mp8? 9.3 mo 63

a was.

41

forms, showing voiced sounds as periodic waves and voiceless sounds as very
attenuated wave forms or flat lines. As Kondo (1997) concluded, devoicing is
not categorial, and I obtained many waveforms that were gradient, but I
chose, for my sample tape, the most devoiced and the most nondevoiced
versions of the same word.

In my data, 41 identical words were pronounced with devoiced and
nondevoiced variants. In addition, 23 words were pronounced either with a
devoiced vowel only or with a nondevoiced vowel only; that is, all the
subjects pronounced the same word with devoiced vowels, or all the subjects
pronounced the same word with nondevoiced vowels. There were 13 words
containing no devoiceable vowel and pronounced in the Tokyo pitch accent,
with the same 13 words in the Kinki pitch accent. Finally, 5 neutral words

were added. Examples of each are given in (6):

(6a) Words pronounced with devoiced and nondevoiced vowels

atafuta 'hurriedly' Speaker] devoiced
atafuta 'hurriedly' Speaker K nondevoiced

(6b) Words pronounced with either a devoiced or a nondevoiced vowel
only

nadeshiko 'a type of pink flower' Speaker K devoiced
michihide 'by high and low tides' Speaker M nondevoiced

(6c) Words with regionally different accents and with no devoiceable vowel

kawari 'replacement' Speaker H Tokyo (LHH)
kawari 'replacement' Speaker G Kinki (HHH)

42

(6d) Neutral words, with no accent difference and no devoiceable vowel

tabun 'probably' Speaker E HLL

The total 136 words and phrases were randomly arranged in terms of
devoicing variation, regionally different accents, neutral words, and speakers,
and a test tape was prepared. A list of the words in the order presented in the
tape is shown in Appendix II.

I coded speaker, speaker's gender, pitch accent pattern, position of the
devoiceable vowel in terms of boundary, position of the devoiceable vowel in
terms of mora counting from the beginning, preceding consonant, identity of
devoiceable vowel, following consonant, and voicing status or type of words
with no devoiceable vowel (Tokyo / Kinki accent or neutral) for each word.
Examples are given in Table 7. In the 'Type' column, 'd' stands for 'devoiced'
and 'n' stands for 'nondevoiced.’ In the same column, 't,’ 'k,‘ and 'x' indicate
types of words with no devoiceable vowel: 't' stands for 'Tokyo pitch accent,‘
'k' stands for 'Kinki pitch accent,‘ and 'x' indicates a neutral word that has no
devoicing variation or accent difference. The words coded as 'd,’ 'n,‘ and 'x'
contain no devoiceable vowel, so they have a '0' in the 'Position' (in terms of
mora counting) column and an 'x' in 'Boundary,‘ 'Precede'(ing consonant),
'Identity' (of devoiceable vowel), and 'Follow'(ing consonant) columns. In
the 'Boundary' column, 'w,‘ 'm,‘ and 'i' stand for 'at a word boundary,‘ 'at a

morpheme boundary,‘ and 'morpheme-internally,’ respectively.

43

 

 

 

 

 

 

 

x x x x x o 15d.“ 2 M msmmfifi
x x x x x 0 mm: m U @833
u x x x x o 1F: m < 5.8an
a n m x E m 1524 2 2 amfixmmo
u U. a c m m SE m _ 33%
at? Bozom . “33> wtoomum .mumucdom common" Emuu< .8950 .86?on 3.83

 

 

 

 

 

 

 

 

 

 

 

Amoafimxmv v.83 Sumo 8w wEUoU

n 2an

44

3.2. Response collection

The test tape was presented to subjects, male and female native speakers of
Japanese. They listened to each of 136 words twice and were asked to judge
whether or not the speaker was from the same region as their own. They
listened while looking at a list of the words written in Japanese (Kanji and
Hiragana). This written list helped the subjects concentrate on making
judgments without taking time to recognize, in particular, unfamiliar words.
They responded on a five-point scale: 0=The speaker is certainly from the
same region, 1=The speaker is probably from same region, 2=Don't know,
3=The speaker is probably not from the same region, and 4=The speaker is
certainly not from the same region.

Demographic data, gender, age, and hometown of the subject were
collected. 'Hometown' was based on the place where the respondents spent
their elementary and junior high school days, assuming that the dialect
learned during those days is the native dialect for them. I chose 'elementary
and junior high school days' because children seem to eventually learn the
dialect of the region they live in, rather than their parents' dialects, to
communicate with their peers. Kitamura (1952) conducted research on about
500 children who moved from Tokyo to Shirakawa, Fukushima prefecture,
and reports that those who moved before the age of six learned the Shirakawa
dialect while those who moved at a later age did not learn it completely. I

eliminated respondents who lived in two places during this crucial period.

45

Besides demographic questions, it was asked whether or not the respondents
could personally identify any of the voices on the tape, and respondents who
did so correctly were eliminated because they may have known the
hometowns of those speakers, and that may have affected their judgments.

I defined 'Tokyo' as the Tokyo metropolitan area, the area which
people would commute for work in downtown Tokyo from, and 'Kinki' as 6
prefectures by governmental region. 'Kinki' used in Japanese dialectology is
not identical with the 6 prefectures (Hirayama et. a1, 1982), but all the Kinki
respondents in my data are also from Kinki by dialect definition.

I collected analyzable responses from 206 subjects, and after eliminating
those who do not fill the conditions stated above, I used the responses by 47
from the Tokyo Metropolitan area and by 73 from Kinki for the analysis in

this paper.

3.3. Preparation for analysis
I eliminated tokens to the stimuli that were not perfectly pronounced as I
intended in the segment after the devoiceable vowel. I also eliminated other
invalid tokens (e.g., no response, marking two responses to one token),
leaving 6,344 tokens judged by Tokyo respondents and 9,785 tokens judged by
Kinki respondents.

With these data, first I flipped the ratings for words of nondevoiced

versions of devoiceable vowels and words with the Kinki accent in responses

46

by Tokyo people, and the ratings for words of devoiced version of devoiceable
vowels and words with the Tokyo accent in responses by Kinki people,
changing 0 to 4, 1 to 3, 3 to 1 and 4 to 0. This reflects the decision to use as
expectations those responses which indicated that a speaker is judged as a
member of the same region as the region the respondent belongs to
depending on the devoicing variation and the pitch accent patterns. The
responses expected from Tokyo people were that a speaker who devoices
devoiceable vowels or uses the Tokyo pitch accent pattern would be judged as
a Tokyo speaker, and that a speaker who nondevoices them or uses the non-
Tokyo pitch accent pattern would be judged as a non-Tokyo speaker. In order
to find out whether or not the data reflect that respondents judge the speakers
as expected depending on the experimental conditions (devoicing,
nondevoicing, Tokyo pitch accent, or Kinki pitch accent), I needed to have a
single scale so that 'certainly from the same region (from Tokyo)‘ for the
tokens with devoicing and the Tokyo accent and 'certainly not from the same
region (not from Tokyo)’ for the tokens with nondevoicing and Kinki accent
would have the same value. I needed to have these four types of tokens
show the same value and be placed at one end of the scale, and at the other
end I needed to have 'certainly from Tokyo‘ for the tokens with
nondevoicing and the Kinki accent and 'certainly not from Tokyo' for the
tokens with devoicing and the Tokyo accent. For example, with the two data

sets given below, (7a) fits the hypothesis and (7b) does not.

47

(7a)

(713)

Word A
Word B
Word C
Word D

Word E
Word F
Word G
Word H

devoiced

nondevoiced
Tokyo accent
Kinki accent

devoiced

nondevoiced
Tokyo accent
Kinki accent

score 0 (certainly from Tokyo)
score 4 (certainly from non-Tokyo)
score 0 (certainly from Tokyo)
score 4 (certainly from non-Tokyo)

score 4 (certainly from non-Tokyo)
score 0 (certainly from Tokyo)
score 4 (certainly from non-Tokyo)
score 0 (certainly from Tokyo)

With this coding, the experiment cannot be treated statistically because it does

not have a single scale, which would require that the scores of words A, B, C,

and D have the same value on one end and the scores of words E, F, G, and H

the same value on the other end. If the scores are flipped as in (8) below

(italic scores are flipped), the results are shown on a single scale, and (8a) will

show that the expectations on responses are right by the score 0 while (8b) will

show that they are wrong by the score 4 in contrast to 0. After flipping these

ratings, 0 and 1 are judgments that confirm my expectations, and 3 and 4 are

judgments that contradict them.

(88)

(819)

Word A
Word B
Word C
Word D

Word E
Word F
Word G
Word H

devoiced

nondevoiced
Tokyo accent
Kinki accent

devoiced

nondevoiced
Tokyo accent
Kinki accent

score 0 (judged as expected, Tokyo)
score 0 (judged as expected, non-Tokyo)
score 0 (judged as expected, Tokyo)
score 0 (judged as expected, non-Tokyo)

score 4 (judged not as expected, non-Tokyo)
score 4 (judged not as expected, Tokyo)
score 4 (judged not as expected, non-Tokyo)
score 4 (judged not as expected, Tokyo)

48

Similar procedures were applied to the responses by Kinki people. For Kinki
people, 'the same region' as their own is Kinki, and the expectation in
responses from Kinki people is that a speaker who devoices devoiceable
vowels or uses the Tokyo pitch accent pattern is judged as a non-Kinki
speaker, and that a speaker who nondevoices them or uses the Kinki pitch
accent pattern is judged as a Kinki speaker. Examples of their raw data sets
can be illustrated as in (9a), confirming the expectations, and (9b),

contradicting them.

(9a) Word A devoiced score 4 (certainly from non-Kinki)
Word B nondevoiced score 0 (certainly from Kinki)
Word C Tokyo accent score 4 (certainly from non-Kinki)
Word D Kinki accent score 0 (certainly from Kinki)

(9b) Word E devoiced score 0 (certainly from Kinki)
Word F nondevoiced score 4 (certainly from non-Kinki)
Word G Tokyo accent score 0 (certainly from Kinki)
Word H Kinki accent score 4 (certainly from non-Kinki)

In order to compare the responses by Tokyo people and Kinki people, I
needed a scale with the responses confing the expectations placed on one
end with the value 0 and the responses contradicting them placed on the
other with the value 4. That is, I needed a scale so that the tokens judged as
'non-Kinki' with devoicing and the Tokyo accent and as 'Kinki' with
nondevoicing and the Kinki accent would have the value 0, and the tokens

judged as 'Kinki' with devoicing and Tokyo accent and as 'non-Kinki' with

49

nondevoicing and Kinki accent would have the value 4. I can obtain such a
scale by flipping the scores of the tokens with devoiced vowels and Tokyo
accent.

I will present and discuss the results in two tables, one with Tokyo data
and the other with Kinki data, that show frequencies and percentages of each
of 5 ratings in each of the token types. Then, I will present the results in two
tables in which I collapsed the ratings into a three-point scale and eliminated
the response 'don't know.‘ Having two-way results is necessary for the
analysis using the statistics program GoldVarb, which I used in the later
analysis, and as I will show, this collapse did not distort the overall picture.
Specifically, I collapsed O and 1 to O, 2 to 1, and 3 and 4 to 2. After collapsing, O
is a judgment confirming the expectations and 2 is a judgment contradicting
the expectations based on my hypothesis, and 1 is 'don't know,‘ which was

eliminated after looking at the five-way results.

3.4. Analysis

I analyzed the responses that confirm or contradict the expectations to
determine which factor(s) affect judgments on the speaker and how each
factor affects such judgments. Imai (1997) examined preceding and following
consonants to find the promoting and demoting factors for devoicing in

speech production. Using my data, it is possible to compare the same factors

50

in perception. I examined the following phonological and social factors as in

(10) below.

(10) Between devoicing variation and regional accents

Position of a devoiceable vowel in terms of mora count

Position of a devoiceable vowel in terms of morpheme boundaries

Pitch accent pattern

Preceding consonant

Identity of devoiceable vowel

Following consonant

Speaker's identity

Gender of the speaker

Gender of the respondent

Age of the respondent
I used the statistics program GoldVarb not only to determine the significance
of each factor group and the rates of expected responses, but also to examine
the significance of each factor relative to all other factors.

GoldVarb is a multivariate logistic regression program, which
identifies insignificant factor groups by an up and down regression test. After
those groups are excluded, the 'weight' of each factor of each group is
obtained by the logistic regression process. The weight indicates how heavily
the factor promotes (above .5) or demotes (below .5) the results (in this case,
the extent of how strongly respondents' judgments confirm or contradict the
expectations of the hypotheses). Then if factors with similar weights in a

factor group are combined (when phonologically or socially justified), and if

the results of a new run are not significant, those factors can be put together

51

permanently as a new factor. This examination makes it possible to identify
which factor groups and factors within groups have the greatest influence on

respondent judgments.

52

4. Results

With the data prepared for analyses, I first examined the overall frequency
and percentage of each score on the basis of the devoicing variation and
regional accents and then compared the results of responses by Tokyo people
to the ones by Kinki people. Then I analyzed the Tokyo data and the Kinki
data separately to examine how the phonological factors and the social factors
affected their judgments of whether or not the speaker is from the same
region as theirs. I also analyzed the devoiced tokens and the nondevoiced
tokens separately for the Tokyo and Kinki data, respectively, in order to

examine the phonological factors and the social factors.

4.1. Overall results by token type

Overall results are examined by the five token types, that is, tokens with
devoiced vowels, with nondevoiced vowels, with the Tokyo accent, with the
Kinki accent, and neutral tokens, separately, and then by the two factor

groups, that is devoicing variation and regional accents.

4.1.1. Overall results by the five token types
The next three tables illustrate the results by the token types. Table 8 shows
how tokens are tabulated in Tables 9 and 10 after flipping the scores of the

respondents based on the hypotheses and assumptions as explained in the

53

 

 

 

 

 

 

 

 

 

 

 

 

 

80%....“ ~88 65 89a «0: :3me m8mm 9.: 88% ~93sz
:2me 988 9.: 89¢ no: 80%?” ~83 65 89a <m 38M
:3me 68mm 65 89c «08 80%8 «88 05 89a 3896982 EEM
common 688m 65 89a common 688m 9% 89a no: <m 9”on
82mg ~88 65 89a comma ~88 9: 89a Ho: @8880
:3me 683 m5 80¢ “on 80mm?" «88 65 89¢ 3.562
:3me m8mm 65 89c 8W8 «.88 9.: 89¢ Ho: <m 28M
common m8mm 65 89a comma 688 m5 898 .6: 3396—802 05.8.
:8me «89. 9: 89a no: 80%8 «88m 65 89a <m 933.
:2me 68mm 65 89a no: :3me 688 «5 89a 88880
1:83.89 _ b33989 .333on. _ 383.80.
UmuummxmaD Umfiomxm mg Gov—oh “Gmucommom

 

 

 

 

 

8:238 w ~38.

54

previous chapter. In the "Token Type" column, "Devoiced" and
"Nondevoiced" refers to the tokens that have devoiced vowels and
nondevoiced vowels, respectively, while "Tokyo PA" and "Kinki PA" refer to
the tokens that are pronounced in the Tokyo pitch accent and in the Kinki
pitch accent, respectively.

In the top row "Expected" and "Unexpected" indicate whether the
tokens are judged confirming the hypotheses or contradicting the hypotheses,
respectively, assuming that vowel devoicing and the Tokyo pitch accent are
Tokyo features while nondevoicing of vowels and the Kinki pitch accent are
Kinki features. In other words, "Expected" responses by Tokyo people are the
judgment "certainly or probably from the same region (Tokyo)" for the tokens
with devoicing and the Tokyo pitch accent and the judgment "certainly or
probably not from the same region (not Tokyo)" for the tokens with
nondevoicing and the Kinki pitch accent, while "Unexpected" responses by
them are the judgment "certainly or probably from the same region" for the
tokens with nondevoicing and the Kinki pitch accent and the judgment
"certainly or probably not from the same region" for the tokens with
devoicing and the Tokyo pitch accent. "Expected" responses by Kinki people,
on the other hand, are the judgment "certainly or probably from the same
region (Kinki)" for the tokens with nondevoicing and the Kinki pitch accent
and the judgment "certainly or probably not from the same region (not

Kinki)" for the tokens with devoicing and the Tokyo pitch accent, while

55

"Unexpected" responses by them are the judgment "certainly or probably
from the same region" for the tokens with devoicing and the Tokyo pitch
accent and the judgment "certainly or probably not from the same region" for
the tokens with nondevoicing and the Kinki pitch accent.

"Neutral" refers to tokens that do not contain devoiceable vowels and
are pronounced in the same pitch accent in Tokyo and Kinki. In this study,
the neutral tokens are expected to sound like the respondent's own
pronunciation because of the lack of salient clues. Following this expectation,
the neutral tokens judged as "certainly or probably from the same region" are
tabulated in the column "Expected" and the ones judged as "certainly or
probably not from the same region" are tabulated in the column
"Unexpected," regardless of whether the respondent is from Tokyo or Kinki.

Tables 9 and 10 show the overall frequency and percentage of each
rating by devoicing variants and regional accents in the Tokyo data and in the
Kinki data, respectively. The figures in the parentheses are percentages. The
results show that the respondents tend to make judgments confirming the
hypotheses, in other words, they tend to make judgments as expected — that
is, devoiced and Tokyo accent versions tend to be rated as 'Tokyo speaker' by
Tokyo respondents and as 'non-Kinki speaker' by Kinki respondents, while
nondevoiced and Kinki accent versions tend to be rated as 'non-Tokyo
speaker' by Tokyo respondents and as 'Kinki speaker' by Kinki respondents.

This is indicated by the higher percentages in the "Expected" columns than in

56

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

so: $3 33: $2 823 ME: 33: 32 33s $9 $23 88 38.
so: mom :3 Va 83 In 83.: B 839 a: $33.12 3.3st
so: 95 :3. Km 33 Em $3 Km 83% SN €me mam .2 25M
83 8% $3 Sm 83$ owe $03 Rm 3.3 E 633 85 885302
so: as :3 Va. :3 VS a; vow 83d emu 633 one .8 938.
8c: 8% no.3 Rm 833 3: 3.3 mmw Rama «8 $3: 3 32969
.3838? .33395. 333.95. .3838?
:35. Umuuon meD 15905 3.809.. Umfimmxm our? 830.5.
35% 89:. 8x8. .3 £838 $296 3 same
so: 3% 83: man 33: E: a; v mwm awed 82 some mg Boa
so: mm 83 v a saw: on $35 mm 9.0di E 83 mw 3:82
so: :0 Es VS 63 v3. GS. v3 5.3 mm 825 «a. .3 25M
so: S: 63¢ wwm 3..va man a? v «8 figs 8... sass mwm Buszncoz
so: :0 3.2.: kw gm: .3 a; V8 3.3 3: 329 m2 5 8on
so: RR and v SN 63: oh. and; «.3 news own 839 «mm 8233
.3883? 33395. .33395. .3838?
Each @3qu meD :3OGM ranch? @38me mega. GmXOH.

 

 

 

 

 

 

3350.5 395. 5on 3.. $885 =8m>0

m 2an

57

the "Unexpected" columns, except for the Kinki devoiced tokens. In the
Kinki devoiced tokens, too, the total percentage of "Expected" columns
(18.37% + 23.57% = 41.94%) is higher than that of "Unexpected" columns
(24.43% + 14.02 = 38.45%), showing similar tendencies. These general
tendencies mean devoicing variation contributes to judgments of whether
the speaker is from the same region just as the Tokyo and Kinki accent
patterns do, although, as noted above, devoicing and nondevoicing are
allophonic phenomena and many people are not aware of them in detail.
Tables 9 and 10 also show some differences between the Tokyo results
and the Kinki results and between token types. The results of regional accents
are more distinct than those of devoiced and nondevoiced versions for
"Expected" and "Unexpected" in both Tokyo and Kinki data, and the results of
devoiced and nondevoiced versions are much more distinct in the Tokyo
data than in the Kinki data. In the bottom rows in both tables, the percentage
of the expected response with respondents' confidence 'certainly' in the
leftmost column is highest and the percentages decrease toward the rightmost
column, skipping the 'Don't Know' (35.04, 26.83, 16.41, and 12.53 in Tokyo,
and 27.39, 24.18, 19.19, and 12.92 in Kinki). In the Tokyo data (Table 9), the
results of each token type generally follow the decreasing pattern as in the
'Total' row, and the decreasing in regional accents and neutral tokens is

sharper than in devoiced and nondevoiced tokens.

58

In the Kinki data (Table 10), the results of the tokens of regional accents
and neutral tokens show decreasing patterns similar to the Tokyo results.
The results of devoiced and nondevoiced tokens show higher percentages in
both expected and unexpected 'probably' responses and lower percentages in
both 'certainly' responses. Devoiced tokens are unexpectedly rated as
'probably from Kinki' with the highest percentage (24.43%). 'Don't Know'
responses in devoiced and nondevoiced tokens are relatively high compared
to other token types (and 'Don't Know' in all token types are rated relatively
higher compared to the Tokyo results as well), and the percentages of all 5
responses in devoiced and nondevoiced are mostly in the range of 20% plus
or minus 5%. This is a small range compared to the other token types in the
Kinki results and all types in the Tokyo data. Nonetheless, tendencies similar
to the ones in results for regional accents are indicated in the devoiced and
nondevoiced tokens in that more tokens are rated expectedly than
unexpectedly.

Tables 9 and 10 also show that the Kinki pitch accent contributes to the
expected judgment more strongly than the Tokyo accent does, both in the
Tokyo and Kinki data. In particular, the 'Certainly' expected response in the
Tokyo data is very high (70.70%) and the difference between 'Certainly' and
'Probably' is very large. Among devoicing variants, on the other hand,
devoiced tokens contribute to the expected judgment more than

nondevoicing does in the Tokyo data, and the Kinki data show the opposite

59

results. In other words, the presumed Kinki features (nondevoicing and the
Kinki accent) contribute more in the Kinki data, while in the Tokyo data, the
non-Tokyo feature contributes more in pitch accent pattern, but the Tokyo
feature contributes more in devoicing variation. I will offer explanations for
these results in the discussion chapter.

The results of neutral tokens should also be noted. In my hypothesis,
neutral tokens were expected to be judged as 'from the same region,‘ that is,
'from Tokyo' by Tokyo respondents and 'from Kinki' by Kinki respondents,
due to the lack of any salient clue, and, in fact, respondents from both areas
tended to make judgments as expected.

Tables 11 and 12 show the two-way results by token types for the Tokyo
and Kinki data, respectively. They are obtained by collapsing the frequency
and percentages of 'Certainly' and 'Probably' expected responses into one
column, collapsing unexpected responses similarly into another column, and
eliminating 'Don't Know' responses.

Tables 11 and 12 preserve the general tendencies observed in the five-
way results in Tables 9 and 10, although the distribution between ’Certainly'
and 'Probably' responses and that of ’Don't Know' responses are not visible.
Tables 11 and 12 still show that the respondents tend to make judgements as
expected — devoiced and Tokyo accent as Tokyo or non-Kinki, and

nondevoiced and Kinki accent as non-Tokyo or Kinki.

6O

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 11 Overall two-way results by token types (Tokyo)
Token Type Expected Unexpfcted Total
Devoiced 1712 (70.11) 730 (29.89) 2442 (100)
Tokyo PA 377 (67.20) 184 (32.80) 561 (100)
Nondevoiced 1174 (59.90) 786 (40.10) 1960 (100)
Kinki PA 507 (86.22) 81 (13.78) 588 (100)
Neutral 155 (73.81) 55 (26.19) 210 (100)
Total 3925 (68.13) 1836 (31.87) 5761 (100)
Table 12 Overall two-way results by token types (Kinki)
Token Type Expected Unexpected Total
Devoiced 1765 (52.17) 1618 (47.83) 3383 (100)
Tokyo PA 724 (86.84) 131 (15.32) 855 (100)
Nondevoiced 1505 (54.69) 1247 (45.31) 2752 (100)
Kinki PA 777 (87.60) 110 (12.40) 887 (100)
Neutral 275 (88.42) 36 (11.58) 311 (100)
Total 5046 (61.63) 3142 (38.37) 8188 (100)

 

 

 

 

 

 

 

Tables 11 and 12 also preserve some difference between the Tokyo
results and the Kinki results and between token types. The results of regional
accents are more distinct than those of devoiced and nondevoiced versions in
both the Tokyo and Kinki data, and the results of devoiced and nondevoiced
versions in the Kinki data show tendencies similar to other tokens but more

weakly, as shown in the percentages close to 50% in both expected and

61

unexpected responses. Tables 11 and 12 also preserve the different token types
that contribute to expected responses; the presumed Kinki features contribute
more in the Kinki data, while in the Tokyo data the non-Tokyo feature
contributes more in regional accents, but the Tokyo feature contributes more
in devoicing variation.

Thus, it seems reasonable to use the collapsed tables for further analysis
to examine how phonological factors and social factors affect the respondents'
judgments, knowing that they preserve important tendencies for further

analysis, although the distribution of responses in five ratings is invisible.

4.1.2. Overall results by devoicing variation and regional accents

The results shown in the previous section suggest that the pitch accent
pattern contributes to expected judgments more strongly than the devoicing
variation does. Tables 13 and 14 show the results by the two factor groups

with the Tokyo data and the Kinki data, respectively.

 

 

 

 

Table 13 Results by regional accents and devoicing variation (Tokyo)
Token Type Expected Unexpected Total
Devoicing / Nondev. 2886 (65.56) 1516 (34.44) 4402 (100)
Tokyo/Kinki PA 884 (76.94) 265 (23.06) 1149 (100)
Total 3770 (67.92) 1781 (32.08) 5551 (100)

 

 

 

 

 

 

62

Table 14 Results by regional accents and devoicing variation (Kinki)

 

 

 

 

 

Token Type Expected Unexpected Total
Tokyo/Kinki PA 1501 (86.17) 241 (13.83) 1742 (100)
Devoicing/Nondev. 3270 (53.30) 2865 (46.70) 6135 (100)
Total 4771 (60.57) 3106 (39.43) 7877 (100)

 

 

 

 

 

The Chi-square probabilities of Tables 13 and 14 are 0.000, but it is clear that
the differences of pitch accent patterns affect judgments more powerfully than
the devoicing factor.

The following analyses focus on what phonological and social factors
do or do not affect such judgments when the respondent hears tokens with

devoicing and nondevoicing of a devoiceable vowel.

4.2. Insignificant factors for overall results
Before analyzing factors that affect judgments and comparing the results
between Tokyo respondents and Kinki respondents, it is necessary to find out

which factors are not significant.

4.2.1. Insignificant factor groups

The factor groups that were examined in this study, listed in (10) in the

previous chapter (in 3.4.), are repeated here in (11) for convenience:

63

(11) Between devoicing variation and regional accents
Position of a devoiceable vowel in terms of mora count
Position of a devoiceable vowel in terms of morpheme boundaries
Pitch accent pattern
Preceding consonant
Identity of devoiceable vowel
Voicing status (devoiced or nondevoiced) of devoiceable vowel
Following consonant
Speaker's identity
Gender of the speaker
Gender of the respondent
Age of the respondent

The results by devoicing variation and regional accents were shown in the
previous section (4.1.2). Insignificant factor groups are found by running the
up-and-down regression test in GoldVarb (the logistic regression test used to

analyze these data). The insignificant factor groups in the Tokyo data and in

the Kinki data are listed in (12).

(12a) The insignificant factor groups in the Tokyo data
Position of a devoiceable vowel in terms of mora count
Pitch accent pattern
Identity of devoiceable vowel
Gender of the speaker
(12b) The insignificant factor groups in the Kinki data
Identity of devoiceable vowel
Gender of the respondent

After the initial factor groups were selected for this study, it became

apparent that it would also be necessary to examine whether some differences

64

not focused on in this study were significant. There are two such factors that
were not singled out for this study but could yield significant results. One is
the differentiation between devoicing and nondevoicing at the end of the
stimulus, and the other is the length of the word. I examined these factors in
the data of the Tokyo and Kinki respondents separately, as reported in the

following sections (4.2.2. and 4.2.3.).

4.2.2. Devoicing and nondevoicing at the end of the stimulus
First, I examined whether or not the words that have devoiceable vowels at
the end of the stimulus, which I did not eliminate completely, should be
included for the analysis. The two words pachipachi and pakupaku were
pronounced with either devoiced or nondevoiced vowels at the end of the
word as well as internally. The significance of differentiating the tokens with
one devoiced vowel (between consonants) only and the tokens with two
devoiced vowels was determined by running the logistic regression test of
GoldVarb before and after the combining these two types of tokens.

The difference of log likelihood times two, which is to be compared to
the significance level in a chi-square test, is 2.572 for the Tokyo data, and 0.254
for the Kinki data. Both are smaller than 3.84, which is the significant level at
one degree of freedom. This means that devoicing and nondevoicing at the
end of the stimulus does not make a significant contribution to judgments

about the speaker as examined in this study and is ignored in what follows.

65

These results were determined before the overall results reported in
the previous section, and these tokens with two devoicings were included in

the factor 'devoiced' in the overall results.

4.2.3. Words of different length and accent patterns

The accent pattern of the stimulus which contains a devoiceable vowel in the
raw data was either HLL, HLLL, HLLH, or LHLL. HLL is a three-mora word,
such as dékuha 'finish reading.‘ The others are the beginnings of words and
phrases containing four or more moras. For example, inchikina 'fake' has
five moras but is coded as HLLL, and osékihan 'red bean rice' also has five
moras but is coded as LHLL. HLLH occurs when two words are combined as
in katsu#tsumetai 'also cold' (shown in the previous section 3.1). It was
already determined for the Tokyo data that the accent pattern was not a
significant factor group (see (12a) above). That means combining HLL, HLLL,
and HLLH and even LHLL does not make significant difference. In addition,
the log likelihood difference times two when combining HLL, HLLL, and
HLLH for the Kinki data is 4.59, which is smaller than 5.99, the significance
level at two degrees of freedom. With the Kinki data, therefore, the tokens of
HLL, HLLL, and HLLH can be combined as one new factor group (HLL), and
compared with the tokens of LHLL to examine the different effects between

the tokens starting with a high pitch and the tokens starting with a low pitch.

66

4.3. Overall results by phonological and social factors

After discarding insignificant factor groups, I analyzed the effects of each
factor group on judgments about the speaker. I eliminated the "Don’t know"
responses to focus the comparison on expected responses and unexpected
responses more sharply. I did not take the factor group of the speaker's
identity into account either, in order to maintain statistically more
meaningful results, because the number of speakers was only fourteen, and
the tabular investigations did not show particularly interesting effects except

for one or two respondents (discussed below).

4.3.1. Position of devoiceable vowels in terms of mora count

The position of a devoiceable vowel is either in the second mora (in HLL) or
the third mora (in HLL or LHL) from the beginning of the stimulus. This was
already found to be an insignificant factor group for the Tokyo data in the
initial run, but it is significant for the Kinki data. Table 15 shows the final
results by position for the Kinki data, after removing the insignificant factor
groups and repeating the run with some combined factors (explained more
fully below). In GoldVarb, the "weight" indicates how heavily a factor
promotes (above .5) or demotes (below .5) the likelihood of the dependent
variable, in this case "expected response." It is clear that for Kinki
respondents voicing variability promotes expected judgments when that

variation occurs in the second mora.

67

Table 15 Final results by mora count (Kinki)

 

 

 

 

 

Position Weight
2nd mora 0.532
3rd mora 0.474

 

 

4.3.2. Position of devoiceable vowels in terms of boundary types
A devoiceable vowel is either in a morpheme internal position or at a
morpheme or word boundary. Tables 16 and 17 show the initial results for

the Tokyo data and the Kinki data, respectively.

Table 16 First results by boundary type (Tokyo)

 

 

 

 

 

 

Position Weight
Word boundary 0.608
Morpheme internally 0.571
Morpheme boundary 0.464

 

Table 17 First results by boundary type (Kinki)

 

 

 

 

 

 

Position Weight
Word boundary 0.636
Morpheme internally 0.535
Morpheme boundary 0.477

 

 

 

The results in Tables 16 and 17 are very similar. Positions at the word
boundary and morpheme internally are promoters of the expected response,
with the position at the word boundary being the strongest promoter, while
the morpheme boundary is a demoter. For the Tokyo data, the tokens with
devoiceable vowels at the word boundary or morpheme internal position can
be combined as a new category, identifying this new category as that of a
devoiceable vowel at the outermost or innermost in the structure of the
word. The results of combining them do not create a statistically significant
difference (by means of the log-likelihood test described above). With the
Kinki data, on the other hand, no combining of factors is allowed because the
differences of the weights between any two are too large. The final results for

the Tokyo data are shown in Table 18.

Table 18 Final results by boundary type (Tokyo)

 

 

 

 

 

 

Position Weight
Innermost/ outermost 0.571
Morpheme boundary 0.466

 

4.3.3. Pitch accent pattern
The pitch accent pattern is either HLL or LHL at the beginning of the

stimulus. It was already found to be an insignificant factor group for the

69

Tokyo data. Table 19 shows the final results by accent pattern for the Kinki

data, where it is clear that HLL promotes expected judgments.

Table 19 Final results by accent pattern (Kinki)

 

 

 

Accent Weight
HLL 0.511
LHL 0.439

 

 

 

 

4.3.4. Preceding consonant
There are 8 possible consonants that precede a devoiceable vowel. Table 20

shows the first results by preceding consonant for the Tokyo data.

Table 20 First results by preceding consonant (Tokyo)

 

 

 

 

 

 

 

 

 

 

Consonant Place of art. Manner of art. Weight

k velar -continuant 0.607
tI palatal icontinuant 0.544
s alveolar +continuant 0.495
4) labial +continuant 0.440

palatal +continuant 0.416
ts alveolar icontinuant 0.411
p labial -continuant 0.401
I prepalatal +continuant 0.396

 

 

 

 

 

7O

There are two ways to generalize on this table: by looking at place of
articulation or manner of articulation.

Looking at manner of articulation, both promoting consonants have
the {-continuant} feature, and all the demoting consonants except for [p] have
[+continuant] .

On the basis of place, the promoting consonants [k] and [t[] are palatal
and velar consonants; that is the consonants whose places of articulation are
more back. Among the 6 demoting consonants [s], [4)], [9], [ts], [p], and [I], only
[9] belongs to the same 'back' consonant group as the promoting consonants.
The others are labial, alveolar, and prepalatal — that is, consonants whose
place of articulation is more forward (describing [I] as prepalatal follows Vance
(1987: 20) and Bloch (1970:131)).

It seems reasonable to interpret these results as an interaction of the
place and manner of articulation. Backness (velar or palatal) and
{-continuant} promote expected judgments, while frontness (labial, alveolar,
or prepalatal) and [+continuant] demote them, but place is more salient than
manner. [k] and [t[] are back consonants and have {-continuant}, so they are
undoubtedly promoters, and [s], [(1)], and [l] are front consonants and
[+continuant], so they are undoubtedly demoters. [ts] and [p] have
{-continuant}, but their frontness cause them to be demoters. [c] is palatal and
[+continuant], but is a demoter for some reason which will require further

interpretation. Table 21 shows the final results for the preceding consonants

71

after they were regrouped by the log-likelihood statistical procedure outlined

above.

 

 

 

 

Table 21 Final results by preceding consonant (Tokyo)
Consonant Place of Articulation Weight
k, t! Back 0.592
S, (I), t5, [3,] Front 0.422
9 Back 0.419

 

 

 

 

I will discuss the behavior of [c] in the following section where I look at

separate results for devoicing tokens and nondevoicing tokens.

The results with the Kinki data are different. Table 22 shows the first

results by the preceding consonant for the Kinki data.

First results by preceding consonant (Kinki)

 

 

 

 

 

 

 

 

 

 

 

 

Table 22
Consonant Place of art. Manner of art. Weight

p labial -continuant 0.636

velar -continuant 0.568
ts alveolar tcontinuant 0.514
I prepalatal +continuant 0.477
c palatal +continuant 0.461
t] palatal :tcontinuant 0.394
s alveolar +continuant 0.348
(I) labial +continuant 0.335

 

 

72

 

 

In this case, it seems that manner of articulation plays a more important role
than place. It seems reasonable to say that {-continuant} is generally a
promoting feature and [+continuant] is a demoting feature. The only
exception is [t]], which has the feature {-continuant} but nevertheless turns
out to be a demoter. Table 23 shows the final combined results by preceding

consonant for the Kinki data.

Table 23 Final results by preceding consonant (Kinki)

 

 

 

 

 

 

 

 

 

 

Consonant Manner of Art. Weight
p -continuant 0.638
k, ts -/ :continuant 0.557
I, c +continuant 0.470
t[ :tcontinuant 0.389
s, (I) +continuant 0.344

 

Combining [p] with [k] and [ts] turns out to be significant because the weights
of [p] and the weights of [k] and [ts] are too different, and so does combining [I]

and [9] with [S] and [(M-

4.3.5. Following consonant
There are ten possible consonants that follow a devoiceable vowel. Tables 24
and 25 are the results by following consonant for the Tokyo and Kinki data,

respectively.

73

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 24 Final results by following consonant (Tokyo)
Consonant Place of Art. Manner of Art. Weight
t alveolar -continuant 0.647
I prepalatal +continuant 0.537
s alveolar +continuant 0.519
k velar -continuant 0.498
p labial -continuant 0.496
h glottal +continuant 0.426
ts alveolar :tcontinuant 0.420
c palatal +continuant 0.413
t] palatal icontinuant 0.321
(I) labial +continuant 0.244
Table 25 Final results by following consonant (Kinki)
Consonant Place of Art. Manner of Art. Weight

k velar -continuant 0.638
t alveolar -continuant 0.569
h glottal +continuant 0.562
s alveolar +continuant 0.520
I prepalatal +continuant 0.501
t] palatal icontinuant 0.466
p labial -continuant 0.377
c palatal +continuant 0.367
(I) labial +continuant 0.341
ts alveolar :tcontinuant 0.283

 

 

 

 

74

There is such considerable variation among these consonants that there does

 

 

not seem to be any reasonable linguistic basis on which to put factors together.
There are stops and continuants and front and back consonants at both ends
of the spectrum. The results by following consonant with the devoiced
tokens and with the nondevoiced tokens separated will be shown in sections
4.4. and 4.5. in hopes that this more detailed analysis will reveal some
interesting patterning. It is perhaps worth noting here, however, that some
consonants are rather strong demoters of expected responses, much more so

than any for preceding consonants.

4.3.6. Identity of devoiceable vowels
The devoiceable vowels I selected for this study are / i / and /u/ . It is an
insignificant factor group for both the Tokyo data and the Kinki data as

determined by the up and down regression run.

4.3.7. Voicing of devoiceable vowels

The frequencies and percentages of expected and unexpected judgments about
the speaker by the token type were shown in Tables 9, 10, 11, 12, 13, and 14 in
section 4.1. They show the results by devoicing variants and by Tokyo and
Kinki accent patterns. These variants (voiced and devoiced) are shown in
Tables 26 and 27 for the Tokyo data and the Kinki data, respectively, within

this logistic regression format.

75

Table 26 Final results by voicing of the vowel (Tokyo)

 

 

 

 

 

 

Voicing Weight
Devoiced 0.552
Nondevoiced 0.436

 

Table 27 Final results by voicing of the vowel (Kinki)

 

 

 

Voicing Weight
Nondevoiced 0.521
Devoiced 0.483

 

 

 

 

In the Tokyo data, a devoiced vowel contributes to the expected judgments

more than a nondevoiced vowel does, while in the Kinki data the results are

opposite.

4.3.8. Gender of the speaker

There are seven male speakers and seven female speakers on the tape.
Gender of the speaker was found to be an insignificant factor for the Tokyo
data in the initial run. The final results with the Kinki data are shown in

Table 28 below.

76

Table 28 Final results by gender of the speaker (Kinki)

 

 

 

 

 

Gender Weight
Male 0.521
Female 0.479

 

Male voices contribute more to the expected judgments than female voices do

for the Kinki data.

4.3.9. Gender of the respondent
Gender of the respondent was found to be an insignificant factor group for the
Kinki data in initial run. The final results for the Tokyo data are shown in

Table 29, in which female respondents are shown to be more likely to make

expected responses.

Table 29 Final results by gender of the respondent (Tokyo)

 

 

 

 

 

Gender Weight
Female 0.524
Male 0.466

 

4.3.10. Age of the respondent

The Tokyo respondents were from 18 to 49 years old, and the Kinki

77

 

 

respondents were from 18 to 59. Tables 30 and 31 show the first results by age

of respondent with the Tokyo data and with the Kinki data, respectively.

Table 30 First results by age of the respondent (Tokyo)

 

 

 

 

 

 

 

 

Age group Weight
18-22 0.467
23-29 0.512
30-39 0.527
4049 0.477

 

Table 31 First results by age of the respondent (Kinki)

 

 

 

 

 

 

 

 

Age group Weight
18—22 0.467
23-29 0.532
30-39 0.489
40-49 0.503
50-59 0.514

 

 

For the Tokyo data, the youngest and the oldest groups, who made expected
judgments more often, can be put together, as can the two groups in the
middle, a typical "age-graded" sociolinguistic pattern. For the Kinki data, on
the other hand, all age groups except for the youngest may be combined. The
final results for the Tokyo and the Kinki data are shown in Tables 32 and 33,

respectively, after combining the age groups.

78

Table 32

Final results by age of the respondent (Tokyo)

 

 

 

 

 

Age group Weight
23-29, 30-39 (middle) 0.517
18-22, 40-49 (youngest and oldest) 0.469

 

 

 

 

Table 33 Final results by age of the respondent (Kinki)
Age group Weight
23-29, 30-39, 40—49, 50-59 (middle and oldest) 0.506
18—22 (youngest) 0.468

 

 

 

In the next two sections, I will show the results for devoiced vowels
and nondevoiced vowels separately. Since devoicing variation is a significant
factor, there may be different tendencies among which factors promote and

demote expected judgments for the devoiced and nondevoiced tokens.

4.4. Results for Tokyo devoiced data and Tokyo nondevoiced data

The insignificant factor groups found in the first run of the up-and-down
regression test, as done for the overall data, are accent and gender of the
respondent for the devoiced data. Number of devoicing (within the word
only or within the word and at the end of the word), which was explained in
the section 4.2.2. and turned to be insignificant with all the data, is also

insignificant. The insignificant factor groups for the nondevoiced data are

79

 

 

identity of the vowel and differentiation among three accent patterns starting
with a high pitch (HLL, HLLL, and HLLH). Number of devoicing is irrelevant
with nondevoiced data. The results for each factor group are shown in the

following sections.
4.4.1. Position of devoiceable vowels in terms of mora count
Tables 34 and 35 below show the final results for the devoiced and

nondevoiced data, respectively.

Table 34 Final results by mora count (Tokyo, devoiced)

 

 

 

 

 

 

Position Weight
3rd mora 0.567
2nd mora 0.409

 

Table 35 Final results by mora count (Tokyo, nondevoiced)

 

 

 

 

 

 

Position Weight
2nd mora 0.605
3rd mora 0.398

 

These results show the 2nd mora and 3rd mora positions have opposite

effects on expected judgments for devoiced and nondevoiced tokens.

80

4.4.2. Position of devoiceable vowels in terms of boundary types

The outermost and innermost positions among the boundary in both
devoiced and nondevoiced data are promoters, as in the overall results. The
nondevoiced data also show similar results when combining the outermost
and innermost positions. Interestingly, with the devoiced data, however, the
weight of the factors of the boundary changed while re-running the logistic
regression test with the factors put together. The final results for the devoiced
data and with the nondevoiced data are shown in Tables 36 and 37,

respectively.

Table 36 Final results by boundary type (Tokyo, devoiced)

 

 

 

 

 

 

Position Weight
Outermost/ innermost 0.621
Morpheme boundary 0.440

 

Table 37 Final results by boundary type (Tokyo, nondevoiced)

 

 

 

 

 

 

Position Weight
Outermost / innermost 0.601
Morpheme boundary 0.455

 

81

4.4.3. Pitch accent pattern

Pitch accent pattern was found to be insignificant in the initial run of the

devoiced data. The final results for the nondevoiced data are shown in Table

38.

Table 38 Final results by accent pattern (Tokyo, nondevoiced)

 

 

 

 

 

Accent Weight
HLL 0.516
LHL 0.378

 

4.4.4. Preceding consonant

The orders and values of factors among preceding consonants except for [t[]
and [c] are very similar to the overall results, both for the devoiced and
nondevoiced data. Tables 39 and 40 show the final results using the same

criteria for combining factors as in the overall results, with the devoiced data

and the nondevoiced data, respectively.

82

 

Table 39 Final results by preceding consonant (Tokyo, devoiced)

 

 

 

 

 

 

 

 

 

Consonant Place of art. Weight
tl Back 0.743
k Back 0.590
c Back 0.545
5, (I), l, p Front 0.406
ts Front 0.300

 

Table 40 Final results by preceding consonant (Tokyo, nondevoiced)

 

 

 

 

 

 

 

 

Consonant Place of art. Weight
k Back 0.655
ti Back 0.496

l, S. (I), p, ts Front 0.428
9 Back 0.214

 

With the devoiced data, all and only back consonants are promoters while all

and only front consonants are demoters. This is perfectly consistent,

compared to the overall results, which have the exception of [c]. With the
nondevoiced data, on the other hand, all front consonants are demoters just

as with the devoiced data, while only [k] remains as a promoter. [t[] and [c],

palatal consonants, are both demoters.

4.4.5. Following consonant

The results by the following consonant with the devoiced data and with the

83

 

 

nondevoiced data turned out to be very different, and those different results
explain the complicated overall results. The first results with feature
information with the devoiced data and with the nondevoiced data are

shown in Tables 41 and 42, respectively.

Table 41 First results by following consonant (Tokyo, devoiced)

 

 

 

 

 

 

 

 

 

 

 

Consonant Manner of art. Weight
g +continuant 0.813
s +continuant 0.693
t -continuant 0.650
I +continuant 0.579
h +continuant 0.541
(IJ +continuant 0.483
p -continuant 0.441
tl i-continuant 0.347
k -continuant 0.283

icontinuant 0.244

 

 

 

 

 

84

Table 42 First results by following consonant (Tokyo, nondevoiced)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Consonant Manner of art. Weight
k -continuant 0.910
p -continuant 0.649
ts icontinuant 0.621
t -continuant 0.550
I +continuant 0.468
s +continuant 0.423
c +continuant 0.326
tl :tcontinuant 0.302
h +continuant 0.220
(I) +continuant 0.077

 

There seems to be a tendency that [+continuant] (in particular without
[-continuant]) are promoters and {-continuant} are demoters for the devoiced
data, but the tendencies are opposite for the nondevoiced data. [t] and [tl],
however, behave differently from other {—continuant} consonants: [t] is a
promoter in both data sets, and [ti] is a demoter in both. [1)] behaves differently
from other [+continuant] consonants, too. The weight of [1)] for devoiced data
is higher than any demoting consonant, and can be joined to the promoting
factor group after combining factors based on appropriate linguistic and
statistical grounds and re-running the logistic regression test. On the other
hand, it is the demoter with the lowest weight for the nondevoiced data.

Tables 43 and 44 show the final results for the devoiced and nondevoiced

85

data, respectively.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 43 Final results by following consonant (Tokyo, devoiced)
Consonant Manner of art. Weight
g +continuant 0.804
s +continuant 0.687
t —continuant 0.652
I +continuant 0.577
h, (I) +continuant 0.517
p -continuant 0.440
tl, k, ts t/ -continuant 0.285
Table 44 Final results by following consonant (Tokyo, nondevoiced)
Consonant Manner of art. Weight
k -continuant 0.909
p, ts, t -/ icontinuant 0.585
I, s +continuant 0.443
h, c +continuant 0.291
t] icontinuant 0.255
(I) +continuant 0.066

 

 

 

 

 

4.4.6. Identity of devoiceable vowels
Identity of the devoiceable vowel was found to be insignificant in the initial

run for the nondevoiced data. Table 45 shows the final results for the

86

devoiced data.

Table 45 Final results by vowel identity (Tokyo, devoiced)

 

 

 

 

 

Vowel Weight
u 0.571
i 0.426

 

 

4.4.7. Gender of the speaker
Tables 46 and 47 show the final results by gender of the speaker for the

devoiced and nondevoiced data, respectively.

Table 46 Final results by gender of the speaker (Tokyo, devoiced)

 

 

 

 

 

Gender Weight
Female 0.542
Male 0.444

 

Table 47 Final results by gender of the speaker (Tokyo, nondevoiced)

 

 

 

 

 

Gender Weight
Female 0.588
Male 0.436

 

87

 

 

4.4.8. Gender of the respondent
Gender of the respondent was found to be insignificant in the initial run for

the devoiced data. Table 48 shows the results for the nondevoiced data.

Table 48 Final results by gender of the respondent (Tokyo, nondevoiced)

 

 

 

 

 

 

Gender Weight
Female 0.556
Male 0.421

 

4.4.9. Age of the respondent
Tables 49 and 50 show the first results with the devoiced data and the

nondevoiced data, respectively.

Table 49 First results by age of the respondent (Tokyo, devoiced)

 

 

 

 

 

 

 

 

Age group Weight
18-22 0.369
23-29 0.601
30-39 0.514
40-49 0.425

 

88

Table 50 First results by age of the respondent (Tokyo, nondevoiced)

 

 

 

 

 

Age group Weight
18-22 0.577
23-29 0.414
30-39 0.543
40-49 0.517

 

 

 

 

The youngest group and the oldest group can be combined as a demoting
group, and the two groups in the middle can also be combined in the
devoiced data as a promoting group. With the nondevoiced data, the group
of 23 to 29 years old can stand alone, and the youngest group and the two
older groups can be combined. That is, the weight of the 23-29 year olds is the
heaviest with the devoiced data and the lightest with the nondevoiced data.
Tables 51 and 52 below show the final results with the devoiced data and with

the nondevoiced data, respectively.

Table 51 Final results by age of the respondent (T okyo, devoiced)

 

 

 

 

 

 

 

Age group Weight

18-22, 40-49 0.382
23—29 0.601
30-39 0.513

 

89

Table 52 Final results by age of the respondent (Tokyo, nondevoiced)

 

 

 

 

 

 

Age group Weight
18-22, 30-39, 40-49 0.555
23-29 0.414

 

4.5. Results for Kinki devoiced data and Kinki nondevoiced data

The insignificant factor groups found by running the up-and-down regression
test for the devoiced data were accent, identity of the vowel, and gender of the
speaker. Number of devoiced vowels (within the word only or within the
word and at the end of the word), which was explained in section 4.2.2. and
proved to be insignificant with all the data, is also insignificant. The
insignificant factor groups in the nondevoiced data were accent, position of a
devoiceable vowel — both in terms of mora count and in terms of the
boundary type, identity of the devoiceable vowel, and gender of the speaker.
Number of devoiced vowels is irrelevant for the nondevoiced data. The

results by each significant factor group are shown in the following sections.

4.5.1. Position of devoiceable vowels in terms of mora count
The position of a devoiceable vowel in terms of mora count is insignificant
for the nondevoiced data. With the devoiced data, the weights of the second

and third mora in the first results are 0.545 and 0.468 respectively, but this

90

factor group turned out to be insignificant on repeating the statistical

procedures.

4.5.2. Position of devoiceable vowels in terms of boundary types
The position of a devoiceable vowel in terms of the morpheme boundary is
insignificant with the nondevoiced data. Table 53 below shows the final

results for the devoiced data.

Table 53 Final results by boundary type (Kinki, devoiced)

 

 

 

 

 

 

Position Weight
Word boundary 0.791
Morpheme internal, morpheme boundary 0.487

 

Morpheme internal position is a promoter in all the Kinki data, whereas it is

a demoter in the devoiced data.

4.5.3. Pitch accent pattern

Accent is insignificant both with the devoiced data and the nondevoiced data.

4.5.4. Preceding consonant
Table 54 shows the first results by the preceding consonant with the devoiced

data.

91

Table 54

First results by preceding consonant (Kinki, devoiced)

 

 

 

 

 

 

 

 

 

 

Consonant Place of art. Manner of art. Weight
p labial -continuant 0.656
k velar -continuant 0.555
ts alveolar tcontinuant 0.514
tl palatal :tcontinuant 0.478
I prepalatal +continuant 0.449
c palatal +continuant 0.430
(I) labial +continuant 0.393
s alveolar +continuant 0.276

 

 

 

 

 

{-continuant} is a promoter and [+continuant] is a demoter, as seen in the
Kinki overall results. The order of the factors is consistent based on sonority:
the two stops are the promoters with heaviest weight; the two affricates

follow them; and the four fricatives are demoters. The final results are

 

 

 

 

 

 

 

 

shown in Table 55.
Table 55 Final results by preceding consonant (Kinki, devoiced)
Consonant Manner of art. Weight
p -continuant 0.654
k, ts, tl -/ tcontinuant 0.528
I, 9, (I) +continuant 0.439
s +continuant 0.299

 

92

 

Table 56 shows the first results for the nondevoiced data.

 

 

 

 

 

 

 

 

 

Table 56 First results by preceding consonant (Kinki, nondevoiced)
Consonant Place of art. Manner of art. Weight

p labial -continuant 0.656
ts alveolar icontinuant 0.588
k velar -continuant 0.546
s alveolar +continuant 0.509

prepalatal +continuant 0.499
c palatal +continuant 0.476
tl palatal icontinuant 0.331
(I) labial +continuant 0.297

 

 

 

 

 

 

The tendencies are not so clear-cut as in the devoiced data, but generally

{-continuant} is a promoter and [+continuant] is a demoter. Table 57 shows

the final results.

 

 

 

 

 

 

 

 

 

Table 57 Final results by preceding consonant (Kinki, nondevoiced)
Consonant Manner of art. Weight
p -continuant 0.653
ts, k :t / -continuant 0.555
s, I, c +continuant 0.495
t! tcontinuant 0.329
(I) +continuant 0.296

 

93

 

4.5.5. Following consonant
Table 58 shows the first results by the following consonant with the devoiced

data.

Table 58 First results by following consonant (Kinki, devoiced)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Consonant Place of art. Manner of art. Weight
g palatal +continuant 0.677
k velar -continuant 0.626
I prepalatal +continuant 0.553
s alveolar +continuant 0.543
t alveolar -continuant 0.516
tl palatal tcontinuant 0.489
(I) labial +continuant 0.408
h glottal +continuant 0.339
ts alveolar icontinuant 0.331
p labial -continuant 0.282

 

Some factors can be combined, but it is hard to find general tendencies with
both phonological and statistical justification. Table 59 below shows the first

results with the nondevoiced data.

94

 

 

 

 

 

 

 

 

 

 

 

Table 59 First results by following consonant (Kinki, nondevoiced)
Consonant Place of art. Manner of art. Weight

h glottal +continuant 0.681
k velar -continuant 0.664
p labial -continuant 0.634
t alveolar -continuant 0.621
s alveolar +continuant 0.477

prepalatal +continuant 0.476
tl palatal icontinuant 0.397
c palatal +continuant 0.314
(I) labial +continuant 0.301
ts alveolar :tcontinuant 0.264

 

 

 

 

 

Stop consonants are all promoters while the two affricates and all the

 

fricatives except for [h] are demoters. That is, it is more consistent to interpret
{-continuant} as a promoter and [+continuant] as a demoter. The final results

are shown in Table 60 below.

 

 

 

 

 

 

 

 

 

 

Table 60 Final results by following consonant (Kinki, nondevoiced)
Consonant Manner of art. Weight

h +continuant 0.679

k, p, t -continuant 0.637

s, l +continuant 0.479

tl icontinuant 0.389

g, d) +continuant 0.302

ts :tcontinuant 0.266

 

95

 

4.5.6. Identity of devoiceable vowels
Identity of devoiceable vowel is insignificant both for the devoiced and

nondevoiced data.

4.5.7. Gender of the speaker
Gender of the speaker is insignificant for both the devoiced data and

nondevoiced data.

4.5.8. Gender of the respondent
Tables 61 and 62 show the final results by the gender of the respondent for the

devoiced data and for the nondevoiced data, respectively.

Table 61 Final results by gender of the respondent (Kinki, devoiced)

 

 

 

Gender Weight
Male 0.543
Female 0.470

 

 

 

 

Table 62 Final results by gender of the respondent (Kinki, nondevoiced)

 

 

 

 

 

 

Gender Weight
Female 0.528
Male 0.459

 

96

4.5.9. Age of the respondent
Tables 63 and 64 show the first results by the age of the respondent for the

devoiced and nondevoiced data, respectively.

Table 63 First results by age of the respondent (Kinki, devoiced)

 

 

 

 

 

 

 

 

Age group Weight
18-22 0.430
23-29 0.551
30-39 0.459
40-49 0.557
50-59 0.543

 

Table 64 First results by age of the respondent (Kinki, nondevoiced)

 

 

 

 

 

 

 

 

Age group Wight
18-22 0.511
23-29 0.516
30-39 0.524
4049 0.435
50-59 0.490

 

 

 

The 50-59 group can be combined with the 40-49 group, and it is comparable to
the Tokyo data, but the final results look very different from the ones for
Tokyo data. With the Kinki data, the weight of the group 23—29 is neither the

heaviest nor lightest. Tables 65 and 66 show the final results.

97

Table 65 Final results by age of the respondent (Kinki, devoiced)

 

 

 

 

 

 

 

 

Age group Weight
18-22 0.431
23-29 0.551
30-39 0.459

4049, 50-59 0.554

 

Table 66 Final results by age of the respondent (Kinki, nondevoiced)

 

 

 

 

 

 

Age group Weight
18-22, 23-29, 30-39 0.519
40-49, 50-59 0.445

 

4.6. Summary of the results

In this study, responses by Tokyo and Kinki people were analyzed to
determine how phonological and social factors affect judgments of whether
or not the speakers on the test tape are from the same region as the
respondent's own. It was expected that a speaker who devoices vowels or
uses the Tokyo accent is judged as 'Tokyo' or 'non-Kinki,‘ while a speaker
who nondevoices vowels or uses the Kinki accent is judged as 'non-Tokyo' or
'Kinki} The results can be summarized as follows.

(a) Devoicing variation contributes to the expected judgments by the

respondents, confirming the hypotheses. A speaker who devoices vowels

98

tends to be judged as 'Tokyo' or 'non-Kinki' and a speaker who nondevoices
them as 'non-Tokyo' or 'Kinki.‘ Voicing variation contributes just as
different accent patterns in Tokyo and Kinki (regional pitch accents) do, but
less distinctly, and the contribution of voicing variation in the Kinki results is
even less distinct than in the Tokyo results. The 'Don't Know' responses in
the devoiced and nondevoiced tokens are relatively high in the Kinki results,
and this causes the distribution of the 5 responses in the five-point scale
within a small range (mostly 20% plus or minus 5%). These general
tendencies, which are indicated in the overall five-way results by token types,
seem to be preserved in the overall two-way results as well.

(b) The Kinki accent contributes to the expected judgments more strongly
than the Tokyo accent in both Tokyo and Kinki data, while devoicing
contributes more strongly than nondevoicing in the Tokyo data but
nondevoicing contributes more strongly than devoicing in the Kinki data. As
for devoicing variation, in other words, presumed local features contribute
more strongly to judgments than outside features. These tendencies,
indicated in the five-way results, are also preserved in the two-way results.

(c) The neutral tokens, containing no devoiceable vowel and pronounced
in the same pitch accent in Tokyo and Kinki, are judged as expected. They
tend to be judged as 'Tokyo' by Tokyo respondents and 'Kinki' as Kinki
respondents. These tendencies are also indicated in the five-way results as

well as in the two-way results.

99

(d) In the Tokyo data, the place of articulation and the manner of
articulation of the consonant that precedes a devoiceable vowel interact, with
more salience of the former. In the Tokyo devoiced data, all the back
consonants (velar and palatal) promote the expected judgments and all the
front consonants (labial, alveolar, and prepalatal) demote them, and, at the
same time, all the promoting consonants except for one have the feature
{-continuant} (a stop and an affricate), and all the demoting consonants except
for one have the feature [+continuant] (fricatives and an affricate). In the
nondevoiced data, [tl] and [c] proved to be demoters, leaving [k] as the only
promoter.

In the Kinki data, in contrast, all the consonants that have
{-continuant} (stops and affricates) are promoters and all the [+continuant]
consonants (fricatives) are demoters in the devoiced tokens, and only [tl]
turned out to be a demoter, leaving the other promoters as promoters and the
other demoters as demoters in the nondevoiced tokens. The Kinki overall
data show similar tendencies as in the nondevoiced data.

(e) The consonants that follow devoiceable vowels have opposite effects
on the judgments with devoiced tokens and nondevoiced tokens in the
Tokyo data. When the vowel is devoiced, all the [+continuant] consonants
(fricatives) promote the expected judgments while all but one that have
{-continuant} (stops and affricates) demote them, and when the vowel is

nondevoiced, all but one that have {-continuant} are promoters while all but

100

one that have [+continuant] are demoters. In other words, just like [t1] and [9]
among preceding consonants, the feature [+continuant] and {-continuant}
generally switch sides when the devoiced and nondevoiced tokens were
examined separately. These opposite results make the overall results look
inconsistent.

In the Kinki data, on the other hand, it is hard to find general
tendencies with both phonological and statistical justification in the devoiced
data as well as in the overall data. In the nondevoiced data, all the
{-continuant} consonants (stops) promote the expected judgments, and all but
one that have [+continuant] (fricatives and affricates) demote them.

(f) Devoiceable vowels at the word boundary or at the morpheme internal
position, in other words, ones at the outermost or innermost in the word
structure, promote the expected judgments in the Tokyo data. In the Kinki
overall data, the results show similar tendencies, but word boundaries are the
only promoters in the devoiced data, and the position of the devoiceable
vowel in terms of morpheme boundary turned out to be an insignificant
factor in the nondevoiced data.

(g) The devoiceable vowel in the third mora from the beginning, rather
than the one in the second mora, contributes to the judgement 'Tokyo'
regardless of the actual devoicing or nondevoicing in the Tokyo data. In the
Kinki data, the position of the devoiceable vowel in terms of mora count is

not a significant factor when devoiced tokens and nondevoiced tokens were

101

examined separately, and the devoiceable vowel in the second mora
contributes to the expected judgments but not strongly in the overall data.
(h) The difference of pitch accent, namely HLL or LHL, generally does not
affect the judgments. It was significant only in the Kinki overall data and
Tokyo nondevoiced data, with .511 and .516 as the promoters' weights.

(i) The identity of the devoiceable vowel generally does not affect the
expected judgments. It was significant only in the Tokyo devoiced data, in
which /u/ is a promoter and /i/ is a demoter.

(j) Female speakers contribute more than male speakers to the expected
judgments in Tokyo devoiced and nondevoiced data when examined
separately, but the gender of the speaker is an insignificant factor group in the
overall data. In the Kinki data, male speakers contribute more in overall
data, but the gender of the speaker is an insignificant factor group when
devoiced and nondevoiced data were examined separately.

(k) Female respondents contribute more than male respondents to the
expected judgments in Tokyo overall data and nondevoiced data. In the
Kinki data, male respondents contribute more in devoiced data while female
respondents do more in nondevoiced data, and the gender of the respondent
turned out to be insignificant in the overall data.

(1) Mild age-grading is indicated in the Tokyo overall data, and sharper
age-grading is indicated in the devoiced data while nondevoiced data shows a

reverse pattern of age-grading; that is, youngest and older age groups are

102

promoters and the young adult group is the demoter. Age-grading is not
indicated in the Kinki data. In the Kinki overall data, the youngest group is
the only demoter and the rest of the age groups are promoters. Kinki
nondevoiced data shows that middle or young adult age groups are
promoters while older groups are demoters. In the Kinki devoiced data,
promoting age groups and demoting age groups show up alternately, which

makes interpretation of patterning difficult.

103

5. Discussion

In Chapter 4, the results of the perception experiment I conducted showed
how devoicing variants affect judgments on whether or not the speaker is
from the same region as the respondent's. The respondents are speakers of
the Tokyo and Kinki dialects. The assumption I made was that devoicing as
well as the Tokyo pitch accent is a Tokyo feature while nondevoicing as well
as the Kinki pitch accent is a Kinki feature. It was expected that a speaker who
devoices vowels or uses the Tokyo pitch accent would be judged as a Tokyo
person by Tokyo respondents and as a non-Kinki person by Kinki
respondents and that a speaker who nondevoices vowels or uses the Kinki
pitch accent would be judged as a non-Tokyo person by Tokyo respondents
and as a Kinki person by Kinki respondents. I called these responses
"expected," and I considered both responses that start with 'certainly' and
'probably' in the same judgment for the same token, "expected."

The results were shown, in the previous chapter, in the tables in which
responses in a five-point scale were tabulated (Tables 9 and 10), and also
shown in tables in which 'Don't Know' responses were eliminated and the
other responses were collapsed into two scales (Tables 11 and 12). The
GoldVarb statistics program was applied after finding that similar tendencies

were indicated in both tables.

104

5.1. Comparison of saliency among token types
In the overall results, in particular in Tables 9 and 10 (five-way results) and in
Tables 11 and 12 (two-way results), it is shown that both Tokyo and Kinki
respondents tend to make judgments confirming the general hypothesis. As
shown in Tables 9 and 10, the tokens pronounced with a Tokyo accent tend to
be rated 'Certainly' or 'Probably' 'Tokyo speaker' by Tokyo respondents
('Certainly' 31.59% + 'Probably' 30.11%) and 'Certainly' or 'Probably' 'non-
Kinki speaker' by Kinki respondents (49.95% + 26.99%), while the tokens with
the Kinki accent tend to be rated 'Certainly' or 'Probably' 'non-Tokyo speaker'
by Tokyo respondents (70.70% + 12.27%) and ‘Certainly' or 'Probably' 'Kinki
speaker' by Kinki respondents (57.79% + 24.60%). These tendencies are
preserved in Tables 11 and 12. Tokyo accents are rated as 'Tokyo' by Tokyo
respondents at 67.20% and as 'non-Kinki' by Kinki respondents at 86.84%,
while Kinki accents are rated as 'non-Tokyo' by Tokyo respondents at 86.22%
and as 'Kinki' by Kinki respondents at 87.60%. It was expected that making
judgments based on pitch accent would be easy because such accent patterns
are phonemic and overtly recognized in the speech communities, although
people do not always describe the differences accurately.

The results by devoicing variation also confirmed the hypothesis.
The five-way results show that the tokens pronounced with devoiced vowels
tend to be rated 'Certainly' or 'Probably' 'Tokyo speaker' by Tokyo

respondents (34.20% + 28.62%) and 'non-Kinki' by Kinki respondents (18.37%

105

+ 23.57%), while the tokens with nondevoiced vowels tend to be rated 'non-
Tokyo speaker' by Tokyo respondents (26.97% + 27.34%) and 'Kinki speaker'
by Kinki respondents (21.95% + 23.24%). The results of the Kinki respondents
are not very distinct, but compared to 'unexpected' responses (24.43 % +
14.02% for devoiced tokens, 20.59% + 16.84% for nondevoiced tokens) they
show tendencies that confirm the hypotheses. A discussion of the indistinct
character of the Kinki devoiced and nondevoiced results follows in this
section.

Devoicing variation, therefore, contributes to the judgments, just as
the differences between the Tokyo and Kinki accent patterns do. As noted
above, however, devoicing and nondevoicing are allophonic phenomena,
and it is interesting, though perhaps not surprising, that such allophonic
phenomena can also be used as criteria in making judgments because most
people are not overtly aware of such differences. A long history of language
attitude studies has provided many examples of speech community
sensitivity (in, for example, rating tasks) to low-level features which are not
available to speaker/heater awareness. In this case, however, it is more likely
than not that the respondent sensitivity to devoicing variation does not
reflect the actual pronunciation differences between Tokyo and Kinki. An
important question, then, is why these two groups of respondents tend to
make judgments based on devoicing variation although they are likely to

devoice vowels at about the same rate, at least in the phonological

106

environments that appeared in the tape for this study (i.e., in low pitched
moras that are not in consecutive devoicing environments).

Before approaching that more general question, I will examine some
differences between the Tokyo and Kinki results. The relative effects of accent
pattern and devoicing variation can be compared for Tokyo (Tables 13) and
Kinki (Table 14). In both, the accent pattern contributes significantly more to
the expected judgments than devoicing variation does, but there are
differences between the two sets of results. In the Kinki results, the difference
of percentages of the expected and unexpected responses is very small for
devoicing variation (53.30% - 46.70% = 6.6%), and this is very different from
the Tokyo results (65.56% - 34.44% = 31.12%). Returning to the five-way tables
(Tables 9 and 10) including 'Don't Know' responses, the differences of
percentages of the five responses are smaller in the Kinki results than in the
Tokyo results. In the Kinki results the range is between 24.43% and 14.02%
for devoiced tokens and between 23.24% and 16.84% for nondevoiced tokens,
while in the Tokyo results the range is between 34.20% and 9.54% for
devoiced tokens and between 27.34% and 9.34% for nondevoiced tokens.

It is reasonable to interpret these results to mean that Kinki
respondents used devoicing much less in their judgments than Tokyo
respondents did, and this may aid in our understanding why Kinki and
Tokyo respondents tend to use the same 'stereotypical' criterion in spite of

the similarity in their actual devoicing rates. Devoicing may not be a very

107

good criterion in making expected judgments for Kinki respondents because
they do devoice vowels, and their response to this criterion is minimal. The
difficulty in making decisions is also reflected in the comparatively high rates
of 'Don't Know' responses in the Kinki results (19.61% for devoiced tokens
and 17.38% for nondevoiced tokens). Tokyo respondents, however, who are
inclined to believe that their speech is "standard" or the national "norm,"
may regard devoicing as the norm and regard any token which reflects such
behavior as 'Tokyo.’ In short, although both responses are based on covert
categories, the Kinki responses appear to be more attuned to actual behavioral
norms and the Tokyo responses appear to be more attuned to ideal norms. I
explore these possible interpretations in more detail as I consider further
differences in the Tokyo and Kinki response patterns. In this particular case,
however, it is possible that when devoicing variation is the only factor, Kinki
people assume that devoicing is a non-Kinki feature based on the higher
nondevoicing rate in other phonological environments (for example, in an
accented mora), but confirmation of that interpretation would require a
different study in which such environments could be compared to the ones
actually presented here.

Tables 11 and 12 (two-way results), as well as Tables 9 and 10 (five-way
results), also show that the distributions of effects are different in the Tokyo
and Kinki results. The Kinki accent contributes to expected judgments more

strongly than the Tokyo accent does, in both Tokyo and Kinki results. On the

108

other hand, devoicing contributes to expected judgments more than
nondevoicing in the Tokyo results, but the Kinki results show the opposite.
In other words, Kinki features consistently contribute more in the Kinki
results, while in the Tokyo results the non-Tokyo feature contributes more in
pitch accent patterns, and the Tokyo feature contributes more in devoicing
variation. In short, both Tokyo and Kinki respondents use the Kinki feature
for pitch accent patterns, and both respondents use their own features (or
stereotypes of them) in devoicing variation, to make judgments.

One reason for these opposite results may lie in the hypothetical
ordering or ranking of criteria in making a judgment. It seems that the non-
Tokyo (Kinki) pitch accent appears to be the most powerful criterion in
reaching a judgment, and the criterion of devoicing is applied only when the
speaker passes such higher-level screening and is not rejected or accepted as
local by Tokyo or Kinki respondents respectively. That is, when Tokyo
respondents hear a Kinki pitch accent, they confidently judge the speaker
'non-Tokyo,‘ and, similarly, when Kinki respondents hear a Kinki pitch
accent, they confidently judge the speaker 'Kinki.’ The powerful criterion, in
particular the 'confidence' by Tokyo respondents to the Kinki accent, could be
interpreted from the strikingly high rate of 'Certainly from non-Tokyo'
(70.70%) in the five-way results (Table 9). On the other hand, when both
respondents hear the Tokyo accent, they may withhold the judgment 'Tokyo'

or 'non-Kinki' and pay attention to other features.

109

It is quite reasonable to assume that respondents can tell someone is
not from the same region as their own when they hear a different feature
from the one they usually use. Therefore, Tokyo respondents judge a speaker
'non-Tokyo' when they hear the Kinki accent at a higher rate than they judge
a speaker 'Tokyo' when they hear the Tokyo accent. On the other hand, the
Kinki results do not show this expected pattern. Why do Kinki respondents
judge a speaker 'Kinki' when they hear the Kinki accent more easily than
they judge a speaker 'non-Kinki' when they hear the Tokyo accent?

I believe the explanation lies in the same area of language ideology as
that offered earlier — the different positions and values of the language
(dialect) in Tokyo and Kinki. Kyoto was a former capital (from 794), and
Osaka was the capital before 710, when Nara (another city in Kinki) became
the capital. Therefore, it is very natural for present-day people to share the
Kinki dialect as a native one inherited from many previous generations.
Tokyo, on the other hand, became the capital in 1868, and people started
moving in after that. There still are many people who were born in Tokyo
but whose parents were not. Consequently, people living in Tokyo may well
have a smaller sense of solidarity based on language, placing less value on
local language. The Tokyo dialect is prestigious because the standard
language is based on it, but it seems that it does not function well in bonding
people together when compared to the Kinki dialect. Ryan, Giles and

Sebastian (1982) have noticed a general tendency for nonprestigious areas to

110

have their speech variety more highly evaluated for solidarity than status,
and Preston (1996) described this dichotomy as follows:

Areas with greater linguistic insecurity focus on regional

solidarity to express local identity. Areas with considerable

security do not use local speech to express such identity, for its

'uniqueness' is already taken up in the expression of status

rather than solidarity matters. (317)

In this case, however, although the Kinki dialect has only a
historically-based prestige, some Kinki people keep using it because they
refuse to use the standard language (i.e., the Tokyo dialect) and find it natural
to speak the Kinki dialect, even when they visit or work in Tokyo (Sakamoto,
Onoe, Hayase, and Saji in Kawauchi 1993; Inoue 1995). These facts seem to
support the interpretation of the Kinki dialect as a "high solidarity" variety
(in addition to whatever historical standardness, and, hence, prestige, it may
have had).

The ideological difference may help explain why nondevoicing (a
supposed Kinki feature) contributes to making expected judgments more
than devoicing (a supposed Tokyo feature) does in the Kinki results. Kinki
respondents can tell someone is from Kinki when they hear nondevoicing
more easily than they can tell someone is not from Kinki when they hear

devoicing because of the value they place on the Kinki dialect and the

"recognition" bonus this valuation gives them.

111

In the case of Tokyo results, there are further reasons why the Kinki
accent contributes judgments more strongly than the Tokyo accent does, other
than the explanation that a different feature from one's own is easy to detect.
When Tokyo respondents heard the Tokyo accent, they let such tokens pass
the first screening (accent pattern) by not judging the speaker 'non-Tokyo,‘
but they withhold the judgment ’Tokyo' and pay attention to other features,
perhaps recognizing features that do not sound 'Tokyo' other than accent and
devoicing. Since this is not a matched-guise study, that possibility was not
controlled for.

The second reason may be that the accent patterns are changing and
that the Tokyo pattern did not induce the judgment 'Tokyo' when I expected
it. Younger generation speakers in Tokyo pronounce words differently in this
regard (Inoue 1998). For example, densha 'train' used to be pronounced as
dénsha (HLL) but now it is densha (accentless), and sekaishi 'world history' is
pronounced as sekdishi (LHLL) but younger people have started to pronounce
it sekaishi (accentless) (pp. 177-178). In my data, for example, I accepted
oyayubi 'thumb' in LHLL based on Kindaichi (1958) and confirmation by
Tokyo persons whom I consulted, but some of my respondents pronounced it
as an accentless word when recording, another example of a word in the
process of accent change. It is quite possible for some respondents to have

judged the LHLL pronunciation of oyayubi as not the correct 'Tokyo' accent in

112

spite of my expectation. These two circumstances may have lowered the
effect of the Tokyo accent on correct judgments.

Neutral tokens, which are pronounced with the same accent patterns
in Tokyo and Kinki and contain no devoiceable vowel, tend to be judged as
'Tokyo' by Tokyo respondents and 'Kinki' by Kinki respondents, as shown in
Tables 9 and 10, as well as Tables 11 and 12. Those are reasonable results and
confirm the fact that the respondents seem to have made judgments based on

whether the word is pronounced with the same accent pattern as their own.

5.2. Significant and insignificant factors in overall results

After examining the general tendencies in the overall five-way results by
token types and finding that they are preserved in the collapsed two-way
results as well, I ran the multivariate logistic regression program on the two-
way results in order to investigate how phonological and social factors affect
the respondents' judgment of the speaker's region.

In the Tokyo overall results, significant factor groups are the position
of a devoiceable vowel in terms of the morpheme boundary, the preceding
consonant, the following consonant, voicing of the devoiceable vowel, and
the gender and age of the respondent, while insignificant ones are the
position of a devoiceable vowel in terms of mora count, the pitch accent
patterns, the identity of the devoiceable vowel, and gender of the speaker.

These results seem to be reasonable considering the phonetic facts in vowel

113

devoicing in production. There is a tendency that the factor groups which
have been studied previously and described with regard to devoicing
preferences (Vance 1988, and others) are significant, and gender and age of the
respondent are common, influential sociolinguistic factors. In other words,
these seem to be reasonable results, which show that the respondents rely on
the conditions which promote and demote devoicing in actual production in
making the judgments sought here.

In the Kinki results, on the other hand, there are only two insignificant

factors: identity of the devoiceable vowel and gender of the respondent.

5.3. Phonological factors

5.3.1. Voicing of devoiceable vowels

The focus of this study is on devoicing, and Tables 26 (Tokyo) and 27 (Kinki)
show opposite results as regards expected responses. Devoiced vowels
promote the expected judgment and nondevoiced vowels demote it in the
Tokyo results, while nondevoiced vowels promote the expected judgment
and devoiced vowels demote it in the Kinki results. In other words, the
supposed features of their own varieties are promoters. It is also shown that
the range of the weight in the Kinki results is smaller than that in the Tokyo
data, strengthening the suggestion that devoicing variation is not as powerful

a criterion for the Kinki respondents.

114

5.3.2. Preceding consonant

As stated in Chapter 4, it seems reasonable to interpret the Tokyo results as an
interaction of the place of articulation and the manner of articulation, with
greater saliency of the place, as shown in Tables 20 and 21. In particular, the
devoiced data are perfectly consistent showing back consonants (velar and
palatal consonants) as promoters and front consonants (prepalatal, alveolar,
and labial consonants) as demoters, as shown in Table 39. The order of the
weights for the devoiced data and the nondevoiced data (final results are in
Table 40) is very similar, which means promoting consonants and demoting
consonants have the same tendencies, except for [t1] and [9]. Those two
consonants were demoters with the nondevoiced data, and because the
weight of [t[] is close to .5, only [g] turned out to be a demoter in the overall
results.

If the results with devoiced tokens are natural ones, there must be
some reason why [t1] and [c] with nondevoiced tokens lower the weights so
much and demote the expected judgment, therefore changing the overall
results such that [c] turned out to be a demoter. Perhaps [t1] and [g] promote
the expected judgment with devoiced vowels and demote it with
nondevoiced vowels. That means they always help the Tokyo respondents
"hear" devoiced vowels and judge the speaker as 'Tokyo.‘ These two
consonants are among those which Imai (1997) found as the best promoters of

devoicing in production. According to her, / i / with a preceding fricative and

115

a shared feature between the preceding consonant and the devoiceable vowel,
that is, palatal for / i / and labial for /u/, are the strongest promoters of
devoicing. [c] is exactly such a consonant, that is, a palatal fricative preceding
/ i / , and [t[] is a palatal consonant with [+continuant] feature preceding / i / .
The results mean that when the respondents heard the best combination of a
consonant and a vowel for devoicing, they judged the speaker as 'Tokyo,‘
regardless of whether or not the vowel was devoiced. It seems that the
respondents know, although they are not overtly aware of it, which
combinations are the best for devoicing vowels and do not pay attention to
actual devoicing or nondevoicing, simply assuming that they are devoiced.
The results for the Kinki data are different. In both the devoiced data
and nondevoiced data (Tables 55 and 57, respectively), {-continuant} is
generally a promoting feature and [+continuant] is a demoting feature. In the
devoiced data, in particular, the results clearly show that stops and affricates
are promoters and fricatives are demoters with no exception. The only
exception in the nondevoiced data is [t1]. It has the feature {-continuant} but
turned out to be a demoter in the nondevoiced data and in the overall Kinki
data (Table 23) as well. The exception in the Kinki results is again what Imai
found as a consonant in the best environment for devoicing. Kinki
respondents judge the speaker as 'non-Kinki' whether the vowel is devoiced
or not; apparently they assume that the vowel is devoiced. This suggests that

Kinki respondents also know the best environments for devoicing. This is

116

consistent if Kinki speakers devoice vowels with (at) the same frequency as
Tokyo speakers, because it suggests that Kinki respondents have the
knowledge of the best environment for devoicing in their own dialect and
that it is the same as in the Tokyo dialect.

In short, the Kinki respondents rely on different features of the
preceding consonant ([-continuant] or [+continuant]) from what the Tokyo
respondents rely on (front or back) to make judgements, but both Kinki and
Tokyo respondents seem to know the best environment for devoicing in
general and use the knowledge to judge the speaker as 'Tokyo' regardless of

whether the vowel is devoiced or not, assuming that it is devoiced.

5.3.3. Following consonant

The results by the following consonant with the Tokyo devoiced data and
with the Tokyo nondevoiced data turned out to be very different as shown in
Tables 43 and 44, respectively, and those different results seem to have created
the complicated overall results as shown in Table 24.

There seems to be a tendency for the feature [+continuant] to be a
promoter and the feature {-continuant} to be a demoter with the devoiced
data, but the tendencies are opposite with the nondevoiced data. In other
words, the respondents tend to judge the speaker as 'Tokyo' when they hear
fricatives in the consonant following devoiceable vowels whether they are

devoiced or not.

117

The different effects of preceding consonants and following consonants
seem to be consistent with a temporal order of perception. After noticing a
preceding consonant, the respondent still attends to the following vowel,
assessing its devoicing or nondevoicing, and judges the speaker accordingly.
Different consonants have different effects, but the orders of the consonants
by weight are almost the same in the devoiced data and in the nondevoiced
data; that is, the same consonants help the respondent make judgments that a
devoicer is from Tokyo and a nondevoicer is not from Tokyo. With
following consonants, on the other hand, the respondents cannot retrieve
their perception of the vowel. So they cannot make judgments based on
devoicing variation, and assume the vowel has the character predicted by the
production effect of the following consonant, in this case, that it was devoiced
when the consonant is [+continuant]. They make judgments based on the
type of consonant only in the case when a voiceless fricative comes after a
high vowel (that comes after another voiceless consonant), assuming that it is
devoiced and go on to reason that a devoicer is from Tokyo.

The two exceptions are the two consonants with {-continuant}: [t] and
[t1]. The former is a promoter in both data and the latter is a demoter in both
data. The reason why they behave differently from the other {-continuant}
consonants is not clear.

The overall results for the Kinki data, shown in Table 25, do not seem

to show any clear tendencies of which features give rise to the expected and

118

unexpected judgments. It might be possible, though, to say that the weights of
back consonants are heavier than those of front consonants and that the
weights of the consonants with the feature {-continuant} are heavier than
those of the ones with [+continuant].

In the results of the devoiced data (Table 58), it is hard to find general
tendencies with both phonological and statistical justification. In the results
of the nondevoiced data (Table 60), on the other hand, show that the
{-continuant} consonants (stops) are promoters and the feature [+continuant]
is a demoter except for [h], which is a promoter. The results of the
nondevoiced data show very similar tendencies to those in the Tokyo
nondevoiced data. It is reasonable to have similar tendencies, though that is
only in the nondevoiced data, considering the limitation of perception in
time. Kinki respondents cannot retrieve their perception of the vowel, just as
Tokyo respondents cannot, and both respondents may well have the same
knowledge of the best environments for devoicing, if their devoicing rate is
almost the same. The Kinki results with the devoiced data are not as
consistent as the Tokyo results, though, and I cannot find any explanation for
this.

In general, however, for both Tokyo and Kinki respondents, there
seems to be a tendency for hearers to believe that vowels are devoiced in
high-frequency devoicing environments, i.e., with regard to both preceding

and following consonants. This, of course, skews the expected results as I

119

have outlined in detail above and shows the necessity in sociolinguistic
perception experiments such as this one of considering speech perception as
well as production facts. The fact that my respondents "hear" a devoiced
vowel in high-frequency devoicing environments is what is known in speech
perception studies as a "context effect" and is well-documented (e.g., Strange
1995). Although many such studies are based on actual phonetic cues to
perception, most conclude that knowledge of the allophonic regularities of a
system by its speakers rather than only acoustic detail plays an important role
in perception. For example, Beddor, et a1. (2002) show that the perception of
vowels by English speakers and Shona speakers are based on language-specific
patterns of coarticulation, and that they "hear" coarticulation when there is
actually none in the coarticulatory context.

Of course, this study is not a straightforward investigation of native
speaker perceptions of vowel devoicing since it interposes a social factor
(region) between the acoustic signal and the judgment. This study is,
therefore, hampered by not having data from such a study available.
Nevertheless, the large body of speech perception work (e.g., that summarized
in Strange 1995) makes the interpretation here — that respondents believed
they heard devoiced vowels in a high-frequency devoicing environment —

very likely.

120

5.3.4. Position of devoiceable vowels in terms of boundary types
Among three positions, at the word boundary, at the morpheme boundary,
and at the morpheme internal position, the outermost position and the
innermost position of the word structure (word boundary and morpheme
internal position) are promoters while the morpheme boundary is a demoter,
in both Tokyo and Kinki results, as shown in Tables 18 and 17, respectively.
These results can be interpreted as a straightforward outcome following
the tendencies of using and avoiding devoicing in production. According to
Vance (1987), devoicing is avoided when a devoiceable vowel is at a
morpheme boundary, as it is before a pause. There is an interaction between
devoicing and a pause, and it is reasonable to assume a morpheme boundary
is less likely to have a pause than a word boundary, and more likely to have a
pause than a morpheme-internal position, because a morpheme boundary
may be in a word internal position. Therefore it seems natural to expect that
devoicing variability is greater at the morpheme boundary because devoicing
occurs most frequently at a morpheme internal position and is avoided most
frequently at a word boundary. The respondents in this study tend to make
the expected judgment at such "innermost" and "outermost" positions,
where occurrence and avoidance are highly expected, respectively, but tend to
make the unexpected judgment with vowels that are not in such positions.
The Tokyo results show the same tendencies in the devoiced data, the

nondevoiced data (Tables 36 and 37, respectively), and the overall results,

121

while the Kinki results are quite different. Boundaries are not significant
factors in the Kinki nondevoiced data, and the Kinki devoiced data show, in
Table 53, that the morpheme internal position turned out to be a demoter
while it is a promoter in the overall results and Tokyo results. The weight of
the word boundary in the Kinki results is the heaviest among all the
phonological and social factors in the Kinki devoiced and nondevoiced
tokens in this study. It is close to the weights of the Kinki pitch accent and the
Tokyo pitch accent in the results by the token type. It means that the Kinki
respondents tend to make the judgment 'non-Kinki' very decisively when
they hear devoicing at the word boundary, just as they hear the Tokyo and
Kinki pitch accent. Considering the word boundary and the morpheme
boundary are the only factors in my data that are reported in previous studies
as positions in which devoicing is often avoided, it might imply that the
heavy weight of the word boundary has some connection with frequencies of
devoicing and nondevoicing in the Tokyo dialect and in the Kinki dialect.
Data on production in the Kinki dialect must be obtained to further discuss

this possibility.

5.3.5. Pitch accent pattern and position of devoiceable vowels in terms of
mora count
The pitch accent pattern is a significant factor only in the Kinki overall data

(Table 19) and in the Tokyo nondevoiced data (Table 38). HLL at the

122

beginning of the word is a promoter for making the expected judgment and
LHL is a demoter in both cases.

The position of a devoiceable vowel, either in the second mora or the
third mora from the beginning of the stimulus, is also significant. In the
Tokyo data, the devoiced data and the nondevoiced data show opposite
results. The third mora is a promoter and the second mora is a demoter in
the devoiced data (Table 34), while the second mora is a promoter and the
third mora is a demoter in the nondevoiced data (Table 35). In other words,
the respondents tend to make the judgment 'Tokyo' whether the vowel is
devoiced or not. The reason is not clear. Due to the opposite results, the
overall results show that the position of a devoiceable vowel in terms of
mora count is not significant. In the Kinki data, on the other hand, the
position of a devoiceable vowel in terms of mora count is not significant in
the final results with both devoiced and nondevoiced data. The overall
results turned out to show that the second mora is a promoter and the third
mora is a demoter (Table 15). The reason for these results is not clear.

Both accent patterns and mora count are factors that play important
roles in Japanese phonology, such as conjugation of verbs and adjectives,
building compounds, and accentuation of loanwords (Kindaichi 1958, Vance
1987, and others). It seems, however, interaction between devoicing and the
accent pattern of the word or devoicing in the position of the word in terms

of mora count has not been described or reported. As stated in section 5.2.,

123

such factors as preceding consonants and morpheme boundaries, whose
interaction with vowel devoicing in production, in contrast, have been
described and investigated in previous studies, tend to show significant
results in this perceptual study. There is, at least at present, little motivation
to pursue reasons for the results by accent patterns and by mora count,

although some data sets show significant results.

5.3.6. Identity of devoiceable vowels

The identity of the devoiceable vowel is significant only with the Tokyo
devoiced data, showing that /u/ is a promoter and / i / is a demoter (Table 45).
It seems reasonable to conclude that the identity of the devoiceable vowel
does not heavily affect the respondents' judgments. This is a rather
unexpected result, actually, considering the coincidence of the phonological
factors of environments that promote devoicing in production (Imai 1997)
and that affect respondents' judgments so that they seem to assume devoicing
regardless of actual devoicing or nondevoicing. It would not be surprising if
/ i / would have been a promoter because it is devoiced more frequently than
/ u/ (Imai 1997), and it might well promote the respondents' perception of

devoiced vowels.

124

5.4. Social factors

5.4.1. Gender of the speaker and gender of the respondent

Female voices promote the expected judgment in the Tokyo devoiced and
nondevoiced results (Tables 46 and 47, respectively), but the overall results
show that the gender of the speaker is not a significant factor. In the Kinki
data, on the other hand, both the devoiced and nondevoiced data show that
the gender of the speaker is not a significant factor, but the overall results
show that male voices promote the expected judgment (Table 28).

The gender of the respondent is not significant in the Tokyo devoiced
data, while female respondents contribute to the expected judgment more
than male respondents do in the overall data and the nondevoiced data
(Tables 29 and 48, respectively). In the Kinki data, on the other hand, male
respondents contribute more in the devoiced data (Table 61) and female
respondents do more in the nondevoiced data (Table 62), and the overall
results show that the gender of the respondent is not a significant factor.

In other words, female voice and female respondent are promoters and
contribute to the expected judgment when gender is a significant factor in the
Tokyo data, while there is no such consistency in the Kinki results. The
greater sensitivity by women in the Tokyo results parallels many such
production and perception findings in sociolinguistics in general (e.g.

Chambers 1995) and it confirms the idea put forth earlier that "standardness"

125

is the overwhelming concern in the Tokyo speech area (rather than

solidarity).

5.4.2. Age of the respondent
In the Tokyo data, the youngest and the oldest groups (18-22 and 40-49 years
old) can be combined statistically (a demoter), as can the two groups in the
middle (23-29 and 30-39 years old, a promoter). This is a typical "age-graded"
sociolinguistic pattern, as shown in Table 32. The Tokyo devoiced results
show an even clearer age-graded pattern, with the youngest and oldest groups
with a lighter weight, and with the 23 to 29 year old group with a heavier
weight, as shown in Table 51. Twenty-three to twenty-nine year old
respondents in this study are generally those who have graduated from
college and started graduate studies or have several years of work experience
or both. The devoiced data show that those respondents, who need to learn
standards of social behavior to be accepted as members of the adult society,
most sensitively perceived devoicing. The nondevoiced data, on the other
hand, show exactly opposite results. The group of 23 to 29 year olds is the
only demoting group, as shown in Table 52. The overall results show an age-
graded pattern, but it is rather weak because of these opposite results for the
devoiced and nondevoiced data.

The opposite results in the devoiced and nondevoiced data are also

shown in palatal consonants preceding devoiceable vowels, as discussed in an

126

earlier section. Just as with those palatal consonants, the opposite results are
interpreted as a case in which the respondents tend to make the judgment
‘Tokyo' when they hear a devoiceable vowel regardless of whether it is A
devoiced or nondevoiced. In other words, the Tokyo results show an age-
grading pattern in their tendency to make the judgment 'Tokyo' when they
hear a devoiceable vowel in both devoiced and nondevoiced data, and the
group of 23 to 29 year olds is the strongest promoter.

The results with the Kinki data, in contrast, do not show such an age-
grading pattern. In the overall results, only the youngest group is a demoter,
and all the other age groups were conflated, as shown in Table 33. The weight
of the promoting age group is 0.506, very close to 0.5; that is, the respondents
in the promoting group did not strongly make the expected judgment. The
separate results of the devoiced data and the nondevoiced data are more
complicated and show no specific pattern (Table 65 and 66, respectively),

although some age groups can be combined based on statistical reasoning.

5.5. Speaker's identity

I did not take the factor group of the identity of the speaker into account in
the multivariate analysis, as explained in an earlier chapter. Instead, I am
going to show the frequency and percentage of responses for each speaker and
describe the observations made during the experiment. It is probably

reasonable to look at the frequencies and percentages of all tokens, instead of

127

the tokens with devoicing variation only, to reflect the respondents' general
perception about the speakers. Tables 67 and 68 show the results of all of the
tokens eliminating the responses 'Don't know' for the Tokyo data and for the

Kinki data, respectively.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 67 Results by speaker (T okyo, all types of tokens)
Speaker Expected Unexpected Total
I 396 (88.00) 54 (12.00) 450 (100)
N 270 (87.38) 39 (12.62) 309 (100)
F 187 (74.21) 65 (25.79) 252 (100)
E 325 (74.03) 114 (25.97) 439 (100)
K 458 (71.56) 182 (28.44) 640 (100)
G 240 (71.01) 98 (28.99) 338 (100)
L 358 (67.17) 175 (32.83) 533 (100)
B 317 (65.63) 166 (34.37) 483 (100)
I 197 (65.02) 106 (34.98) 303 (100)
D 507 (63.14) 296 (36.86) 803 (100)
H 226 (61.41) 142 (38.59) 368 (100)
A 206 (62.61) 123 (37.39) 329 (100)
C 26 (60.47) 17 (39.53) 43 (100)
M 212 (45.01) 259 (54.99) 471 (100)
Total 3925 (68.13) 1836 (31.87) 5761 (100)

 

 

 

 

128

 

Table 68 Results by speaker (Kinki, all types of tokens)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Speaker Expected Unexpected Total

G 393 (79.88) 99 (20.12) 492 (100)
483 (74.31) 167 (25.69) 650 (100)

J 330 (71.58) 131 (28.42) 461 (100)

L 526 (69.67) 229 (30.33) 755 (100)

K 574 (63.64) 328 (36.36) 902 (100)

A 313 (62.98) 184 (37.02) 497 (100)

E 365 (60.83) 235 (39.17) 600 (100)

N 266 (58.21) 191 (41.79) 457 (100)

F 204 (57.79) 149 (42.21) 353 (100)

M 373 (56.52) 287 (43.48) 660 (100)

C 36 (56.25) 28 (43.75) 64 (100)

H 300 (54.05) 255 (45.95) 555 (100)

D 563 (51.79) 524 (48.21) 1087 (100)

B 320 (48.85) 335 (51.15) 655 (100)
Total 5046 (61.63) 3142 (38.37) 8188 (100)

 

 

All the speakers but one is judged in the "expected" way in each respondent
group. Only the speaker M in the Tokyo data and the speaker B in the Kinki
data were judged oddly. The respondents frequently mentioned after the
experiment that some speakers sounded conspicuous, some saying that was
because of the pitch accent or speech rate, and this caused them to judge these
voices as non-Tokyo (for Tokyo respondents) or Kinki (for Kinki
respondents). The comment about the accent seems quite natural among

Tokyo respondents based on a stereotypical assumption that a non-Tokyo

129

person is conspicuous when he or she does not change the pitch accent
patterns upon visiting Tokyo. However, the respondents' description does
not always match the actual pronunciation in the test tape.

Respondents particularly mentioned speaker K as the one most
confidently judged as a non-Tokyo person or Kinki person, and actually he is
from Kinki. There is no token, however, of the Kinki accent among the
words he pronounced. Furthermore, when he devoiced vowels, 53.75% of
the Tokyo respondents judged him as Tokyo, although this is a rather weak
recognition. His neutral tokens, which are expected to be judged as Tokyo by
Tokyo respondents because of lack of clues by pitch accent or voicing
variations, were also judged as Tokyo, as expected, at 72.09%. His
nondevoiced tokens were judged as 'non-Tokyo,‘ as expected, by 74.47% of
them. Similar respondents descriptions of pronunciation, which do not
reflect actual facts, were observed in previous studies (e.g., Kerswill 1985). A
possible interpretation of these incorrect descriptions by the respondents is
that they perceived nondevoicing and judged the speaker as non-Tokyo, but
could not verbalize the linguistic fact and simply reported an obvious non-
Tokyo feature, non-Tokyo accent.

In contrast, only 31.40% of Kinki respondents judged Speaker K as non-
Kinki when he devoiced vowels while 65.24% judged his nondevoiced
tokens as Kinki. As seen in the overall results by token types, it seems that

Kinki respondents did not use devoicing as efficiently as Tokyo respondents

130

to judge whether or not a speaker is from Kinki. From another perspective, it
can be said that Kinki respondents can tell this speaker is from the same
region as theirs in spite of the Tokyo feature.

Speaker M, on the other hand, was heavily judged in the unexpected
way; fewer than 50% of Tokyo respondents gave the expected responses. His
nondevoiced tokens were judged as non-Tokyo by only 27.93% of Tokyo
respondents while his devoiced tokens were judged as Tokyo by 84.44% of
them. Just as Speaker K for the Kinki respondents, Speaker M is judged as
Tokyo regardless of the presumed Kinki feature in his pronunciation. The
two speakers K and M illustrate cases in which respondents can tell whether
or not the speaker is from their own area accurately in spite of non-local
features in his or her pronunciation. The reason is not clear. Their results
indicate, at least, that they may have some stronger factor, such as voice
quality, as suggested by Laver and Trudgill (1979), than different pitch accent
patterns and devoicing variations.

Laver and Trudgill (1979) discuss the possibility that voice qualities of
extralinguistic, paralinguistic, and phonetic features are social markers, which
signal such membership as regional affiliation, social class, and age. They
give a number of examples from previous studies, and also point out that
attitudinal variation depends on phonological environment, whose
importance varies in different social groups, and that the variation is not a

conscious process.

131

Some possible influencing factors in my data are (i) vowel length, (ii)
pitch contour in a single vowel, and (iii) other quality of sounds. Horii (1982)
suggests that vowels in a one-mora word (with a suffix) and in word final
position tend to be longer in Kinki dialect, and it is widely known that the
Kinki pronunciation has a rising or falling pitch contour within a single
vowel while the Tokyo one does not commonly have such a contour (Tahara
et a1. 1998, Warner 1997, Kokugogakkai 1980, Makimura 1979). Finally, Maeda
(1977) and Horii (1982) state that consonants are pronounced less deliberately
in Kinki than in Tokyo. Horii gives such examples of deletion of [sh] and
weakening of [sh] to [h]. I notice that /s/ is not perceived sometimes by non-
native speakers of Japanese when a Kinki person devoices sentence-final /u/
as in /desu/ (copula). When I prepared the test tape, I selected words which
were pronounced most 'Tokyo-like' to my ears to eliminate other non-Tokyo
features. I may not have done a perfect job, however, because of phonetic
differences that I could not detect or could not avoid due to limited samples
despite having detected them. Those differences can be a possible topic to
investigate in the future. The final chapter summarizes these detailed

results.

132

6. Conclusion

In this study I conducted a perception experiment, using vowel devoicing,
which is a common allophonic phenomenon in Japanese. Perception
experiments focusing on features at the allophonic level have rarely been
conducted. The object of this study, vowel devoicing, is popularly behaved to
occur less frequently or not at all in the Kinki dialect, although empirical
studies show that it occurs as frequently, at least in the most general
devoicing environments, as in the Tokyo dialect. Because of this mismatch
between popular belief and the facts of devoicing in the two regions, one
could not predict what local reactions to devoicing and nondevoicing might
be. In my experiment, Tokyo and Kinki respondents were asked to judge
whether or not the speakers on a test tape were from the same region as their
own, and the results were examined to show how phonological factors
(devoicing and environments) and the social factors of both speakers and
respondents (gender and age) would affect judgments. In the analyses of the
results, devoicing as well as the Tokyo pitch accent were treated as Tokyo
features while nondevoicing as well as the Kinki pitch accent were treated as
Kinki features. The comparison was based on the working hypothesis that a
speaker who devoices vowels or uses the Tokyo pitch accent is more likely to

be judged as a Tokyo (or non-Kinki) person while a speaker who nondevoices

133

vowels or uses the Kinki pitch accent is more likely to be judged as a non-
Tokyo (or Kinki) person.

The results confirmed this hypothesis. They suggest that the voicing
variants of a devoiceable vowel control the judgments by both Tokyo and
Kinki respondents about where a speaker is from, supporting the idea that a
low-level linguistic variant can have social meaning, as in other
sociolinguistic studies. The results of this study can be said to have moved
the field from suprasegmental but phonemic features (pitch accent patterns)
as in Warner (1997) into allophonic features, and show that such lower level
features can be recognized and used as clues in making judgments.

Vowel devoicing is not noticed in natural conversation; consequently,
the respondents could confidently identify some speakers as Tokyo or Kinki
but could not describe the linguistic features of those speakers accurately as
the basis for their judgments. Instead they attributed their judgments to such
recognizable features as phonemic differences as in pitch accent patterns.

Another finding similar to those in other sociolinguistic studies is that
Tokyo results show an age-graded pattern and greater sensitivity by female
respondents. The age group of 23 to 39 gave more expected judgments than
younger or older age groups. Since that age group has finished school and
gone to work in the wider society, the results by age, as well as by gender,
match the well-known greater middle-age and female sensitivities to the

linguistic 'norm' of the community.

134

Other significant and insignificant factors turned out to be ones
reported in some of the earlier literature on devoicing production; the factors
that interact with devoicing judgment variation, such as the position of
devoiceable vowels in terms of morpheme boundary, tend to be significant,
while the factors for which no such previous interaction had been reported,
such as the position of devoiceable vowels in terms of moraic position, were
insignificant.

Interesting findings in this study, however, are not simply based on the
respondents' use of devoicing and nondevoicing in making their judgments.
Both Tokyo and Kinki respondents seem to use obviously covert knowledge
in identifying the best phonological environments for vowel devoicing in
production, assuming that the vowel in those environments should be
devoiced whether they actually were or not. These results suggest that the
respondents do not solely rely on the linguistic information presented to
them but use some prior knowledge of how likely vowels are to be devoiced
in production.

The influencing features of the consonants that precede devoiceable
vowels and the ones that follow them also suggest a combination of sources
of information in making judgments. For the preceding consonant,
{-continuant} (as well as back consonants in the Tokyo data) is a feature
promoting the expected judgments. For the following consonant, on the

other hand, [+continuant], generally, promotes perception of devoiced vowels

135

and, therefore, judgment of 'Tokyo' and 'non-Kinki' regardless of actual
devoicing. It seems that the respondents pay attention to the continuancy of
the preceding and following consonant, take the best environments for vowel
devoicing into account, and assume the vowel is devoiced based on such
continuancy. Thus, the results illustrate the reasonable coordination of the
perception of linguistic information with utilization of knowledge of vowel
devoicing in production.

The results of this study also show some differences between Tokyo
and Kinki respondents. The Tokyo results clearly show the use of devoicing
or nondevoicing in making judgments, while the Kinki ones show similar
tendencies but more weakly. The weak tendencies in the Kinki results can
support the speculation that vowel devoicing occurs as frequently in the
Kinki dialect as in the Tokyo dialect, at least in the most general devoicing
environments and that is why devoicing variation is not efficient as a
criterion for Kinki respondents to make judgments. In other words, the
Tokyo respondents made their judgments based only on the stereotype that
they devoice vowels more frequently.

Comparison of the Tokyo and Kinki results of the tokens by different
pitch accent patterns also reveals differences. Both dialects have prestige in
the local areas, but the Tokyo pitch accent does not immediately induce the
judgment 'Tokyo' for Tokyo people while the Kinki pitch accent

overwhelmingly causes the Kinki people to make the judgment 'Kinki.‘

136

These results suggest that the Tokyo accent does not work well in
maintaining 'solidarity' among people in the region although it has prestige
as standard, compared to the Kinki accent.

From the speakers' point of view, the results show that they can make
a Tokyo person believe they are from Tokyo by using both Tokyo pitch accent
and devoicing (and from non-Tokyo by using the Kinki pitch accent and
nondevoicing) while they can less easily make a Kinki person believe that
they are from Kinki by using nondevoicing (and from non-Kinki by using
devoicing).

In spite of these differences, I have shown statistically that Tokyo and
Kinki people perceive devoicing and nondevoicing of vowels and use that
feature to make judgments on whether a speaker is from the same region as
theirs. The results show similar patterns on certain points as other
sociolinguistic and phonological studies and support the idea that devoicing
variants, which are allophonic and not consciously produced or perceived in
natural conversations, can determine such judgments, although their criteria
in making judgments do not reflect actual production of vowel devoicing in
Tokyo and Kinki. The results also seem to illustrate nicely that the
respondents utilize covert knowledge about devoicing in production (based
on the preceding and following segments) as well as the voicing status of the

vowels themselves.

137

Now the next natural questions are how devoicing and nondevoicing
are distributed and how phonological and social factors affect their variation
in the Kinki dialect. The results of this study match the suggestion that
vowel devoicing occurs in the most general environments as frequently in
the Kinki dialect as in the Tokyo dialect, about 80%, based on the very limited
tokens in previous studies, and that nondevoicing occurs in other
environments more frequently in the Kinki dialect. It is necessary to collect
more controlled comprehensive natural data and examine vowel devoicing
in order to determine the factors that determine its distribution and variation
in non-Tokyo dialects before such a comparison of vowel devoicing in the
Tokyo and non-Tokyo dialects can be completed.

Without that comprehensive study, however, it is clear that vowel
devoicing is a covert regional stereotype in Japanese. It interacts not only
with the social categories of age, gender, and region, moreover, but also with
such subtle linguistic clues as promoting and demoting linguistic
environments, suggesting that, like other advances in the general area,
sociophonetics is of considerable concern to studies of variety perception as

well as production.

138

APPENDICES

139

Appendix I List of the subjects for recording

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Speaker ID Gender Hometown Dialect
(Prefecture) Region

A F Hyogo Kinki
B F Yamaguchi Chfigoku
C M Tochigi North Kant?)
D F Hyogo Kinki
E F Kyoto Kinki
F M Ibaragi North Kanto
G F Miyagi South Ou
H M Okayama Chfigoku
I F Osaka Kinki
J F Osaka Kinki
K M Hyogo Kinki
L M Aichi Tokai
M M Chiba East-west-south Kanto
N M Wakayama Kinki

 

 

 

 

 

 

Based on I-Iirayama et al. (1992):

East-west-south Kanté includes the Tokyo metropolitan area. South On and
Tokai are in the north and in the west of Kanto, respectively. South Ou,
North Kanto, East-west-south Kanto, and Tokai belong to the Eastern dialects.
Chfigoku and Kinki belong to the Western dialects. Chfigoku is in the west of
Kinki.

140

Appendix II The list of the words

By devoicing variation: D=Devoiced ND=Nondevoiced

By accent: K=In the Kinki accent T=In the Tokyo accent N=Neutral

001 atafuta (D) 039 muhitsude (ND)
002 hekichide (D) 040 tekipakiyaru (D)
003 jifusuru (D) 041 tenkitokion (D)
004 nadeshiko (D) 042 pakupaku (ND)
005 neokisuru (ND) 043 mahishi (ND)

006 soshishiyo (ND) 044 kyobashini (K)

007 rikishiwa (ND) 045 kochifuite (ND)

008 pakupaku (D)

046 kyofuka (D)

009 katsutsumetai (D) 047 akusude (D)

010 kaihisuru (D) 048 kinkitokanté (D)
011 taue (K) 049 kawari (T)

012 herikutsumo (D) 050 masaka (K)

013 sekitsui (D) 051 shufutoseikatsu (D)
014 nornisuke (D) 052 kamakura (K)

015 asatsukio (D) 053 hiroshikun (ND)
016 yasushidono (D) 054 kaihisuru (ND)
017 hochisuru (ND) 055 michitonosogfi (D)

018 jimushitsuni (D)
019 tohitomohatsuni (D)

056

kubetsusuru (ND)

057 boyakeru (K)

020 kazokukana (D) 058 nansai (T)

021 tekipakiyaru (ND) 059 sanpitsuwa (ND)
022 risutokuma (ND) 060 jifusuru (ND)
023 kyobashini (T) 061 ninputoakachan (D)
024 konishisan (D) 062 mahishi (D)

025 taberu (T) 063 odeko (N)

026 yasuha (ND) 064 osekihan (ND)
027 tanpukude (D) 065 semotare (N)
028 atsushidono (ND) 066 dokuha (ND)
029 kajitsutokajfi (D) 067 hatashite (D)
030 nanakusa (ND) 068 potsupotsu (D)
031 tabezakari (T) 069 tanukikana (D)
032 boyakeru (T) 070 karupisuwa (D)

033 inchikina (D)
034 toshichan (D)

071
072

ninchisuru (ND)
genshutowa (D)

035 himeji (K) 073 torihikimo (D)
036 shakufuwa (D) 074 kowai (K)

037 neokisuru (D) 075 maekake (T)
038 maekake (K) 076 asatsukio (ND)

141

077
078
079
080
081
082
083
084
085
086
087
088
089
090
091
092
093
094
095
096
O97
098
099
100
101
102
103

gatapishinaru (D)
tabezakari (K)
rikishiwa (D)
masaka (T)
hochisuru (ND)
ninputoakachan (ND)
bakufuwa (ND)
tamashii (N)
kokuhide (ND)
katsutsumetai (ND)
sekitsui (ND)
karupisuwa (ND)
kawari (K)
ninchisuru (D)
michitonosogfi (ND)
himeji (T)

akushude (ND)
kubetsusru (D)
tabun (N)

soshishiyo (D)
osekihan (D)
hekichide (ND)
pakupaku (D)
michihide (ND)
shufutoseikatsu (ND)
nanakusa (D)
tenkitokion (ND)

104 bakufuwa (D)

105
106

taue (T)
genshutowa (ND)

107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136

142

kamakura (T)
saishusuru (ND)
kirikuchini (D)
kinkitokanto (ND)
oyayubi (K)
atafuta (ND)
kokuhide (D)
pachipachi (D)
nisensee (N)
senshuka (D)
herikutsumo (ND)
kowai (T)

kyofuka (ND)
konishisan (ND)
yasuha (D)
henpikana (D)
kochifuite (D)
kazokukana (ND)
taberu (K)
gatapishinaru (ND)
risutokuma (D)
yasushidono (ND)
shakufuwa (ND)
oyayubi (T)
torihikimo (ND)
nansai (K)
atsushidono (D)
tenpusuru (D)
hiroshikun (D)
dokuha (D)

REFERENCES

143

REFERENCES

Beddor, Patrice Speeter, James D. Hansberger, and Stephanie Lindemann.
2002. Language-specific patterns of vowel-to-vowel coarticulation:
Acoustic structures and their perceptual correlates. Journal of
Phonetics 30. 591-627.

Bloch, Bernard. 1970. Studies in Colloquial Japanese, IV. Phonemics [from
Language 2686-125 (1950)]. Bernard Bloch on Japanese, ed. by Roy
Andrew Miller, 113-65. New Haven: Yale University Press.

Chambers, J. K. 1995. Sociolinguistic theory: Linguistic variation and its
social significance. Oxford, UK: Blackwell.

Graff, David, William Labov, and Wendell Harris. 1983. Testing listeners'
reactions to phonological markers of ethnic identity: A new method for
sociolinguistic research. Diversity and diachrony, ed. by David
Sakoff, 45-58. Amsterdam/ Philadelphia: John Benjamins.

Han, Mieko Shimizu. 1962. Unvoicing of vowels in Japanese. Onsei no
Kenkyfi [Study of Sounds] 10. 81-100. Nihon Onsei Gakkai.

Hirayama, Teruo, et a1. (32 co-authors) 1982. Zenkoku hogen jiten
[Dictionary of dialects in the whole country]. Tokyo: Kadokawa Shoten.

Hirayama, Teruo, Ichiro Oshima, Masao Otsuka, Makoto Kuno, Mariko
Kuno, and Takao Sugimura (ed.) 1992. Gendai nihongo hogen dai-
jiten [Dictionary of Japanese dialects]. Tokyo: Meiji Shoin

Horii, Reichi. 1972. Kinki hogen no gaisetsu [An introduction to Kinki
dialect]. Koza Hogengaku 7: Kinki chiho no hégen [Lecture on
dialectology 7: Dialects in the Kinki region], ed. by Kiichi ltoyo,
Sukezurni Hino, and Ry6ichi Sato, 1-26. Tokyo: Kokushokankokai.

Imai, Terumi. 1997. Vowel devoicing in Japanese. East Lansing, MI:
Michigan State University MA Thesis.

Inoue, Fumio. 1998. Nihongo wotchingu [Japanese watching]. Tokyo:
Iwanami Shinsho.

144

Inoue, Shoichi. 1995. Kansai-jin no shotai [Reality of Kansai people].
Tokyo: Shogakukan.

Jun, S.-A. and Beckman, M. 1993. A gestural-overlap analysis of vowel
devoicing in Japanese and Korean. Paper presented at the 1993 Annual
Meeting of the Linguistic Society of America, Los Angeles, January 7-10.

Kawauchi, Atsuro. 1993. Kansai-hen tanken: Kawauchi Atsuré taidan-shfi
[Exploring Kansai dialect: Interviews by Atsuro Kawauchi]. Osaka:
Toho Shuppan.

Kerswill, Paul E. 1985. Native dialect and dialect mixing in Bergen: A
perception experiment. Cambridge Papers in Phonetics and
Experimental Linguistics 4.

Kindaichi, Haruhiko. 1958. Mékai Akusento jiten [Accent dictionary].
Tokyo: Kadokawa Shoten

Kitahara, Mafuyu. 1998. The interaction of pitch accent and vowel devoicing
in Tokyo Japanese. Japanese Korean Linguistics 8, ed. by David J. Silva,
303-15. Stanford, CA: CSLI. Also presentation at the 8th Japanese/
Korean Linguistics Conference at Cornell University on July 25-27,
1997.

Kitamura, Hajime. 1952. Kodomo no kotoba wa ijfi ni yotte d6 kawaru ka
[How children's language change after moving]. Gengo Seikatsu Vol.
8. 2027. Tokyo: Chikuma Shobo.

Kokugogakkai. 1980. Kokugogaku daijiten [A comprehensive dictionary of
Japanese linguistics]. Tokyo: Tokyodo Shuppan.

Kondo, Mariko. 1994. Mechanisms of vowel devoicing in Japanese.
Proceedings of 1994 International Conference of Spoken Language
Processing, 1, 61-66. Yokohama, Japan.

Kondo, Mariko. 1997. Mechanisms of vowel devoicing in Japanese. Ph. D.
dissertation at the University of Edinburgh.

Kondo, Mariko. 1999. Syllable weight and syllable structure in Japanese.

Presentation at the 9th Japanese / Korean Linguistics Conference. The
Ohio State University.

145

Labov, William. 1972. Sociolinguistic patterns. Philadelphia: University of
Pennsylvania Press.

Laver, John and Peter Trudgill. 1979. Phonetic and linguistic markers in
speech. Social markers in speech, Ch.1, ed. by Klaus R. Scherer and
Howard Giles, 1-32. Cambridge, UK: Cambridge University Press.

Lovins, J. B. 1976. Pitch accent and vowel devoicing in Japanese: A
preliminary study. Annual Bulletin of Research Institute of
Logopedics and Phoniatrics 10. 113-25.

Maeda, Isamu. 1977. Osaka-ben [Osaka dialect]. Tokyo: Asahi Shinbun-sha.

Maekawa, Kikuo. 1983. Kyotsfigo ni okeru boin no museika no kakuritsu
ni tsuite [On probabilities of vowel devoicing in the language for
mutual understanding]. Gengo no Sekai vol. 1, no. 2. 69-81.

Makimura, Shiyo. 1979. Osaka kotoba jiten [Dictionary of the Osaka dialect].
Tokyo: Kodansha.

Mogami, Katsuya. 1994. Hyojungo, kyotsfigo, hogen, chiikigo [Standard
language, common language, dialect, and local language]. Nihongoron
vol. 2, no. 8. 26-31. Tokyo: Yamamoto Shobo.

Nakai, Yukihiko. 1991. Terebi shutsuen-ji no akusento: Kinki chihé no
shogakusei ga Kinki rokaru no bangumi ni shutsuen suru baai
[Accents in appearing on a TV program: The case of elementary school
children in Kinki appearing in a program broadcast locally in Kinki].
Script for the presentation at the 52nd Conference of N ihon Hogen
Kenkyfikai [Japan Dialect Studies Association].

Niedzielski, Nancy. 1999. The effect of social information on the perception
of sociolinguistic variables. Journal of Language and Social Psychology,
Vol. 18, No. 1, March 1999. 62-85.

Peng, Fei. 1993. Osaka kotoba no tokucho [Characteristics of Osaka dialect].
Osaka: Izumi Shoin.

Preston, Dennis R. 1993. Folk dialectology. American Dialect Research, ed. by
Dennis R. Preston, 333-77. Amsterdam/ Philadelphia: John Benjamins.

146

Preston, Dennis R 1996. Where the worst English is spoken. Focus on the
USA, ed. by Edgar W. Schneider, 297-360. Amsterdam: John
Benjamins.

Purnell, Thomas, William Idsardi, & John Baugh. 1999. Perceptual and
phonetic experiments on American English dialect identification.
Journal of Language and Social Psychology, Vol. 18, No. 1, March 1999.
10—30.

Ryan, Ellen Bouchard, Howard Giles, and Richard J. Sebastian. 1982. An
integrative perspective for the study of attitudes toward language
variation. Attitudes towards language variation, eds. by Ellen
Bouchard Ryan and Howard Giles, 1-19. London: Edward Arnold.

Sanada, Shinji. 1991. Hyojungo wa ikani seiritsu shita ka [How the standard
language was established]. Tokyo: Kaitakusha.

Shibatani, Masayoshi. 1990. The languages of Japan. Cambridge: Cambridge
University Press.

Shibuya, Kasumi. 1995. Shinjo to wakimae ishiki no shototsu suru tokoro
[A place where emotion and discretion clash]. Gengo, Vol. 24, no.
12, Special Issue in November. 110-21. Tokyo: Taishfikan Shoten.

Strand, Elizabeth A. 1999. Uncovering the role of gender stereotypes in
speech perception. Journal of Language and Social Psychology, Vol. 18,
No. 1, March 1999. 86-100.

Strange, Winifred. 1995. Cross-language studies of speech perception: A
historical review. Speech Perception and Linguistic Experience, ed. by
Winifred Strange, 345. Baltimore: York Press.

Sugito, Miyoko. 1962. Kansai-ben ni okeru onsei-jo no tokucho oyobi butaigo
no shomondai [Phonetic characteristics and problems in dramatic
language in the Kansai speech]. Onsei no Kenkyfi [Study of Sounds] 10.
199-214. Nihon Onsei Gakkai.

Sugito, Miyoko. 1969. Tokyo, Osaka ni okeru musei-boin ni tsuite [A study

of voiceless vowels in Tokyo and Osaka]. Onsei no Kenkyfi [Study of
Sounds] 14. 249-64. Nihon Onsei Gakkai.

147

Sugito, Miyoko. 1988. Nihon no 8 toshi ni okeru boin no museika [Vowel
devoicing in 8 cities in Japan]. Osaka Shoin Women's College
Collected Essays 25. 1-10.

Tahara, Hiroshi, et al. 1998. Jinmonkon Database Vol. 2 (CD-ROM).
Kanagawa, Japan: 56gb Kenkyfi Daigakuin Daigaku.

Tse, Peter. 1993. Kansai Japanese. Rutland, VT: Charles E. Tuttle Company.

Tsujimura, Natsuko. 1996. An introduction to Japanese linguistics.
Cambridge, MA: Blackwell.

Umesao, Tadao. 1995. Keisei katei ni aru kyotsfigo [The standard
language, a language in the process of establishment]. Osaka-ben no
Sekai [The World of the Osaka Dialect], ed by NHK Osaka-ben project,
260-73. Tokyo: Keiei Shoin.

Vance, Timothy. 1987. An introduction to Japanese phonology. Albany:
SUNY Press.

Warner, Natasha. 1997. Recognition of accent patterns across dialects in
Japanese. Proceedings of the 23rd Annual Meeting of the Berkeley
Linguistics Society, in February, 364-75.

Yoshioka, Hirohide. 1981. Laryngeal adjustments in the production of the
fricative consonants and devoiced vowels in Japanese. Phonetica 38.
236-51.

Yoshioka, Yasuo. 1995. Shuto hogen no ikc’i to gengo ishiki, gengo henka
[Prestige in the Tokyo dialect, language consciousness, and language
change]. Gengo, Vol. 24, no. 12, Special Issue in November. 48-61.
Tokyo: Taishfikan Shoten.

148