 

I

WW

I
)

 

'

f

=
=
2
—
—
—"
-'—
'—_—
'——
"—
"'
_
2

949

 

THEGS

lllllllllllllllllIlllllllllillllllllllllllllllllllllllllllllll

3 1293 01570 2503

This is to certify that the

thesis entitled
VOWEL DEVOICING IN JAPANESE
presented by

Terumi Imai

has been accepted towards fulﬁllment
of the requirements for

M.A. degree in _Linguis.tics

 

 

 

_, 1 ,
ll ajor pro or
Date 5- / 7 / Q 7
07639 MS U is an Afﬁrmative Action/Equal Opportunity Institution

 

 

 

LIBRARY

Michigan State
University

 

 

 

PLACE N RETURN BOX to romovo this chockout from your rocord.
TO AVOID FINEQ rotum on or bdoro dot. duo.

DATE DUE DATE DUE DATE DUE

    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU In An Afﬂrmotivo AetiorVEquol Opportunity lnotituﬂon
Wanna-9.1

 

VOWEL DEVOICING IN JAPANESE
By

Temmi Imai

A THESIS

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

MASTER OF ARTS

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

1997

ABSTRACT

VOWEL DEVOICTNG [N JAPANESE

By

Terumi Imai

Japanese close vowels are devoiced when they occur between voiceless
consonants or between a voiceless consonant and a pause. The goal of this study is to
ﬁnd the promoting and demoting environments for the devoicing of each vowel, in terms
of place and manner of articulation of the adjacent segments. Natural conversation was
recorded on a cassette tape and an acoustic analysis program “Si gnalyze” was used to
determine the voicedness of all the vowels in the data. The encoded data were analyzed
using a statistical program “VARBRUL.” It was found that each vowel has different
promoting and demoting environments of its own, and the best promoter for the
devoicing of /i/ is a preceding fricative, but the best promoter for the devoicing of other
vowels is a preceding stop or the feature “stop.” It was also found that although the

preceding pause promoted devoicing, the following pause demoted it.

ACKNOWLEDGMENTS

My sincere appreciation is extended to the following people for their
contributions to this study:

Dr. Dennis R. Preston, my advisor, who supported me throughout the process of
completing this study. Without his patience and most invaluable advice, I would not
have been able to complete this study.

Dr. Grover Hudson and Dr. Mutsuko Endo Hudson for their most helpful advice.

Dr. Timothy Vance for his kind suggestions, and Dr. Mary Beckman of Ohio
State University for providing me with valuable articles, which I would not have been
able to obtain without her generosity.

My family and my friends for their support.

iii

TABLE OF CONTENTS

LIST OF TABLES .............................................................................................................. vi
LIST OF FIGURES .......................................................................................................... viii
1. mTRODUCTION ............................................................................................................ 1
2. PREVIOUS STUDIES ..................................................................................................... 4
2.1 Vowel Identity ........................................................................................................... 4
2.2 Preceding Segment .................................................................................................... 5
2.3 Following Segment .................................................................................................... 6
2.4 Following Vowel ....................................................................................................... 8
2.5 Accent ....................................................................................................................... 8
2.6 Gender ..................................................................................................................... 10
2.7 Data Type ................................................................................................................ 10
2.8 Summary ................................................................................................................. ll

3. METHODS .................................................................................................................... 12
3.1 Subjects ................................................................................................................... 12
3.2 Data Collection ....................................................................................................... 12
3.3 Encoding Data ......................................................................................................... l3
Codes .......................................................................................................................... 14

Si gnalyze ..................................................................................................................... 16

4. RESULTS ...................................................................................................................... 19
4.1 Cross-Tabulation ..................................................................................................... 19
4.1.1 Interaction .......................................................................................................... 21
4.1.1.1 Vowel Identity .............................................................................................. 21
4.1.1.2 Accent ........................................................................................................... 22

4.2 Knockouts ................................................................................................................ 23
4.2.1 First Knockouts .................................................................................................. 23
4.2.2 Second Knockouts ............................................................................................. 24
4.2.2.1 Vowel identity and place of articulation of the preceding segment ............. 24
4.2.2.2 Vowel identity and place of articulation of the following segment ............. 26
4.2.2.3 Vowel identity and manner of articulation of the preceding segment ......... 27
4.2.2.4 Vowel identity and manner of articulation of the following segment ......... 29

iv

4.2.2.5 Vowel identity and the identity of the next vowel ....................................... 30

4.2.2.6 Accent and place of articulation of the following segment .......................... 31
4.2.2.7 Accent and manner of articulation of the following segment ...................... 32
4.2.2.8 Summary of knockouts, and infrequent applications and occurrences ........ 34
4.3 Binomial Step Up & Down ..................................................................................... 35
4.4 Binomial One Level ................................................................................................ 37
4.4.1 Accent and Manner of Articulation of the Following Segment ......................... 39
4.4.2 Vowel Identity and Manner of Preceding Segment ........................................... 39
4.4.2.1 Fricatives ...................................................................................................... 42
4.4.2.2 Stops ............................................................................................................. 43
4.4.2.3 /r/s ................................................................................................................. 44
4.4.2.4 Vowels .......................................................................................................... 44
4.4.2.5 Summary ................................................................................... ' .................... 44
4.4.3 Voicedness of the Preceding and Following Segments ..................................... 44
4.4.4 Accent ................................................................................................................ 47
4.4.5 Vowel Length ..................................................................................................... 48
4.4.6 Gender ................................................................................................................ 49
4.5 Other Observations .................................................................................................. 50
4.6 Summary ................................................................................................................. 52
5. CONCLUSION .............................................................................................................. 57
APPENDICES .................................................................................................................... 59
APPENDIX A. Sample Transcription ........................................................................... 59
APPENDIX B. Sample Cross-Tabulation with No Interaction (Subject Identity and
Vowel Identity) ............................................................................................................. 60
APPENDIX B. Sample Cross-Tabulation with Interaction (Vowel Identity and Place
of Articulation of the Preceding Segment) ................................................................... 61
BIBLIOGRAPHY ............................................................................................................... 62

LIST OF TABLES

Table 1 - Sample Coding ................................................................................................... 15
Table 2 - Cross-Tabulation (Subject Identity and Vowel Identity) ................................... 20
Table 3 - Cross-Tabulation (Vowel Identity and Following Place) ................................... 21

Table 4 - Knockouts and Infrequent Applications and Occurrences
(Vowel Identity and Preceding Place; Applications and Total Tokens) ........... 25

Table 5 - Infrequenst Applications and Occurrences
(Vowel Identity and Following Place; Applications and Total Tokens) ........... 27

Table 6 - Knockouts and Infrequent Applications and Occurrences
(Vowel Identity and Preceding Manner; Applications and Total Tokens) ....... 28

Table 7 - Knockouts and Infrequent Applications and Occurrences
(Vowel Identity and Following Manner; Applications and Total Tokens) ....... 29

Table 8 - Knockouts and Infrequent Applications and Occurrences
(Vowel Identity and Next Vowel; Applications and Total Tokens) ................. 30

Table 9 - Knockouts and lnﬁequent Applications and Occurrences
(Accent and Following Place; Applications and Total Tokens) ....................... 31

Table 10 - Knockouts and Infrequent Applications and Occurrences

(Accent and Following Manner; Applications and Total Tokens) .................... 33
Table 11 - Summary of Knockouts and Infrequent Applications ...................................... 35
Table 12 - Binomial One Level (Vowel Identity and Preceding Manner) ......................... 40
Table 13 - Binomial One Level (Vowel Identity and Preceding Manner - 2) ................... 42

Table 14 - Binomial One Level (Voicedness of the Preceding Segment) ......................... 45

vi

Table 15 - Binomial One Level (Voicedness of the Following Segment) ......................... 46

Table 16 - Binomial One Level (Voicedness of the Following Segment - 2) .................... 46
Table 17 - Binomial One Level (Accent) ........................................................................... 47
Table 18 - Binomial One Level (Accent - 2) ..................................................................... 48
Table 19 - Binomial One Level (Vowel Length) ............................................................... 48
Table 20 - Binomial One Level (Gender) .......................................................................... 49

vii

LIST OF FIGURES

Figure l - Sample Wave Forms .......................................................................................... 18

viii

1. INTRODUCTION

It is a widely known phenomenon that the Japanese close vowels (/i/ and /u/) are
devoiced or deleted when they occur between voiceless consonants or between a
voiceless consonant and a pause (Sakuma 19292264, Kindaichi, eta]. 1988296 and 247-8,
Maekawa 19882136, Sugito and Hirose 1988221, Sugito 198821, Jun and Beckman
19942479, Kondo 1994261, and 19952238, Nagano-Madsen 19952564). It was long
assumed that the devoiced vowel was not accented, which turns out not to be correct
(Han 1962, Sugito and Hirose 1988).

In the literature it has been noted that open vowels (/a/, /e/ and /o/) can also be
devoiced in certain cases, although rarely (Shimizu 19292229, Han 1966282, Kindaichi,
et. a1. 1988:248, Maekawa 19882136, Sugito 198829). The devoicing of open vowels,
however, seems to happen more frequently than is discussed in the literature, especially
in casual conversation. In most of the literature it is neglected because of the low
frequency of occurrence, partially because earlier researchers used a reading passage to
elicit their data, which is not a natural speech style. I hypothesize that, even within such
low frequency, we could still ﬁnd the promoting and demoting factors for the devoicing
of open vowels.

On the other hand, the devoicing of close vowels does not occur categorically even
in the environment described above. In the most favored environment, devoicing appears
to be variable. There are some earlier reports on the environments that seem to promote

and demote the devoicing of close vowels (Han 1966:88-89, Maekawa 1983:70-71,

2
Maekawa 1988:144-149, Sugito 1988:] and 9, Jun and Beckman 19942479, Nagano-

Madsen 19952564); however, most of them were not statistically well supported. There
are also earlier reports of the devoicing of close vowels after a voiceless consonant but
before a voiced segment. For example, Maekawa (1983 and 1988) and Sugito ( 1988)
mentioned cases in which close vowels were devoiced before a voiced segment, but some
details of their ﬁndings, discussed below, contradict one another. It is assumed that
devoicing usually occurs after a voiceless consonant. Although there is no integrated
report on the environments that promote and demote the devoicing of each vowel
separately, I hypothesize that there may be such environments.

Thus, although the study of vowel devoicing in Japanese has been discussed
frequently in the literature, it is still a controversial issue in terms of the promoting and
demoting factors. Furthermore, a careful study of the devoicing of open vowels has not
been done. In this research I investigated the preceding and following environments of
all the vowels in terms of manner and place of articulation, voicedness of the segment,
and the place of an accent within a phrase, as well as the identity of the vowel itself,
length of the vowel, and gender difference, utilizing a sophisticated statistical program
“VARBRUL.” I used natural conversation to investigate these environments since it is
the style that seems to promote the overall occurrence of the devoicing but has not been
used for this kind of study before. The goal of this study is to ﬁnd the promoting and
demoting factors for each vowel and the relative signiﬁcance of those factors.

In the next section, I will review previous research on vowel devoicing in Japanese

and recent ﬁndings of the researchers who have investigated this phenomenon. Section 3

3
describes the methodology, and Section 4 describes and explains the results of the

statistical treatment in detail. Finally, Section 5 concludes with a summary of this

research.

2. PREVIOUS STUDIES

The devoicing of the Japanese close (or high) vowels /i/ and /u/ has been well
studied (Sakuma 1929:227-234 and 262-269, Miyajima 1961, Arisaka 1964, Shimizu
1965, Maekawa 1983 and 1988, Kindaichi et. al. 1988296 and 247-8, Sugito 1988, Sugito
and Hirose 1988, Jun and Beckman 1993 and 1994, Kondo 1995, Nagano-Madsen 1995,
Beckman 1996). However, the devoicing of the open vowels (/a/, /e/ and /o/) has not
been fully discussed. This section provides a review of the discussion in the previous

research.

2.1 Vowel Identity

In most of the literature the devoicing of open vowels is not discussed (Kindaichi
1967) or it is mentioned that the devoicing of open vowels is so rare that it could be
ignored (Han 1962, Maekawa 1988, Sugito 1988, Jun and Beckman 1994). Kindaichi, et.
a]. (19882248) note the devoicing of open vowels, but only the fact that it can be observed
in such words as /k§taéi/ and /kgkoro/ (underline indicates the devoiced vowel), and no
explanation is given. Maekawa (1983) states that no devoicing of open vowels was
found in his study.

However, Maekawa (19882136) states that the devoicing of /a/ and /o/ is also

possible in such words as /kakaji/ (‘a scarecrow’), lkgkoro/ (‘spirit’), Argka/ (‘a grave’)

and /h_qkori/ (‘dust’), but that /e/ is less likely to be devoiced than /a/ and /o/, although he

5
states that Sakuma (1929) provides such examples as /k§j'jite/ (‘never’) and /sgkkaku/

(‘especially, purposely’) in which the vowel /e/ can be devoiced.

Even within discussion of the close vowels, there are some conﬂicting arguments.
Han (1962: 84) argues that /u/ is more readily devoiced than /i/ in standard Japanese.
However, Sugito (198826) argues that /i/ is more frequently devoiced than /u/ in western
Japan. On the contrary, Maekawa (1983 and 1988) states that he found no difference
between the devoicing of /u/ and that of /i/ (1983:71and74, 19882144).

Maekawa (19882142) discusses the social importance of the devoiced close
vowels. For example, if a person does not devoice the ﬁnal vowel /u/ in /so:desg/ (‘it is
so’), which did not occur in this study because the data were taken from casual
conversation, speakers of the Tokyo dialect would not think the person as a speaker of
the Tokyo dialect. In other words, the ﬁnal vowel must be devoiced in the Tokyo dialect.
The degree of importance of devoicing, however, is very low, if any, for the Open vowels
(Maekawa 19882137). Therefore, it may be reasonable to draw a line between the
devoicing of close vowels and that of open vowels according to some social requirement.
The social consequences of devoicing (or failure to devoice) are not taken up in this

study.

2.2 Preceding Segment
Han (1962:89) argues that fricatives have a greater effect on the devoicing of the
vowels than affricates, which, in turn, have a greater effect than stops. Particularly, /s/

has the strongest effect on the devoicing of /u/. In general, she argues that the fricatives

6
/s/ and /j/ have such a strong effect that following close vowels are devoiced even in an

accented syllable, as in ljiku/ (‘four by nine’). However, she does not mention the

possibility of other environments promoting the devoicing of /i/ and /u/ differently.

While Han ( 1962291) groups affricates and fricatives as non-stops as opposed to
stops and argues that non-stops promote devoicing, Maekawa (1983271 and 77) groups
stops and affricates together as opposed to fricatives, which, like Han, he argues promote
devoicing most. Maekawa ( 19882145) reconﬁrms that fricatives promote devoicing more
than stops and affricates do. In any case, the common conclusion is that fricatives
promote the devoicing of vowels most, but the relation of this argument to vowel identity
has not been mentioned.

In all the literature it is assumed that the segment preceding a devoiced vowel is a
voiceless consonant. In fact, Maekawa (1988:136-7) states that the only condition for
vowel devoicing is that the preceding consonant be voiceless; I will show in a later

section that this is not always the case.

2.3 Following Segment
Han (1962:87) argues that when a vowel is preceded by a voiceless consonant and
followed by a voiced consonant, the vowel is not devoiced, and the only exception is

when a close vowel is preceded by /s/ or /j/ and followed by a semi-vowel /y/. However,

Maekawa (1983277 ) mentions the devoicing of vowels followed by a voiced consonant.

According to Maekawa, following nasals promote the devoicing of close vowels in the

7
Tokyo dialect, and he reconﬁnns this in Maekawa 19882145 in other dialects. Maekawa

(19882145) also states that the following /r/ also promotes the devoicing of the vowel /i/.

On the other hand, Sugito (198829) argues that following nasals demote the
devoicing of close vowels. She further argues that the devoicing of close vowels when
preceded by a voiceless consonant and followed by a voiced consonant is so rare that it
can be ignored (198827).

Contrary to Maekawa and Sugito’s argument, Nagano-Madsen (1995) states that
most studies agree that the following consonant is the most decisive factor for vowel
devoicing and argues that a following stop promotes the devoicing more than fricatives
do. Nagano-Madsen further argues that there are differences among fricatives and that /s/
and /h/ behave the same as stops and affricates. Jun and Beckman ( 1994) also argue that
the following consonant is one of the factors that determine the devoicing of high vowels
in Korean; stops promote devoicing more than fricatives and fortis consonants promote
devoicing more than lenis ones.

Thus, there are conﬂicting arguments regarding the importance of the following
segment to vowel devoicing in the literature. It seems that there are different arguments
in different linguistic subﬁelds, namely phonetics and sociolinguistics. Phoneticians
(e.g., Nagano-Madsen, Jun and Beckman) argue that the following segment is the
decisive factor for devoicing while sociolinguists (e. g., Maekawa and Sugito) argue no
such importance for the following segment. Most phonetic studies used words in
isolation, and sociolinguistic studies used a reading passage. Since this study used

natural conversation, which is closer in style to a reading passage than to a word list, it is

8
predicted that this study will show results similar to those obtained by the sociolinguistic

01165.

2.4 Following Vowel

Maekawa (19882145) argues that the vowel in the following syllable has an effect
on the devoicing of close vowels, and it is open vowels that promote the devoicing of the
close vowels in the preceding syllable. Therefore, the identity of the vowel in the

following syllable is included in this study.

2.5 Accent

Japanese has a pitch accent system (high and low) (Kindaichi 1966240-47,
Shibatani 1990:177-184) and stress (or an accent) is usually considered to be on a
syllable which has a high pitch followed by a syllable that has a low pitch (where a drop
of pitch accent occurs). Once the drop of a pitch in a word is known, the pitch accent
(high or low) of the remaining syllables is predictable (Shibatani 1990:177-184).

It has been assumed in the literature that devoicing does not occur in accented
syllables (Sakuma 1929:227-234, Han 1962, Shibatani 19902161). Han (1962:85—6)
argues that pitch accent has a strong effect on vowel devoicing, and when a close vowel
is in an accented (or stressed) syllable, devoicing is not common. However, Maekawa
( 1983) also examines the effect of pitch accent and argues that it does not have effect on

vowel devoicing except for /j‘i/ and /su/. He also argues that /ji/ and /su/ in copula forms

are devoiced at a high frequency. Han considers only accented syllables, where a drop of

9
pitch accent occurs; however, it is not clear if Maekawa means “an accented syllable” by

“high pitch” or any high pitched syllable, and “a syllable after the accent” by “low pitch”
or any low pitched syllable.

Maekawa (1988) states that although it is believed that vowel devoicing cannot
occur in accented syllables (or that when it occurs, the accent is shifted to the following
syllable), it has been observed that accented syllables can be devoiced and still bear the
accent (Sugito and Hirose 1988). They argue that the reason that we can perceive the
accent in devoiced syllables is that there is a sharp descending FO—contour in the
following vowel (i.e., of the fundamental frequency or “pitch”). Nagano-Madsen (1995 )
states that the inﬂuence of the accent is quite strong, and the accent is shifted to either
the preceding or the following vowel if the underlyingly accented syllable is devoiced, or,
once the devoiceable vowel is accented, it resists devoicing, at least in the studies using
words in isolation. Other studies, which use “longer words and sentence databases”
(Kondo 1993, Nagano-Madsen 1994) on the contrary, found no inﬂuence of the accent in
resisting devoicing.

Thus, there are contradictory statements in the literature between those studies
which used isolated words and those which used more natural data. Since this study used
natural data, it was predicted that the accent will not have a strong inﬂuence on

devoicing.

10
2.6 Gender

Since most of the sociolinguistic studies of vowel devoicing have only female or
male subjects, there is no literature that deals with gender difference in this phenomenon.
It is well known that female speech and male speech are different (Wardhaugh 19922312-
345) and, especially, that women are more advanced (i.e., the promoters of a change)
than men in certain vowel shifts (Labov 19942156). Women are also “leaders” in the use
of language standards and the adoption of new ones (Wardhaugh 19922312-345).
Therefore, it is reasonable to hypothesize that there may be a gender difference in the
frequency of vowel devoicing and that those differences may help us determine the

“standardness” of vowel devoicing in general.

2.7 Data Type

In most of the sociolinguistic studies of vowel devoicing, the data come from
reading passage style (including TV news programs) (Maekawa 1983 and 1988, Sugito
1988). On the other hand, most of the phonetic studies use more manipulated data such
as words in isolation, nonsense words, and controlled dialogue (Jun and Beckman 1993
and 1994, Kondo 1994, Nagano-Madsen 1995). Natural data, however, has not been
used in the study of vowel devoicing in Japanese. All the data in this study were taken
from spontaneous speech between friends, which is most likely to be casual and rapid It
is known that in careful speech, which is rather slow, vowel devoicing is less likely, but

in rapid casual speech, it is more likely (Han 1966, Kondo 1994).

1 1
2.8 Summary

As shown in the previous subsections, there are conﬂicting statements in the
literature regarding the devoicing of open vowels, the frequency difference in the
devoicing of close vowels, the importance of the following segment, and the inﬂuence of
accent. There seems to be an agreement that the preceding fricative promotes the
devoicing of close vowels. However, we do not know to what degree preceding
fricatives promote devoicing in comparison to preceding stops and affricates. We do not
know if the preceding fricative is a promoter for the devoicing of the vowel /i/ or /u/, or
for both. This study aims to investigate the promoting and demoting factors for the
devoicing of each vowel and the degree of signiﬁcance of those factors in the natural

data.

3. METHODS

3.1 Subjects

The subjects in this study were two females and two males in their late twenties’
from Tokyo and Yokohama, who speak the Tokyo dialect. Maekawa (19882151) found
that younger subjects had more devoicing than older respondents in some dialects, but he
did not ﬁnd this age difference in others. I chose younger subjects because the younger
generation may provide evidence of linguistic change (Labov 1994247), and Maekawa
conﬁrms this age effect in at least one dialect.

I chose as my subjects those who speak the Tokyo dialect because it is one of the
dialects in which devoicing is likely to occur, although it is not the one with the greatest
frequency]. The frequency of devoicing seems to vary in different dialects (Maekawa

1988, Sugito 1988).

3.2 Data Collection

Casual conversations of the subjects were recorded on a cassette tape using a
walkman-type recorder, SONY TCM-47 Cassette—Corder. The subjects were asked to
carry on a conversation with their friends or with the researcher. Therefore, the

conversations are informal, casual ones and seem to have generally avoided the

“observer’s paradox” (Labov 19722209).

1. According to Sugito (1988:6-7), devoicing occurs most frequently in Nagoya.

12

13
Each conversation used as data is about ﬁve minutes in length. There are

approximately 500 tokens (vowels) per subject.

3.3 Encoding Data

First, each conversation was transcribed (refer to APPENDIX A for a sample
transcription). Then each potentially inﬂuencing environment was coded into a
spreadsheet (Microsoft Excel), for every vowel in the data, in terms of the following

categories.

0 Subject identity
0 Gender of the subject
0 Vowel identity

0 Place and manner of articulation and voicedness of the
preceding segment

0 Place and manner of articulation and voicedness of the
following segment

0 Identity of the following vowel
0 Location of accent
o Voicedness of the target vowel

0 Length of the vowel

14

Codes (Table 1 illustrates a sample coding.)

Subject identity

Since there are four subjects, the codes are the numbers 1, 2, 3 and 4.
Vowel identity

The codes are the 5 vowels; a, e, i, o and u.
Place of articulation of the preceding and following segment

There are four codes; ‘1’ (labial), ‘a’ (alveolar), ‘p’ (palatal), and ‘v’ (velar).
When the syllable is preceded or followed by a pause, it was left blank, which
was later coded as ‘x.’

Manner of articulation of the preceding and following segment

There are twelve codes for this category; ‘5’ (stop), ‘f’ (fricative), ‘a’ (affricate),
‘n’ (nasal), ‘g’ (glide), ‘r’ (/r/), ‘p’ (pause), and the numbers 1, 2, 3, 4 and 5 for
the vowels (a = 1, i = 2, u = 3, e = 4, o = 5). /r/ is treated as an independent
category in Maekawa 1988, and he reports that it promotes the devoicing of the
preceding /j'i/. Therefore, it is treated separately as one category here.

Voicedness of the preceding and following segment
Codes are ‘-‘ and ‘+’.
Identity of the following vowel
Same as Vowel Identity above; a, e, i, o and u.
Location of accent

There are six codes; 1 = accent on this segment (i.e., on the segment which is
potentially devoiced), 2 = accent on the preceding segment, 3 = accent on the
following segment, 4 = none of above (accent not on this segment nor on the
preceding or following segment), 5 = accent on the preceding segment and
following segment, 6 = accent on this segment (in phrase ﬁnal position). The
location of the accent is based on the actual utterances recorded on the tape, not
on the lexical word accent.

15

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

.« + N a + a a u m > a v
m + ~ a .. m > + p a s v
m + m s + u a + m o v
m + v o + v - m > o v
m - w o - m > - m > _ v
m + v m + m > u m > o m
m - v o - m > - m > o m
a + N a + m _ m
m + E _ + m + m > o m
a - m o + m > - m m s m
+ E + E o - m m - m a o m
E + m o + a > t m a o N
E .+. V 0 r u— s + w _ m N
E + v a + w _ - m > a m
E + v a - m > - m a a m
E + v _ + m - m a a E
E + v e .. m a m s E
E + m + E > + m _ a _
E .. E a + m _ - a a _ ﬂ
E + m _ - a a m _ _
law—BA 5980 36> Eooo< _oBo> 36> 6582 33m 86> BEBE com—m 3:52 Suﬁsm
32:5 :82 EoEwom wage—Em Eo§m 9:385 _o>»o>

 

9:80 295E - E 2%;

 

16
o Voicedness of the target vowel

Same as the voicedness of the preceding and following segment; ‘-‘ and ‘+.’
0 Gender of the subject

There are two codes; ‘m’ (male) and ‘f (female).
0 Length of the target vowel

When the vowel was long, ‘+’ was coded and when it was short, it was left blank.

Signalyze

The data were transferred into the computer, and acoustic analysis was done on
each vowel to identify its voicedness using the speech analysis program “Signalyze.”
The recorded sounds were transformed into sound wave forms in Signalyze (see sample
wave forms in Figure 1). Although vowels have a more sinusoidal (“ﬁshbone”) shape
than obstruents, nasals and semi-vowels also have a sinusoidal shape which resembles
that of vowels. Figure 1 shows sample wave forms for the phrases; 1. /~tatte#kara/

(“after ~ period of time has passed”), 2. /taijita#koto#nai/ (“it is nothing”), and 3.

/kikoeru#kana/ (“I wonder if this can be heard”). In 1, the second vowel is between
voiceless consonants and in a position in which devoicing is likely to occur. However,
the vowel /e/ was pronounced, although very softly, as we can see from the small

sinusoidal peak. On the other hand, the vowel /i/ in the syllable /ji/ in 2 is completely

devoiced. There is not even a trace of a voicing. The same can be said of the syllable

/to/. We can barely see the wave. The ﬁrst syllable in 3, /ki/, is also devoiced. Although

17
we can see some amplitude, it came from the stop /k/, not from the vowel /i/, because we

cannot see the sinusoidal shape.

Whenever the judgment between the sinusoidal shape contributed by a vowel or a
nasal or a semi-vowel was difﬁcult, spectral analysis was done to obtain F1 and F2
numbers using LPC (Linear Predictive Coding), because vowels usually have good peaks
of F1 and F2, whereas non-vowels, including nasals and semi-vowels, do not have such
clear peaks. Spectrographic analysis was not used since it is too time-consuming to
obtain a spectrogram of each vowel, although a few samples were examined to conﬁrm
the above techniques.

The resulting data were interpreted by “Goldvarb,” a statistical analysis program,
to determine the factors that have signiﬁcant effects on vowel devoicing in Japanese and
the precise degree of those effects. The next section discusses the procedures and results

of the statistical operation as well as the ﬁndings obtained through the operation.

18

av.

 

mEuom 3a? 295% - _ Bsmﬁ

E o

B

86v.

3

9.

 

HE

4. RESULTS

“Goldvarb” is a Macintosh version of the statistical analysis program
“VARBRUL,” which makes it possible to identify the inﬂuence of various factors
(arranged in “factor groups”) on a dependent variable. When one feature of the
dependent variable is chosen (here devoicing) it is said that the “rule” applies. The
degree to which individual elements within a factor group inﬂuence rule application
(i.e., the probability) is identiﬁed on a “weight” range from 0.0 to 1.0. Weights below
.50 “disfavor” or “retard” application of the rule; those above .50 “favor” or “promote”
it. Goldvarb also permits the study of interactions among the factor groups requiring
them to be rearranged as a new (single) complex group. In the ﬁrst Goldvarb run, there
were several factors that were “knocked out.” If a factor shows no variation, that is, no
devoicing occurred or devoicing always occurred in that environment, the factor is

excluded. I will discuss these knockouts in more detail in 4.2.

4.1 Cross-Tabulation

After the initial knockouts were eliminated, Cross-Tabulation was done to
discover possible interaction between factor groups. Every possible combination of the
twelve factors was examined for possible interaction. For example, the Cross-Tabulation
of factor groups 1 (subject identity) and 2 (vowel identity) showed no interaction (Table
2). The relative order of the frequency of devoicing (derived from percentages) is given

in Table 2, instead of the actual number of occurrences and percentages, for

19

20
simpliﬁcation. (See APPENDIX B for the actual numbers and percentages.) Number 1

identiﬁes the most frequent and number 5 the least. If the table shows (generally) the

same order, there is no interaction between the two factor groups under consideration.

Table 2 - Cross-Tabulation (Subject Identity and Vowel Identity)

 

Subject #1 Subject #2 Subject #3 Subject #4
i 2 2 2 1
a 5 5 5 5
u 1 1 l 2
e 3 3 4 3
o 4 4 3 4

 

 

 

 

The vowel /u/ was most frequently devoiced by all of the subjects except Subject
#4, and it was in second place for her. /i/ was next, then /e/ and /o/, and /a/ was most
unlikely to be devoiced for all the subjects, which conﬁrms Han’s (1966) claim that /u/ is
more often devoiced than /i/, but contradicts Maekawa’s (1988) statement that /e/ is less
likely to be devoiced than /a/ and /o/. Subjects showed very similar patterns for the
devoicing rate of each vowel (which was expected from a linguistic point of view). If the
nature of the vowel itself is one of the factors for devoicing, then it should not vary
individually. Therefore, it is reasonable to conclude that there is no interaction between
vowel identity and subject identity.

However, the Cross-Tabulation of factor groups 2 (vowel identity) and 3 (place of
articulation of the preceding segment) showed interaction (Table 3). Palatals promote

the devoicing of /i/, but not the other vowels. Labials are the second most powerful

21
promoter of devoicing for the vowel /u/, but it demotes the devoicing of other vowels.

We do not see any similarity in the order; therefore, we suspect an interaction between

vowel identity and the place of articulation of the preceding segment.

Table 3 - Cross-Tabulation (Vowel Identity and Following Place)

 

 

 

 

 

 

1 a u e 0

none 4 3 5 3 4
palatal 1 5 4 4 3
labia] 5 4 2 5 5
affricate 3 2 3 1 1
velar 2 1 1 2 2

It is interesting that the Cross-Tabulation showed that vowel devoicing might be
inﬂuenced by the interaction of the place of articulation of the preceding segment and
vowel identity, a possibility which has never been reported in the literature. However,

the complex factor group based on this interaction turned out not to be signiﬁcant.

4.1.] Interaction

4.1.1.1 Vowel identity As a result of these Cross-Tabulations, several
interactions were observed. Vowel identity interacted with place of articulation of the
preceding segment, as illustrated above. It also interacted with manner of articulation of
the preceding segment. The importance of the manner of articulation of the preceding
segment has been discussed in the literature, but there are no reports on the manners of

articulation inﬂuencing the vowels differently.

22
Place and manner of articulation of the following segment also showed an

interaction with vowel identity. Again, there are some reports and claims in the literature
about the importance of manner of articulation of the following segment to vowel
devoicing (see pages 6 and 7), but there is no report on the place of articulation of the
following segment which promotes or demotes devoicing. Also, although following
stops and fricatives are claimed to promote vowel devoicing, the Cross-Tabulation
implies that there is a possibility that each vowel is promoted and demoted differently in
terms of place and manner of articulation in the following segment.

The last group that interacted with vowel identity is the identity of the following
vowel. It is noted in Maekawa 1988 that the devoicing of close vowels is likely to occur
when the vowel in the next syllable is an open vowel. This tendency was obtained in the
present research, but the difference among the ﬁve vowels is not large and it does not

seem signiﬁcant.

4.1.1.2 Accent Accent was found to interact with two factor
groups; place and manner of articulation of the following segment. This is rather
surprising since it is not very easy to imagine the inﬂuence of the accent on the place or
manner of articulation of the following segment from a linguistic point of view. I’ll

leave discussion of this possibility for a later section.

Whenever a possible interaction was observed, the two factor groups were

combined. Each factor within a group was combined with each factor in the other group.

_ 23
Thus, if one group (X) has ﬁve factors in it and another (Y) has four, the result of the

combination of each factor should be twenty new factors in the new factor group, i.e., a
combination of factor groups X and Y. The new factor groups underwent a second
statistical run and a new Result ﬁle was obtained, in which quite a few additional
knockouts were found. The next subsection explains the knockouts and infrequent

applications, which are all demoting environments for vowel devoicing.

4.2 Knockouts
4.2.] First Knockouts

Those factors that showed no variation are knocked out. There were several
knockouts in the ﬁrst Result ﬁle (before Cross-Tabulation). These knockouts were all
instances of the environment in which no devoicing occurred The following is a list of

all the knockouts in the ﬁrst run.

0 Preceding /i/ 2 Devoicing never occurred after the vowel /i/ out of 22 total
occurrences.

o Preceding /u/ 2 Devoicing never occurred after the vowel /u/ out of 19 total
occurrences.

o Preceding /o/: Devoicing never occurred after the vowel /0/ out of 32 total
occurrences.

0 Following /a/ 2 Devoicing never occurred before the vowel /a/ out of 20 total
occurrences.

0 Following /u/ 2 Devoicing never occurred before the vowel /u/ out of 13 total
occurrences.

0 Following /e/ 2 Devoicing never occurred before the vowel /e/ out of 8 total

OCCIIITBI'ICCS.

24

0 Accent Type 6: Devoicing never occurred on a stressed phrase ﬁnal vowel out of
115 total occurrences.

Thus, vowel devoicing is not likely to occur next to the most prominent sound,
namely a vowel, and in the syllable that has a phrase ﬁnal stress. Both results naturally
follow from the fact that the environment in which devoicing typically occurs is between
voiceless consonants and that a vowel with phrase ﬁnal stress is usually associated with
the elongation of the vowel, which is not likely to be devoiced. These factors were

eliminated before Cross-Tabulation

4. 2.2 Second Knockouts
After Cross-Tabulation was done and interacting factor groups were combined,
there were quite a few new knockouts. I will discuss the speciﬁc environments in which

no devoicing was observed (i.e., knockouts) in the following subsections.

4. 2. 2. l Vowel identity and place of articulation of the preceding segment

Table 4 shows the actual number of devoiced vowels and the number of
occurrences of the environment for the complex factor group which consists of vowel
identity and place of articulation of the preceding segment. (The ﬁrst number is the
number of devoicings, and the second number is the total occurrences. The number in
parentheses is the percentage.) “x” in the table means either a pause or another vowel,
which I did not code for place of articulation. Darkly shaded boxes are the knockouts,

lightly shaded boxes are the environments with a very small number of devoicings (5 or

25
below), and boxes with diagonal shading are the environments with a low frequency of

occurrences (20 or below).

Table 4 - Knockouts and Infrequent Applications and Occurrences
(Vowel Identity and Preceding Place; Applications and Total Tokens)

alveolar l 5.
velar 1 8/1 1 l 2
73/ 1 67.6

labial 3/1
x

 

When the vowels /u/, /e/ and /0/ were preceded by either a pause or another
vowel, no devoicing occurred. However, I did not obtain the same results for /a/ and /i/,
although the number of rule-applications (i.e., devoicings) is very small. Further, if /e/ or
/0/ was preceded by a labial consonant, no devoicing occurred. Also, when /a/ and /i/
were preceded by a labial consonant, the devoicing was retarded as we can see from the
small number of rule-applications. Since all these environments have a good number of
occurrences, except “x” for /e/ and /u/ and “labial” for /i/, we can reasonably conclude
that a preceding pause and vowel are demoting environments for the devoicing of all
vowels and a preceding labial is also a demoting environment for the devoicing of all
vowels, except /u/.

The last knockout is the vowel /e/ with a preceding palatal. However, since there

were no occurrences of the environment in my data, we cannot conclude that a preceding

palatal is a demoting environment for the vowel /e/. If we look at the number of

26
rule-applications for the other vowels, however, it is small, except for /i/, and it indicates

that a preceding palatal might demote devoicing, although the amount of data for /u/ is
unfortunately small.

As for the other environments, velars do not demote the devoicing of /a/, /i/ and
/u/, but somewhat demote the devoicing of /e/ and /o/, and alveolars demote the
devoicing of /i/ only.

The interesting fact in the Table 4 is that the preceding palatal demoted the
devoicing of all the vowels except /i/, and the preceding labial demoted the devoicing of
all the vowels except /u/. The vowel /i/ is usually considered to be associated with the
feature “palatal” and the vowel /u/ with the feature “labial.” Therefore, it seems that
there is a correlation between the vowel and the preceding segment, namely, when a
vowel shares a place feature with the preceding consonant, the devoicing of the vowel is

demoted.

4. 2. 2.2 Vowel identity and place of articulation of the following segment
Although no knockouts were found in this complex factor group, there were some
factors that had a very small number of applications and/or occurrences. Table 5 shows

the number of rule-applications and the total tokens.

27

Table 5 - Infrequent Applications and Occurrences
(Vowel Identity and Following Place; Applications and Total Tokens)

e o
alveolar l . . 23/91 .3 24/1 1
velar 13/14] 22/5 40. ll 1 14.]

10 1
labial 1 9
x

 

A following pause, vowel, labial, or palatal all seem to demote the devoicing of
all the vowels. A following velar also demotes the devoicing of the vowel /e/, but
following alveolars and velars, in general, do not demote devoicing.

It is interesting that the orders of the factors, which I deliberately ordered
according to the numbers of devoicings with higher numbers on the top and smaller
numbers on the bottom, in Tables 4 and 5 are the same. It might indicate the general
tendency of the place of articulation to promote and demote vowel devoicing in both

preceding and following environments.

4. 2.2.3 Vowel identity and manner of articulation of the preceding segment

No vowels were devoiced when preceded by another vowel, except the vowel /i/.
(See Table 6; please note that vowels /i/, /u/, and /0/ were knocked out in the ﬁrst run;
therefore, they are not included in the table. See page 20.) Even though /i/ was devoiced
when preceded by /a/ or /e/, the number of the rule-applications is very small. Therefore,

it may be safe to conclude that a preceding vowel demotes devoicing, which is the same

28
result obtained from the place of articulation of the preceding segment. However, it

should be noted that the vowel /i/ was devoiced even after the strongest demoter.

Table 6 — Knockouts and Infrequent Applications and Occurrences
(Vowel Identity and Preceding Manner; Applications and Total Tokens)

r
1 40.4
fricative 4
affricate 21/41 51
nasal

/r/

 

Also, no vowel was devoiced alter a pause, except /a/. Again, the devoicing of /a/
after a pause occurred only once, and it is reasonable to say that a preceding pause also
demotes vowel devoicing, which is consistent with the results obtained from the place of
articulation of the preceding segment. A preceding glide is the same as a preceding
pause; it demotes the devoicing of all the vowels.

In addition, a preceding sonorant consonant (i.e., nasal and /r0 seems to demote
or knockout devoicing. Since vowels, glides and sonorant consonants are all voiced, it is
natural to think of them as demoters in the manner of articulation of the preceding
segment. Also, it is obvious from the table that only a preceding stop does not demote

the devoicing of any vowel, but preceding fricatives and affricates do not demote the

29
devoicing of /i/ and /u/. This reﬂects the fact that the close vowels (/i/ and /u/) are more

often devoiced than the open vowels (/a/, /e/, and /o/).

4. 2.2.4 Vowel identity and manner of articulation of the following segment

Knockouts in this complex factor group are mainly by a following vowel and
pause, with the exception of the vowel /i/ which is also knocked out by a following glide.
(See Table 7; please note that vowels /a/, /u/, and /e/ caused knockouts in the ﬁrst run;
therefore, they are not included in the table. See page 20.) However, if we include the
environments in which vowels were devoiced 5 or fewer times, we can arrive at a better

generalization.

Table 7 - Knockouts and Infrequent Applications and Occurrences
(Vowel Identity and Following Manner; Applications and Total Tokens)

u 6
35/81 43 15.] 19/1 12.]
fricative 1 1/41 9/21 1 1
/r/ l 1
nasal

 

A following vowel, pause, glide or affricate demotes the devoicing of all the

vowels. A following nasal or /r/ also demotes the devoicing of la/ and /e/, but not of /i/,

30
/u/ and /o/, and a following fricative demotes the devoicing of /a/ only. Finally, a

following stop does not demote the devoicing of any vowel.

4. 2. 2.5 Vowel Identity and the identity of the next vowel

Only two lmockouts were observed in this complex factor group; /a/ and /e/
before a pause. (See Table 8; please note that “x” means a pause.) However, if we again
include “marginal” devoicing (5 or below), a following pause demotes the devoicing of
all the vowels. This conﬂicts with the claim that following pause promotes devoicing
(Maekawa 19882147-8, Sugito 198825). However, it is reported that the /u/ in /~desg/
(copula), before a sentence ﬁnal pause, is almost categorically devoiced, but a phrase
ﬁnal pause demotes devoicing (Maekawa 1988:147-8, Sugito 198825). Since the data in
this study is taken from casual conversation between friends, they do not contain the
formal sentence ﬁnal copula, /~desu/. This seems to be the reason that a following pause

retards devoicing in this study but is found in others.

Table 8 - Knockouts and Infrequent Applications and Occurrences
(Vowel Identity and Next Vowel; Applications and Total Tokens)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

a l 1.1 e 0
/a/ 15/198(7.6) 27/115035) 19/62(30.6) 11/71055) 16/96(16.7
/O/ 3/73(4.1) 18/71(25.4) 24/59(40.7) 17/76(22.4) 13/120(10.8
/e/ 7/95(7.4) 33/62(53.2) 6/25(24.0) 2/27(7.4) 6/59(10.2
/i/ 9/151(6.0) 16/45(35.6) 9/40 22.5 3/35(8.6) 7/74(9.5)
/u/ 3/54(5.6) 4/24(16.7) 8/42(19.0) 3/32(9.4)
x jig/45(0), 4/25(16.0) 5 0/26(0) 3/58(5.2)

 

 

 

31
Following /u/ seems to demote the devoicing of /a/, /i/ and /o/. Also, the vowel /e/

is not likely to be devoiced when the next vowel is /e/ or /i/, and the vowel /a/ is not likely
to be devoiced before /o/. However, these observations do not tell us about the demoting
environments for the following vowel. We cannot ﬁnd any signiﬁcant generalization
about following vowels from this table. Therefore, I will leave the discussion open until

the section in which the results of VARBRUL are discussed.

4. 2. 2. 6 Accent and place of articulation of the following segment

If the syllable under investigation is not accented and neither the preceding nor
the following syllable has an accent (i.e., accent type 4), devoicing is not demoted by any
environments. (See Table 9; please note that the accent type 6 [phrase ﬁnal stress] was
knocked out in the ﬁrst run and, therefore, not included in this table. See page 20.) On
the contrary, if the syllable is between accented syllables (accent type 5), devoicing was

always demoted.

Table 9 - Knockouts and Infrequent Applications and Occurrences
(Accent and Following Place; Applications and Total Tokens)

2
alveolar 12/] 8.6 34/11
velar 9/1 . 1 8/71

labial 5/21
x

 

32
It is interesting that the other accent types (1, 2 and 3) show variation according to

the place of articulation of the following segment. If the accent is on that syllable,
devoicing never occurred before a pause, and the following palatal and labial also
demoted devoicing. If the accent is on the preceding syllable, a following pause, palatal
and labial demoted devoicing. However, if the accent is on the following syllable, a
following pause and labia] demoted devoicing, but a following palatal did not demote the
devoicing. Finally, a following alveolar or velar did not demote devoicing in any accent
type, except 5 and 6.

There is no good linguistic reason, however, to assume that accent location is
related to the place of articulation of the following segment and this interaction will turn

out to be insigniﬁcant.

4. 2. 2. 7 Accent and manner of articulation of the following segment

The interaction of accent and manner of the following segment is very similar to
that of the accent and place of the following segment discussed in the previous
subsection. (See Table 10; please note that the vowels /a/, /u/, and /e/, and the accent
type 6 [phrase ﬁnal stress] were knocked out in the ﬁrst run; therefore, they are not

included in this table. See page 20.)

33

Table 10 - Knockouts and Infrequent Applications and Occurrences
(Accent and Following Manner; Applications and Total Tokens)

36/1 . .3
1 . 21/] 19.3

8/5] 15. 19/1 11.
l 17.

 

If the accent is not on the syllable under investigation and neither the preceding
nor the following syllable is accented (i.e., accent type 4), only a following vowel or a
glide demoted devoicing. If the syllable is between accents (accent type 5), devoicing
was always demoted. A following affricate and pause demoted the devoicing of the
vowel of accent type 1, 2, and 3, but not 4, and a following /r/ demoted the devoicing of
the vowel of accent type 1 and 2, but not 3 and 4. Also, a following nasal and fricative
demoted the devoicing of the vowel of accent type 1, but not 2, 3, and 4. Finally, a
following stop did not demote the devoicing of the vowel of any accent type.

We may conclude that when the syllable has a phrase ﬁnal stress (accent type 6),
devoicing is demoted most, when the syllable is between accents (accent type 5),
devoicing is also demoted, and when the syllable is not accented and neither the
preceding nor the following syllable has an accent (or an accent is not around it; accent

type 4), devoicing is not demoted

34
However, if we look at the numbers of occurrences of accent type 4, the number

of devoicings is large because the number of occurrences of the environment is large. If
we look at the knockouts, the number of occurrences is small, except /i/ with a following
/i/. Also, if we look at the ordering of the environments, it is very similar to that of Table
7. Therefore, it may be the case that the manner of articulation of the following segment
does not interact with accent location but just reﬂects its inﬂuence on vowel devoicing.
The same can be said of the factor group of accent and place of articulation of the
following segment (Table 9). Since the ordering in Table 9 is identical to that of Table 5,
we can say that the inﬂuence of the following segment in terms of the place of
articulation is reﬂected in this factor group, and they do not interact linguistically. They
will not be treated as interactive groups in the following statistical work. (See 4.4.]

below.)

4. 2. 2. 8 Summary of knockouts, and tnﬁequent applications and occurrences

Whether vowels, glides or a pause were in preceding or following environment,
they were either knocked out or had a very small number of applications; therefore, they
can be considered as the principal demoters for vowel devoicing. If we include nasals
and /r/, which were also always knocked out or had a very small number of applications
when they were in the preceding segment (and also were demoters for the vowels /a/ and
/e/ when they were in the following segment), all voiced segments were demoters for
vowel devoicing. Other factors, however, showed some variation. In place of

articulation, palatals seem to be the principal demoter for devoicing, and labials are the

35
second. Table 11 is a summary of all the knockouts and small number of applications

and occurrences. (‘x’ means a knockout or a small number of applications [i.e.,
demoting environments].) As indicated above, the complex factor groups of accent and
place and manner of the following segment, and of vowel identity and the next vowel do

not seem to make linguistic sense or to be signiﬁcant. Therefore, they are not included in

 

 

 

 

 

 

 

 

 

 

 

 

 

 

the table.
Table 11 - Summary of Knockouts and Infrequent Applications
When the following Precedes the vowel Follows the vowel
factor a i u e o a i u e o
vowel x x x x x x x x x x
pause x x x x x x x x x x
glide x x x x x x x x x x
nasal x x x x x x x
/r/ x x x x x x x
affricate x x x x x x x x
ﬁicative x x x x
palatal x x x x x x x x x
labial x x x x x x x x
velar x x x
alveolar x

 

 

 

 

 

 

 

 

 

 

 

 

4.3 Binomial Step Up & Down
After all the knockouts were eliminated, a Binomial Step Up & Down was done
to see which factor groups were signiﬁcant out of the 12. The groups selected were the

following:

36

0 complex factor group of vowel identity and manner of articulation of the preceding
segment,

0 voicedness of the preceding segment,

0 complex factor group of accent and manner of articulation of the following segment,
0 voicedness of the following segment,

0 vowel length, and

0 gender.

The factor groups that were not selected by Binomial Step Up & Down as

signiﬁcant were excluded. They were the following:

0 subject identity,

0 complex factor group of vowel identity and place of articulation of the preceding
segment,

0 complex factor group of vowel identity and place of articulation of the following
segment,

0 complex factor group of vowel identity and manner of articulation of the following
segment,

0 complex factor group of vowel identity and the next vowel, and

0 complex factor group of accent and place of articulation of the following segment.

Although there were many complex factor groups that consist of interacting
factors, most of them turned out to be not signiﬁcant. In Table 8, we were not able to

make any generalizations about vowel identity and the next vowel, and this factor group

37
was not signiﬁcant. Also, we suspected that the complex factor groups that include

accent showed interaction between place and manner of the following segment, but it
merely reﬂected the inﬂuence of the place and manner of the following segment, and the
complex factor group of accent and place of the following segment turned out not to be
signiﬁcant. However, the complex factor group of accent and manner of the following
segment survived the Binomial Up & Down. The discussion on this factor group is
provided in 4.4.].

Although we found interesting interactions between vowel identity and place and
manner of the following segment and place of the preceding segment, these factor groups
turned out not to be signiﬁcant. Finally, subject identity was not signiﬁcant, which is
natural from a linguistic point of view.

The next subsection discusses the results of the actual VARBRUL run, called

“Binomial One Level.”

4.4 Binomial One Level

As a result of the Binomial One Level, we get a list of the environments of the
factor groups with a weight for each factor. If the number is above .50, the environment
“promotes” devoicing, and if the number is below .50, the environment “demotes”
devoicing. The higher the number, the more the environment promotes or favors
devoicing, and the lower the number, the more the environment demotes or disfavors

devoicing.

38
“Input” is the overall application of the devoicing “rule,” regardless of

environment or identity. The input obtained in this study was 0.079. This means that
about 8 percent of all the vowels in the data were devoiced. The chi-square in the ﬁrst
Result ﬁle was 850.69, which suggested considerable interactions, but the ﬁnal chi-
square after combining factors, which is discussed below, was 118.37, which was still not
optimal, because it should be below 24.00 to be not signiﬁcant, but much better than the
ﬁrst one. We signiﬁcantly improved the chi-square by combining factors to make better
generalizations. Also, the “chi-square per cell” must be under 1.00 to indicate that we
are studying the right factors controlling the variation. The smaller the number, the
better. In the ﬁrst Result ﬁle, the chi-square per cell was 1.14, but the ﬁnal one was 0.77,
which is well under 1.00.

There are several possible reasons for the bad chi-square and chi-square per cell.
First, since we had only 4 subjects, there might have been idiosyncratic effects from each
subject. Secondly, we had only a small number of cells. The total tokens (vowels) were
over 2,000, but, since only 8 percent of the vowels were devoiced, there were less than
200 tokens for rule-applications (i.e., devoicing). These factors caused some large errors,
in which the environment occurred only once or twice. If we had more tokens, these
errors could have been eliminated to a certain degree. Third, there may be some other
factors that were not included in this study. We basically focused on the potentially
devoiced vowel and the nearby environments; i.e., preceding and following segments,
and did not consider the syllable structure (i.e., if the following consonant is a coda or

not), the prosodic structure, or the effect of devoicing in successive syllables. These

39
factors may have effects on devoicing. However, even though the total chi-square was

signiﬁcant, we obtained reasonable generalizations by combining factors, and by doing
so, improved the chi-square signiﬁcantly. This indicates that we are on the right track.
The following subsections discuss each factor group that was selected as

signiﬁcant.

4. 4. 1 Accent and Manner of A rticulation of the Following Segment

The complex factor group of accent and manner of articulation of the following
segment was selected as signiﬁcant. However, if we consider the linguistic plausibility
and signiﬁcance of the interaction of these two factors, we cannot ﬁnd any. Therefore,
these groups were nm as separate factor groups. A Binomial Up & Down was done
again to choose factor groups that are signiﬁcant, and accent was selected as signiﬁcant
but manner of the following segment was thrown out. Manner of the following segment

was deleted from subsequent tests.

4. 4.2 Vowel Identity and Manner of Preceding Segment

Table 12 is a list of the factors in the complex factor group of vowel identity and
manner of articulation of the preceding segment. Promoting factors are at the top and
demoting ones at the bottom. Note that the vowels from position 1 to 8 are either /i/ or
/u/. We can improve in the general nature of this detail, however, by combining the
environments that have common linguistic characteristics and similar weights, reducing

the number of factors and improving the generalizations.

40

Table 12 - Binomial One Level
(Vowel Identity and Preceding Manner)

Simbol Number Precedi Manner
.994 fricative
.914 affricate
.895 nasal
.859 affricate
.835

.81 1

.750 fricative
.635 ‘ /r/
.586

.581 affricate
.571

.565 /e/
.564 /r/
.482 /r/
.401

.355 nasal
.306 fricative
.276

.272 ide
.267 fricative
.240 nasal
.225 la/
.22 1 fricative
.219 /r/
.214 nasal

”TJOOwI'Tlcr-w

I

1
2
3
4
5
6
7
8
9
10
1]
12

”:3
wmaoooor—r,wmo

NI—I—ﬁh—I—I—lhd
OOOOQONM-b

NNN
WNi—t

 

omn—-mommo-moc—-ooo—-c--::::::-.--

NM
MA
Ui<-1G.

41
First, we combined factors 2 through 6, because the weights were close, all had a

stop feature in the preceding segment, and the vowels were either /i/ or /u/ (close vowel).
By re-coding factors 2 through 6 as “”,b we reduced the number of total factors by 4.
Whenever we combined factors, a Binomial One Level was re-run, and we subtracted the
old log likelihood from the new one. For example, the log likelihood that was obtained
in the ﬁrst Binomial One Level run was 500.919, and the one in the second run, after re-
coding factors 2 through 6, was 502.929. The difference was 2.01, and it was multiplied
by 2 (4.02). We checked the chi-square at 4 degrees of freedom, and found that 4.02 was
not signiﬁcant (at .05). Since there was no signiﬁcant difference when these factors were
combined, combining 2 through 6 was justiﬁed. This procedure was followed whenever
possible to arrive at greater generalization.

Second, we combined factors 9 through 11, because the weights were close, all
have a stop feature, and the vowels were either /e/ or /0/ (mid vowel). Again, the result
of combining these factors was not signiﬁcant. Then, factors 12 through 14 were
combined, since the preceding segment was either a vowel or /r/, although the vowels
were /i/, /u/ and /o/. However, the result was not signiﬁcant.

Then factors 18, 19, 21, 23, and 24 were combined, because all the vowels were
/a/. This group is /a/ with all the preceding environments except stop, which seems to
promote even the devoicing of /a/ (number 15 in Table 12). Again, the result was not
signiﬁcant. Then, 17 and 20 were combined because the preceding manner was the same
(ﬁ'icative) and the vowel was a mid vowel. Again, the result was not at all signiﬁcant.

Finally, the new factor ‘h’ (12, 13 and 14) and 8 were combined because the weights

42
were close and the preceding manner included /r/. The result was not signiﬁcant. Thus,

we could reduce the number of factors from 25 to 11. Table 13 is the new list.

Table 13 - Binomial One Level
(Vowel Identity and Preceding Manner - 2)

S bol Number ’ Manner Vowel
.943 fricative i
.866 + close vowels
.755 fricative u
.607 + mid vowels
.534 /e/ i

/r/ ' o
.412 a
.335 nasal i
.326 fricative mid vowels
.230 a
.227 nasal o
.208 /a/ i

 

I would like to discuss the promoting and demoting environments for each vowel
in terms of the manner of articulation of the preceding segment according to each

feature.

4. 4.2.1 Fricatives

It was conﬁrmed that a preceding fricative promotes the devoicing of close
vowels, but it was found that it promotes the devoicing of /i/ more than /u/ (1 and 3 in
Table 13). Although a preceding fricative promotes the devoicing of /i/ and /u/ (close

vowels), it demotes the devoicing of other vowels (8 and 9 in Table 13).

43
4. 4. 2.2 Stops

A preceding stop, including nasals and affricates, promotes the devoicing of close
vowels (2 in Table 13). A preceding stop or affricate also promotes the devoicing of /o/
more than a preceding fricative does, and a preceding stop relatively promotes the
devoicing of /a/, even within the small devoicing rate for that vowel. However, a
preceding fricative rather retards the devoicing of la/ (23 in Table 12), ranked below
pause, glide, and nasal (18, 19 and 21 in Table 12). Therefore, a preceding stop can be
considered as the best promoter for all the vowels except /i/; the best promoter for /i/ is a
preceding ﬁicative, and the second best is an affricate, which has a fricative feature. On
the other hand, the best promoter for /u/ is a nasal, which has a stop feature, and the
second best is an affricate, which also has a stop feature, and ‘stop’ itself comes in the
third place for the vowel /u/. Therefore, we can conclude that a stop feature promotes the
devoicing of /u/ most. A preceding stop also promotes the devoicing of /a/ and /e/ most.
Actually, a preceding stop is the only promoter for the devoicing of /e/ in Table 12.

Since affricates have both fricative and stop features, they are ranked high for
both close vowels. A preceding affricate is also the best promoter for /o/. However,
affricates were knocked out for the vowels /a/ and /e/, although the best promoter for
those vowels is a stop.

It is interesting that, although the preceding nasal promotes the devoicing of /u/, it
retards the devoicing of other vowels, including /i/. A preceding nasal for /e/ was even

knocked out.

44
4. 4.2.3 :‘r's

It was also found that the preceding /r/ slightly promotes the devoicing of /i/, /u/
and /o/, but retards the devoicing of /a/ and /e/. The inﬂuence of the preceding /r/ on the

devoicing of /i/ has never been reported in the literature.

4.4.2.4 Vowels

It should be noted that all the preceding vowels were knocked out for all the
vowels except for /i/. Preceding /e/ is in the 5th place and preceding /a/ is in the 11th
place in Table 13, but these are the only instances of a preceding vowel inﬂuencing
devoicing. It is interesting that the vowel /i/ could be devoiced even after the strongest

demoter, namely another vowel.

4. 4. 2.5 Summary

Although it was noted in the literature that a preceding fricative promotes vowel
devoicing, this study shows that that is true only for close vowels, not for open vowels. It
is also found in this study that a preceding stop (or a stop feature) promotes the devoicing

of open vowels more than a preceding fricative.

4. 4.3 Voicedness of the Preceding and Following Segments
Previous research shows that most typical environment in which vowel devoicing

occurs is between voiceless consonants. We can expect, therefore, that voiceless

45
consonants will promote vowel devoicing, and this was conﬁrmed in the present study.

Table 14 provides the weight of each factor for the voicedness of the preceding segment.

Table 14 - Binomial One Level
(Voicedness of the Preceding Segment)

Voicedness Wei
Voiceless .666

Pause .576
Voiced .333

 

A preceding voiceless consonant promotes devoicing most, as expected, but a
preceding pause also promotes devoicing. The important thing to be noted here is that a
vowel can be devoiced even after a voiced segment. This conﬂicts with Maekawa’s
claim that a necessary condition for vowel devoicing is a preceding voiceless segment
(1988:136-7). However, it should be noted that in many cases the preceding voiced
segment was actually devoiced before the devoiced vowel. In other words, the devoiced
vowel had an inﬂuence on the preceding voiced segment and made it voiceless, but we
had calculated i voice on the basis of the “underlying” segment in all cases, except, of

course, for the dependent variable itself.

Table 15 provides the weights of each factor of the voicedness of the following

segment.

46

Table 15 - Binomial One Level
(Voicedness of the Following Segment)

Voicedness Wei
Voiceless .727

Voiced .354
Pause .338

 

Here, a following voiceless consonant promotes vowel devoicing even more than
a preceding one. A following pause demotes vowel devoicing as much as a voiced
segment does. A following pause has the same effect as a following voiced segment on
vowel devoicing, although a preceding pause behaves more like a preceding voiceless
segment.

Since a following pause and a following voiced segments are very similar in
weight, these two factors were combined. The log likelihoods were checked and the
result was not signiﬁcant. Therefore, we have only two factors for this group; voiceless

and voiced + pause. Table 16 shows the ﬁnal weight of each factor.

Table 16 - Binomial One Level
(Voicedness of the Following Segment - 2)

 

Voicedness I Weight
Voiceless I .727
Voiced + Pause I .353

—L—

47
Thus, a following voiceless segment promotes vowel devoicing and a following

voiced segment or a pause demotes devoicing. However, it should be noted that

devoicing can occur before a voiced segment or a pause since neither was knocked out.

4. 4.4 Accent
Accent turned out to be signiﬁcant as an independent factor group. Table 17

shows the weight of each factor.

Table 17 - Binomial One Level

 

 

 

 

 

 

 

(Accent)
Accent Type Weight
4 (not on this syllable nor preceding or following syllable) .642
2 (on thejreceding syllable) .578
3 (on the following syllable) .530
5 (on both precedirg and following syllable) .462
1 (on this syllable) .264

 

 

Accent type 4 promotes vowel devoicing slightly more than accent type 2, which
promotes it slightly more than accent type 3, but there is not much difference. The only
factor that strongly inﬂuences devoicing is accent type 1, which demotes vowel
devoicing. It is easy to understand that the accented syllable resists vowel devoicing
because, if the vowel is devoiced, it is difﬁcult for it to carry the accent. It was long
assumed in the literature that an accented syllable could not be devoiced, or if an
underlyingly accented syllable is devoiced, the accent is shifted to the following syllable,

until Sugito and Hirose (1983) showed the mechanism of an accented devoiced vowel.

 

48
Since there is not much difference among the promoting factors, they were combined.

The result was not signiﬁcant, and now we have only two factors for this group; 1 (accent
is on this syllable) and 2 (accent is not on this syllable). Table 18 shows the ﬁnal weight

of each factor.

Table 18 - Binomial One Level

 

 

 

(Accent - 2)
I Accent Type I Weight I
I l (on this segment) I .264 I
I 2 (not on this segmerm I .561 I

 

Thus, vowel devoicing is demoted if the vowel is accented, but is not demoted if

the vowel is not accented, and it does not matter where the accent is.

4. 4. 5 Vowel Length
Vowel length showed the strongest demoting inﬂuence on vowel devoicing. The

weight of each factor is shown in Table 19.

Table 19 - Binomial One Level

 

 

 

(Vowel Length)
I Vowel Length I Weight I
I Short | .530 l
r Long L .098 I

 

49
The weight of a long vowel is the smallest of all the factors considered in this

study. It is reasonable to expect that a long vowel should not be devoiced, but there was
one instance of devoicing of a long vowel in the data of subject #2. The second long

vowel /ii/ was devoiced in /nyuuiﬁrando/ (“New Zealand”). The preceding /2'/ was also

devoiced and became /I/, and the long vowel /ii/ was shortened and became more like /i/.

Since /i/ is a close vowel and the preceding segment is a fricative, which is the best
promoter for /i/, this is a promoting environment for devoicing. However, the important

thing to note here is that the environment was altered.

4. 4.6 Gender
Although gender differences did not seem signiﬁcant, the Binomial Up & Down

run selected them as signiﬁcant. The weights are provided in Table 20.

Table 20 - Binomial One Level

 

 

 

(Gender)
Gender I Weight I
Male | .555 I
Female I .435 I

 

Males devoice their vowels more than females do. If this phenomenon is a
“standard” feature, a gender difference is expected, in which case females should be the

promoters of the phenomenon. However, it is not the case here. Therefore, if any

50
“standardness” is involved in this phenomenon, it should be “nonstandardness” since

men promote the phenomenon rather than women.

4.5 Other Observations

There are several things to be mentioned before concluding this section. First, the
avoidance of vowel devoicing in successive syllables is often discussed in the literature
although it was not dealt with in this study. Han (1966291) even states that “in normal
speech, the unvoicing of vowels in two succesive syllables is rare, and the unvoicing of
three succesive vowels do not occur.” Maekawa (1988: 148-9) also discusses the
tendency to avoid devoicing in successive syllables, mentioning that when there are two
or more devoiceable vowels in sequence, it is not known what determines which vowels
are devoiced. Maekawa reports instances of devoicing in two successive syllables, even
though in small numbers. In this study, devoicing of two successive syllables was seen

frequently, as in /topikkgsu/ (“topics”) and /tsggi#no#tr_)jj(#ka)/ (“next year”; “#”

indicates a morpheme boundary), and devoicing of three successive syllables was also

observed several times, as in /tanajj#ji#51_rki/ (“I like Tanashi city”) and /(mabuta#)ng#
fgc'i(#ka)/ (“edge of [eyelid]”). Even devoicing of four successive syllables, as in
/hont9_#kisgtsg#roodooja/ (“a real seasonal worker”), was observed once per subject

except for subject #3, whose overall devoicing rate was the lowest among the four

subjects.

5 1
The second point is that devoicing even occurred between voiced segments as in

/~kedg#z'jssai/ (“~, but actually”). The second vowel /o/ is between voiced consonants,

but still it was devoiced. As I discussed above, the preceding and the following voiced
consonants were devoiced and became voiceless. This is also an instance of the
inﬂuence of the devoiced vowels on the preceding voiced segment. A possible
explanation is that the third vowel /i/ was devoiced due to the following voiceless

consonant /s/; the preceding voiced consonant /i/ was devoiced by the inﬂuence of the
devoiced /i/ , then the devoiced /z’/ inﬂuenced the preceding mid vowel /o/ and devoiced

it, and ﬁnally the preceding voiced consonant /d/ was devoiced because of the devoiced
/o/. If this is the case, a devoiced vowel has an inﬂuence on the preceding segment,
which, in turn, has an inﬂuence on the preceding vowel. However, it is important to note
that those consonants are underlyingly voiced, and those sounds are perceived as voiced

sounds.

These phenomena may have been observed because the present study used natural
conversation, which is very rapid and casual, and in which vowel devoicing is most likely

to OCCUR

52

4.6 Summary

The followings are the lists of the factors that promote and demote vowel
devoicing in Japanese. The most signiﬁcant one is at the top and the least signiﬁcant one

is at the bottom.

(1) Promoting factors for overall vowel devoicing (The complex factor group of vowel
identity and manner of articulation of the preceding segment id excluded here.)

Following voiceless segment
Preceding voiceless segment
Preceding pause
Unaccentedness

Male

Short vowel

9}”:BP’I‘JT‘

(2) Demoting factors for overall vowel devoicing (The complex factor group of vowel
identity and manner of articulation of the preceding segment is excluded here.)

Phrase ﬁnal stress

Long vowel

Accentedness

Preceding voiced segment
Following pause and voiced segment
Female

9MPP’N?‘

(3) Promoting factors for each vowel (In terms of manner of articulation of the preceding
segment.)

A. /i/

fricative

+stop (affricate and stop)
/r/ and /e/

nasal

/a/

.V'PP’N.”

. +stop (nasal, affricate and stop)
. fricative
3. /r/

N1—

53

C. /o/

+stop (affricate and stop)
/r/

fricative

nasal

PP’N!‘

D. /e/
1. stop
2. fricative

E. /a/
. stop
. others (pause, glide, nasal, fricative, and /r/)

Nr—i

The most signiﬁcant factor for vowel devoicing in Japanese is the length of the
vowel. Long vowels most strongly demote devoicing. On the other hand, the vowel /i/
with a preceding fricative most strongly promotes devoicing. This conﬁrms the claim
made by previous researchers that preceding fricatives promote the devoicing of close
vowels, but adds that this is true only for the vowel /i/, since the greatest promoter for the
vowel /u/ is the feature ‘stop.’ This fact seems to have caused some confusion in the
literature regarding the status of fricatives, afﬁ’icates and stops as promoters for
devoicing. However, this study clearly showed their different rankings in signiﬁcance for
each vowel. The feature ‘stop’ also seems to promote the devoicing of mid vowels (/e/
and /o/), not only /u/. Finally, the best promoter for the devoicing of /a/ is also a
preceding stop. In general, preceding stop is the best promoter for the devoicing of all
the vowels, except /i/, whose principal promoter is a preceding fricative.

Another fact that was observed in this study, but has never been reported, is that

the preceding voiceless segment more strongly promoted overall vowel devoicing than

54
the following voiceless segment. This fact was also relevant to the observation that a

devoiced vowel, which was devoiced due to the preceding voiceless consonant, had an
inﬂuence on the following voiced consonant and devoiced it, and the devoiced

consonant, in turn, devoiced the vowel that followed it, as in /~tgk2_1#iibun(#de)/ (“[she

said] something like ~ by [her]self’). It was also observed, however, that a devoiced
vowel, which was devoiced due to the following voiceless segment could have an
inﬂuence on the preceding voiced segment and devoice it. A devoiced vowel could have
an inﬂuence in both direction. Scherre and Naro (199221) claim the “serial effect on
internal and external variables,” which is “a preference for similar forms to co-occur in
sections of discourse, even if this results in redundancy or lack of economy.” Although
Scherre and Naro claim this effect in morphology, we can also expect the same effect in
phonology. However, this conflicts with the claim made by the previous researchers of
Japanese vowel devoicing, which is the avoidance of devoicing in successive syllables.
To determine which effect is at work in this phenomenon, we need to include this in the
factors.

Accent was found to be signiﬁcant in this study, conﬁrming most of the previous
researchers’ report that accented syllables strongly resist devoicing, although this study
used natural data, in which accent does not seem to have a strong effect according to
previous research. However, the place of accent (before and/or after the segment in
question) was not found to be signiﬁcant.

Although the following vowel is claimed to have an inﬂuence on the devoicing of

the close vowels, no signiﬁcance was observed in this study. Also, the following

55
segment was not found to be signiﬁcant in terms of place and manner of articulation,

contrary to previous researchers’ reports. Although it was not signiﬁcant, some variation
was found which makes it seem that following stops and fricatives promote devoicing.

Gender difference was also found to be signiﬁcant, which has never been reported
in the literature. “Males” promote devoicing and “females” demote it, which implies the
nonstandardness of this phenomenon. However, whether or not this phenomenon is a
“nonstandard” feature is still a question. To determine this, we need more subjects. The
phenomenon of vowel devoicing was probably ﬁrst mentioned in Collado 1632 (“Art
Grammaticae Iaponicae Linguae”), and does not seem to be a recent change. To see if
the younger generation devoices more vowels than the older generation, in which case it
could be interpreted as a slow but ongoing change, we need data from the older
generation. Furthermore, it is not clear that the reason why we had extremely rare
instances of devoicing, such as devoicing in four successive syllables, is because we used
spontaneous casual data, because the subjects were from the younger generation, or
because of the “serial effect” mentioned above. To answer this question as well, we need
data from different generations.

Dialect difference in this phenomenon is well known, and there are many studies
focused on such variation (e. g., Sugito 1988 and Maekawa 1988). It was not dealt with
in this study because the purpose of this study was to ﬁnd out the basic promoting and
demoting environments for vowel devoicing and such variation would have made such

generalizations very difﬁcult to come by. However, differences in dialects should be

56
investigated to support the general tendency of the promoting and demoting

environments, or to ﬁnd out different environments for other dialects.

5. CONCLUSION

This study showed that each vowel has different factors which promote and
demote their devoicing rates, which was the researcher’s hypothesis. A secondary
hypothesis, that open vowels also have promoting and demoting environments even
within their small number of occurrences, was also conﬁrmed.

One of the major ﬁndings in this study is that it is the preceding segment that
have a signiﬁcant effect on vowel devoicing in casual speech in Japanese. Although the
following segment is claimed to be signiﬁcant, it turned out not to be true in this study.

Another ﬁnding that is of interest, which has never been reported in the literature,
is that the common feature of the preceding consonant and the vowel demotes devoicing;
namely, the feature ‘palatal’ for the vowel /i/ and the feature ‘labial’ for the vowel /u/.
This might indicate that phonology is also involved in this phenomenon, not only
phenotics. The answer to this question needs a further study.

Another interesting ﬁnding is that, although a preceding pause was found to
promote devoicing, a following pause was found to demote it, as discussed in 4.4.3. This
conﬁrms Sugito’s statement that the sentence ﬁnal pause promotes devoicing, but the
phrase ﬁnal pause demotes it (198829). This makes a good sense if the high frequency of
devoicing before the sentence ﬁnal pause is caused by the sentence ﬁnal copula /~desu/,
because the data in this study were taken from the casual conversation that soes not

contain formal sentence endings (/~desu/).

57

58
Guy 1980 claims a dialect difference in the effect of pause on ﬁnal /t,d/ deletion.

A following pause was like a following consonant, and promoted the deletion for the
New Yorkers, but for the Philadelphians, a following pause was even more conservative
than a following vowel, and strongly demoted it. In this study, the difference in the
position of a pause (i.e., preceding or following) showed an inﬂuence on devoicing, and
dialect difference can also be expected. Now that we have set a statistical standard from
young Tokyo speakers, it is hoped that more careful comparison (based on acoustically
measured data and treated in a statistically sophisticated way) can be made within the
social variation of Tokyo Japanese and among the regional dialects of the language as

well.

APPENDICES

APPENDIX A

R4:

R3:

R4:

R3:

R4:

R3:

R4:

Note:

APPENDIX A

Sample Transcription

J aa, - kyoo no t0pikkusu wa ((pause)) nani kashira.=
well today’s topics TOP what I-wonder
“Well, I wonder what today’s topic is.”

l 1
Un.

Yes.

= Nani kashira yappari atashi no me no shujutsu no koto kana. -- (Moo) =
what I-wonder after-all I of eye of surgery of matter I-guess
“I wonder what (it is). It should be about the surgery of my eyes, I guess.”

[

Aa, soo soo 500.
Yeah, right, right, right.

= itakattanda kara hontoni.
painful-PAST P truely
“It was painful, truely.”

I

Nanda masui nai no?=
what-cop. anesthesia NEG Q
“What! No anesthesia?”

=(Ya), masui mo itai shi, demo masui (utsu kara)
(why) anesthesia also painful SFP, but anesthesia (shoot because)
“Why, anesthesia was also painﬁrl, but (since I was given) anesthetic

[

Ne, masui tte me ni sasu no?
listen, anesthesia eye to stick Q
“Wait, did they give anesthetic in your eye?”

cop. = copula

P = particle
SFP = sentence ﬁnal particle

59

APPENDIX B

 

APPENDDI B

Sample Cross-Tabulation with No Interaction

(Subject Identity and Vowel Identity)

 

L
“T erum12.Cel”

o 4123.39702 : 49 PH

0 Token file: TerumiS-Goldvarb
0 Conditions: Terun12.Cnd

Group #1 -- horizontally.

Group #2 -- vertically.

 

 

 

 

 

 

1 x 2 X 3 X 4 X E x
i - 37 32 27 36 12 15 26 37 102 30
+ 77 68 48 64 7O 85 45 63 240 70

E 114 75 82 71 342
a - 11 6 8 6 9 5 9 8 37 6
+ 180 94 123 94 166 95 110 92 579 94

2 191 131 175 119 616
u - 20 35 25 38 13 27 11 25 69 32
+ 37 65 41 62 35 73 33 75 146 68

2 57 66 48 44 215
e - 14 17 12 21 7 13 8 10 41 15
+ 70 83 46 79 46 87 74 90 236 85

2 84 58 53 82 277
o - 15 11 13 10 13 15 7 8 48 11
+ 125 89 112 90 76 85 78 92 391 89

E 140 125 89 85 439
Z - 97 17 85 19 54 12 61 15 297 16
+ 489 83 370 81 393 88 340 85 1592 84

E 586 455 447 401 1889

 

 

 

 

 

 

60

 

61

APPENDIX B

Sample Cross-Tabulation with Interaction
(Vowel Identity and Place of Articulation of the Preceding Segment)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

‘k

g] “TerumiZJZel”

o 4t23l9702249 PM

o Token tile: Tertmi3-Goldyarb

0 Conditions: TeruniZ.Cnd

Group #2 -- horizontally.

Group #3 -- vertically.

i x a x u x e x o x Z x

x - 3 3 1 4 0 0 0 0 0 0 4 2
+ 102 97 22 96 15 100 6 100 24 100 169 98

2 105 23 15 6 24 173

p - 73 68 1 2 1 7 0 -- 3 4 78 33
+ 35 32 42 98 14 93 0 -- 68 96 159 67

2 108 43 15 0 71 237

1 - 1 5 3 3 12 44 0 0 0 0 16 7
+ 19 95 111 97 15 56 23 100 45 100 213 93

2 20 114 27 23 45 229

a — 4 7 14 5 32 30 36 17 40 18 126 14
+ 55 93 265 95 76 70 173 83 185 82 754 86

2 59 279 108 209 225 880

2v - 21 42 18 11 24 48 5 13 5 7 73 20
+ 29 58 139 89 26 52 34 87 69 93 297 80

2 50 157 50 39 74 370

2 - 102 30 37 6 69 32 41 15 48 11 297 16
+ 240 70 579 94 146 68 236 85 391 89 1592 84

Z 342 616 215 277 439 1889

 

 

BIBLIOGRAPHY

BIBLIOGRAPHY

Arisaka, Hideyo. 1964. Gosei Enkaku Kenkyu. Tokyo: Sanseido.

Beckman, Mary and Atsuko Shoji. 1984. Spectra] and Perceptual Evidence for CV
Coarticulation in Devoiced /si/ and /syu/ in Japanese. Phonetica 41, 61-71.

Beckman, Mary. 1996. Phonological Structure and Language Processing: Cross-
Linguistic Studies. Edited by Takashi Otake and Anne Cutler. 95-123. Berlin, New
York: Mouton de Gruyter.

Guy, Gregory. 1980. Variation in the group and the individual: The case of ﬁnal stop
deletion. In Locating Language in Time and Space. edited by William Labov. 1-36.
New York: Academic Press.

Jun, Sun-Ah and Mary E. Beckman. 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.

Jun, Sun-Ah and Mary E. Beckman. 1994. Distribution of Devoiced High Vowels in
Korean. Proceedings of the 1994 International Conference on Spoken Language
Processing, Vol. 22 479-482.

Kindaichi, Haruhiko. 1967. Nihongo Onin no Kenkyu. Tokyo: Tokyodo Shuppan.

Kindaichi, Haruhiko, et a]. 1988. Nihongo Hyakka Daijiten (An Encyclopedia of the
Japanese Language). Tokyo: Taishukan.

Kondo, Mariko. 1993. The effect of blocking factors and constraints on consecutive
vowel devoicing in Standard Japanese. Poster presented at Labphon 4, Oxford, 11-14
August.

Kondo, Mariko. 1994. Mechanisms of Vowel Devoicing in Japanese. Proceedings of the
1994 International Conference on Spoken Language Processing, Vol. 22 479-482.

Kondo, Mariko. 1995. Temporal Adjustment of Devoiced Morae in Japanese.
Proceedings of the 13th International Congress of Phonetic Sciences, Vol. 32 61-64.

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

62

63

Labov, William. 1994. Principles of Linguistic Change. Volume 1: Internal Factors.
Cambridge: Blackwell.

Maekawa, Kikuo. 1983. Kyotsugo ni okeru Boin no Museika no Kakuritsu ni tsuite
[Probabilities for the occurrence of vowel devoicing in Common Japanese]. Gengo
no Sekai 1-2269-81.

Maekawa, Kikuo. 1988. Boin no Museika, Koza Nihongo to Nihongo Kyoiku.
Edited by Miyoko Sugito, Volume 2, 135-153, Tokyo: Meiji Shoin.

Nagano-Madsen, Y. 1994. Vowel devoicing rates in Japanese from a sentence corpus.
Working papers 42, 117-127, Department of Linguistics and Phonetics, Lund
University.

Nagano-Madsen, Y. 1995. Effect of Accent and Segrnenta] Contexts on the Realization
of Vowel Devoicing in Japanese. Proceedings of the 13th International Congress of
Phonetic Sciences, Vol. 32 564-567.

Sakuma, K. 1929. Nihon Onseigaku. Tokyo: Kyoubunsha.

Scherre, Maria Martha Rereira and Naro, Anthony J. 1992. The serial effect on internal
and extema] variables. Language Variation and Change, Volume 4: 1-13.

Shibatani, Masayoshi. 1990. The Languages of Japan. Cambridge.

Shimizu, Han Mieko. 1965. Unvoicing of Vowels in Japanese. Onsei no Kenkyu 10,
Nihon Onsei Gakkai.

Sugito, Miyoko. 1988. Nihon no 8 Toshi ni okeru Boin no Museika. Osaka Shoin
Women's College Collected Essays 25: 1-10, Osaka.

Sugito, Miyoko and Hajime Hirose. 1988. Production and Perception of Accented
Devoiced Vowels in Japanese, Annual Bulletin 22 (Research Institute of
Logopedics and Phoniatrics), No. 22, 21-39, Tokyo: University of Tokyo.

Vance, Timothy. 1987. An Introduction to Japanese Phonology. Albany: SUNY
Press.

Wardhaugh, Ronald. 1992. An Introduction to Sociolinguistics: Second Edition.
Cambridge: Blackwell.