Cultural and Educational Contributions to Recent Phonological Changes in Japanese2

Home , Ainu language, Bad Friends, Daijirin, Do Your Best, Hi (kana), Japanese pitch accent

CULTURAL AND EDUCATIONAL CONTRIBUTIONS TO RECENT PHONOLOGICAL CHANGES IN JAPANESE

Item Type text; Electronic Dissertation

Authors Watanabe, Seiji

Publisher The University of Arizona.

Rights Copyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction or presentation (such as public display or performance) of protected items is prohibited except with permission of the author.

Download date 03/10/2021 20:28:31

Link to Item http://hdl.handle.net/10150/195117

CULTURAL AND EDUCATIONAL CONTRIBUTIONS TO RECENT PHONOLOGICAL CHANGES IN JAPANESE by Seiji Watanabe

A Dissertation Submitted to the Faculty of the GRADUATE INTERDISCIPLINARY DOCTORAL PROGRAM in SECOND LANGUAGE ACQUISTION AND TEACHING In Partial Fulfillment of the Requirements For the Degree of

DOCTOR OF PHILOSOPHY In the Graduate College THE UNIVERSITY OF ARIZONA

2009

THE UNIVERSITY OF ARIZONA GRADUATE COLLEGE

As members of the Dissertation Committee, we certify that we have read the dissertation prepared by Seiji Watanabe entitled Cultural and Educational Contributions to Recent Phonological Changes in Japanese and recommend that it be accepted as fulfilling the dissertation requirement for the

Degree of Doctor of Philosophy

______Date: 03/12/09 Timothy J. Vance

______Date: 03/12/09 Adam Ussishkin

______Date: 03/12/09 Kimberly Jones

Final approval and acceptance of this dissertation is contingent upon the candidate’s submission of the final copies of the dissertation to the Graduate College.

I hereby certify that I have read this dissertation prepared under my direction and recommend that it be accepted as fulfilling the dissertation requirement.

______Date: 03/12/09 Dissertation Director: Timothy Vance 3

STATEMENT BY AUTHOR

This dissertation has been submitted in partial fulfillment of requirements for an advanced degree at the University of Arizona and is deposited in the University Library to be made available to borrowers under rules of the Library.

Brief quotations from this dissertation are allowable without special permission, provided that accurate acknowledgment of source is made. Requests for permission for extended quotation from or reproduction of this manuscript in whole or in part may be granted by the copyright holder.

SIGNED: Seiji Watanabe

ACKNOWLEGEMENTS

I have been studying linguistics for years, but I have never realized how insufficient language is to express my gratitude to those who helped me finish this dissertation. I will do my best.

I would like to thank my committee members. I owe my deepest gratitude to Dr. Timothy Vance, who has exerted a great influence on my perspective on linguistics. Dr. Vance has always been extremely patient with my slow progress, and has spent tremendous amounts of time and energy reviewing and commenting on my drafts. Without his continuous and strong encouragement and support, I could not have completed this dissertation. Dr. Adam Ussishkin gave insightful, encouraging, and challenging comments which significantly improved the quality of this dissertation. Dr. Kim Jones has always been supportive, provided detailed feedback and asked insightful questions. I am indebted to her for many of her heart-warming comments that motivated me to make this dissertation better. I am very grateful, as well as very proud, that I completed this dissertation under these wonderful committee members.

I have received tremendous amounts of help from people outside of my committee as well. I owe a debt of gratitude to Dr. Chris Maloney, the head of the Philosophy department. He repeatedly hired me as a TA for the general education class Mind, Matter, and God. Finishing this dissertation would have been impossible without so many years of support from Dr. Maloney and the Philosophy department.

I would like to thank Dr. Mario Montalbetti for a long-term friendship and inspiring discussions. Without his help and encouragement, I would not have even started studying at the University of Arizona.

My friends from SLAT, East Asian Studies, the Linguistics department and the Philosophy department not only supported me, but also put up with discussions about my dissertation topic and provided me with helpful information. I would like to thank Nathan Ballantyne, Dan Brenner, Maggie Camp, Ian Evans, Brian Fiala, Itsumi Ishikawa-Peck, Cole Mitchell, Chieko Nakano, Atsuko Oyama, Emil Salim, Tetsuya Sato, Yasumasa Shigenaga, and especially I would like to thank Daniel Sanderman for patiently correcting my English and providing me with comments as a general reader before I turned in my manuscripts to committee members.

My brother Hirochika and his wife Asami have supported me by taking care of my parents when they were sick. I didn’t give up my study because of their support. My nephew Hiroki and my niece Yuki helped me keep in touch with the hottest Japanese popular culture, which is one of essential aspects of this dissertation.

Finally, I would like to thank my late parents, Tadao and Tae, for their support and understanding. I regret that I could not finish writing this dissertation while they were around. 5

To my parents 6

TABLE OF CONTENTS LIST OF FIGURES ...... 9! LIST OF TABLES...... 10! ABSTRACT...... 12! CHAPTER 1! INTRODUCTION ...... 13! 1.1! Statement of the Problem...... 13! 1.2! The Goals of this Dissertation...... 16! 1.3! This Study is not Limited to Loanword Phonology...... 19! 1.4! Japanese Sounds...... 20! 1.4.1! Japanese Vowels...... 20! 1.4.2! Sound Sequences of Modern Standard Japanese...... 20! 1.4.3! Innovative Sound Sequences ...... 22! 1.5! Data: Non-Marginal Words and Marginal Words ...... 23! 1.6! The Use of Japanese Orthography in this Dissertation...... 24! 1.6.1! Voicing of Obstruents in Katakana Representation...... 26! 1.6.2! Geminate Consonants in Katakana Representation...... 26! 1.6.3! Vowel Substitution ...... 26! 1.7! Problem with the Use of Katakana Spellings...... 27! 1.7.1! Approximants ...... 27! 1.8! Other Issues That Will Not Be Discussed In This Dissertation...... 28! 1.8.1! Accent and Intonation...... 28! 1.8.2! Vowels ...... 28! 1.9! Previous Studies...... 29! 1.9.1! Hattori (1960): Seams and Gaps—Structural Asymmetry ...... 29! 1.9.2! McCawley (1968): Generative Phonology ...... 32! 1.9.3! Ito and Mester (1995a, 1995b, 1999): Optimality Theory (OT)...... 33! 1.10! Lexical Stratification...... 36! 1.11! Structure of Dissertation ...... 37! CHAPTER 2! GEMINATE VOICED OBSTRUENTS...... 39! 2.1! Introduction...... 39! 2.2! Overview of the Issues in Geminate Voiced Obstruents in Japanese...... 40! 2.3! Previous Studies...... 41! 2.3.1! Ito and Mester (1999) ...... 41! 2.3.2! Nishimura (2006)...... 45! 2.3.3! Kawahara (2005, 2006) ...... 47! 2.4! Problems with Data Description in Previous Analyses ...... 52! 2.4.1! Geminate ‘b’ (BB) ...... 52! 2.4.2! DDj and DDz ...... 55! 2.4.3! DVRVDDV-type words ...... 57! 2.5! Exceptions...... 62!

TABLE OF CONTENTS - Continued

2.5.1! [!"#okki$] ‘groggy’ ...... 62! 2.5.2! [(d)zetto] ‘z’...... 63! 2.5.3! ky!piddo ‘cupid’ and kata paddo ‘shoulder pad’ ...... 63! 2.5.4! [deb%!!%$] ‘debugger’...... 64! 2.6! Changes in Voicing Patterns when DVDDV Words Are Compounded...... 64! 2.7! Devoicing Patterns in the Early 20th Century ...... 66! 2.8! Sociolinguistic Study: Quackenbush (1989)...... 66! 2.9! Does Lyman’s Law Really Affect DD?...... 72! 2.10! Origins of Geminate Voiced Obstruents...... 74! 2.10.1! Syncope and Assimilation (Two Morphemes) ...... 75! 2.10.2! Emphatic Expressions (Within a Morpheme)...... 76! 2.11! Discussion...... 81! CHAPTER 3! ISSUES OF JAPANESE HA-GY! AND INNOVATIVE FA-GY!...... 86! 3.1! Introduction...... 86! 3.2! Articulation of Ha-gy" Consonants ...... 87! 3.3! The Phonetic Quality of [h] ...... 93! 3.4! Historical Background ...... 95! 3.5! The Innovative /fa fi fe fo/...... 103! CHAPTER 4! VOICELESS CORONAL FRICATIVES ...... 111! 4.1! Introduction...... 111! 4.2! Innovative [&e] in the Dialect of Tokyo...... 112! 4.3! The [si] Sequence...... 113! 4.4! Japanese ‘/s/’ and ‘/'/’...... 118! 4.5! The Overview of Historical Changes of Japanese Coronal Fricatives ...... 123! 4.6! The Methods of Reconstruction of Sa-gy" and Sha-gy" ...... 125! 4.6.1! Reconstruction of Sa-gy" through Chinese characters and Sanskrit characters (8th century ~ 1333) ...... 126! 4.6.2! Emergence of Sha-gy"...... 128! 4.6.3! Reconstruction Through the Roman Alphabet ...... 131! 4.7! Variations of Sa-gy" ...... 132! 4.8! Two Scenarios...... 134! 4.8.1! The First Extreme Scenario: Consonants of Sa-gy" Were [s] ...... 134! 4.8.2! The Second Extreme Scenario: Consonants of Sa-gy" Were [&] ...... 136! 4.9! /s/ in Ainu...... 137! 4.10! Articulatory And Acoustic Account ...... 138! 4.11! Why Is [&e] Prevalent While [si] Is Still a Foreignism? ...... 142! 4.12! Why [si] Is Still a Foreignism...... 146! 4.13! Conclusion ...... 148! 8

TABLE OF CONTENTS - Continued

CHAPTER 5! JAPANESE [ti], [t"], [di], AND [d"]...... 149! 5.1! Introduction...... 149! 5.2! The Diversity of Ta-gy" /Da-gy" Consonants and Confusion in the Writing System...... 151! 5.3! The Recent Development of Japanese [ti], [t"], [di], and [d"]...... 155! 5.3.1! The Distribution of [ti], [t"], [di], and [d"] ...... 163! 5.3.2! Phonological Variants with [ti]/[di]...... 168! 5.3.3! Phonological Variants with [t"] and [d"] ...... 171! 5.4! Have Innovative [ti] and [t"] Replaced Ta-gy" [c&i] and [c&"]?...... 176! 5.4.1! Akima (Gap) Theory and the Innovative Sound Sequences [ti] and [t"] .. 176! 5.5! Vowel Devoicing and Innovative Sound Sequences ...... 179! 5.6! Reasons for Vowel Devoicing ...... 182! 5.7! Why vowels in [ti] and [t"] do not devoice ...... 183! 5.8! [ti] and [t"] as co-articulation ...... 183! 5.9! Evidence From New-type Gairaigo ...... 189! 5.10! Distributions of [ti]/[di] and [t"]/[d"] in Marginal Items ...... 192! 5.11! Theoretical Account...... 198! 5.12! Cultural Account...... 201! 5.13! Differences in Generations ...... 205! 5.14! Difference in the Timing of Borrowing...... 206! 5.15! Difference in Cultural Context...... 206! 5.16! The Complex Structure of Cultural Values and [ti], [di], [t"], and [d"]...... 207! 5.17! Importer’s Influence on the Phonological Forms of Loanwords...... 209! 5.18! The Intelligentsia, the General Public, and the Pseudo-intelligentsia ...... 210! 5.19! Conclusion ...... 211! CHAPTER 6! DISCUSSION...... 212! 6.1! Stratification or No Stratification...... 212! 6.2! The Constraint *NT and Modern Standard Japanese ...... 212! 6.3! Historical Account ...... 217! 6.4! Problems with Lexical Stratification ...... 222! 6.4.1! Definition of Lexical Stratification...... 222! 6.4.2! Accessibility to Lexical Stratification ...... 224! 6.5! Interactions of Grammars ...... 225! 6.6! Integration of Linguistic Theories ...... 227! 6.7! Implications for Second Language Acquisition and Teaching...... 232! 6.8! Concluding Remarks...... 234! REFERENCES ...... 235! 9

LIST OF FIGURES

Figure 2.1. Recursive local conjunction ...... 48!

Figure 2.2. Degrees of devoicing in [dd] ...... 68!

Figure 2.3. Degrees of devoicing in [!!] ...... 69!

Figure 3.1. The change of Yamato ‘light’ and Mimetic ‘light’ ...... 97! Figure 4.1. Historical development of sibilants: time table of mergers...... 135!

LIST OF TABLES

Table 1.1. Sound sequences of Japanese ...... 21! Table 1.2. Innovative sound sequences sorted in terms of onset...... 22! Table 1.3. Goj#-on-zu ‘fifty-sound-display’...... 25! Table 1.4. Seams and Gaps...... 30! Table 1.5. Lexical stratification by McCawley...... 32! Table 1.6. The structure of strata based on constraints...... 35! Table 2.1. Non-marginal DVRVDDV words ...... 58! Table 2.2. Ratio of voicing and devoicing of DD...... 70! Table 2.3. General distribution of geminate obstruents in non-foreign items ...... 79! Table 3.1. Distribution of current standard Japanese ha-gy"...... 86! Table 3.2. Distribution of sound sequences in fa-gy"...... 86! Table 3.3. Historical development of ha-gy", pa-gy", and fa-gy" ...... 101! Table 3.4. Numbers of words with innovative /fV/ sequences that are listed in Japanese dictionaries...... 103! Table 3.5. Historical development of ha-gy", pa-gy", fa-gy", Mimetic pa-gy", and Mimetic ha-gy"...... 105!

Table 4.1. Distribution of [s] and [&] ...... 112!

Table 4.2. Sequences involving [s] and [&] in terms of the general phonetic allophonic analysis...... 119!

Table 4.3. Sequences involving [s] and [&] based on the writing system ...... 120!

Table 4.4. Sequences involving [k] and [k(] based on the writing system...... 121! Table 4.5. Overview of known historical facts on Japanese coronal fricatives...... 123!

Table 4.6. The historical shift of the sound [&a] between sa-gy" and sha-gy"...... 129! Table 4.7. The change of the distribution in the complete sa-gy" and sha-gy" ...... 130! 11

LIST OF TABLES - Continued Table 4.8. The development of ka-gy" and kya-gy"...... 130! Table 5.1. Distribution of ta-gy" and da-gy" sound sequences...... 149! Table 5.2. Distribution of the sound sequences in sa-gy" and ta-gy" ...... 153! Table 5.3. Distribution of the sound sequences in za-gy" and da-gy" ...... 153! Table 5.4. Development of yotsugana ...... 154! Table 5.5. The history of standardization of katakana spellings of loanwords...... 157!

Table 5.6. Distribution of [ti] / [di] and [t"] / [d"] in non-marginal Items...... 164!

Table 5.7. List of non-marginal words with innovative sound sequences [t"] or [d"]. 166!

Table 5.8. Defective distribution of ta-gy" and [ti] and [t"] as gap fillers ...... 177! Table 5.9. Cross-tabulation of the voicing of /i/ in the /CoVcCo/ environment by C1 and C2...... 181!

Table 5.10. Distribution of consonants that precede either [i] or ["]...... 186! Table 5.11. Examples of the comparisons between traditional loanwords and New-type gairaigo...... 190! Table 5.12. List of the titles of songs recorded by the Beatles that contain, or potentially contain [ti], [t"], [di], or [d"] ...... 194!

Table 5.13. Distribution of [ti], [t"], [di], and [d"] in the titles of songs recorded by the Beatles...... 197! Table 5.14. Samples of stereotypical pronunciations of [ti] and [di] by the older speakers ...... 205!

Table 5.15. General tendencies of cultural values associated with [ti], [di], [t"], and [d"] ...... 208! Table 6.1. Sporadic postnasal voicing in counter phrases ...... 218! 12

ABSTRACT

This study investigates innovative sound sequences in Japanese. A relatively large number of phonological changes have occurred in the short period of time since WWII, mainly due to an influx of loanwords from English. However, innovative sound sequences have not been accepted in Japanese uniformly. This fact raises two questions.

Why are some innovative sound sequences fully accepted in Japanese while others are still foreignisms? Why are certain sound sequences acceptable in one situation, but not so in others?

Previous studies on innovative sound sequences in modern standard Japanese have tried to solve these problems by establishing innovative lexical strata, such as

“Assimilated Foreign” and “Unassimilated Foreign.” However, this study found that the distribution of innovative sound sequences is much more complex than previously believed. Furthermore, in many cases, the acceptance of innovative sound sequences is word-by-word or speaker-by-speaker. This suggests that the cause of the distribution of innovative sound sequences in Japanese is better described as an intricate interaction among various extra-grammatical factors, such as processes of borrowing, speakers’ socioeconomic status, influence of English education, acoustic and articulatory phonetics, the writing system, and historical linguistic factors. 13

CHAPTER 1

INTRODUCTION

1.1 Statement of the Problem

Language change has been well-studied in linguistics. Studying language change involves many factors. In this dissertation, I roughly divide these factors into two categories: grammatical factors and extra-grammatical factors. According to Ohala

(2003), some linguists believe that language change can be attributed to grammar—“the psychological representation of language” (Ohala 2003:638). Many linguists who support this view maintain that language change caused by altering the grammar improves communication in terms of speech production and perception.1

Other linguists do not take the position that language change is caused solely by grammatical factors. Rather, changes in grammar are viewed as resulting from the influence of extra-grammatical factors—such as phonetic and sociolinguistic factors.2

Historically and typologically recurrent and regular sound changes can often be explained by phonetics (Ohala 1992, 2003; Guion 1998; Blevins 2004). Irregular and language- specific changes must be explained by extra-grammatical factors, as well. Extra- grammatical factors of language change frequently have something to do with language contact and social factors such as socioeconomic class, ethnicity, gender, age, and social

1 See Ohala (2003:683) for a more detailed discussion and a list of linguists who support this view. 2 Blevins (2004) claims that some aspects of phonetics are part of universal grammar (UG): innate knowledge of language. I do not object to her position; however, for the sake of my argument, I contrast phonetics and traditionally believed grammatical aspects, such as constraints used in Optimality Theory (Prince and Smolensky 2004). 14 class (Labov 1972, 2001; Blevins 2004; McMahon 1994). Foreign language education also affects language change. According to McMahon (1994), language contact and lexical borrowing depend on bilingualism, and the degree of bilingualism required for lexical borrowing varies from “very restricted” to “far-reaching.” McMahon’s definition of bilingualism, thus, includes foreign language education (or second language education).

The cause of language change is the combination of various factors described above, and any change in any of the factors may affect the result. Since no sound recording before 1877 is available, the direct cause of language change in the past is often difficult to determine—in particular when the written records do not reflect the cultural background. In contrast, the causes of ongoing changes are detectable if the cultural backgrounds are carefully examined. Labov’s (1972) study on variations of English in

Martha’s Vineyard is a classic example of such a case where social factors are identified as the causes of language change. Labov collected tokens of the centralized /ay/ and /aw/ from various age groups and social and ethnic backgrounds in Martha’s Vineyard. The results demonstrate that younger age groups (except for 30 and under)3 produced higher numbers of centralized vowels, and the distribution patterns differed among the different social and ethnic groups.

In Japan, a relatively large number of phonological changes have occurred in the short period of time since WWII. These changes are well-documented. Recent innovative

3 Labov’s age groups are: 14-30, 31-45, 46-60, 61-75, 75+. 15 sound sequences can be found almost exclusively in loanwords. Consequently, some

Japanese language researchers believe that innovative sound sequences in Japanese are merely the results of language contact. Indeed, it is apparent that lexical items that include innovative sound sequences are etymologically related to English and other foreign languages, such as German, French, Italian, Spanish, Portuguese, and Russian.

Nonetheless, the distribution of these innovative sound sequences raises several substantial questions. One question is why some sound sequences have been prevalent, whereas others have not. Innovative sounds sequences have not been borrowed uniformly into Japanese. For example, the [ti] sequence and [si] sequence are both innovative. It has been believed that Japanese lacked these sound sequences because /ti/ and /si/ in Japanese native items are realized as [c&i] and [&i] respectively (Shibatani 1990). Both [ti] and [si] can be found in source words, such as Citibank [sitibæ"k]. However, while [ti] in a source word is likely to be realized as [ti] in a loanword, [si] in a source word is never realized as [si]. Consequently, the loanword based on Citibank is pronounced as

[&itib%"k"] in Japanese (Ito and Mester 1999). Some researchers give a grammatical account for why the sequence [ti] is prevalent but [si] is still a foreignism. For example,

Hattori (1960) claims that there are structural differences between [ti] and [si]. Ito and

Mester (1995b, 1999, 2006) assert that there are hierarchical differences in constraints that prohibit [ti] and [si]. However, if we scrutinize the available data, the distribution of these sound sequences is much more complex than can be explained on grammatical grounds alone. 16

In this dissertation, I take the position that the distribution of innovative sound sequences cannot be explained without considering extra-grammatical factors, such as sociolinguistic factors, historical linguistic factors, and language education. Language change can be best described as the interaction between grammatical factors and extra- grammatical factors.

1.2 The Goals of this Dissertation

The goals of this dissertation are twofold. The first goal is to present descriptions of innovative sound sequences in Japanese that are more accurate than in previous accounts.

The data that are presented in traditional studies of Japanese phonology are frequently biased. In many of these studies—synchronic studies in particular—the data function as evidence for the existence of certain regularities or constraints. Consequently, the data that do not support the claim are likely to be excluded from the study. As a result, some of the discussions among Japanese linguists are preoccupied with the identification of exceptions to the rules and modifications to the rules.4 In order to avoid this situation, I propose that we investigate all words in the dictionary. As I will describe below, it is relatively easy to do so if we use electronic versions of dictionaries. By examining all the words that contain innovative sound sequences, it is possible to detect all the exceptions.

If an exception to a regularity is found, the exception must be explained. If it is unexplainable, the proposed regularity should be questioned. Using this method, I will

4 For example, see discussions between Rice (1997, 2005) and Ito and Mester (1995b, 2003) and Ito, Mester and Padgett (1999) regarding the effectiveness of the constraint *NT, which prohibits post-nasal voicing. 17 present some exceptions and irregularities that have not been discussed in previous studies. I believe that the method I propose here will make the data of future studies of

Japanese more universal and less biased, which will reduce the unnecessary debate over the data and will let the discussion concentrate on the theoretical issues.

The second goal is to provide explanations for the exceptions and irregularities that I find as a result of the method I described above. I evaluate previous synchronic analyses of innovative sound sequences by proposing alternative views based on historical and sociolinguistic evidence. Although I criticize the synchronic analyses, I do not deny that certain regularities and constraints exist. I do, however, object to the claim that regularities and constraints can explain all phonological phenomena that are related to innovative sound sequences. This study found that extra-grammatical factors play significant roles in the case of innovative sound sequences in Japanese. As discussed above, language change involves many factors—grammatical factors, sociocultural factors, historical factors, and phonetic factors—and one of the duties of linguists is to identify how each factor affects grammar. Blevins (2004) claims that historical accounts should precede synchronic accounts because recurrent sound changes have phonetic motivations, and, as I discussed above, I consider phonetics extra-grammatical.5

Similarly, if a language-specific phenomenon can be explained by a sociocultural factor, there is no reason to provide synchronic analyses. Conversely, if a phonological

5 Blevins (2004) believes that certain phonetic features that motivate recurrent sound changes, as well as distinctive features, are parts of universal grammar. Therefore, she might disagree with my claim that phonetics is extra-grammatical. I use the term “extra-grammatical factors” to mean factors that are not considered grammar by most synchronic linguists. 18 phenomenon cannot be explained by any extra-grammatical factors, it should be analyzed synchronically. Therefore, the reader should not assume that I disregard or deny the importance of synchronic linguistics. Rather, I suggest that extra-grammatical studies can contribute to identifying genuine synchronic issues.

Studying appropriate linguistic data is essential for synchronic linguistics.

However, determining whether a phonological phenomenon is a subject of synchronic linguistics or not may be much more difficult than previously believed. Chomsky and

Halle (1968) claim that the object of generative linguistics is “the potential performance of an idealized speaker-hearer who is unaffected by” grammatically irrelevant factors that include “nonlinguistic knowledge and beliefs” (1968:3).6 Previous studies on Japanese phonology, such as Ito and Mester (1995a, 1995b, 1999, 2003, 2008), Kawahara (2006), and Nishimura (2006), are based on the assumption that their data reflect the object of generative linguistics described above. However, the study in this dissertation reveals that modern standard Japanese contains many variants created by extra-grammatical factors.

The discrepancy between the claims in previous studies and the study in this dissertation demonstrates that the identification of appropriate objects of synchronic linguistics, as proposed by Chomsky and Halle, is much more difficult than previously believed.

Assuming that all synchronic linguistic data in a homogeneous community are free from extra-grammatical factors may result in incorrect synchronic analyses. Therefore, linguistic data must be carefully examined by historical linguists and sociolinguists

6 See Labov (1972) for a counter argument to this view. 19 before synchronic linguists can claim that the data are truly suitable for the synchronic study. Consequently, successful synchronic analyses depend on successful extra- grammatical studies.7

1.3 This Study is not Limited to Loanword Phonology

Since most innovative sound sequences are found in recently borrowed items, this dissertation discusses the phonology of loanwords extensively. However, my focus is not solely loanword phonology. I am more interested in the interactions among various factors in language change. Loanword phonology is merely one of these factors—albeit an important one. That is to say, traditional Japanese phonology and loanword phonology interact with each other and eventually will create innovative phonology. For example, most geminate voiced obstruents are found in recently borrowed items. But they are also found in some literature as sound effects. Moreover, recent slang expressions have geminate voiced obstruents as well. These facts suggest that phonology that applies only to innovative sound sequences, namely loanword phonology, will eventually influence the phonology of native items by creating a single phonology for Japanese.

7 Blevins (2004) also asserts the significant roles of phonetics, historical linguistics and sociolinguistics. Blevins’s major interest is in historically and typologically recurrent sound changes that are based on phonetics, whereas this dissertation focuses on changes of a particular language in a limited time. Phonetics may not play a significant role in analyses on non-recurrent particular language changes. Therefore, this dissertation emphasizes historical linguistics and sociolinguistics. 20

1.4 Japanese Sounds

In this section, I will briefly illustrate the sounds of Japanese, including the difference between traditional sound sequences and innovative sound sequences.

1.4.1 Japanese Vowels

8 Modern Standard Japanese has five vowels: [%], [i], ["], [e], and [o]. The symbol ["] represents a high back unrounded vowel. In non-standard Japanese, particularly in western Japan where the capital used to be located, a more rounded vowel—closer to

[u]—is used instead of ["].

1.4.2 Sound Sequences of Modern Standard Japanese

Table 1.1 illustrates sound sequences of modern standard Japanese. I present the data in

Table 1.1 as bisegmental sequences rather than as single phonemes, because the sound changes of modern standard Japanese did not occur at the phonemic level, but at the bisegmental or larger level. That is, no particular vowels or consonants are innovative; therefore, listing single phonemes does not help in understanding the sound changes of modern standard Japanese. The data is based on the author’s knowledge as a native speaker. I also refer to Mabuchi (1971, 1996), Ito and Mester (1995b), and Vance (2008).

8 Throughout this dissertation, whenever I display vowels of Japanese, I keep the order of [%], [i], ["], [e], and [o], following the Japanese tradition. 21

Table 1.1. Sound sequences of Japanese9 [%] [i] ["] [e] [o] [k] [k%] [ki] [k"] [ke] [ko] [kj] [kj%] [kj"] [kjo] [!] [!%] [!i] [!"] [!e] [!o] [!j] [!j%] [!j"] [!jo] [s] [s%] [si] [s"] [se] [so] [dz]~[z] [dz%]~[z%] [dzi]~[zi] [dz"]~[z"] [dze]~[ze] [dzo]~[zo] [&] [&%] [&i] [&"] [&e] [&o] [c&] [c&%] [c&i] [c&"] [c&e] [c&o] [)*]~[*] [)*%]~[*%] [)*i]~[*i] [)*"]~[*"] [)*e]~[*e] [)*o]~[*o] [t] [t%] [ti] [t"] [te] [to] [ts] [ts%] [tsi] [ts"] [tse] [tso] [tj] [tj"] [d] [d%] [di] [d"] [de] [do] [dj] [dj"] [h] [h%] [hi] [h"] [he] [ho] [+] [+%] [+i] [+"] [+e] [+o] [+j] [+j"] [p] [p%] [pi] [p"] [pe] [po] [pj] [pj%] [pj"] [pjo] [b] [b%] [bi] [b"] [be] [bo] [bj] [bj%] [bj"] [bjo] [n] [n%] [ni] [n"] [ne] [no] [nj] [nj%] [nj"] [njo] [m] [m%] [mi] [m"] [me] [mo] [mj] [mj%] [mj"] [mjo] [#] [#%] [#i] [#"] [#e] [#o] [#j] [#j%] [#j"] [#jo] [j] [j%] [j"] [je] [jo] [,] [,%] [,i] [,e] [,o]

9 Shaded cells indicate innovative sound sequences. [&] is a voiceless alveolo-palatal fricative. [)*] is a voiced palatal affricate. [+] is a voiceless bilabial fricative. The issue of [h"] and [+"] will be discussed in Chapter 3. [#] is an alveolar flap. [,] is a velar approximant. The table lacks [-], which is a uvular nasal. This sound appears only in a syllable-final position; therefore, it is excluded from the table. All non-palatal consonants are palatalized before [i] or [j], but [(] has been omitted from the transcriptions.

1.4.3 Innovative Sound Sequences

In this dissertation, I define innovative sound sequences as those sound sequences which appeared in or after 1868—the first year of the Meiji period (1868~1912). Innovative sound sequences are shown in Table 1.2.

Table 1.2. Innovative sound sequences sorted in terms of onset

BILABIAL ALVEOLAR ALVEO-PALATAL VELAR P LOSIVE [ti] [t"] [di] [d"]

[tj"] [dj"] F RICATIVE [+%] [+i] [+e] [+o] [si] [zi] [&e] [ ]

+j"

A [*e/)*e] FFRICATE 10 [ts%] [tsi] [tse] [c&e] [tso]

A PPROXIMANT

[je] [,i][,e][,o]

10 The sequence [ts%] may not be innovative because some native items have this sound sequence. (e.g. [ototts%-] slang form of [oto$ts%-] ‘father’) See Hattori (1960:361), Akamatsu (2000:94), and Vance (2008:84) for further discussion. 23

I will concentrate on the following sound sequences: [+%], [+i], [+e], [+o], [ti],

[t"], [di], [d"], [si], and [&e]. In addition, I will discuss geminate voiced obstruents, such as [!!] in [b%!!"] ‘bag’ and [dd] in [beddo] ‘bed’.

1.5 Data: Non-Marginal Words and Marginal Words

The primary data used in this dissertation are taken from two major Japanese language dictionaries: K"jien (Shinmura 2008) and Daijirin (Matsumura 2006). I use DVD-ROM versions of these dictionaries to examine all words that include innovative sound sequences. In this dissertation, I call words found in at least one of these dictionaries non- marginal words. The advantage of using dictionaries is that they enable collecting less biased data. The disadvantage is the arbitrariness of the dictionaries themselves.

However, using two dictionaries may reduce this risk. One might raise concerns about the use of dictionaries as data sources because they often include words that are not actually used by average speakers. However, the comprehensiveness and authenticity of the words in dictionaries are unsurpassed by any other source. Another advantage of using dictionaries is that if a word has an alternative pronunciation, it is usually noted in the text of the dictionaries. In addition to K"jien and Daijirin, if circumstances require, I use

Nihon Kokugo Daijiten (Nihon Kokugo Daijiten Dainiban Hensh! Iinkai, ed. 2000-

2002), the largest Japanese language dictionary.

As secondary data, I use brand names, band names, titles of songs, movies, TV programs, and expressions used in manga (Japanese comics) and anime (Japanese animation). These data tend to be collected more haphazardly than non-marginal words; 24 therefore they may be biased. For this reason, I use these words as secondary or supplemental data. I call these words marginal words.

1.6 The Use of Japanese Orthography in this Dissertation

Transcription of speech sounds is always problematic because it is impossible for any writing system to adequately capture acoustic and articulatory information in the speech sounds of various languages. In order to minimize the inadequacy of representation in any particular writing system, traditionally, the International Phonetic Alphabet (IPA) is used for phonological analyses. IPA transcription is the best method for indicating places and manners of articulation. Nevertheless, in this dissertation, I will also use katakana and hiragana, two of the subsystems of Japanese orthography, as well as the goj#-on-zu

‘fifty-sound-display’, which arranges the kana syllabaries on a grid with the consonants in the columns and vowels in the rows. The columns are called “gy"” [!jo$] in Japanese.

Table 1.3 illustrates the goj#-on-zu.

Table 1.3. Goj#-on-zu ‘fifty-sound-display’ wara-gy" ya-gy" ma-gy" ha- na- ta-gy" sa-gy" ka-gy" a-gy" gy" gy" gy" わワらラやヤまマはハなナたタさサかカあア [,%] [#%] [j%] [m%] [h%] [n%] [t%] [s%] [k%] [%] りリみミひヒにニちチしシきキいイ [#i] [m(i] [h(i] [n(i] [c&i] [&i] [k(i] [i] るルゆユむムふフぬヌつツすスくクうウ [#"] [j"] [m"] [h"] [n"] [ts"] [s"] [k"] ["] れレめメへヘねネてテせセけケえエ [#e] [me] [he] [ne] [te] [se] [ke] [e] ろロよヨもモほホのノとトそソこコおオ [#o] [jo] [mo] [ho] [no] [to] [so] [ko] [o]

In each cell in Table 1.3, the letter on the top-left is hiragana and the letter on the top-right is katakana. Highlighting marks cases in which the consonant is phonetically markedly different (i.e., differs by more than just palatalization) from others in the same column. Most recently borrowed items are written with katakana.

Some phonologists might consider the use of Japanese orthography problematic for the following reasons. One popular criticism against the kana syllabaries is that they cannot indicate a consonant and a vowel in a syllable independently (Tsujimura 1996).

The goj#-on-zu ‘fifty-sound-display’ is criticized for not reflecting allophones accurately by including as many as three different consonants in a column (Hattori 1979).

Although katakana, hiragana, and the goj#-on-zu do not always reflect phonetic facts, I will use them as necessary, since they indicate some significant phonetic and phonological features primarily based on historical facts. Such features include voicing of obstruents, consonant gemination, and vowel substitution. These features are described 26 better with Japanese-specific orthography than with the International Phonetic Alphabet

(IPA).

1.6.1 Voicing of Obstruents in Katakana Representation

Voiced obstruents are marked with two dots called dakuten: daku and ten literally mean

‘muddy’ and ‘points/dots’ respectively. All voiced obstruents, except for the first half of a voiced geminate consonant, are marked with dakuten in the Japanese kana system, as illustrated in (1).

(1) Voicing contrast of カ /ka/ and ガ /ga/

カ /ka/ + ､､ [+voice] = ガ /ga/

1.6.2 Geminate Consonants in Katakana Representation

Geminate consonants are also referred to as long consonants. Consonant gemination is noted by a reduced-size ツ [."] in katakana—as shown in (2b), and a reduced size つ

[."] in hiragana.

(2) a. バグ /bagu/ ‘bug’

b. バッグ /baggu/ ‘bag’

1.6.3 Vowel Substitution

Kana cannot represent a consonant and a vowel in a syllable independently (Tsujimura

1996). Thus, in order to represent a new CV sequence, existent kana need to be modified. 27

For example, traditional Japanese did not have a [ti] sequence. The [ti] sequence is represented by a combination of テ [te] and a reduced sized イ [i], as shown in (3).

(3) テ [te] + イ [i] = ティ [ti]

Traditionally, kana letters are believed to indicate moras, such as a vowel, or a combination of a consonant and a vowel. But vowel substitution shows that this is not always the case.11

1.7 Problem with the Use of Katakana Spellings

1.7.1 Approximants

Kana letters are not suitable for representing the following sound sequences with approximants: [,i], [,e], [,o], and [je]. It is possible to attempt to represent these sound sequences, for example, by combining ウ/u/ and reduced sized イ/i/, but this attempt does not always succeed. The problem is that it is not always clear just by looking at the katakana spelling whether ウィ is pronounced as [,i] or ["i]. The problem involves approximants are illustrated in (4).

11 Hansell (2002:132) calls the combination of two syllabaries to indicate one syllable “syllable telescoping.” He indicates that the situation of syllable telescoping requires the same number of letters as the Roman alphabet does. 28

(4) ウィ/wi/ = ウ/u/ + イ/i/ ([,i]? ["i]?)

ウェ /we/ =ウ/u/ + エ/e/ ([,e]? ["e]?)

ウォ /wo/ =ウ/u/ + オ/o/ ([,o]? ["o]?)

イェ /je/ = イ/i/ + エ/e/ ([je]? [ie]?)

It is worth investigating the distribution of sound sequences shown in (4).

However, since kana spelling does not represent them reliably, I will not discuss innovative sound sequences with approximants in this dissertation.

1.8 Other Issues That Will Not Be Discussed In This Dissertation

1.8.1 Accent and Intonation

Accent is one of the significant aspects of Japanese phonology. Linguists from various fields—such as historical linguists, sociolinguists, and theoretical linguists—have extensively studied Japanese pitch accent (McCawley 1968; Mabuchi 1971; Vance 1987,

2008; Ito and Mester 2003; Katayama 1998; Pierrhumbert and Beckman 1988).

Nevertheless, I will not discuss accent in this dissertation because the dictionaries I use do not allow me to. For the same reason, I will not discuss intonation.

1.8.2 Vowels

Changes in vowel quality are also well-investigated topics in Japanese historical linguistics (Hashimoto 1950; Mabuchi 1971; Miyake 2003; Takayama 2003). However, I will not discuss vowels in this dissertation because the writing system does not represent 29 such differences. Researching changes in vowels requires empirical studies. The only exceptions are devoiced vowels, which will be discussed in Chapter 5.

1.9 Previous Studies

In this section, I will review some previous studies that discuss innovative sound sequences in Japanese.

1.9.1 Hattori (1960): Seams and Gaps—Structural Asymmetry

Hattori claims that the distribution of traditional Japanese moras is asymmetrical. As we have seen, traditional Japanese lacks some sound sequences, such as [si], [ti], and [t"]. In traditional Japanese phonology, instead of these sound sequences, [&i], [c&i], and [ts"] are believed to appear, respectively (Shibatani 1990). Hattori argues that pronouncing /si/ as [&i] is phonetically well-grounded: the consonant of [&i] appears as a result of assimilation to the following high front vowel. In contrast, he claims that there are no phonetic reasons for [c&] and [ts] to appear instead of [t] when followed by a high vowel.

Hattori concludes that the consonant of [&i] is phonemically the same as the consonants of [s%], [s"], [se], and [so], but that the consonants of [c&i] and [ts"] belong to a separate phoneme /c/, which is proposed by Hattori himself. Hattori calls the empty spots created by establishing a new phoneme /c/ “gaps”. According to Hattori, the innovative sound

12 sequences /ti/ [ti] and /tu/ [t"] fill in the gaps. On the other hand, he states that

12 However, Hattori does not directly argue that the appearance of the innovative sound sequences [ti] and [t"] in Japanese is caused by the motivation of filling the gaps. He is certainly aware of the fact that Japanese has innovative sound sequences. However, he appears to be focusing on the description of the asymmetric distributions of traditional Japanese and to be indifferent to the innovative sound sequences in general. 30 pronouncing [si] is not impossible for Japanese speakers, but the appearance of [si] does not fill in a gap because the consonants of [si] and [&i] share the same phoneme /s/.

Hattori calls the spot where [si] fits a “seam”.13 Table 1.4 illustrates the difference between seams and gaps.

Table 1.4. Seams and Gaps

% i " e o

&i /s/ s% s" se so si (SEAM)

/t/ t% ti (GAP) t" (GAP) te to

/c/ c&i ts"

Hattori’s explanation depicts the asymmetric distribution of traditional Japanese, but raises some questions. First, he claims that the affrication of [t] followed by [i] is phonetically ungrounded; however, typologically, it is not uncommon (see Chapter 5).

For this reason, the difference between a “gap” and a “seam” seems rather unclear.

Second, the levels of acceptance of the sequences [si], [ti], and [t"] do not correspond to typological tendencies. Typologically and historically, palatalization of Japanese /s/ in front of a high front vowel is not special at all, since palatalization of /s/ when adjacent to a high front vowel happens in many languages (Blevins 2004:142). Palatalization and

13 Hattori calls seam and gaps “sukima” !"# and “akima” $"# respectively. 31 affrication of /t/ followed by a high front vowel is also observed in other languages, such as Brazilian Portuguese and Cairene Arabic (Yoshida 2001). However, affrication of /t/ followed by a high back vowel is not common. Thus, the order of phonetic or typological naturalness is as illustrated in (5).

(5) #More natural Unnatural$

[&i] for /si/ > [c&i] for /ti/ > [ts"] for /tu/

However, the order of acceptance is different, as shown in (6).

(6) #Foreignism Well accepted$

[si] > [t"] > [ti]

This comparison of (5) and (6) suggests that phonetic and typological naturalness alone cannot explain the disproportional distribution of innovative sound sequences. Third, it is still debatable whether the sequences [ti] and [t"] have really filled in the gaps or not.

It is not fair simply to reject Hattori’s theory of seams and gaps, because it is not intended to explain innovative sound sequences. It was developed in order to account for the synchronic structural asymmetry of Japanese. In addition, Hattori does not directly mention the cause of the emergence of innovative sound sequences. However, it is probably not a good idea to use his theory to explain innovative sound sequences if it is not intended to do so. 32

1.9.2 McCawley (1968): Generative Phonology

McCawley (1968) indicates that possible sound sequences are different among native

Japanese vocabulary items, Sino-Japanese vocabulary items, and foreign vocabulary items, as illustrated in Table 1.5.14

Table 1.5. Lexical stratification by McCawley Sino-Japanese and Foreign Native Onomatopoeia Cu % % %

Cyu ! % % Co % % % Cyo ! % % Ca % % % Cya % % % Ce % % % Cye ! ! % Ci ! ! % Cyi % % %

(Cy denotes sharp consonant, C plain consonant, % occurrence, ! non-occurrence)

(Source: McCawley 1968:65)

In order to differentiate lexical items among strata, McCawley introduces the diacritical features (Chomsky and Halle 1968) [± Native], [± Sino], and [± Foreign].

14 McCawley claims that mimetic items (he calls them “onomatopoeia” reluctantly, noting that they differ from English onomatopoeia) have the same sound sequences as Sino-Japanese items. However, he also provides extensive lists of distinctive features (McCawley 1968:91-93). The lists differentiate all four strata. 33

McCawley stresses that these diacritical features are based on synchronic grammar and

15 not on historical considerations (1968:75). For example, the loanword [mi#"k"] ‘milk’ is [+ Native] because it does not have either Cyu, Cyo, Cye, or Ci. It had been known that

Japanese has four lexical strata and items in each stratum behave phonologically slightly differently. But these differences were interpreted as etymological facts until McCawley

(1968). McCawley’s definition of lexical strata based on synchronic grammar is innovative. Nonetheless, if the synchronic definition is decided by observing the phonological distribution of etymological strata, then it is not perfectly synchronic.

In McCawley’s analyses, the diacritical features [± Native], [± Sino], and

[± Foreign] function as if they are similar to distinctive features. However, the nature of diacritical features and distinctive features must be different because the former are language-specific (or exceptional), while the latter are universal and innate to all human beings. Language stratification and use of diacritical features have influenced succeeding studies of synchronic Japanese phonology, but it is not clear where these diacritical features as synchronic grammatical knowledge come from.

1.9.3 Ito and Mester (1995a, 1995b, 1999): Optimality Theory (OT)

The approaches of Optimality Theory (OT) and traditional generative phonology differ.

The former is a constraint-based approach, while the latter is rule-based (Prince and

Smolensky 2004; Archangeli and Langendoen 1997; Kager 1999). In the analyses of

15 Chomsky and Halle (1968) also claim that while it is possible to provide a historical explanation for the classification of the lexicon, such explanation is irrelevant because speakers do not have to know such historical facts in order to acquire grammar. 34 innovative sound sequences in Japanese, both approaches use a notion of lexical stratification that is based on synchronic grammar. However, OT includes more recent information and is more elaborated to describe the situation of innovative sound sequences in Japan.

Ito and Mester (1995b) review constraints in Japanese phonology and create a structure based solely on constraints. They first propose a core that consists of three strata

(Yamato, Mimetic, and Sino-Japanese). According to Ito and Mester (1995b) these strata are related in an intricate manner. Yamato and Sino-Japanese items allow /p/ only in a geminated or partially geminated form (*P); in Yamato and Mimetic, post-nasal obstruents must be voiced (*NT);16 and in all of Yamato, Mimetic, and Sino-Japanese, geminate voiced obstruents are disallowed (*DD). These constraints can differentiate the strata as demonstrated in Table 1.6. Ito and Mester stress that Foreign items do not belong to a homogeneous stratum. Rather, the constraints that govern Foreign items develop a continuous and gradual structure around the core.

16 Asano (2004) indicates that the Mimetic stratum has items that violate *NT. Kawahara, Nishimura and Ono (2002) state that “postnasal voicing fails to take effect at stem-initial syllables” in Mimetic items. Rice (1997) questions the effectiveness of *NT in both Yamato and Mimetic items. 35

Table 1.6. The structure of strata based on constraints Yamato *P *NT *DD

Sino-Japanese *P -- *DD

Mimetic -- *NT *DD

Foreign ------

(Source: Ito and Mester 1995b:820)

Ito and Mester (1995b) demonstrate that the structure of constraints in Japanese is gradual and complex. How speakers access the correct output by applying the correct constraints has been a challenging question for synchronic linguists. As we have seen, in the traditional generative approach, McCawley (1968) tries to solve the problem by introducing the diacritical features [± Native], [± Sino], and [± Foreign]. In the framework of OT, Ito and Mester (1995a) propose a model in which a Faithfulness constraint reranks among strata. This model, which is later called the co-grammar approach (Ito and Mester 2008) or cophonology approach (Inkelas and Zoll 2003), requires at least four co-grammars in a language. Ito and Mester (2008) claim that the co- grammar approach is problematic because it requires reranking of faithfulness constraints. They also indicate that the co-grammar approach may not be able to evaluate compound forms composed of morphemes from different lexical strata.

In order to overcome these problems, Ito and Mester (1999) introduce the indexed faithfulness approach. In the indexed faithfulness approach, faithfulness constraints that rerank among co-grammars do not exist. Instead, the strata, such as [YAMATO] and [SINO 36

JAPANESE], are indexed to the faithfulness constraints and they are ranked in the appropriate positions so that the correct output will be produced. The stratum indices are applied to the input as well, and non-correspondent faithfulness constraints are ignored.

1.10 Lexical Stratification

As we have seen, all synchronic studies of innovative sound sequences depend on the presupposition that lexical stratification is accessible to the speakers. But where does the knowledge of lexical stratification come from? Ito and Mester (1999) claim that Japanese speakers are aware of lexical stratification because it is reflected in the writing system.

Yamato items use hiragana and kanji, the Sino-Japanese vocabulary items use kanji, and the Foreign vocabulary items use katakana. Perceptual experiments by Gelbart and

Kawahara (2007) demonstrate that Japanese speakers use their knowledge of lexical stratification in speech perception. Vance (2002) and Rice (1997), on the other hand, argue that lexical stratification is not grammatical, but etymological knowledge.17

Indeed, most Japanese adult speakers are able to classify many words into the correct stratum. On the other hand, it is also true that there are many words for which speakers have trouble determining the correct strata. Moreover, young children who do not have knowledge of lexical stratification or the writing system of Japanese do not have any trouble pronouncing a compound word of a Foreign item and a native item, such as

17 Rice (1997) is specifically critical about the validity of the constraint against postnasal voicing in modern standard Japanese (see Chapter 6). However, Rice (1997) does not reject the idea that lexical strata specific grammars might exist. 37

18 [kiti c&%-] ‘(Hello) Kitty’. Adult Japanese speakers may have knowledge of lexical stratification and they may use this knowledge in speech perception. However, doing so helps semantic recognition, but does not affect phonology.

One may claim that the items belonging to different strata have various statistically reliable properties that separate one stratum from another. However, I do not think there is any set of properties that can reliably assign items to the strata that have been suggested for Japanese. See Ohala (1992) on English (review of SPE).

1.11 Structure of Dissertation

Chapter 2 discusses geminate voiced obstruents. Geminate voiced obstruents appear to follow certain grammatical rules. However, there are many exceptions due to extra- grammatical factors. The situation of geminate voiced obstruents depicts how grammatical factors and various extra-grammatical factors interact. In this chapter, I review both synchronic analyses and a sociolinguistic study. I also present wide varieties of data that contradict previous studies. I conclude that geminate voiced obstruents involve extra-grammatical factors such as phonetics, second language education, writing systems, and socioeconomic status. These issues also show intricate interactions between factors that create subtle exceptions.

In Chapter 3, I challenge a widely believed proposition that the initial sound of the sound sequence /fu/ in modern standard Japanese is a voiceless bilabial fricative.

18 See Rice (1997, 2005) for more extensive arguments against language stratification in terms of learnability and the writing systems. 38

Then, I show that the issues surrounding Japanese /f/ are largely historical, not synchronic.

Chapter 4 discusses the voiceless coronal fricatives in [si] and [&e]. I explain the distributional difference between [si] and [&e] by considering historical facts, as well as acoustic and articulatory facts. I then provide sociocultural reasons why [si] is still a foreignism in Japanese, considering various situations when [si] could be used.

Chapter 5 deals with Japanese [ti] and [t"] and their voiced counterparts. First, I reject the claim that the appearance of [ti] and [t"] is merely filling in systematic gaps by presenting the fact that the vowels of Japanese [ti] and [t"] are never devoiced. Then I discuss how sociocultural factors affect the levels of acceptance of Japanese [ti] and [t"].

Finally, Chapter 6 summarizes the findings and discusses the conclusions.

CHAPTER 2

GEMINATE VOICED OBSTRUENTS

2.1 Introduction

Geminate obstruents are long obstruents with an extra mora (Vance 1987). They are also called ‘long consonants’ (Blevins 2004). In this dissertation, I represent geminate consonants by repeating the same consonant. For example, the geminate voiced velar stop is represented as [!!]. It is important to note that there is not an oral closure release in between the two stops (Vance 1987). Geminate voiceless obstruents are observed in all strata of Japanese vocabulary. On the other hand, geminate voiced obstruents are found almost exclusively in recently borrowed foreign vocabulary items in Japanese. The only native word with a geminate voiced obstruent I can find in the dictionaries is the mimetic item [z%bb"] ざっぶ (< [z%mb"#i] ざんぶり) ‘a sound of diving into water’, which is used mainly for ky"gen (a traditional Japanese farce).19 In addition to non-marginal words, geminate voiced obstruents can be found in emphatic expressions, such as

[s"!!oi] (< [s"!oi] ‘amazing’) (Vance 1987), [j%bb%i] (< [j%b%i] ‘no good’) (Hirayama

2005), and onomatopoeia, such as gunshot sound effects used in manga (Japanese comics).

Geminate voiced obstruents show somewhat irregular distributions, and scholars have been trying to find regularities by proposing various models (Ito and Mester 1999;

19 K"jien lists [z%bb"] ざっぶ, whereas Daijirin does not. 40

Nishimura 2006; Kawahara 2006). Nevertheless, no model can satisfactorily account for the distribution of geminate voiced obstruents. In this chapter, I argue that no synchronic grammar can account for the distribution of geminate voiced obstruents because that distribution is a reflection of an ongoing diachronic process. Instead of proposing a single model, I examine the distribution more carefully, dividing it into several patterns, and offer multiple accounts to explain each pattern.

2.2 Overview of the Issues in Geminate Voiced Obstruents in Japanese

Japanese has had geminate voiceless obstruents since at least the 16th century and single voiced obstruents since at least the 8th century (Mabuchi 1971). Some Yamato vocabulary items acquired geminate voiceless obstruents because of the influence from emphatic expressions (Vance 1987). Consequently, both geminate voiceless obstruents and single voiced obstruents can be found in all vocabulary strata today. Geminate voiced obstruents, on the other hand, are found only in emphatic expressions, mimetic words, or recent loanwords. Rice (2006) indicates that the emergence of geminate voiced obstruents means the creation of a new contrast. Rice focuses on the contrast between geminate voiced obstruents and geminate voiceless obstruents. However, there is another contrast between geminate voiced obstruents and single voiced obstruents. Consider the contrast between [b%!!"] ‘bag’ and [b%kk"] ‘back’, and between [b%!!"] ‘bag’ and [b%!"]

‘bug’. The former pair contrasts in terms of voicing, and the latter pair contrasts in terms of gemination. The problem with geminate voiced obstruents is that their voicing values are sometimes unstable, making the contrast, such as the one in [b%!!"] ‘bag’ and 41

[b%kk"] ‘back’, ambiguous. Interestingly, in the case of the gemination contrast, such as

20 [b%!!"] ‘bag’ and [b%!"] ‘bug’, there seems to be little confusion. Thus, Japanese speakers may confuse [b%!!"] ‘bag’ and [b%kk"] ‘back’, but they do not confuse

[b%!!"] ‘bag’ and [b%!"] ‘bug’. Kawahara (2005, 2006) and Nishimura (2006) indicate the instability of voicing in geminates, providing numerous examples of voicing variations in geminate obstruents. They conclude that voicing is particularly unstable when the word with a geminate voiced obstruent contains another voiced obstruent. The previous studies lead to the following questions. Is the instability of gemination uniform, as previous studies suggest? And if not, what are the causes of nonuniformity?

2.3 Previous Studies

2.3.1 Ito and Mester (1999)

Geminate voiced obstruents have been studied since the middle of the last century (for example, Lovins 1975; Vance 1987 and work cited there), but current studies on geminate voiced obstruents were probably inspired by the work of Ito and Mester (1999).

The cophonology approach, or co-grammars approach (Anttila 2002; Inkelas, Orgun and

Zoll 1997; Ito and Mester 1995) presupposes multiple different grammars that exist simultaneously in one language in order to explain different output depending on lexical strata. Instead, Ito and Mester (1999) proposed the indexed faithfulness approach, in which inputs and faithfulness constraints are indexed according to the lexical strata they

20 Since [s"!!oi] (emphatic expression) and [s"!oi] ‘amazing’ have the same basic semantic value, geminate voiced obstruents in emphatic expressions ([!!]) and single voiced obstruents ([!]) may not contrast. 42 belong to, preventing the creation of multiple grammars. For example, a constraint applied only to ASSIMILATED FOREIGN words is FAITH/ASSIMILATED FOREIGN, which includes the index ASSIMILATED FOREIGN. The constraint FAITH/ASSIMILATED FOREIGN affects only inputs with the index ASSIMILATED FOREIGN. In this way, the indexed faithfulness approach can avoid multiple grammars in one language. Ito and Mester

(1999) applied the indexed faithfulness approach to account for the voicing variations of geminate voiced obstruents. In addition to the faithfulness constraints

FAITH/UNASSIMILATED FOREIGN and FAITH/ASSIMILATED FOREIGN, the markedness constraint NO-DD (Ito and Mester 1995, 1999), which prohibits geminate voiced obstruents (DD), is introduced. NO-DD is ranked in between FAITH/UNASSIMILATED

FOREIGN and FAITH/ASSIMILATED FOREIGN, as shown in (1).

(1) FAITH/UNASSIMILATED FOREIGN » NO-DD » FAITH/ASSIMILATED FOREIGN

If the input has a DD and is indexed with /UNASSIMILATED FOREIGN/, the constraint

FAITH/UNASSIMILATED FOREIGN prohibits devoicing; thus the output maintains voicing in

DD. If the input has a DD and is indexed with /ASSIMILATED FOREIGN/, the highest ranked constraint FAITH/UNASSIMILATED FOREIGN does not evaluate DD, but the second highest constraint NO-DD prohibits voicing of DD. In this way, the DD in the input is devoiced in the output. 43

(2) f Unassim F f Assim F oreign oreign

AITH AITH / / ilated

ilated

No No No - - DD NT ------P

/beddo/ ! [beddo] * Unassimilated [betto] *! (IDENT-F) foreign

/baggu/ [bagg"] *! Assimilated foreign ! [bakk"] * (IDENT-F)

[pabb"] *! * /pabbu/ ! [pab"] * (IDENT-µ) * Assimilated foreign [hab"] **! (IDENT-F,µ) [habb"] *! * (IDENT-F)

(Source: Ito and Mester 1999:74)

In the tableau shown in (2), the constraint NO-DD is dominated by the faithfulness constraint UNASSIMILATED FOREIGN, and dominates the faithfulness constraint

ASSIMILATED FOREIGN; thus, DD occurs only for an input indexed as UNASSIMILATED

FOREIGN. Three candidates without geminate voiced obstruents are excluded: [bedo] for

/beddo/, [b%g"] for /baggu/ and [p%pp"] for /pabbu/. Ito and Mester claim that [bedo] violates “another higher-ranking constraint ALIGN-R (STEM, &), requiring that the right edge of a stem and a syllable coincide (Kitahara 1996), or a sympathetic faithfulness constraint requiring /d/ to maintain its syllable role (here: as a coda, see Katayama

(1998))” (Ito and Mester 1999:97). Instead of [p%pp"], [p%b"] is discussed as a 44 candidate because parsing the b non-moraically is one way to avoid voiced obstruent gemination. However, no clear reasons were provided for the exclusion of [p%pp"]. If

[p%pp"] is a valid candidate, it should be as good as the winning candidate [p%b"], according to the tableau shown in (2), since it violates the same constraint,

FAITH/ASSIMILATED FOREIGN. This issue will be discussed later.

Another problem in Ito and Mester (1999) is the use of the term assimilated in the constraints. Readers with no background knowledge of Japanese phonology might assume that the geminate consonants in ‘baggu’ were once pronounced as [!!], and now they are likely to be devoiced because the word ‘baggu’ went through the process of assimilation, just as the Chinese loanword kai$a ‘corporation’ underwent Rendaku21 later.

However, it is likely that the probability of devoicing in the word baggu has not increased since it was first borrowed.22 Thus, it is not appropriate to describe this phenomenon as

“assimilation”.

The most crucial problem in Ito and Mester (1999) is that the data in the analysis are not reliable: beddo and baggu are not always realized as [beddo] and [b%kk"] respectively. In fact, no other researchers would agree that there is a difference in devoicing patterns of beddo and baggu. In their later work, Ito and Mester (2008) treat both beddo and baggu as Assimilated Foreign. Ito and Mester (1999) explain the instability of the voicing of DD in beddo and baggu as follows:

21 Rendaku is the replacement of a morpheme initial unvoiced obstruent with a voiced obstruent when the morpheme is not in an initial position of a compound or a word. (Vance 1987) 22 Today’s Japanese dictionaries list bakku (バック) only as an optional transcription. 45

It is unsurprising that we find a considerable amount of variation in this area of the lexicon, with some speakers treating the loanword for ‘bed’ as ASSIMILATEDFOREIGN (i.e., betto), and the loanword for ‘bag’ as UNASSIMILATEDFOREIGN (i.e., baggu). (Ito and Mester 1999:97)

It is still not clear to me how speakers “treat” the index of an input. Analyses by the indexed faithfulness approach in Ito and Mester (1999) are indeed landmarks in

Optimality Theory, and they obviously inspired other researchers to investigate the issues relevant to geminate voiced obstruents. However, the data described by Ito and Mester are not as invariant as they suggest.

2.3.2 Nishimura (2006)

Nishimura (2006) gives more detailed analyses, claiming that optional devoicing is allowed in words with a DD and another voiced obstruent (“DVDDV-type” stem), but that it is not allowed in words with a DD but no other voiced obstruents (“TVDDV-type” stem). Nishimura indicates that the behavior of the TVDDV-type words and the

DVDDV-type words is similar to phenomena regulated by Lyman’s Law. The self- conjoined constraint [VOP2]stem disallows co-occurrence of more than one voiced obstruent within a stem. Nishimura (2006) claims that the difference in voicing between the DVDDV-type words and the TVDDV-type words is due to the locally conjoined constraint [[VOP2]stem &*DD]stem, which is violated if both [VOP2]stem and *DD are violated.23 The conjoined constraint dominates Ident (voice)-Foreign, VOP2, and *DD, as illustrated in (3).

23 *DD = No-DD 46

(3) [VOP2 & IDENT (voice) VOP2 *DD *DD] - Foreign a. DVDDV *! * * $ b. DVTTV * i. /DVDDV/Foreign c. TVDDV * *! d. TVTTV **! $ a. TVDDV * ii. /TVDDV/Foreign b. TVTTV *!

(Source: Nishimura 2006:87)

Tableaux (3i) and (3ii) show why the TVDDV-type words do not devoice the geminate, while the DVDDV-type words do. In tableau (3i), candidate (3ib) is selected because candidate (3ia) violates the conjoined constraint [[VOP2]stem &*DD]stem, which is ranked higher than the other constraints, and candidates (3ic) and (3id) violate more constraints than candidate (3ib) does. In tableau (3ii), candidate (3iia) is selected because IDENT (voice)-Foreign, which is violated by candidate (3iib), is ranked higher than *DD.

Nishimura further claims that since geminate devoicing is optional in the

DVDDV-type words, the ranking between [[VOP2]stem &*DD]stem and Ident (voice)-

Foreign is not fixed. The interchangeable rankings are depicted in (4). When Ident

(voice)-Foreign is ranked higher than [[VOP2]stem &*DD]stem, as illustrated in (4i), then the winning candidate is DVDDV. When the ranking order is reversed, as shown in 47

(4ii), the winning candidate is DVTTV, which is the same as in (4ii). Either case would not affect the result of the TVDDV-type words.

(4)

Tableau (i) IDENT (voice) - [VOP2 & *DD] VOP2 *DD Foreign $ a. DVDDV * * * b. DVTTV *! Tableau (ii) [VOP2 & *DD] IDENT (voice) - VOP2 *DD Foreign a. DVDDV *! * * $ b. DVTTV * (Source: Nishimura 2006:88)

Nishimura treats the issue as purely phonological, and describes the problems of geminate voiced obstruents more accurately than Ito and Mester. To date, Nishimura’s descriptions of the distribution of geminate voiced obstruents seem to be the most accurate. Nevertheless, Nishimura’s descriptions are still incomplete and inadequate. This issue is discussed later in this chapter.

2.3.3 Kawahara (2005, 2006)

Kawahara (2005, 2006) embraces Nishimura’s descriptions of the distributions of voicing varieties and provides a phonetic account of the distribution based on Nishimura (2006).

However, Kawahara (2006) claims that the conjoined constraint [[VOP2]stem

&*DD]stem (Nishimura 2006) is too powerful since it is a recursive local conjunction 48 that prohibits more than two occurrences of a particular structure: VOP2 is the self- conjunction of *VoiObs, and *DD is the local conjunction of *VoiObs and *Gem; therefore, the conjoined constraint [[VOP2]stem &*DD]stem has the structure illustrated in Figure 2.1.

[[VOP2]stem &*DD]stem

/ \

[VOP2]stem *DD

/ \ / \

*VOIOBS *VOIOBS *VOIOBS *GEM

Figure 2.1. Recursive local conjunction (Source: Kawahara 2006:548; with slight modifications)24

Kawahara claims that languages do not count more than two occurrences of a particular structure (Chomsky 1965, McCarthy and Prince 1986, cited in Kawahara

2006); therefore, recursive self-conjoining *X is undesirable since “it is possible to derive a language that counts the instances of X within a domain D via a constraint like

{{*X & *X}D & *X}D (Itô and Mester 1998:n. 17)” (Kawahara 2006:548).

24 The original diagram by Kawahara (2006:548) is illustrated below.

{{*VoiObs&*VoiObs}Stem &{*VoiObs&*Gem}Seg}Stem / \

{*VoiObs&*VoiObs}Stem {*VoiObs&*Gem}Seg / \ / \ *VoiObs *VoiObs *VoiObs *Gem

Instead of the conjoined constraint, Kawahara uses the Obligatory Contour

Principle (OCP), which prohibits two adjacent identical features (Goldsmith 1973; Leben

1973; McCarthy 1986; Odden 1986; Ito and Mester 1986, cited in Kawahara 2006). By following the analysis by Ito and Mester (1986), Kawahara proposes the constraint

OCP(+voi), which prohibits the occurrence of more than one voiced obstruent within a morpheme. Also, Kawahara (2006:545) introduces the new constraints

Ident(+voi)Sing(leton), which prohibits devoicing of singleton obstruents, and

Ident(+voi)Gem(inate), which prohibits devoicing of geminate obstruents. By ranking

Ident(+voi)Sing higher than Ident(+voi)Gem, Kawahara tries to account for the difference between the voicing patterns of DVDDV words and TVDDV words. In tableau (5),

OCP(+voi) is ranked higher than Ident(+voi)Gem in order to show that DVDDV words have a winning candidate with TT. But Kawahara claims that OCP(+voi) and

Ident(+voi)Gem can be unranked so that a candidate with DD can be an optional winner.

(5)

OCP(+voi) » Ident(+voi)Gem

/baggu/ Ident(+voi)Sing OCP(+voi) Ident(+voi)Gem a. [baggu] *! b. ! [bakku] * c. [paggu] *! d. [pakku] *! *

(Source: Kawahara 2006:546)

Since TVDDV words do not violate OCP(+voi), their winning candidates always have a DD, as illustrated in (6).

(6) Ident(+voi)Sing, Ident(+voi)Gem » *VoiObs.

/eggu/ Ident(+voi)Sing Ident(+voi)Gem *VoiObs a. ! [eggu] * b. [ekku] *! /bagu/ a. ! [bagu] ** b. [pagu] *! * c. [baku] *! * d. [paku] *!*

(Source: Kawahara 2006:547)

Kawahara (2005, 2006) also claims that geminate voiced obstruents (DD) are more likely to be devoiced than single voiced obstruents (D) because voicing contrasts are perceptually more confusable in DD than in D due to the weakened acoustic cues for

[+voice]. He conducted a production experiment and examined (1) closure voicing duration, (2) duration of the preceding vowel, (3) closure duration, (4) F0 of the surrounding vowels, and (5) F1 of the surrounding vowels of T, D, TT, and DD. He also conducted perceptual experiments. In the perceptual experiments, Kawahara covered the stimuli with cocktail party noise to confuse the listeners and measured the perceptibility differences among T, D, TT, and DD. The results of Kawahara’s experiments and his discussion are somewhat puzzling. In his production experiment, he claims that some 51 acoustic cues for [+voiced] are weakened in DD. For example, the closure voicing ratio

(NOT the closure voicing duration), closure duration, and spirantization are different.

However, all of the primary data he collected show that the voicing distinctions are maintained by statistically significantly differing values of acoustic cues. On the other hand, his claim that acoustic cues for [+voiced] in DD are weaker than in D is based on his impression of the data and no statistical analysis is provided. Moreover, in his perceptual experiment, all the words he used were TVDDV words or TVTTV words, yet the results show that DD in TVDDV were misperceived more frequently. These results seem to contradict the claim by Nishimura (2006) and Kawahara (2005, 2006) that devoicing of TVDDV words is impossible: if the instability of DD is due to poor perception, TVDDV words could have the devoicing option. Although the arguments in

Kawahara (2005, 2006) are weak, I do not deny the possibility that the perception of DD is poorer than that of TT or D. This may be one of the reasons that DD has not been found in Japanese vocabulary items until quite recently. Nevertheless, the more important fact is that Kawahara’s data in his production experiments demonstrate that Japanese speakers are capable of producing DD with statistically significantly differing values of acoustic cues. This fact suggests that although DD is innovative in Japanese, it has the potential to be more stable in the future. But since Kawahara did not use any DVDDV words in his perceptual experiment, the difference between the stabilities of DD in

TVDDV and DVDDV is still unknown. 52

2.4 Problems with Data Description in Previous Analyses

As mentioned above, Nishimura’s description of DD is more accurate than the one provided by Ito and Mester (1999), but it still needs substantial adjustment. I will describe more complex cases of the behavior of DD. In order to describe the behavior of words with DD, I will primarily depend on my observations and my own grammaticality judgments, which may be very similar to Nishimura’s or Kawahara’s. However, I will also consider a sociolinguistic study by Quackenbush (1989, cited in Quackenbush and

'so 1990) described later in §2.8. Occasionally, I use the Roman alphabet to indicate the transliteration of katakana spellings in order to demonstrate discrepancies between katakana spellings and actual pronunciations.

2.4.1 Geminate ‘b’ (BB)

The DD in the word webbu ‘web’ tends to be pronounced as a single D [,eb"], as geminate ‘b’ (BB) is extremely rare. In fact, it seems that all BB have a tendency to degeminate, and none seems to have the devoicing option.25 In contrast, this

“degeminination” process is not observed in other words. Researchers have been aware of the low number of BB in real words and the existence of the degemination process

(Kawahara 2005, 2006; Hirayama 2005; Rice 2006), but none have pointed out the lack of the devoicing option of BB.26 The examples in (7) illustrate the lack of the devoicing option of BB.

25 One exception I can think of is the brand name ヴェポラッブ [bepor%bb"~bepor%pp"] *[bepor%bb"~bepor%b"] ‘(Vicks) VapoRub’, which is of course a marginal item. 26 Nishimura (2006) and Kawahara (2005, 2006) list [!ebbe#"s"~!eppe#"s"] ‘Göbbels (proper name)’ as an example of BB having the devoicing option. But this is not an appropriate example since those who 53

(7) [,ebb" ~ ,eb"] *[,ebb" ~ ,epp"] ‘web’ (with a ‘bb’ gemination) [bobb" ~ bob"] *[bobb" ~ bopp"] ‘bob’ (with a ‘bb’ gemination) *[ba!!" ~ ba!"] [ba!!" ~ bakk"] ‘bag’ *[beddo ~ bedo] [beddo ~ betto] ‘bed’

Kawahara (2005) indicates that the low number of BB is Japanese-specific, since labials are the easiest to maintain voicing in (Hayes and Steriade 2004). Hirayama (2005) proposes that the rarity of BB in Japanese loanwords can be explained by structural markedness. Hirayama claims that the universal ranking of markedness constraints according to phonetic difficulty (*gg>>*dd>>*bb>>*g>>*d>>*b; Hayes & Steriade

2004) does not match the frequency pattern observed in Japanese loanwords. Rather,

Hirayama claims that complexity in feature geometry prevents BB from occurring. That is, in addition to the Doubly-linked position (i.e., being geminate), ‘dd’ has a Place specification, and ‘gg’ has a Place and a Peripheral specification, while ‘bb’ has a Place, a Peripheral, and a Labial specification, as illustrated in the diagrams in (8).

pronounce this word as [!eppe#"s"] may simply not know that the correct pronunciation is [!ebbe#"s"]. In fact I didn’t know that it is [!ebbe#"s"] until I read Nishimura (2006) and Kawahara (2005, 2006); I had believed that [!eppe#"s"] was the correct pronunciation. Before I found out that [!ebbe#"s"] is the correct pronunciation, [!eppe#"s"] was the only possible pronunciation, but now [!ebbe#"s"] is the only choice. Accordingly, this is not a phonological issue. 54

(8)

a. Labial b. Coronal c. Dorsal *X X X X X X !" !" !" Place Place Place | |

Peripheral Peripheral |

Labial (Source: Hirayama 2005:151)

According to the diagrams in (8), the labial geminate consonants BB are the most complex in structure. Hirayama argues that in order to have a BB in a loanword, the underlying form also must be geminated; thus, structural markedness plays a significant role. Hirayama’s account is consistent with the fact that many Japanese speakers produce words with emphatic BB, such as, [j%b%i]~[j%bb%i] ‘no good’ (slang), whose underlying representations are not geminated and whose surface forms are derived from an underlying singleton b; thus, they are not prohibited by the structural markedness proposed by Hirayama.

Hirayama’s explanation may account for the low number of ‘bb’ sequences, but it does not explain why ‘bb’ geminate consonants undergo ‘degemination’ instead of devoicing. This may be related to the fact that, due to historical processes, the devoiced counterpart of [b] is not [p], but [h] or [+] in current Japanese phonology. That is,

Japanese speakers may not be able to devoice [bb] because its phonological (or 55

orthographical) counterpart is either [hh] or [++], while its phonetic counterpart is [pp].

This issue needs more discussion.

2.4.2 DDj and DDz

Like the geminate voiced obstruent BB, DDj and DDz also behave differently from other geminate voiced obstruents. I claim that the DDj and DDz do not have voiceless counterparts. There are only two exceptions, as shown in (9). Japanese dictionaries also note that these two words have the devoicing option.

(9) a. [b%jj*i]~[b%cc&i] ‘badge’ バッジ～バッチ b. [dojj*ibo$#"]~[docc&ibo$#"] ‘dodgeball’ ドッジボール～ドッチボール

As the examples in (10) below show, words with the geminate voiced obstruents

DDz and DDj— other than the two exceptions shown in (9) above—never have the devoicing option or the degemination option.

(10) a. [kemb"#ijj*i] ‘Cambridge’ ケンブリッジ b. [b"#ijj*i] ‘bridge’ ブリッジ c. [bi#ejj*i] ‘village’ ビレッジ d. [j*%jj*i] ‘judge’ ジャッジ e. [dojj*i #%iN] ‘Dodge line’ ドッジライン 27 f. [!"ddz"] ‘goods’ グッズ g. [d%jj*i] ‘Dodge’ ダッジ (a brand name of automobiles: marginal) h. [!%#ejj*i se$#"] ‘Garage Sale’ ガレッジセール (Japanese comedians: marginal)

I argue that DDj and DDz words do not have the devoicing option for historical reasons (and possibly for orthographic reasons as well). Historically, [j*i], [*i], [dz"], and [z"] were the voiced counterparts of [c&i], [&i], [."], and [s"] respectively. By

1600, the Japanese language (except for some dialects) had lost the distinction between

[j*i] and [*i], and [dz"] and [z"] (Toyama 1972:198-202, cited in Vance 1987).

Consequently, when voicing is required, such as when Rendaku is applied, [c&i] and [&i] share the same voiced counterpart [j*i], while [."] and [s"] share the same voiced

28 counterpart [dz"]. This causes a problem when devoicing is required, since both [j*i] and [dz"] have two candidates—namely [c&i] and [&i], and [."] and [s"]. Phonetically,

[c&i] and [."] should be the only choices, but phonologically, this is not apparent.

Orthographically, [j*i] and [dz"] in loanwords are transcribed as ジ (the voicing

27 Nishimura (2006) claims that the [!"ddz"] ~ [!"tts"] alternation occurs. Some young Japanese speakers actually pronounce [!"tts"] and I do not deny the possibility that pronouncing [!"tts"] might come to be as popular as pronouncing [!"dds"] in the future. But at this moment, [!"tts"] sounds unacceptable to those who know the correct katakana spelling of the word.

28 [*i] and [z"] may occur intervocalically, but it does not affect the fact that [c&i] and [&i], and [."] and [s"] share the same voiced counterparts. 57

counterpart of [&i] シ) and ズ (the voicing counterpart of [s"] ス), respectively. This might have some effect on Japanese speakers’ devoicing choices because, orthographically, the devoicing counterpart of [j*i] (ジ) is [&i] (シ), but, phonetically, it should be [c&i] (チ). Notice the katakana transcriptions of [j*i] (ジ) and [c&i] (チ) in the two exceptions I listed: (9a) [b%jj*i]~[b%cc&i] (‘badge’ バッジ～バッチ) and (9b)

[dojj*ibo$#"]~[docc&ibo$#"] (‘dodgeball’ドッジボール～ドッチボール).

This apparent inconsistency between the orthographic system and phonetic facts might make Japanese speakers hesitate about devoicing [j*i] and [dz"]. As discussed in the previous section, this inconsistency is also observed in BB.

2.4.3 DVRVDDV-type words

Kawahara (2005, 2006) and Nishimura (2006) divide the words with DD into two classes: DVDDV (deddo-type) and TVDDV (teddo-type). They conclude that within each group the devoicing patterns are similar. I argue that more careful observation is required because there are some systematic exceptional groups in DVDDV—

DVRVDDV words. DVRVDDV words and DVDDV words are differentiated in the following ways. DVDDV type words include DD and additional voiced obstruents

(Kawahara 2005, 2006; Nishimura 2006). Thus, by definition, words with an additional syllable in between DV and DDV, such as [do#%!!"] ‘(either) drag/drug’, are considered

DVDDV as well. I differentiate these types of words from the ones without an additional 58 syllable and call them DVRVDDV temporarily.29 I further claim that DVRVDDV words should be classified as TVDDV words, not as DVDDV words.

Nishimura and Kawahara claim that what I call DVRVDDV words and DVDDV words behave the same. However, many DVRVDDV words seem to behave very differently from the regular DVDDV-type words. Table 2.1 lists all the DVRVDDV-type words that are found in the major Japanese dictionaries K"jien and Daijirin.

Table 2.1. Non-marginal DVRVDDV words IPA Transcriptions Gloss K D Japanese Transcriptions !!" [%ntedo#%!!"] antedrug ! アンテドラッグ [iho$ do#%!!"] illegal drug ! 違法ドラッグ [es$/&%#" do#%!!"] essential drug ! エッセンシャルドラッグ [o$+%n do#%!!"] orphan drug ! ! オーファンドラッグ [!"#o!!"] grog ! グロッグ [!o$ho$ do#%!!"] legal drug ! 合法ドラッグ [d%ppo$ do#%!!"] illicit drug ! 脱法ドラッグ [do!!"#e!!"] dogleg ! ! ドッグレッグ [do#%!!"] drag ! ! ドラッグ [do#%!!"] drug ! ! ドラッグ [do#%!!" %ndo drag and drop ! ! ドラッグアンドドロップ do#opp"] [do#%!!"s"to%] drugstore ! ! ドラッグストア [do#%!!" de#ib%#i$ drug delivery ! ! ドラッグデリバリーシス system &is"tem"] テム [do#%!!"b%nto] drag bunt ! ! ドラッグバント

29 The additional syllable can be anything but DV and cannot be limited to RV. But I could not find any example other than RV. In emphatic expressions, however, a Sino-Japanese item zenzen [dzenzen] ‘absolutely, totally, (not) at all’ can be pronounced as [dzenddzen], which is DVRVDDV by definition, but includes N instead of RV. 59

IPA Transcriptions Gloss K D Japanese Transcriptions [do#%!!"#e$s"] drag race ! ドラッグレース [+is"k%#" do#%!!"] fiscal drag ! フィスカルドラッグ [b"#%!!"] Bragg ! ! ブラッグ [b"#i!!"] brig ! ブリッグ [b"#i!!"z"] Briggs ! ブリッグズ [p"#o do#%!!"] prodrug ! ! プロドラッグ ))*i [o$k"&om b"#i))*i] auction bridge ! オークションブリッジ [okks"b"#i))*i] Oxbridge ! オックスブリッジ [k%b%#e))*i] coverage ! カバレッジ [!"#ini))*i] Greenwich ! ! グリニッジ [!"#ini))*i çjo$)*"n)*i] Greenwich ! グリニッジ標準時 Mean Time [!"#ini))*i bi#e))*i] Greenwich ! ! グリニッジビレッジ Village [!"#o$b%#" bi#e))*i] global village ! グローバルビレッジ [kemb"#i))*i] Cambridge ! ! ケンブリッジ [kemb"#i))*i !%k"h%] Cambridge ! ! ケンブリッジ学派 school [kemb"#i))*i d%i!%k"] Cambridge ! ! ケンブリッジ大学 University [kemb"#i))*i p"#%to" Cambridge ! ケンブリッジプラトン学 Platonist !%k"h%] 派 [konto#%k"to b"#i))*i] contract bridge ! コントラクトブリッジ [s"no$ b"#i))*i] snow bridge ! ! スノーブリッジ [seb"m b"#i))*i] (

IPA Transcriptions Gloss K D Japanese Transcriptions Bridges

[b"#i))*it%"-] Bridgetown ! ! ブリッジタウン [b"#i))*i b%"k"] bridge bank ! ブリッジバンク [b"#i))*im%-] Percy Williams ! ! ブリッジマン Bridgman [beb%#i))*i ho$kok"] the Beveridge ! ベバリッジ報告 report (see ビバリッジ報告) [bo$di"!" b"#i))*i] boarding bridge ! ! ボーディングブリッジ [m%ib"#i))*i] Eadweard ! マイブリッジ Muybridge [jokoh%m% bei Yokohama Bay ! 横浜ベイブリッジ Bridge b"#i))*i] [#eb%#e))*i] leverage ! レバレッジ [#eb%#e))*ido #i$s"] leveraged lease ! ! レバレッジドリース [)*e$ bi#e))*i] (

IPA Transcriptions Gloss K D Japanese Transcriptions [do#eddo#okk"s"] dreadlocks ! ドレッドロックス [h%ib"#iddo] hybrid ! ! ハイブリッド [h%ib"#iddo en)*i-] hybrid engine ! ハイブリッドエンジン [h%ib"#iddo k%$] hybrid car ! ! ハイブリッドカー [h%ib"#iddo hybrid computer ! ハイブリッドコンピュー komp(j"$t%$] ター [h%ib"#iddo çin&"] hybrid variety ! ハイブリッド品種 [h%ib"#iddo] hybrid rice ! ! ハイブリッド米 [b%#%ddo] ballad ! ! バラッド [b"#%ddo] blood ! ブラッド [b"#%ddobe#i$] Ray Bradbury ! ! ブラッドベリ [b"#%ddo#i$] Bradley ! ! ブラッドリー [b"#eddo] bread ! ! ブレッド [ho$#" hoi$to b"#eddo] whole-wheat ! ホールホイートブレ bread ッド [m%ik"#o !"#iddo] microgrid ! マイクログリッド Note: K=K"jien, D=Daijirin

Some words listed in Table 2.1 contain the sound sequences DDj and DDz, which do not have the devoicing option. Thus, I exclude them from the discussion. Let us now focus on one of the words on the list, [b"#eddo] ‘bread’, which does not have the devoicing option. Few native speakers would accept the devoicing counterpart [b"#etto].

The word [b"#eddo] is as common as [s%#%b"#eddo] ‘thoroughbred’. It is not clear why the optionality of [s%#%b"#eddo] ‘thoroughbred’ has been discussed in the previous studies and [b"#eddo] ‘bread’, which might be a crucial counterexample, has never been discussed. I argue that most of the words in Table 2.1 do not allow devoicing except for 62

the words that include [do#%!!"] ‘drug/drag’, [h%ib"#iddo] ‘hybrid’, and [s%#%b"#eddo]

‘thoroughbred’.

2.5 Exceptions

There are still some apparent exceptions to the tendencies discussed so far. The words shown in (11) suggest that devoicing patterns of DD are not as simple as previously thought. I will first describe them and then discuss what these exceptions indicate.

(11) a. gurokk( [!"#okki$] ‘groggy’ グロッキー b. zetto [(d)zetto] ‘z (the letter “z”30)’ ゼット c. ky!piddo [kj"$piddo] ~ [kj"$pitto] ‘cupid’ キューピッド~キューピット d. kata paddo [k%t% p%ddo] ~ [k%t% p%tto] ‘shoulder pad’ 肩パッド～肩パット e. debagg) [deb%!!%$] ‘debugger’ デバッガー

2.5.1 [!"#okki$] ‘groggy’

(11a) [!"#okki$] ‘groggy’ is a very special case. Since its alternative pronunciation

[!"#o!!i$] is rare, but does exist, it should be classified as a DVRVDDV type word in my analysis and a DVDDV type word in analyses by Nishimura (2006) and Kawahara

(2005, 2006). Nevertheless, its entry word in Japanese dictionaries is always transcribed with a geminate voiceless obstruent (i.e. グロッキー [!"#okki$]). The alternative

30 British pronunciation [z0d] 63

transcription グロッギー [!"#o!!i$], which is more faithful to the source word pronunciation in terms of voicing, is not found in dictionary entries. Instead, the body of the definition of the lexical item gurokk% explains that the pronunciation [!"#okki$] is a

“corruption” of [!"#o!!i$]. This fact is unusual and noteworthy because all the other words listed in dictionaries that could be TVDDV or DVDDV words are transcribed with a geminate voiced obstruent.

2.5.2 [(d)zetto] ‘z’

(11b) [(d)zetto] is another exceptional case that does not have the voicing option.

Dictionaries list only the devoiced version as a possible pronunciation without even mentioning that [(d)zeddo] (ゼッド) is more faithful to the source word. 31 Since it is the last letter in the Roman alphabet, “z” means ‘the ultimate’, just like the last letter of the

Greek alphabet (*) sometimes means ‘the ultimate’. Consequently, in Japanese popular culture and brand names ’z’ has been widely used.32 In spite of the frequent use of the letter ‘z’, its pronunciation has been consistently [(d)zetto].

2.5.3 ky#piddo ‘cupid’ and kata paddo ‘shoulder pad’

Although there are slight differences in devoicing patterns, both (11c) ky#piddo and (11d) kata paddo have the devoicing option as if they were DVDDV words. As for (11c) ky#piddo, [kj"$pitto] is the dominant pronunciation. (11c) ky#piddo ‘cupid’ and (11d)

31 The American English pronunciation [zi$] was not used as a source pronunciation, probably because the [zi] sequence is still a foreignism in Japanese and the alternative pronunciation [j*i$] is homophonous with another Roman alphabet letter, ‘g’. 32 Nissan’s Z-car (1969) is one example. 64 kata paddo ‘shoulder pad’ are exceptions to the regular devoicing patterns of TVDDV words that do not allow optional devoicing according to Nishimura (2006) and Kawahara

(2005, 2006). One may suspect that the [PVDDV] sound sequence, which is a more specific sound sequence than [TVDDV], might affect the devoicing pattern. However, it is not certain if the PVDDV structure affects the devoicing patterns or not.

2.5.4 [deb%!!%$] ‘debugger’

In spite of its [DVDDV] sequence, devoicing the DD in (11e) debagg& [deb%!!%$]

‘debugger’ seems to be impossible. There are two possible reasons. First, although debagg& is listed in dictionaries, and therefore is a non-marginal word, it is a relatively new word that is used almost exclusively by those who are involved in computer technology. However, Nishimura (2006) claims that debaggu ‘debug’ has the devoicing option. If the timing of borrowing and the DVDDV sound sequences are the only reasons for the variations in devoicing patterns, then the devoicing patterns of debagg& and debaggu should be the same. Second, the cause of the devoicing resistance might be specific to the sound sequence [!!%]. Both K"jien and Daijirin list another word with the sound sequence [!!%]: Haidegg& [h%ide!!%$] ‘(Martin) Heidegger’ does not seem to have the devoicing option either, but it has the degeminate option ([h%ide!%$]).

2.6 Changes in Voicing Patterns when DVDDV Words Are Compounded

DVDDV words may show different devoicing patterns when they are compounded with another word. All scholars agree that beddo ‘bed’ has the [beddo]~[betto] variants (Vance

1987; Ito and Mester 1999, 2008; Nishimura 2006; Kawahara 2005, 2006). However, 65 when the word beddo is compounded with another word, the possibility of devoicing seems to drop. For example, the devoicing probabilities of compound words, such as

[beddot%"-] (

33 odd. Similarly, while the [!"ddo]~[!"tto] alternation is acceptable, the

[!"ddob%i]~[!"ttob%i] alternation in guddo bai ‘goodbye’ is less likely to be accepted.

For example, the geminate [ddo] in the children’s song Guddo bai ‘Goodbye’ is

34 consistently voiced. In terms of pitch accent, [beddo] and [!"ddo] both have a low pitch on the last syllable. On the other hand, the last syllable of ‘[beddo]’ in [beddot%"-] or [beddo#"$m"] carries a high pitch. The last syllable of ‘[!"ddo]’ in [!"ddob%i] also carries a high pitch, although it is not the highest pitch. In the case of compound words, the accent on DDV may be also playing a role in maintaining the voicing of DD.

Compared with beddo and guddo, baggu ‘bag’ rarely becomes the first element of a compound word. The only such compound word I could find in dictionaries is

[b%!!"p%i"p"] ‘bagpipe’. The item [b%!!"p%i"p"] is frequently degeminated as

[b%!"p%i"p"], but it does not have the devoicing option.

33 This applies to other words, such as [beddome$ki1!"] ‘bedmaking’ and [beddo&i$-] ‘bedroom scene’. 34 Guddo bai グッドバイ‘Goodbye’ words by Sat2 Yoshimi, music by Kawamura Y+k+ (1934). 66

2.7 Devoicing Patterns in the Early 20th Century

Nishimura (2006) claims that devoicing of DD in TVDDV words is impossible. However, this has not been so since DD first emerged in the Japanese lexicon. As the examples in

(12) show, it seems that DDs in TVDDV words were once devoiced frequently.

(12) a. [kitto] ‘kid’ (Ichikawa 1930:182, cited in Lovins 1975:87) b. [hettor%ito] ‘headlight’ Ibid. c. [ope#ap%kk"] ‘opera-bag’ Ibid. d. [#etto] ‘red’ (Lovins 1975:87) e. [hetto] ‘head’ Ibid. f. [kj"$pitto] ‘Cupid’ Ibid. g. [!"#okki$] ‘groggy’ (Lovins 1975:88) h. [menc&i ekkis"] ‘mince eggs’ (Lovins 1975:88; Umegaki 1974:421)

The pronunciations of many of these examples have not been “established as the generally accepted versions of the loanwords” (Lovins 1975:130), and may sound unacceptable to younger Japanese speakers. Nonetheless, as we have seen, at least (12f)

[kj"$pitto] ‘Cupid’ and (12g) [!"#okki$] ‘groggy’ are still acceptable. Data from

Ichikawa and Lovins suggest that DD devoicing is an ongoing process.

2.8 Sociolinguistic Study: Quackenbush (1989)

Nishimura’s claim that DD sequences in DVDDV words have the devoicing option may be interpreted differently as showing that the patterns of devoicing of DD sequences vary among social groups. If so, then this is a sociolinguistic issue rather than a theoretical one.

Quackenbush (1989, cited in Quackenbush and 'so 1990) interviewed a total of 40 67

Tokyo residents, following the methods used by Labov (1972) and Trudgill (1974). The informants were: elementary school students (12 years old; 5 male, 5 female), college students (20~25 years old; 5 male, 5 female), salaried workers (35~45 years old; 5 male,

5 female) and retail store owners (35~45 years old; 5 male, 5 female). Quackenbush studied the pronunciations of informants in word-list style, reading passage style, and conversation style. She made use of the words in (13).

(13) a. [h%#i"ddo] ‘Hollywood’ TVDDV ハリウッド b. [m%ddo m%kk"s"] ‘Mad Max’ TVDDV マッドマックス c. [bik"to#i$ zeddo] ‘Victory-Z’ DVDDV ビクトリーゼッド d. [beddo] ‘bed’ DVDDV ベッド e. [beddot%"-] ‘bed town’ DVDDV ベッドタウン f. [b"#"do!!"] ‘bulldog’ DVDDV ブルドッグ g. [bos"tomb%!!"] ‘Boston bag’ DVDDV ボストンバッグ h. [&o#"d%$b%!!"] ‘shoulder bag’ DVDDV ショルダーバッグ i. [h%ndob%!!"] ‘hand bag’ DVDDV ハンドバッグ

According to the following charts, some speakers devoice DD in the TVDDV words (13a) [h%#i"ddo] and (13b) [m%ddo m%kk"s"]. The high devoicing rate of (13c)

[bik"to#i$ zeddo] conforms to my claim that DD in [zeddo] does not have the devoicing option. Other DVDDV words (13d)~(13i) show a higher devoicing rate than in the

TVDDV words. However, the devoicing rates differ word-by-word and speaker-by- speaker. 68

Overall [dd] (13a) [h%#i"ddo] ‘Hollywood’ TVDDV

(13b) [m%ddo m%kk"s"] ‘Mad Max’ TVDDV (13c) [bik"to#i$ zeddo] ‘Victory-Z’ DVDDV

(13d) [beddo] ‘bed’ DVDDV (13e) [beddot%"-] ‘bed town’ DVDDV

Figure 2.2. Degrees of devoicing in [dd] (Source: Quackenbush and 'so 1990:122) # word-list style A-B: elementary school students (12 years old) A: male, B: female $ reading passage style C-D: college students (20~25 years old) C: male, D: female % conversation style E-F: salaried workers (35~45 years old) E: male, F: female G-H: retail store owners (35~45 years old) G: male, H: female 69

Overall [!!]

(13f) [b"#"do!!"] ‘bulldog’ DVDDV (13g) [bos"tomb%!!"] ‘Boston bag’ DVDDV

(13h) [&o#"d%$b%!!"] ‘shoulder bag’ DVDDV (13i) [h%ndob%!!"] ‘hand bag’ DVDDV

Figure 2.3. Degrees of devoicing in [!!] (Source: Quackenbush and 'so 1990:124) # word-list style A-B: elementary school students (12 years old) A: male, B: female $ reading passage style C-D: college students (20~25 years old) C: male, D: female % conversation style E-F: salaried workers (35~45 years old) E: male, F: female G-H: retail store owners (35~45 years old) G: male, H: female 70

The overall results show that more elementary school students devoiced DD and fewer college students and salaried workers devoiced DD. As for style, the word-list style showed the least devoicing effect, while the conversation style showed the most devoicing effect. The devoicing may be related to the degree of the speakers’ knowledge of the source language, i.e. English. Compared with DVDDV words, TVDDV words are less likely to be devoiced in general. However, as shown below, the devoicing rate of DD in TVDDV is different from Nishimura’s study (Nishimura 2006). The rate of devoicing

DD in TVDDV words in Quackenbush’s study is much higher than in Nishimura’s.

Nishimura’s results are shown in Table 2.2.

Table 2.2. Ratio of voicing and devoicing of DD a. # of stems voiced devoiced sum teddo-type 743 39 782 (95.0%) (5.0%) (100%) deddo-type 379 310 689 (55.0%) (45.0%) (100%) sum 1122 349 1471 X2=323.9, P<.001 b. # of speech files voiced devoiced sum teddo-type 299 31 330 (90.6%) (9.4%) (100%) deddo-type 178 178 356 (50.0%) (50.0%) (100%) sum 477 209 686 X2=133.3, P<.001

(Source: Nishimura 2006:90)

Nishimura’s study is based on the data in the Corpus of Spontaneous Japanese

(CSJ), a large-scale speech database. The CSJ contains 3302 speech samples

(approximately 660 hours). Nishimura tallied the numbers of stems with a geminate voiced obstruent or its devoiced counterpart in the CSJ (Table 2.2.a). Nishimura also counted the number of speech files with such stems (Table 2.2.b). The data in the CSJ are based on recordings by researchers, university students and temporary workers sent by temp agencies. Consequently, the data in the CSJ do not include speech by elementary school students and retail store owners, who may devoice DD more frequently than other social groups according to Quackenbush’s study. Thus, it is natural that the rate of devoicing in TVDDV words in Nishimura’s study is much lower than the one in

Quackenbush’s study. The CSJ contains samples from more than 1,400 speakers whose ages range from twenty to eighty,35 whereas Quackenbush studied only 40 speakers from age twelve to forty five. The CSJ has more speakers with a wider age range, but does not include social groups with relatively little English education, i.e., elementary school students and retail store owners. Consequently, Nishimura’s study may reflect the devoicing patterns of certain social groups very well, and yet this pattern may not apply to other social groups.

Another difference between Nishimura’s study and Quackenbush’s study is that the latter examined the devoicing patterns of each word. The extremely high devoicing rate of (13c) [bik"to#i$ zeddo] in Quackenbush’s study supports the idea that the devoicing

35Kokuritsu Kokugo Kenky#jo (The National Institute for Japanese Language) http://www.kokken.go.jp/en/research_projects/kotonoha/csj/ (accessed August 6, 2009) 72 patterns are not solely determined by grammar. Even though the voicing is clearly indicated in the text used in the word-list style method, the voicing mark in the writing did not affect the devoicing of DD in most groups. This shows that many Japanese speakers are familiar with the pronunciation [zetto], and since the British pronunciation

[z0d] is not taught in school, many Japanese speakers do not know that it is actually pronounced as [z0d] in British English. The result may reflect unfamiliarity of the pronunciation [zeddo] in Japanese and [z0d] in British English. Quackenbush’s study also depicts subtle differences in the devoicing patterns, such as the one in (13d) [beddo] and

(13e) [beddot%"-]. Her study suggests that the devoicing patterns depend on the speakers’ social background as well as the words’ cultural and historical background.

Furthermore, such cultural and historical background is affected by socioeconomic or other social factors, such as social identity issues, of those who have access to it.

2.9 Does Lyman’s Law Really Affect DD?

Several researchers suggest that the optionality of DD in DVDDV words involves

Lyman’s Law (Nishimura 2006; Rice, 2006; Ito and Mester 2008). Lyman’s Law prohibits two voiced obstruents in a morpheme (McCawley 1968; Vance 1987; Ito and

Mester 1986, 1995b; Fukazawa and Kitahara 2001; Rice 2005, 2006{Rice, 2006 #192}).

Most Yamato items follow Lyman’s Law, whereas foreign items do not abide by

Lyman’s Law. Nishimura (2006) claims that since DD is stable when preceded by a mora with a voiceless obstruent, and it has the devoicing option when preceded by a mora with a voiced obstruent, DD follows a rule that is similar to Lyman’s Law. Rice (2006) also 73 makes references to Lyman’s Law, but it is not clear whether she thinks it is relevant to the optionality of DD devoicing. Ito and Mester (2006) state that Nishimura’s findings are “surprising” because OCP(+VOI), which is ranked lower than IDENT-FOREIGN, should be inactive for foreign items. Thus, Lyman’s Law should affect Yamato items only.

Nonetheless, Ito and Mester accept Nishimura’s analysis.

In this section, I will argue that Lyman’s Law is irrelevant to the instability of voicing in geminate obstruents. First, as discussed above, since many DVRVDDV words do not have the devoicing option, the conditions for devoicing DD are more specific than the one in Lyman’s Law. The function of Lyman’s Law is to prohibit two voiced obstruents in a morpheme. Thus, whether the moras with a voiced obstruent in a word are next to each other or not, such words are prohibited by Lyman’s Law. For example, a word with a DVDV structure is just as bad as a word with a DVRVDV structure. In contrast, according to the data provided in Table 2.1, the scope of the devoicing condition of DD is not a whole morpheme, but rather the sequence of phonemic segments DVDDV.

This fact suggests that what causes the devoicing tendencies of DD are the local phonetic conditions, rather than phonological rules that govern the whole stem. Second, Lyman’s

Law does not seem to be valid for recent borrowings in Japanese. If this claim is true, there is no reason to make use of Lyman’s Law in order to account for the instability of

DD. There are some pieces of evidence suggesting that Lyman’s Law is no longer valid for new loanwords in Japanese, aside from the traditional explanation that it is only valid for Yamato vocabulary items (Ito and Mester 1986, 1995b, 1999). Some recent loans not only allow two voiced obstruents within a morpheme, but also actually change the 74 voiceless obstruents in the source words into voiced obstruents. For example, the German loanword gipusu ギプス (< Gips) ‘plaster cast’ is usually pronounced as [!ib"s"] ギブ

ス, and K"jien even lists [!ib"s"] as an alternative pronunciation. A more striking example is that the loanword abokado アボカド ‘avocado’ is frequently pronounced as

36 [%bo!%do] アボガド, which contains three consecutive DV moras. Ohno (2005:59) provides another example ジャグジー[)*%!")*i$] ‘jacuzzi’. If Lyman’s Law is still valid for non-Yamato words, then how could these instances of consecutive voicing happen?

And if Lyman’s Law does not apply to FOREIGN items with no gemination, why should it apply only to words with gemination? The fact that some voiceless consonants become voiced despite the existence of structurally adjacent voiced obstruents also suggests that voicing in singleton voiced obstruents may not be as stable as traditionally thought (for example, see Rice 2006 and works cited). Ohno also claims that the discrimination between voiced and voiceless obstruents “is not as stable as normally believed” (Ohno

2005:59). Although the idea that the instability of DD can be attributed to Lyman’s Law would be appealing to many researchers, the unstable and complex distribution of DD cannot be explained by Lyman’s Law.

2.10 Origins of Geminate Voiced Obstruents

In this section, I will turn my attention from synchronic linguistics to diachronic linguistics, and investigate the origins of geminate voiced obstruents. I will also discuss

36 Nihon Kokugo Daijiten, the largest Japanese language dictionary, lists the alternative pronunciation アボガド [%bo!%do]. 75 how historical facts can account for the tendencies and irregularities of geminate voiced obstruents that we have seen so far. Before I discuss the origins of geminate voiced obstruents in particular, I will briefly review the origins of geminate consonants in general. The evolutionary pathways to geminates, as shown in (14), can be found in many languages (Blevins 2004).

(14) a. assimilation in consonant clusters b. assimilation between consonants and adjacent vowel/glides c. vowel syncope d. lengthening under stress (including expressive lengthening) e. boundary lengthening f. reinterpretation of a voicing contrast g. reanalysis of identical C+C sequences

Blevins (2004:170-171)

In particular, the following evolutionary pathways to geminates are well- documented in Japanese: (14a) assimilation in consonant clusters, (14c) vowel syncope and (14d) lengthening under stress (including expressive lengthening, such as consonant lengthening in emphatic expressions). Geminate obstruents can also be divided into those resulting from combining two morphemes and those which are purely emphatic.

2.10.1 Syncope and Assimilation (Two Morphemes)

Many Japanese geminate voiceless obstruents emerged due to consonant lengthening resulting from vowel syncope and consonant assimilation. For example, in the Heian- 76

Kamakura period (794-1333), changes by vowel syncope between identical consonants when two morphemes are combined, such as motite (moti+te) > motte ‘have’, and tatite

(tati +te) > tatte ‘stand’, took place (Mabuchi 1971:90).37 In contrast, no geminate voiced obstruent was created in this manner: geminate voiced obstruents (DD) in loanwords do not require a combination of two morphemes. The lack of DD in native items in Japanese is frequently accounted for by using the contrast between the process of gemination of voiceless obstruents and the process of post-nasal voicing (Ito and Mester 1999;

Kawahara 2005; Hirayama 2005; Rice 2006).38 In order to determine the origin of DD, we need to examine expressive lengthening of voiced consonants in emphatic expressions.39

2.10.2 Emphatic Expressions (Within a Morpheme)

Consonant lengthening in emphatic expressions is another source of geminate obstruents in Japanese. Unlike syncope, geminate obstruents in emphatic expressions do not require more than one morpheme to emerge. Geminate obstruents in emphatic expressions are created by lengthening an obstruent for expressive and emphatic purposes (Blevins

2004:174). According to Blevins, consonant lengthening in emphatic expressions occurs in many unrelated languages (Blevins 2004:174), such as Rotuman (Churchward 1940),

37 In the Heian-Kamakura period, the /ti/ sequence was probably pronounced as [ti], not as [c&i] (Mabuchi 1971:120-121). 38 For example, kat ‘win’ kat-ta ‘win, past’, but yob ‘call’ yoN-da ‘call, past’ (Hirayama 2005). Rice (2006) points out that nasal substitution, such as yoN-da, is expected only in morphologically derived environments in Japanese. 39 DD in non-foreign items due to combinations of two morphemes can be observed in some dialects (e.g., [kok"go] > [ko!!o] ‘national language’, [te."do$] > [teddo$] ‘railroad’: Takayama 2003). 77

West Greenlandic Eskimo (Rasmussen 1979; Woodbury 1987), and Bengali and Marathi

(Masica 1989).

Like the languages discussed above, Japanese consonants are frequently lengthened in emphatic expressions, creating geminate obstruents. Geminate obstruents in emphatic expressions may appear trivial and insignificant because their impact on grammar seems only transitory. However, contrary to this impression, geminate obstruents—in particular geminate voiceless obstruents—in emphatic expressions often influence Japanese phonology. Hamada (1955, cited in Vance 1987) lists several words with emphatic geminate voiceless obstruents that are derived from their non-geminate counterparts. Some of them even took the place of their original non-geminate versions.

For example, [s%kki] ‘a little while ago’ is derived from [s%ki] ‘before’. [m%tt%k"]

‘completely’ has replaced the original non-emphatic form [m%t%k"]. Finally, the emphatic form [t%tt%] ‘only’ and its original form [t%d%] coexist sharing the same basic meaning, but the geminate obstruent in the emphatic form is devoiced. There are many more examples of non-marginal words with geminate voiceless obstruents whose origins are emphatic expressions.

As mentioned above, Japanese also has geminate voiced obstruents in emphatic expressions, such as [s"!!oi] (<[s"!oi] ‘amazing’), [k"dd%#%n%i] (<[k"d%#%n%i]

‘absurd’) (Vance 1987) and [j%bb%i] (<[j%b%i] ‘no good’) (Hirayama 2005). Sometimes emphatic expressions create DVDDV form, such as [b"ddz%m%] or [b"ddz%m%$]

(<[b"(d)z%m%] ‘awkward’). Notice that the sequence [ddz%] is not found in foreign items. 78

More DVDDVs are found in literary works, such as novels and poetry. For example, Miyazawa Kenji’s Kaze no Matasabur" (Matasabur", a Wind Boy) starts with onomatopoetic expressions [doddodo dodo$do dodo$do dodo$] (Miyazawa 1995:5), which contain DVDDV. 40 The poet Kusano Shimpei (1903-1988) created pieces of poetry that consist entirely of onomatopoeia. One of his lines includes an onomatopoeic expression with DVDDV form: [!(j%!(!(j%!(!(j%!(!(j%!(!(j%3] (Kusano 1978:31, cited in Yamaguchi

2002:11). DVDDV forms can also be observed in sound effects used in manga and anime.

In manga, sound effects with DVDDV forms are frequently used to imitate intense sounds, such as explosions and gunfire. These sound effects are now used in everyday conversations as well. For example, Yamaguchi states that the sound sequence

[d%dd%dd%dd%3] is used to describe the sound and appearance of extreme dashing when one is almost late for work (Yamaguchi 2002:79).

So far we have seen that geminate consonants originated from syncope, assimilation, and emphatic expressions. Syncope and assimilation occur when two morphemes are combined, and they result in geminate voiceless obstruents, but not geminate voiced obstruents. In contrast, emphatic expressions do not involve morpheme combination and create both TT and DD. Some words with emphatic expression-origin

TT have become non-marginal, but no DD words except for [z%bb"] have become non- marginal. All types of geminate obstruents can be observed in special literary works, such

40 Miyazawa Kenji 宮沢賢治 (1896~1933) was one of the most popular writers of children’s literature in the early 20th century. His Kaze no Matasabur" 風の又三郎 (Matasabur", a Wind Boy) has been popular ever since it was published in 1934. 79 as poetry and manga. The following table illustrates the origins of geminate consonants in Japanese. Table 2.3 shows that DDs in non-foreign items appear only in words that are used under limited circumstances, such as emphatic expressions and sound effects. Table

2.3 also indicates that pronouncing DD requires more airflow than pronouncing TT, in order to maintain voicing (Hayes and Steriade 2004; Vance 1987; Ohala 1983).

Table 2.3. General distribution of geminate obstruents in non-foreign items TT TVDDV DVDDV

Syncope origin Yes No No

Emphatic Non-Mimetic Yes No origin/Non- (e.g. [m%tt%k"] < No marginal [m t k ]) % % " Mimetic Yes (but rare: Yes No [z%bb"] only)

Emphatic Expressions/ Yes Yes Marginal Yes [s"!!oi] [b"ddz%m%] [j%bb%i]

Other/ Sound effects Marginal in manga, anime (e.g. sounds of Yes Yes Yes gunfire), and poetry

Expected amount of airflow smaller larger

Vance (1987) argues that DDs are “ordinarily confined to emphatic expressions,” and that both pronouncing DDs and pronouncing emphatic forms require more energy than pronouncing unemphatic words in order “to keep air flowing from the lungs and maintain vocal cord vibration while air pressure is building up behind a supraglottal closure” (Vance 1987:43). Therefore, “speakers may feel that they are somehow out of place in unemphatic pronunciations of recent loanwords” that contain a DD (Vance

1987:43). This might explain why [t%dd%], the supposed emphatic form of [t%d%] ‘only’, could not come to be unemphatic and non-marginal: instead, its devoiced version [t%tt%] has become an unemphatic word. Although now DDs in TVDDV forms are believed to be stable and do not have the devoicing option, they might have been as unstable as DDs in DVDDV words in the past. This scenario corresponds with the devoiced versions of

TVDDV words from Lovins (1975), which are not currently used. Accordingly, TVDDV words with the devoicing option, such as [kj"$pitto]~[kj"$piddo] ‘cupid’ can be considered to be living fossils from when even TVDDV words were “out of place”

(Vance 1987:43).

Although there are no explicit empirical studies on the difference between

TVDDV words and DVDDV words in terms of required energy to maintain sufficient airflow, it is reasonable to assume that pronouncing DVDDV words requires more energy for voicing DD than pronouncing TVDDV words. Moreover, emphatic DVDDV words are rare compared with emphatic TVDDV words. Consequently, speakers may hesitate to apply larger pulmonary energy to DVDDV words, as Vance (1987) claims. If this happens, such DVDDV words do not have the voicing option. Exceptional words, such as 81

[!"#ok(k(i$] ‘groggy’ and [(d)zetto] ‘z’, reflect this hesitation. Since Japanese speakers are fully capable of pronouncing DVDDV words, they partially overcame such hesitation later. At this stage, DVDDV words were established phonemically, but were still unstable phonetically. Since there are DVDDV words without the devoicing option, such as

[deb%!!%$] ‘debugger’, it seems likely that hesitation in voicing DDs in DVDDV words will be completely cast aside in the near future.

2.11 Discussion

The problem of geminate voiced obstruents has previously been treated as synchronic phonology, with particular emphasis on Lyman’s Law (Nishimura 2006; Kawahara 2005,

2006; Rice 2006). I argue that this issue is a diachronic one and cannot be accounted for by a synchronic grammar. Nishimura (2006) and Kawahara (2005, 2006) raise the question of why the geminate voiced obstruents in DVDDV words have the devoicing option while TVDDV words are stable. Within the framework of OT, ‘having the devoicing option’ means that it is a question of “free variation” or “optionality” (Kager

1999). Both Nishimura and Kawahara use unranked constraints to account for this optionality. The first problem for this analysis is that, as discussed above, the devoicing of DD has nothing to do with Lyman’s Law. Consequently, constraints that prohibit “the coexistence of a voiced obstruent and DD within a stem,” such as [VOP2]stem

(Nishimura 2006) and OCP(+voi) (Kawahara 2006), must be modified. The second problem is that the optionality of DD devoicing has too many exceptions to be treated naturally within the framework of OT. Moreover, both Nishimura and Kawahara do not 82 deal with the existence of exceptions, which gives the impression to non-native readers that the optionality of devoicing of DD is without exception. Descriptions of exceptions and reasons for exclusion of the exceptions from the data are essential for any linguistic study. For example, the optionality of Finnish genitives described in the study by Anttila

(1995:3-4) is without random exception. When there is a “systematic exception”, it is noted and accounted for (Anttila 1995:5). Too many exceptions to the optionality of devoicing of DD indicate that synchronic grammatical analyses are inappropriate.

The main reasons I consider this issue a diachronic one are the following: (1) there are not all that many lexical items with DD, yet there are too many ‘word by word’ cases, i.e. exceptions; (3) the devoicing tendencies are changing over time; and (3) extragrammatical factors, such as orthographic effects and the speakers’ knowledge of the source words, are playing significant roles. Since DD is an innovative sound sequence and its [+voice] feature is difficult to perceive, according to the experiments by Kawahara

(2005, 2006), it is reasonable to assume that it was equally difficult to pronounce and perceive both TVDDV and DVDDV for Japanese speakers at the time they started using

DD. On one hand, exceptional words such as [!"#ok(k(i$] ‘groggy’, [(d)zetto] ‘z (the letter “z”)’, and [kj"$piddo] ~ [kj"$pitto] ‘cupid’, along with the data listed by Lovins

(1975), suggest higher instability of voicing of DD in the past. On the other hand, the feature [+voice] in DD in more recent DVDDV loanwords, such as [deb%!!%$]

‘debugger’, shows more stability. Therefore, ‘having the option to devoice’ in DVDDV words today is merely one characteristic of instability in diachronic processes that can be easily trumped by extragrammatical factors. For example, as we have seen above, except 83

for [b%jj*i]~[b%cc&i] ‘badge’ and [dojj*ibo$#"]~[docc&ibo$#"] ‘dodge ball’, DVDDV words with DDj or DDz do not have the devoicing option. Similarly, DVDDV words with BB do not devoice. The lack of the optionality of devoicing in DDj, DDz, and BB is probably due to the inconsistency between the orthographic system and the phonetic facts.

Accordingly, this fact is an example where the orthographic system outweighs the apparent optionality. The orthographic effect on the optionality is closely related to the effect of knowledge of the lexical items or how words are learned. Not all lexical items are learned auditorily: many lexical items, particularly new words and loanwords, are learned by reading. It is important to consider the difference between auditory learning and visual learning. If new words with DD are learned exclusively by listening, the learners may not perceive the feature [+voice] clearly, as demonstrated in Kawahara’s perceptual experiments in the ‘cocktail party settings’ (2005, 2006). Consequently, there is a good chance that such words will be learned without the feature [+voice]. However, if the learners learn these words by reading, they know that the feature [+voice] exists even before they hear the words for the first time. Recall that kana, one of the subsystems in Japanese writing, indicates voiced obstruents with a diacritic. Once learners know that the feature [+voice] is implied by orthography, if they hear these words without the feature [+voice] they should find the devoiced DD ‘ungrammatical’ because, according to Kawahara (2005, 2006), perception of TT is stable. This is even more evident for the words with DDj and DDz, whose phonetic voiceless counterparts and orthographic voiceless counterparts show discrepancies. 84

One may still wonder why there are apparent differences in devoicing patterns between DVDDV words and TVDDV words. Perceptually, as reported by Kawahara

(2005, 2006), voicing in TVDDV words is difficult to hear. Voicing in DD in DVDDV words may be even more difficult to hear than in TVDDV words. Unfortunately, since

Kawahara did not study the perception of DVDDV words, it is still uncertain whether the perception of DD in DVDDV words is more difficult than the perception of DD in

TVDDV words. Articulatorily, one may claim that the difference in devoicing patterns reflects the difference in the amount of aerodynamic power required: DVDDV words require more airflow to maintain voicing than TVDDV words. But again, this speculation is scientifically unfounded and requires empirical study.

As we have seen, the issues of geminate voiced obstruents cannot be handled without considering all of the aspects described above: acoustic and aerodynamic properties, orthographic effects, learners’ knowledge, and diachronic pathways. These aspects influence each other, creating various exceptions. But I presume that eventually

DD will be established firmly in Japanese, as described by Ito and Mester (1999), for the following reasons. First, according to Kawahara’s acoustic analyses of DD, Japanese speakers are capable of pronouncing DD and TT with significant differences in the acoustic cues (Kawahara 2005, 2006). Second, the Japanese writing system has a method to indicate DD clearly. Finally, although it is true that pronouncing DD is aerodynamically challenging, it is not impossible at all. When DD is established firmly in

Japanese, then presumably some words will always be pronounced with DD and some words will always be pronounced without. This is exactly like the situation described by 85

Ito and Mester (1999), in which the difference between ASSIMILATED FOREIGN and

41 UNASSIMILATED FOREIGN could be clearly determined by the constraint *DD. However,

DD in Japanese has not established such status: it is still unstable. Therefore, it is premature to give a synchronic analysis of DD in Japanese. The instability of DD that we can observe today is just a point along the path to a completely stable DD. And when DD comes to be stable, then it will be difficult to imagine the unstable situation in the past without referring to thorough descriptions of DD in the 20th century. That is precisely the reason why accurate descriptions and analyses today are important in understanding language change.

41 The legitimacy of strata, such as ASSIMILATED FOREIGN and UNASSIMILATED FOREIGN, will be discussed in Chapter 6.

CHAPTER 3

ISSUES OF JAPANESE HA-GY! AND INNOVATIVE FA-GY!

3.1 Introduction

The traditional account of the distribution of Japanese ha-gy" sound sequences may be the most confusing for non-native speakers because the ha-gy" consonants have three different places of articulation: glottal, palatal, and labial. The confusion is mostly due to historical reasons. Table 3.1 illustrates the distribution of current standard Japanese ha- gy".

Table 3.1. Distribution of current standard Japanese ha-gy" /ha/ /hi/ /hu/ /he/ /ho/ ha-gy" [h ] [çi] [ ] [he] [ho] sequences % +" katakana ハヒフヘホ

Table 3.2 shows the distribution of the innovative fa-gy" sound sequences:

Table 3.2. Distribution of sound sequences in fa-gy" /fa/ /fi/ - /fe/ /fo/ fa-gy" [ ] [ i] ([ ]) [ e] [ o] sequences +% + +" + + katakana ファフィ (フ) フェフォ

The IPA symbols [+] and [ç] refer to a voiceless bilabial fricative and a voiceless dorso-palatal fricative respectively. The traditional ha-gy" and innovative fa-gy" raise several issues. I argue that the initial consonant of ha-gy" /hu/ is very close to [h], although many Japanese linguists claim that it is a voiceless bilabial fricative [+]

(Akamatsu 1997; Sait+ Yumiko 1990; Inozuka, H. and Inozuka, E. 2003; Kashima 2002;

Kase 2001; Sait+ Yoshio 2003). Nonetheless, I admit that the consonants of fa-gy" sequences are [+]. In this respect, the phone [+], which is used in fa-gy" sequences, is an innovative sound. I conclude that the fa-gy" sound sequences are not created by replacing the vowels of existing sound sequences, which can be observed in other innovative sound sequences, such as [ti], [t"], and [&e]. The question is why sound sequences with [+], which was not present in Japanese at the time of emergence, came to be prevalent. I will examine historical evidence and claim that [+] existed in Japanese in the past, and the lack of [+] in non-foreign Japanese items is a lexical problem, not a grammatical one. In addition to the issues of [+], I argue that ha-gy" /hi/ is closer to [h(i] than [çi]. The consonant [h(] is a palatalized [h], which has the tongue shape for [i] but lacks friction in the palatal region. The consonant [ç] has the similar tongue shape with [h(]; however, friction in the palatal region is present in [ç].

3.2 Articulation of Ha-gy" Consonants

Before discussing the issues of innovative fa-gy", I will propose that current standard ha- gy" /hu/, which has been believed to be pronounced as [+"], is actually articulated in a manner that is very close to [h"]. I will also argue that the actual pronunciation of ha-gy" 88

/hi/ is not [çi], but [h(i]. These claims, in particular the former claim, may sound outrageous to many Japanese linguists, because pronouncing /hu/ as [+"] is one of the most well-known “facts” about Japanese even among non-Japanese linguists (e.g., see

Blevins 2004). Nevertheless, I would like to raise this controversial issue in order to untangle the complicated problems revolving around ha-gy" and fa-gy".

I argue that the place and manner of articulation of the consonant of ha-gy" /hu/ is not the same as those of the consonants of innovative fa-gy" sound sequences, at least word initially. The consonant of the ha-gy" /hu/ is commonly described as a voiceless bilabial fricative [+] (Akamatsu 1997; Sait+ Yumiko 1990; Inozuka, H. and Inozuka, E.

2003; Kashima 2002; Kase 2001; Sait+ Yoshio 2002). For example, Akamatsu states that it is essential that “both lips are involved in the articulation of [+] ” (Akamatsu 1997:88).

Many books on Japanese linguistics and textbooks for Japanese language teachers state that the sound of ha-gy" /hu/ is similar to the sound that is produced when blowing out a candle flame (Sait+ Yumiko 1990; Inozuka, H. and Inozuka, E. 2003; Kashima 2002;

Kase 2001), implying that the pronunciation of the consonant of ha-gy" /hu/ is a voiceless bilabial fricative. On the other hand, some other researchers stress the lack of lip activity in pronunciation of the consonant of the ha-gy" /hu/. Using the Roman alphabet, Hashimoto (1950:48) describes that current ha-gy" sound sequences are pronounced as “ha hi hu he ho”; however, he transcribes the ha-gy" sound sequences before the Edo period (1603~1867) as “fa fi fu fe fo.” Hashimoto clearly notes that the pronunciation of “f” is a bilabial sound that is close to the pronunciation of the consonant of current ha-gy" /hu/. By using different letters (“f” for ha-gy" consonants before the 89

Edo period and “h” for current ha-gy" consonants) and adding “close” to the description of the pronunciation of “f” before the Edo period, I interpret Hashimoto as implying that the pronunciations of the consonants of ha-gy" before the Edo period and the consonant of current ha-gy" /hu/ are similar, but still different. Hashimoto (1950:95) notes, by using katakana, that ha-gy" sound sequences were pronounced as ファフィフゥフェフォ until the end of 16th century. Notice that katakana フゥ is used instead of フ to describe the pronunciation of ha-gy" /hu/ at that time.42 The extra ゥ added after フ suggests that

Hashimoto supposes that the pronunciation of ha-gy" /hu/ at that time was somewhat different from the current pronunciation. Hashimoto also indicates the lack of lip activity in pronunciation of the consonants of current ha-gy" sound sequences. Judging from these descriptions, I conclude that Hashimoto considers the pronunciation of current ha- gy" /hu/ different from [+]. Vance (1987) states that /f/ in the Japanese word /fune/ is a bilabial or occasionally a labiodental fricative with very weak frication (Vance 1987:19-

21). The results of experimental studies by Löfqvist (2005) also show that lip movement in the Japanese /f/ is minimal. Sait+ Yoshio (2003) states that [+] appears in careful speech and [h] otherwise. For example, the careful pronunciation of /hune/ ‘boat’ is

[+"ne], but in casual speech, [h"ne] might appear. Sait+ also claims that [+] always appears when followed by a voiceless vowel ["4] (e.g. [+5k"], [+"4k"] /huku/ ‘clothes’;

Sait+ Yoshio 2003:9).43 Uehara and Kiyose (1974) describe the pronunciation of the

42 Unfortunately, Hashimoto’s katakana notation for ha-gy" sound sequences with [+] is not consistent. 43 I have strong doubts about Sait+’s observation. As discussed in a later section, in my observation, [+] appears in /ihuku/ ‘clothes’, but not in /huku/. 90

Japanese /f/ as similar to the consonant in the English who with a spreading of the lips.44

Maddieson (2005) agrees with the claim by Uehara and Kiyose, and maintains that the

Japanese /f/ is not a fricative but an approximant.45

Although [+] may occur under certain circumstances, I claim that the default pronunciation of ha-gy" /hu/ is not a voiceless bilabial fricative as traditionally thought.

Pronouncing [+] requires more airflow and lip activity than pronouncing the consonant of ha-gy" /hu/. Pronouncing ha-gy" /hu/ produces little airflow. Therefore, the description that the sound of ha-gy" /hu/ is similar to the sound that is produced when blowing out a candle flame is clearly wrong. Pronouncing /hune/ ‘boat’ or /hu$hu/ ‘couple’ in front of a candle flame does not move the flame much. In contrast, pronouncing /fa fi fe fo/ will move the flame. Unlike /hu/, pronouncing the innovative sound sequences /fa fi fe fo/ always requires lip activity and more airflow; thus, the /f/ in /fa fi fe fo/ is a voiceless bilabial fricative [+], possibly without lip rounding. It is beyond the scope of this dissertation to describe the phonetic quality of Japanese /f/ experimentally, but I suggest that the acoustic, aerodynamic, and articulatory characteristics of ha-gy" /hu/ and fa-gy" sequences should be studied empirically in the future. Acoustic signals and the amount of airflow could be easily measured. Articulatory characteristics might be measured by the methods used by Löfqvist (2005).

44 The spreading of the lips observed in ha-gy" /hu/ is probably due to the shape of the vocal tract for the vowel ["], which also has lip spreading 45 Maddieson does not consider [h] a fricative, in spite of the fact that the International Phonetic Association classifies [h] as a “voiceless glottal fricative.” This issue will be discussed further in §3.3. 91

The claim that the place and manner of articulation of the consonant of ha-gy"

/hu/ is different from that of fa-gy" sequences can also be confirmed by the following informal method. Find any modern standard Japanese speakers and ask them to pronounce fa-gy" sequences and ha-gy" /hu/ while holding their lips apart with their fingers. They should be able to correctly pronounce nothing but ha-gy" /hu/. This is a piece of evidence that pronouncing ha-gy" /hu/ does not require bilabial frication.

Accepting the description by Uehara and Kiyose (1974), I will propose that [h"] satisfactorily depicts the phonetic characteristics of ha-gy" /hu/.

Let us turn to the articulation of ha-gy" /hi/. The pronunciation of the consonant of ha-gy" /hi/ has been believed to be a voiceless dorso-palatal fricative [ç] by many

Japanese linguists (Mabuchi 1971; Inozuka and Inozuka 2003; Kashima 2002; Kase

2001; Sait+ Yoshio 2003). Sait+ Yoshio (2003) and Vance (1987) both mention that the consonant [ç] exists in German and English. Sait+ maintains that the place of articulation of [ç] in German is sometimes more advanced than the one in Japanese, and to Japanese ears, German [ç] sounds like a phone that is somewhat in between Japanese [ç] and a voiceless alveolo-palatal fricative [&]. Both Sait+ Yoshio (2003) and Vance (1987) indicate that [ç] occurs in the English word huge. According to Sakuma (1929:138-139, cited in Vance 1987), instead of [ç], a “palatalized h” can appear, but Kawakami

(1977:49, cited in Vance 1987) states that it is restricted to [ç] when /h/ is followed by a voiceless high front vowel. Sait+ Yoshio (2003:12) on the whole agrees with Sakuma and

Kawakami. Sait+ reports that a “palatalized h” [h(] appears in casual speech, possibly 92 sometimes accompanied by a weak [ç]. He claims that the mandatory appearance of [ç] followed by a voiceless high front vowel is due to the weak frication of [h].

The descriptions of [ç] followed by a voiceless high front vowel by Kawakami

(1977) and Sait+ (2003) are credible. The issue here is whether /h/ followed by [i] is [ç] or [h(]. As I described in 3.1, the major difference between [h(] and [ç] is the presence of friction in the palatal region in [ç]. In my observation, many standard Japanese speakers pronounce [h(i] instead of [çi]. The sound sequence [çi] may occur depending on the speaker or context, but overall, particularly in casual speech, [h(i] occurs more frequently than [çi]. Except for the condition when /h/ is followed by a voiceless high vowel, I do not find any significant difference between English /hi/ and current standard Japanese /hi/ in terms of the conditions of the appearance of [ç]. Then, why has the consonant of ha- gy" /hi/ been transcribed as [çi] by Japanese phonologists? I suspect that ha-gy" /hi/ once was actually pronounced as [çi], but has changed to [h(i].

It is well-documented that in Shitamachi (a traditional working class area) in

46 Tokyo, people often pronounced ha-gy" /hi/ as [&i] , and some people still do pronounce it this way (Sat+ 2002:340). According to Sat+ (2002), Shitamachi language is a descendant of a traditional dialect (Edo-go: Edo language) spoken in the Tokyo area.

Yamanote language, on the other hand, is spoken by relatively wealthier people in an uptown area in Tokyo. Yamanote language was elaborated by people who moved in from outside Tokyo in the late 19th century and early 20th century. Consequently, in the

46 [&] is a laminal voiceless alveolo-palatal fricative. 93

Tokyo area, Yamanote language contrasted with Shitamachi language. Yamanote language eventually came to be the basis of standard Japanese. The contrast between

Yamanote language and Shitamachi language has been reduced considerably since

WWII; still, some speakers in Shitamachi pronounce ha-gy" /hi/ as [&i] (e.g., [&ib%c&i]

‘hibachi’). The places of articulation of a voiceless alveolo-palatal fricative [&] and a voiceless dorso-palatal fricative [ç] are close, and the manners of articulation are the same. The sound sequence [&i] as ha-gy" /hi/, which appears in Shitamachi language, suggests that there is a possibility that ha-gy" /hi/ may have been [çi] before it came to be

[h(i]. I suspect that ha-gy" /hi/ changed in the following way: [+i] > [çi] > [h(i].

In the next section I will discuss the classification of the place and manner of articulation of [h] by the International Phonetic Association (IPA). At least some linguists consider the classification of [h] by the IPA dubious, and I believe the inaccurate description of ha-gy" /hu/ is partially due to this dubious classification.

3.3 The Phonetic Quality of [h]

Supposedly, the place of articulation of [h] and [+] are different: the former is glottal, while the latter is bilabial. The consonant [h] has been classified as a ‘voiceless glottal fricative’ by the International Phonetic Association (IPA). However, classifying the place of articulation of [h] as glottal, as well as classifying the manner of articulation of [h] as fricative, may not be accurate. Ladefoged (2001:254) questions the classification of [h] by the IPA because of the lack of notable glottal friction when [h] is pronounced.

According to Ladefoged (2001), the place of articulation of [h] depends on its adjacent 94 sound, and the manner of articulation of [h] may not be clearly specified because [h] is usually a voiceless transition into or out of a syllable. Therefore, it is inappropriate to define the place and manner of articulation of [h] as “glottal fricative”. Consequently,

Ladefoged (2001) suggests that instead of classifying [h] as “glottal fricative”, it should be removed from the chart of pulmonic consonants and listed under “Other Symbols.”

Ladefoged and Maddieson (1996) propose that [h] in English words is a segment “that has only a laryngeal specification,” and is “unmarked for all other features” (1996:326).

As discussed in §3.2, Maddieson (2005) regards [h] an approximant.47

I have argued that current phonetic descriptions of some ha-gy" consonants are not satisfactory. It is important for linguists to describe sounds of Japanese as accurately as possible. Actually, the influence of the accuracy of phonetic descriptions is not limited to linguistics. It may also affect Japanese language teaching and learning. Phonological and phonetic descriptions by Japanese linguists are referred to by teachers and learners of the Japanese language. Inaccurate descriptions may confuse the learners. As discussed earlier, the consonant of ha-gy" /hu/ is the same as [h]: its place of articulation depends on its adjacent sound. Most Japanese can pronounce /hu/ without changing the shape of the vocal tract for /u/ ["]. If the consonant of ha-gy" /hu/ is [+], there must be some kind of obstruction of the air at both lips. To cause the obstruction at the lips, it is necessary to make some movement. But I do not notice such movement when I observe ha-gy" /hu/.

In this dissertation, as Uehara and Kiyose (1974) claim, I maintain that current standard

47 By contrast, Johnson (2003) states that “the sound source in [h] is aperiodic turbulent noise produced at the glottis” (Johnson 2003:121). 95 ha-gy" /hu/ is very similar to English who [hu].48 The issue of place and manner of ha- gy" /hu/ is important when considering historical development of Japanese ha-gy" and fa- gy".

3.4 Historical Background

In this section, I will review the history of Japanese ha-gy". I will discuss only word- initial ha-gy" sound sequences in this chapter.49 It is known that consonants of word- initial ha-gy" sequences changed from [p] to [+] and then to [h] except for [çi] and [+"].

Many scholars account for this phenomenon as labial lenition, or weakening (Hashimoto

1950; Inozuka and Inozuka 2003).50 In historical phonology, changes from stops to fricatives are called lenition and can be observed in many languages, such as the change from Latin pater to English father (Campbell 1999). The exact time when ha-gy" [p] was replaced by [+] or when ha-gy" [+] was replaced by [h] is still unknown. Nevertheless, dialectal evidence and historical documents sufficiently prove the changes of the consonants of ha -gy". There are no definite written records that directly prove that the ha-gy" consonant was [p] in Pre-Old Japanese. Ueda (1898) claims that [p] was a ha-gy" consonant in Pre-Old Japanese because: (1) in current standard Japanese, voiceless counterparts of voiced bilabial stops are ha-gy";51 (2) [h] did not exist in old Japanese;

48 Needless to say, it is important to empirically evaluate this claim in future studies. 49 The intervocalic ha-gy" consonants in pre-Old Japanese, which might have been pronounced as [b] or [,], changed into [w] (Takayama 2003; Hamano 1998). The sound [p] remained word internally only after /N/ in Sino-Japanese (some cases) or after /Q/. 50 But see Miyake (2003) for a counterargument. 51 Hashimoto (1950) suggests that this is not direct evidence that proves that ha-gy" consonant was [p], since ba-gy" consonants at that time could have been a voiced bilabial fricative. 96

(3) some Ainu words, which are borrowed from Japanese with ha-gy" sequences, have

[p] instead;52 and (4) [h]~[p] alternation occurs when [h] geminates even in current

Japanese. As dialectal evidence, linguists point out that some Okinawan dialects have words with ha-gy" [p] (Ueda 1989; Ueyama 1991; Morishita and 'no 2001; Karimata

2002). Although current Yamato items and Sino-Japanese items are virtually free from word initial [p], it can be observed in Mimetic items (Ueda 1989; Hamano 1998; Toyama

1972). It is important to note that, although [p] in pre-Old Japanese seems to have changed drastically, the word initial [p] in Mimetic words has been preserved ever since pre-Old Japanese in order to maintain iconicity (Hamano 1998). For example, the pre-Old

Japanese word for ‘light’ is assumed to have been pronounced as [pikari], and its

Mimetic counterpart was also [pika (ri)]. The Yamato item [pikari] went through the labial lenition, and it is pronounced as [çi4k%#i] in Modern Standard Japanese. On the other hand, the Mimetic item [pika (ri)] has not been affected by the labial lenition, and consequently the [p] has been preserved.53 Figure 3.1 compares the historical changes of

Yamato ‘light’ and Mimetic ‘light’.

52 Hashimoto (1950) notes that Ainu has [p], but [+] appears only conditionally. Thus, it is still unclear whether original Japanese words had [p] or [+]. Ueda’s arguments support the idea that ha-gy" consonants were bilabials in the past, but not all of them prove that ha-gy" consonants were once [p] before it was [+]. 53 It is important to note that /ri/ in the Mimetic [pika (ri)] is not the same as the verb stem-final /ri/. 97

Old Japanese Current Japanese

Yamato ‘light’ [pikari] " [+ikari] " [çi4k%#i]

Mimetic ‘light’ [pikari] " [pikari] " [pi4k%#i] Figure 3.1. The change of Yamato ‘light’ and Mimetic ‘light’

Although there are no historical documents that directly prove that ha-gy" consonants were [p] in Pre-Old Japanese, there are many historical documents that demonstrate that ha-gy" consonants were [+] before they were replaced by [h] (see documents cited in Hashimoto 1950; Mabuchi 1971; Toyama 1972; Ito and Mester

1999). As mentioned above, it is still not clear exactly when ha-gy" [+] was replaced by

[h]. Toyama (1972:196) assumes that the phone [h] already existed in exclamations and interjections in the early 17th century, based on the descriptions of exclamations, such as

Ha ‘ha’ and Hum ‘humph’ in Arte da Lingoa de Iapam (1604~08) written by the

Portuguese missionary João Rodríguez.54 In Arte da Lingoa de Iapam, regular ha-gy" consonants were transcribed as “f”. Therefore, the phone [h] existed in exclamations and

55 interjections, but not as a ha-gy" consonant. The traces of the transition from ha-gy" [+] to [h] also can be observed in instruction manuals for traditional oral culture, such as traditional songs. These instruction manuals provide specific notations to indicate phones that had been lost at the time they were written. The instruction manual Ongyoku

54 Arte da Lingoa de Iapam (Art of the Language of Japan) 日本大文典 55 Notice that Japanese speakers still say hun! フン！‘humph!’ as an exclamation to express irritation, and the first consonant is never [+], but [h]. 98

Gyokuen Sh# (1727)56 presents kana notations of ha-gy" sound sequences, as shown in

(1).

(1) Kana descriptions in Ongyoku Gyokuen Sh# (Source: Arisaka 1957:230)

はひふへほ

フハフヒフヘフホ

The top row indicates the hiragana transcription of the traditional songs, and the bottom row shows the notations of old pronunciations written in katakana. Notice that each hiragana character is accompanied by two katakana characters to provide a pronunciation guide. For example, フハ indicates [+%], the combination of フ (probably

[+u] at that time) and ハ (probably [h%] at that time). The notation フハ probably means

“pronounce [h%] (ハ) with the consonant of [+u] (フ),” as illustrated on the lower left- hand corner in (2).

(2) Kana descriptions in Ongyoku Gyokuen Sh# with IPA notation はひふへほ (ハ [h%]) (ヒ [hi]) (フ [+u]) (ヘ [he]) (ホ [ho])

フハフヒフヘフホ [+u+h%]=[+%] [+u+hi]=[+i] [+u+he]=[+e] [+u+ho]=[+o]

56 Ongyoku Gyokuen Sh# 音曲玉淵集 provides phonetic advice for singing old songs authentically. 99

Arisaka (1957) claims that the fact that the ha-gy" pronunciations in traditional songs needed special notations suggests that there was a sound change. The change was most likely [+] to [h] in the consonants of ha-gy" sequences except /hu/. This means that the consonants of ha-gy" /ha hi he ho/ in the early 18th century were pronounced as [h],

57 but the consonant of /hu/ was pronounced as [+].

Word-initial [p] as a separate phoneme from ha-gy" /h/ started to reappear in

Japanese probably around the middle of the 16th century, when Portuguese Jesuit missionaries came to Japan and imported Portuguese vocabulary items, such as /pa-/

(

The innovative sound sequences [+% +i +e +o] probably emerged in the 20th century. However, many English words with /fV/ sequences were borrowed with /hV/ sequences until the end of WWII. Consequently, there are many loanwords with /hV/ sequences (and sometimes /huV/ sequences) whose source words have donor-language

/fV/ sequences instead. The example in (3) illustrates a loanword with a [ho] sequence instead of a [+o] sequence.

(3) ([p"#%tto]) [ho$m"] ‘platform’ (プラット) ホーム

57 It is not clear from the document whether /hi/ was [h(i] or [çi]. 100

Later, the /fV/ sequences [+% +i +e +o] started to be used for recently borrowed items, replacing [h% h(i he ho]. Consequently, there are many variants, as shown in (4).

58 (4) [h"%-] ‘fan’ フアン [ɸ%-] ファン

["eh%$s"] ‘wafers’ ウエハース ["e+%$s"] ウエファース

59 [h"i#"m"] ‘film’ フイルム [+i#"m"] フィルム

[j"/iho:m"] ‘uniform’ ユニホーム [j"/i+o:m"] ユニフォーム

[ho#"m%#i-] ‘formalin’ ホルマリン [+o#"m%#i-] フォルマリン

Some words do not have variants with [+% +i +e +o] sequences, as shown in (5) below.

58 Both [h"%-] and [ɸ%-] may occur when meaning ‘an enthusiast’. However, only [ɸ%-] can occur when meaning ‘a blowing device’. 59 The Japanese name of the company ‘Fujifilm’ is ‘Fujifuirumu’, not ‘Fujifirumu’. Although [h"i#"m"] and [+i#"m"] are nearly the same word, [h"i#"m"] almost exclusively refers to photographic films. [+i#"m"] is used for other items, such as movie films. For example, microfilm is almost always [m%ik"#o+i#"m"], not [m%ik"#oh"i#"m"]. 101

(5) [hi#e] ([nik"]) ‘fillet’ ヒレ(肉) (*[+i#e (nik")])

([p"#%tto]) [ho$m"] ‘platform’ (プラット) ホーム (*[+o$m"])

[te#eho-] ‘telephone’ テレホン ([te#e+o-] may occur)

[heddoho-] ‘headphones’ ヘッドホン ([heddo+o-] may occur, but is less common)

We have reviewed the sound changes that involve ha-gy", pa-gy", and fa-gy".

Table 3.3 approximately summarizes the historical development of the consonants in ha- gy", pa-gy", and fa-gy".

Table 3.3. Historical development of ha-gy", pa-gy", and fa-gy" Old Japanese Heian Edo Pre-WWII Post-WWII

ha-gy" p $ + $ h (çi, +") $ h (h(i)

pa-gy" p $

fa-gy" + 6

As discussed in the previous section, current ha-gy" should be transcribed as [h% h(i h" he ho], instead of traditional [h% çi +" he ho]. However, since all the ha-gy" consonants were [+] in the past, there is a good chance that ha-gy" with sound sequences

[h% çi +" he ho] might have existed before the current ha-gy" was established. The consonant of ha-gy" /hu/ in the late 19th century might have been close to [+"], rather than [h"], based on the fact that ha-gy" /hu/ is transcribed as “fu” in the Hepburn 102

Romanization. The Hepburn Romanization, which was made popular by the third edition of James Curtis Hepburn’s Japanese-English dictionary published in 1886, was loosely based on English spelling (Seeley 1991:140).

Although I claim that current ha-gy" should be transcribed as [h% h(i h" he ho], consonants of some word internal ha-gy" /hu/ are pronounced as [+] even in current standard Japanese. For example, introspection suggests that the word /huku/ 服 ‘clothes’

60 is pronounced as [h"4k"], whereas /ihuku/ 衣服 ‘clothes’ is pronounced as [i+k"].

The phone [+] is often described as the sound of blowing out a candle flame

(Sait+ Yumiko 1990; Inozuka and Inozuka 2003; Kashima 2002; Kase 2001). However, the actual sound of blowing out a candle flame is typically represented by the symbol

[7]—a voiceless labial-velar fricative, which involves lip rounding. Therefore, the sound of blowing should be transcribed as [7$]. Nevertheless, when Japanese speakers express the sound of blowing or puffing in an onomatopoetic manner, they are likely to use [+"$], as shown in (6).

(6) [+"$ tto h"it%] ‘(I) blew (with the sound whew)’ フーッと吹いた

60 As discussed in §3.2, Sait+ (2003) claims that /huku/ has [+] due to the devoiced /u/ preceded by /h/. But /h/ in /huku/ and /ihuku/ are clearly different and pronouncing [+5k"] or [+"4k"] seems very unnatural to me. Sait+ (2003) and Kawakami (1977:49, cited in Vance 1987) both claim that [+ ç] can never be replaced by [h h(] when followed by a voiceless vowel. I agree with the description of [ç], but not [+]. This makes the issue of [+] more complicated. I still cannot correctly predict when [+] appears. 103

The history of ha-gy" can be summarized as follows. All the consonants of ha- gy" were [p] in pre-Old Japanese. Word-initially, they changed into [+] before the Heian period. Then they changed to [h] except for the consonant of /hu/ and /hi/. Finally, all consonants of ha-gy" came to be pronounced as [h], and the [+"] sequence is allowed in special cases only—such as word internally and in onomatopoetic expressions.

3.5 The Innovative /fa fi fe fo/

As shown in Table 3.4, there are many words that contain the innovative sound sequences

/fa fi fe fo/. These words are all recently borrowed from English and other languages.

Table 3.4. Numbers of words with innovative /fV/ sequences that are listed in Japanese dictionaries /fa/ /fi/ /fe/ /fo/

K"jien 390 319 240 294

Daijirin 486 434 303 319

Judging from the numbers in Table 3.4, the innovative /fa fi fe fo/ seem to be well-established in Japanese. If ha-gy" /hu/ is actually [h"] and the consonant of the innovative /fa fi fe fo/ is [+] as I argue, then the emergence of the innovative /fa fi fe fo/ is quite unique for the following reason. Other innovative sequences are all combinations of the phones of existent sound sequences. For example, the innovative sound sequence

[ti] is a combination of the phone [t] in the existent sound sequences [t%], [te] or [to], and 104

61 another existent phone [i]. In contrast, the innovative /fa fi fe fo/ [+% +i +e +o] are not created by simply combining existent phones because [+] did not exist word-initially when /fa fi fe fo/ were introduced. Where did the [+] come from? In order to account for the origin of the innovative [+], it is essential to look into the historical facts concerning ha-gy" and fa-gy".

Table 3.3 presented in the previous section shows that there are cases of innovative sound sequences that were once traditional sound sequences. That is to say, innovative fa-gy" sound sequences that appeared in the 20th century also existed in the

16th century, and pa-gy" sound sequences that were innovative in the 16th century also existed in Old Japanese. Table 3.5 shows that the truly innovative sounds are the ha-gy"

[+] that replaced ha-gy" [p], and the ha-gy" [h (çi)] that replaced ha-gy" [+] (except for

[+"]).

61 Literate Japanese speakers may consider [ti] a combination of [te] and [i], mainly due to the fact that kana transcription of [ti] ティ is a combination of [te] テ and a small version of [i] イ. However, since the vowels of [ti] and [te] are both front, this orthographical solution may not be far fetched (see Inozuka and Inozuka 2003:58 for a more detailed discussion). 105

Table 3.5. Historical development of ha-gy", pa-gy", fa-gy", Mimetic pa-gy", and Mimetic ha-gy" Old Japanese Heian Edo Pre-WWII Post-WWII

ha-gy" p $ + $ h (çi, +") $ h (h(i) $

([+]: truly innovative) ([h]: truly innovative)

pa-gy" p $

(innovative, but existed in the past)

fa-gy" + 6 (innovative, but existed in the past)

Mimetic p $ pa-gy"

Mimetic h $ ha-gy"

The change from ha-gy" [p] to ha-gy" [+] in Yamato items could be explained as an example of labial lenition. However, this change does not involve Mimetic items. The phone [p] in Mimetic items is considered to have been the same since at least the era of

Old Japanese. Does the disappearance of word-initial [p] in non-Mimetic words mean that Japanese speakers at that time lost the grammatical ability to produce and process

[p]? There are several historical and other linguistic facts that should be taken into consideration when discussing the apparent labial lenition in Japanese. First, after [+] completely replaced [p], [p] reappeared in Foreign items later, probably in the 16th century. Second, cross linguistically, [p] is one of the very first phones acquired by 106 children if the language has one (Jakobson 1971). Third, as discussed above, Mimetic [p] has been present since Old Japanese.

Within the framework of Optimality Theory (OT), scholars try to solve the problem of [p] by providing innovative constraints and rankings (Ito and Mester 1995b,

1999). Ito and Mester (1995b:819) propose the constraint *P: a constraint against single

[p], which prohibits [p] unless it is in a geminated or partially geminated form. According to Ito and Mester, this constraint only applies to Yamato and Sino-Japanese items in current Japanese. Non-uniform application of *P across the strata cannot be explained by a single constraint ranking without some modification. As discussed in Chapter 2, some

OT scholars have proposed multiple grammars with different constraint rankings within a language (Anttila 2002; Inkelas, Orgun and Zoll 1997; Ito and Mester 1995b). This approach is called the co-grammar (or co-phonology) approach. On the other hand, the other approach, which has only one grammar, is called the indexed faithfulness approach

(Ito and Mester 1999, 2008). Instead of assuming multiple grammars, the indexed faithfulness approach keeps a single grammar with faithfulness constraints that are indexed with strata, such as FAITH/YAMATO, FAITH/SINO-JAPANESE,

FAITH/ASSIMILATEDFOREIGN, FAITH/UNASSIMILATEDFOREIGN, and so forth. The indexed faithfulness approach does not allow markedness constraints to be indexed. The indexed faithfulness approach is believed to be more advanced than the co-grammar approach because it requires only one grammar with strict domination (Ito and Mester 1999). From the diachronic point of view, both approaches have similar problems. The co-grammar approach requires the speakers to know which grammar is applicable to what word. The 107 indexed faithfulness approach, on the other hand, requires the input to be indexed so that the correct indexed faithfulness constraint is applied. In short, OT solutions require pre- knowledge of language stratification, and they do not explain how the existence of such apparently language-specific, or even stratum specific innate knowledge, such as

FAITH/ASSIMILATEDFOREIGN, is possible. Similarly, in order for the indexed faithfulness approach to operate properly, the input also must be labeled, such as

ASSIMILATEDFOREIGN, UNASSIMILATEDFOREIGN, and so forth. In the case of [p], labeling the constraint FAITH/YAMATO and labeling the input “YAMATO” are problematic because current Japanese does not allow [p] in Yamato items, whereas Old Japanese did allow [p] in Yamato items. The constraint FAITH/YAMATO does not seem to be universal through time.62 Assuming that Japanese speakers have access to pre-knowledge of language strata does not seem to provide convincing explanations of the distribution of certain phones among language strata.

In this section, I argue that language strata are irrelevant to grammar because they are merely historical facts; therefore, the strata-specific constraint *P, which prohibits non-geminated [p], has never been valid in Japanese. Historical facts demonstrate that

Mimetic items and non-Mimetic items took different paths: Mimetic items kept [p], while

Yamato and Sino-Japanese [p] was replaced by [+] in Old Japanese, and that [+] was

62 Ito and Mester (1999) alternately propose an unindexed IO-faithfulness constraint instead of FAITH/YAMATO. This should surely work perfectly with the examples used in Ito and Mester (1999), which do not include Mimetic items. But once Mimetic items are taken into consideration, an unindexed IO- faithfulness constraint does not function. In fact, the indexed faithfulness approach proposed by Ito and Mester (1999) does not work with or without the constraint FAITH/YAMATO once Mimetic items are considered. See also Inkelas and Zoll (2003). 108 replaced by [h]. The [p] in Mimetic has always been present in Japanese in order to maintain iconicity (Hamano 1998). This means that there is a specific, non-grammatical reason for [p] in Mimetic to have existed. Usually, phonetic changes apply to the whole vocabulary (Labov 1994). The resistance to the change of [p] in Mimetic items suggests that the change from [p] to [+] in non-Mimetic items was not grammatical since, if it were grammatical, non-grammatical specific reasons could not have led to resistance to the change. Thus, the change from [p] to [+] in non-Mimetic items does not require the constraint *P. We do not know any specific reason why this change occurred. But whatever the reason is, it was not powerful enough to change all vocabulary items. These historical facts mostly explain why single [p] is present in Mimetic items, but not in

Yamato and Sino-Japanese items in current Japanese. After the change from [p] to [+] in ha-gy" consonants, the phone [p] reappeared in non-Mimetic items as pa-gy" /p/ in the

16th century. The revival of non-Mimetic [p] also suggests that the constraint *P has never been valid. Historical facts can sufficiently explain, without requiring pre- knowledge of language stratification, why [p] does not appear in Yamato and Sino-

Japanese items, but does occur in Mimetic and Foreign items. Judging from the fact that

[p] has been always present in Japanese, it is dubious that the constraint *P actually exists.63

The distribution of [p] in Japanese is now an issue of historical linguistics, not synchronic linguistics. Synchronic linguists no longer have to answer any questions

63 It would be interesting to compare Arabic, which has [b] but lacks [p], to see if similar arguments against *P constraint apply. 109 regarding the distribution of [p] in Japanese. Instead, this means that historical linguists are challenged with the question of why ha-gy" consonants changed from [p] to [+]. It is impossible to answer this question completely, since the historical evidence is limited.

One possible scenario is that [+] emerged accidentally or had existed as a phone used in marginal items at that time and was preferred over [p]. A similar scenario may apply to the change from [+] to [h].

The disappearance of [+] and the revival of [+] in current Japanese can be explained in a manner that is similar to the case of ha-gy" [p]. Before I present the historical account associated with [+], I review the synchronic account by Ito and Mester

(1995b). Ito and Mester propose the constraints *F and -FU. The constraint *F prohibits the fa-gy" sound sequences and the [+"] sequence. The -FU constraint prohibits [h"] from labializing the [h] to create a [+"] sequence. Ito and Mester claim that the constraint -FU is more peripheral than *F because modern standard Japanese allows fa- gy" sound sequences, but not the ha-gy" [h"] sequence. The account by Ito and Mester appears to depict the synchronic grammar of ha-gy" adequately. However, the existence of the constraint -FU is doubtful for the following two reasons. First, to the best of my knowledge, there is no language other than Japanese that disallows [h"] and labializes the [h] to [+]. Second, the constraint -FU had been effective for a very limited period of time. Since [h] did not appear in Japanese at least until the 17th century, -FU started being effective merely less than 400 years ago, which is extremely recent in the history of human language. Moreover, if my hypothesis that ha-gy" /hu/ is pronounced as [h"] in modern standard Japanese is correct, actual active period of -FU is very short. 110

Accordingly, the existence of the constraint -FU is doubtful considering the fact that it seems to play an extremely specific role in Japanese for an extremely limited period of time. It is difficult to believe that -FU is a constraint which is part of universal and innate human knowledge. Rather, the circumstances that appear to require the constraint -FU can be explained due to the processes of language change. Unlike *P or -FU, the constraint *F may exist. The phone [+] is not very common (Ladefoged and Maddieson

1996), and it may have not existed in Japanese until ha-gy" [p] changed to [+].

The disappearance of [+] in the Edo period may be better explained as the result of lexical change, rather than as the result of reranking, because [+"] survived until it was replaced by [h"], and the rest of ha-gy" sequences reappeared later as fa-gy" sequences. Moreover, the reappearance of [+] itself is evidence that its disappearance in the Edo period was not due to the fact that speakers lost the ability to pronounce and perceive [+].

I have argued that while it may be possible to account for the appearance of innovative fa-gy" sequences synchronically, analyses of historical facts provide more convincing explanations. The OT account depicts the current distributions of innovative sound sequences systematically, but at the same time, it requires constraints whose existence is questionable. In contrast, historical analyses provide simple and concrete reasons why innovative fa-gy" sequences appeared and have prevailed so easily.

CHAPTER 4

VOICELESS CORONAL FRICATIVES

Gilead then cut Ephraim off from the fords of the Jordan, and whenever Ephraimite fugitives said, 'Let me cross,' the men of Gilead would ask, 'Are you an Ephraimite?' If he said, 'No,' they then said, 'Very well, say Shibboleth.' If anyone said, 'Sibboleth', because he could not pronounce it, then they would seize him and kill him by the fords of the Jordan. Forty-two thousand Ephraimites fell on this occasion. – Judges 12:5-6, NJB

4.1 Introduction

There are two newly introduced sound sequences that include a voiceless coronal fricative: [si] and [!e]. Japanese [s] is a laminal voiceless alveolar fricative and Japanese

[!] is a laminal voiceless alveolo-palatal fricative. Japanese [!] is often transcribed as ["], a voiceless postalveolar fricative (e.g. Ito and Mester 2007). In this dissertation, however, the symbol ["] is not used for the transcription of modern standard Japanese, because the subtle difference in articulatory gestures between [!] and ["] plays a significant role in solving the issue raised in this chapter. The primary issue involving these sound sequences is that the innovative sound sequence [!e] has been well accepted ever since shortly after it was introduced,64 whereas [si] is still foreignism. Table 4.1 illustrates the phonetic distributions of voiceless coronal fricatives in Japanese.

64 However, the existence of doublets, such as [!ep#$do] ~ [sep#$do] ‘shepherd’ and Makku [!eik%] (or [!e$k%]) ‘McDonald’s milk shake’ ~ miruku[se$ki] ‘milk shake’, suggests that at first there might have been some difficulty in accepting the sound sequence [!e]. 112

Table 4.1. Distribution of [s] and [!]

[s"] [si] (foreignism) [s#] [se] [so]

[!"] [!i] [!#] [!e] (innovative) [!o]

Notice that [si] and [!e] complete the table by filling in the ‘gaps’ (or the empty

65 spots). It is possible to claim that [si] and [!e] were introduced to fill in the systematic gaps in the table. However, doing so does not explain why [!e] has been accepted widely, whereas [si] is still a foreignism.

Although it is not in the scope of this dissertation to reconstruct old speech sounds, I will examine some scenarios for the development of voiceless coronal fricatives that might explain the issue because, historically, the developments of [s] and [!] have close connections (Mabuchi 1971; Arisaka 1957; Miyake 2003; Hashimoto 1950). To evaluate these scenarios, I will refer to historical evidence, dialectal evidence, acoustic and articulatory evidence, and social factors.

4.2 Innovative [!e] in the Dialect of Tokyo

The innovative sound sequence [!e] in current standard Japanese entered in non-marginal words as a result of the influence of European languages, such as English. Daijirin lists

117 words with the [!e] sequence and K!jien lists 97. All of these listed words are

65 Hattori (1960) argues that the spot for [!e] is a ‘gap (akima)’, whereas the spot for [si] is a ‘seam (sukima)’. See Chapter 1 for a more detailed argument. 113

recently borrowed. However, in marginal situations, the [!e] sequence does occur in native words as well. First, Japanese young children often replace [se] with [!e]. Alveolar backing is frequently observed in Japanese first language acquisition (Beckman,

Yoneyama, and Edwards 2003; Tsurutani 2004). This fact suggests that pronouncing [!e] may be easier than pronouncing [se] for most Japanese speakers. Second, in casual settings, Japanese adults also replace [se] with [!e] to project certain images—such as childishness and dependency—by imitating young children’s speech, such as

[s"im#!e$] (<[s"im#se$] ‘I am sorry’). Hamano (1998) claims that palatal sounds in mimetic items have similar effects. Third, /se/ is pronounced as [!e] in certain regions of

Japan.66 Consequently, people in Tokyo have had chances to hear the sound sequence.

Fourth, [!e] is frequently used in manga and anime. Ito and Mester (1995b:830) list [!e%]

(exclamation used by famous cartoon figure)67 as an example of a lexically “native” item that includes phonologically “peripheral” sound sequences.

As we have seen, although [!e] is an innovative sound sequence, it was also found in marginal situations in some cases even before it was introduced into non-marginal words.

4.3 The [si] Sequence

Compared with [!e], the [si] sequence is hard to find in both marginal and non-marginal words. Neither K!jien nor Daijirin contains this sound sequence in their entire text. Even

66 Hattori (1960:757) claims that [!] in dialectal /se/ [!e] is even less palatalized than [!] in /si/ [!i]. 67 The character mentioned by Ito and Mester is Iyami from Osomatsu-kun (Akatsuka 1962-1967) 114 in popular culture, the [si] sequence is difficult to find. Oddly, Mabuchi (1971:20) includes the sequence [si] in his list of moras of modern Japanese, claiming that locution, such as [n!ssi!"#] ‘nothing’ instead of [n!$$i!"#], came to be prevalent (Mabuchi

1971:20-21). Indeed, TV newscasters occasionally pronounce [n!ssi!"#], particularly in broadcasts of baseball games.68 Also, TV newscasters used to pronounce the name of

Roman letter ‘c’ as [si%] in the 1980s (e.g. [si%di%] ‘CD’ [i%si%] ‘EC’). Some TV commentators use the [si] sequence for foreign items, such as [sis#tem#] (< [$is#tem#]

‘system’). However, the [si] sequence is still a foreignism and very rare.

The [si] sequence is used for some foreign vocabulary items in popular culture that are influenced by Western culture. American and European popular songs have greatly influenced Japanese popular songs targeting youth. Consequently, such Japanese songs often contain English words and phrases with [si] in their lyrics. For example, the

Japanese popular vocal duo Pink Lady (1970-1981) pronounce the word [sek#$i%] ‘sexy’ as [seksi%] in their hit song Nagisa no Shindobaddo ‘Sindbad at the Beach’ (Aku and

Tokura 1977) in spite of the fact that the word is written in katakana (セクシー

69 [sek#$i%]). The word [sek#$i%] ‘sexy’ is a non-marginal item listed in both Daijirin and

K!jien. This word is pronounced with [$] in daily conversations. The reason Pink Lady

(or their producer) chose to use the foreignism [si] sequence was probably the desire to

68 Since Japanese does not have the sound [&], it is replaced by either [s] or [$]. Many Japanese feel that [s] is much closer to [&] than [$] is. [n!ssi'"#] ‘nothing’ may also be used when young people are trying to be funny by creating puns, for example, [mond!i n!ssi'"#] ‘there is no problem’, whose authentic form is [mond!i n!$i] (lit. ‘problem, there is not’). 69 Japanese TV programs often subtitle lyrics. 115 create an exotic and fashionable atmosphere. The Pink Lady singers’ command of

English is believed to be very limited. Accordingly, the ability to pronounce [si] intentionally has little to do with English ability. Most current Japanese popular songs targeting youth contain some English words or phrases (see Loveday 1996 for examples).

Therefore, the audience for these songs is familiar with the [si] sequence. It is important to note that popular songs are not only heard but also sung by the audience, particularly in Japan, where karaoke is extremely popular. That is to say, the audience members may also produce the [si] sequence when they sing songs with [si].

The [si] sequence can be heard in certain TV programs as well. In Sekus! Buch"

‘Sexy Manager’ セクスィー部長 (NHK 2007~), a popular recurring sketch in the television program Sarar!man NEO ‘Salaried worker NEO’, the main character is called

[sek!si" b!c#o"] with a [si] sequence. The katakana transcription セクスィー also suggests that the main character’s name is pronounced as [sek!si" b!c#o"], but not

[sek!#i" b!c#o"]. Notice that the [si] sequence is transcribed with the combination of ス

([s!]) and ィ (a small-size イ [i]). The traditional transcription uses シ [#i]. In many episodes of the sketch, if the main character’s name is mispronounced as [sek!#i" b!c#o"], one of his subordinates corrects the mistake. This suggests that the word

[sek!#i"] ‘sexy’ is still pronounced with a [#i] sequence by many speakers. On the other hand, it also suggests that at least the audience can tell the difference between [si] and

[#i]. 116

The katakana transcription スィ [si] is frequently used in introductory foreign language textbooks that are available at regular bookstores.70 As described earlier, in katakana, the [si] sequence can be transcribed as スィ, a combination of ス [s!] and ィ, which is a small-size イ [i], indicating that [i] is the vowel of the syllable that replaces

[!] in ス [s!]. The Cabinet of Japan, which officially establish policies regarding the use of orthography, including katakana, has specific guidelines for the use of katakana for loanwords. In the guidelines, [si] スィ is described as a special sound sequence,71 whereas katakana transcriptions of many other innovative sound sequences, such as シェ

72 ["e], are classified as kana (spelling) that are generally used. This fact suggests that the

Cabinet of Japan, on the one hand, regards シェ ["e] as a general sound sequence and, on the other hand, regards スィ [si] as a special sound sequence. Nonetheless, katakana can adequately and unambiguously transcribe the sound sequence [si], and that is one of the reasons why the katakana transcription スィ [si] is used in foreign language textbooks.

Notwithstanding the fact that the sound sequence [si] is rare in standard Japanese, it does not seem to be difficult for Japanese speakers to pronounce it. Most Japanese junior high school students study English, and many of them do not have trouble pronouncing [si] even if their command of English is not good. Moreover, it is very

70 Authentic language textbooks used in educational institutions rarely use kana transcriptions because of their inability to transcribe phones that Japanese does not have. For example, it is impossible to transcribe [!] using katakana unless some modifications are made. 71 国語施策情報システム国語表記の基準外来語の表記留意事項その１(原則的な事項) Bunkach! (The Agency of Cultural Affairs) http://www.bunka.go.jp/kokugo/frame.asp?tm=20090806232408 (accessed August 6, 2009) 72 国語施策情報システム国語表記の基準外来語の表記「外来語の表記」に用いる仮名と符号の表 Bunkach! (The Agency of Cultural Affairs) http://www.bunka.go.jp/kokugo/frame.asp?tm=20090806232408 (accessed August 6, 2009) 117 common for a speaker of Japanese who does not have a good command of English to

73 mispronounce [!i] ‘she’ as [si]. The sound sequence [si] is also observed in non-

74 standard Japanese. Some dialects have the [s"] sequence (!hashi 2002) and standard

Japanese speakers have chances to hear them on TV programs and in movies. However, since the vowel of [s"] is centralized compared with modern standard Japanese /u/ ([#]), it may not be perceived as a genuine [si] sequence.

Interestingly, many linguists report that young Japanese speakers have a tendency to pronounce sa-gy! /si/ close to [si], rather than [$i] (Hamada 1964; Kunihiro 1983;

Inoue 1989; Unger 2006; Shibata 1988). Kunihiro (1983) reports that the pronunciation of sa-gy! /si/ varies. He claims that while most Japanese speakers use [$i] for sa-gy! /si/, younger women—junior high school to high school students—tend to pronounce

[%&t&si] instead of [%&t&$i] ‘I’. On the other hand, he also states that there are other people who pronounce the word as [%&t&!i]. Kunihiro further claims that those who use

[si] for sa-gy! /si/ tend to pronounce something like [sj&, sj#, sjo] for sha-gy! /!&, !#,

75 !o/, respectively. Inoue (1989) mentions that sa-gy! /si/ is coming to be more like [si].

Unger (2006) and Shibata (1988) indicate that the place of articulation of /s/ in standard

Japanese [$i] has been advancing so that it is more like [si]. Admittedly, some young speakers occasionally pronounce [si] instead of [$i], but I do not think that this is a

73 This kind of hyper-correction indicates that although they can pronounce [si], they might still confuse [si] and [$i] or [!i]. 74 ["] is a high central unrounded vowel. 75 However, Inoue suggests that it is not a perfect [si], but rather something in between [!i]/[$i] and [si]. 118 uniform phenomenon yet. Shibata (1988), who is from Nagoya,76 states that his own pronunciation of /si/ is close to [si]. But ‘close to [si]’ may not mean ‘sounds like [si]’.

Introspection of my own places of articulation of sa-gy! consonants, including sa-gy!

/si/, suggests that my tongue position is so advanced that the tongue tip protrudes from the teeth. Nonetheless, my sa-gy! /si/ still does not sound like [si]. Although the claim that Japanese [!i] has been advancing its place of articulation is intriguing, it needs empirical studies to be justified.

It appears that many Japanese speakers are capable of pronouncing [si], but some speakers may confuse [si] and [!i]. The sound sequence [si] is rarely used in everyday conversation, despite its occasional use in TV broadcasting and popular songs.

4.4 Japanese ‘/s/’ and ‘/"/’

The phonemic distribution of Japanese consonants has been phonologically and phonetically controversial. In particular, the distribution of the two voiceless coronal fricatives in Japanese, /s/ and /"/, is probably the last frontier for any Japanese phonologist. Although this may be an interesting topic, it is not in the purview of this dissertation; therefore, I will only briefly introduce different views. In short, the issue is whether [!] in the sequence [!i] is an allophone of /s/ or /"/. In this dissertation, I will defend a view that is closely related to the former view because it reflects historical facts.

Before comparing the two different views, I will review the articulatory characteristics of Japanese voiceless coronal fricatives. The voiceless coronal fricatives

76 Nagoya is a city located in central Japan. 119

discussed in this chapter and their phonetic characteristics are: [!] (a voiceless alveolo- palatal fricative), [s] (a voiceless alveolar fricative), and ["] (a voiceless postalveolar fricative). In terms of the place of articulation, [!] is somewhat in between [s] and ["].

The following table illustrates the distribution of voiceless coronal fricatives by following the general phonemic and allophonic analysis. The most significant advantage of this distribution is that the phonetic consistency between phonemes and allophones is maintained: phonemes and allophones have a one-to-one correspondence, as shown in

Table 4.2.

Table 4.2. Sequences involving [s] and [!] in terms of the general phonetic allophonic analysis

/s/ [s#] [si] (foreignism) [s$] [se] [so]

/"/ [!#] [!i] [!$] [!e] (innovative) [!o]

The other view, which is mainly based on the Japanese writing system, has dealt with the allophonic classification of voiceless coronal fricatives in a slightly different

77 way. This view treats the consonants of [!#], [!$], and [!o] as /sy/ sequences.

Consequently, the most significant difference is that the sound sequence [!i] is not included in the group /"/ (/sy/), but rather in the group /s/. Table 4.3 demonstrates the phonetic inconsistency of this view.

77 The innovative sound sequence [!e] should be classified into the same group as [!#] [!$], and [!o], although the kana orthography does not indicate the existence of /y/ in [!e], due to the lack of /ye/ sound sequences in modern standard Japanese. 120

Table 4.3. Sequences involving [s] and [!] based on the writing system

/s/ [s"] [!i] [s#] [se] [so]

([!e]) /$/ (/sy/) [!"] [!#] [!o] (innovative)

According to this view, the distribution of the allophones of /s/ and /$/ can be

78 explained as follows: [!] in the sequence [!i] is an allophone of /s/. The phonological rule that is illustrated in Table 4.3 can be described as ‘/s/ becomes [!] before /i/’.

Tsujimura (1996) states that since typically any vowel phoneme can follow any consonant, and the sound sequence [si] simply does not exist in Japanese, it is predictable that [!i] occurs instead of [si]. Another piece of evidence that [!] in the sequence [!i] is an allophone of /s/ is the verbal conjugation rule. For example, the non-past form of the verb /kas/ ‘to lend’ is [k"s-#], while its past form is [k"!-it"]. Thus, it is possible to

79 claim that [!] in [k"!-it"] is derived from /s/ in /kas/.

Another piece of evidence that might support the view that [!] in the sequence [!i] is an allophone of /s/ is the similarity to the distributions of other consonant phonemes. If we consider the distribution of the allophones of the voiceless velar consonant /k/ in current Japanese, we notice that the distributions of the allophones of /s/ and /$/ are

78 Hattori (1960:757) claims that [!] in [!i] is slightly less palatalized than that of [!"], [!#], and [!o]; therefore, treating [!i] as the surface form of /si/ is not problematic. 79 Nevertheless, Tsujimura (1996) also asserts that aside from the clear cases in which [!] is derived from /s/ (as in [k"!-it"], which is the past form of the verb /kas/), [!] may be an allophone of the phoneme /$/. [!] can be followed by any vowel and when [!i] is not derived from a combination of two morphemes, it is also possible to argue that the underlying representation of [!i] is not /si/, but /$i/. 121 similar to situation with /k/ (see also McCawley 1968:76). Like the ancestors of the current [!"], [!#], and [!o] sequences, the consonant clusters in /kya/, /kyu/, and /kyo/ emerged by combining more than one syllable. Table 4.4 illustrates the distribution of sound sequences that start with /k/ in narrow transcriptions that reflect palatalization.80

Table 4.4. Sequences involving [k] and [k$] based on the writing system

/k/ [k"] [k$i] [k#] [ke] [ko]

([k$je]) /ky/ [k$j"] [k$j#] [k$jo] (foreignism)

Although there are differences between Table 4.3 and Table 4.4, such as [k] and

[k$] as opposed to [s] and [!], and the presence of [j] in the realizations of the consonant cluster /ky/, the distribution of the sound sequences with voiceless velar consonants has basically the same structure as the distribution of the sound sequences with voiceless coronal fricatives.81

The synchronic facts discussed above are not incompatible with the idea that [!] in the sequence [!i] is an allophone of /s/, but these facts do not decide the issue. Thus, it is still not clear whether non-derivative [!i] is an allophone of /s/ or /%/. Now, I will turn to historical linguistics because the phonetic inconsistency displayed in Table 4.3 reflects historical facts. In the history of Japanese phonology, the current [!"], [!#], and [!o] sound sequences have different origins from the current [!i]. The ancestors of the current

80 See Ladefoged (2001:218) for palatalization of English /k/ in front of a high front vowel. 81 See also Hamada (1964). 122

[!"], [!#], and [!o] sequences emerged as innovative sound sequences in the 8th century by combining more than one syllable, such as /si+a/, /si + u/, /si + o/ and so forth

(Mabuchi 1971).82 Although the actual phonetic values of these sounds have changed considerably since the 8th century, treating the [!"], [!#], and [!o] sequences differently from the current [!i] sequence is not unreasonable, because typologically and historically the palatalization of [s] before a high front vowel is observed frequently (Blevins 2004).

Since this dissertation needs to address historical facts, I adopt the traditional phonemic distribution that treats [!] in [!i] as an allophone of /s/. It is important to note, however, that I do not claim that the allophonic analysis based on the historical facts is ‘correct’ or

‘better’. Both positions have their own advantages and disadvantages. The sole purpose of using the traditional analysis is to effectively illustrate diachronic facts about Japanese voiceless coronal fricatives, and to demonstrate how these diachronic facts account for the emergence of current innovative sound sequences.

Although I take the position that treats [!] in [!i] as an allophone of /s/, I avoid inferring the phoneme-allophone relationship as much as possible. Instead, I will treat current [s"], [!i], [s#], [se], and [so] as a group. I will call this group sa-gy!: the sa-

83 column of the goj"-on-zu ‘fifty-sound-display’. Similarly, I group [!"], [!#], [!e], and

[!o] together and call the group sha-gy!. Whether [s] and [!] are allophones of the same phoneme or not is insignificant for my purposes. The important fact is that Japanese

82 Actual Japanese vowels when the ancestors of the current [!"], [!#], and [!o] sequences emerged were much more complicated than shown in the examples listed here. 83 The goj"-on-zu ‘fifty-sound-display’ arranges kana syllabaries on a grid with the consonants on the columns and vowels on the rows. 123

continued to lack the [si] sequence even after the innovative sound sequence [!e] came to be widely used. In order to account for the somewhat puzzling distribution of voiceless coronal fricatives, I will review some previous work on the reconstruction of sa-gy! and sha-gy!.

4.5 The Overview of Historical Changes of Japanese Coronal Fricatives

The distribution of Japanese coronal fricatives and their phonetic values are controversial as well (for example, see Miyake 2003 and works cited there). In addition to the fact that the innovative sound sequence [!e] has been added to sha-gy!, the only conclusive historical facts are that: (1) sha-gy! did not exist in old Japanese; and (2) sa-gy! /se/ was pronounced as [!e] or ["e], instead of [se], at least in the early 16th century, when there was no contrast between sa-gy! /se/ and sha-gy! /"e/. It is also important to note that in certain areas of Japan, sa-gy! /se/ is still pronounced as [!e] (Okumura 1972; Sat! 2002).

Table 4.5 presents an overview of known historical facts on Japanese coronal fricatives.

Table 4.5. Overview of known historical facts on Japanese coronal fricatives sa-gy! /sa si su se so/ sha-gy! /"a "u "e "o/

8th century Inconclusive Did not exist

Before 16th century Inconclusive [!#] [!$] [!o]

16th century [s#] [!i] [s$] [!e][so] [!#] [!$] [!o]

19th century [s#] [!i] [s$] [se][so] [!#] [!$] [!o]

Current [s#] [!i] [s$] [se][so] [!#] [!$] [!e] [!o]

124

Notice that sa-gy! /se/, as well as sa-gy! /si/, had a palatal consonant in the 16th century. Two possible simple scenarios can be imagined from the data described above.

One scenario is that all the consonants (possibly except for the consonant followed by /i/) of sa-gy! before the 16th century were not palatal consonants and the sequences with front vowels were “palatalized” later (see Ito and Mester 2007).84 The other scenario is that the consonants of sa-gy! were all palatal consonants, either [!] or [t"], and the consonants of sa-gy! /sa/, /su/, and /so/ were “depalatalized” after sha-gy! sound sequences were introduced (see Mabuchi 1971). However, both scenarios have problems.

As for the former scenario, it is uncertain how a mid-front vowel [e] could behave as a

“palatalizer”. The latter scenario is also problematic because having [!] without having

[s] in the inventory of a language may be typologically implausible. I will call these two scenarios “extreme scenarios” to illustrate the problem, implying that what actually happened was somewhere in between the two extreme scenarios. The phonetic values of the sa-gy! consonants that have historically inconclusive distributions may be more complex than these two extreme scenarios.

84 It is not clear whether Ito and Mester believe that the change from the Pre-Edo system to the Edo system involves the palatalization of /se/ or not. But their diagram suggests that there was such a change (see Figure 4.1). Even if they do not intend to claim so, it is still worthwhile presenting the scenario for demonstrative purposes. 125

4.6 The Methods of Reconstruction of Sa-gy! and Sha-gy!

There are several ways to reconstruct phonetic values in the past based on historical documents. In the case of Japanese historical phonology, the methods described in (1) are generally used.

(1) a. Reconstruction through Chinese characters. b. Reconstruction through descriptions by speakers in the period of their own pronunciation. c. Reconstruction through the Roman alphabet. d. Reconstruction through Hangul (the Korean orthography). e. Reconstruction through the Sanskrit characters. (Mabuchi 1971:24)

Among the reconstruction methods described in (1), I will concentrate on all of these methods, with the exception of (1d). Aside from these reconstruction methods that depend on historical documents, a method that compares dialectal varieties (Nakata 1972) will also be discussed. The obvious yet crucial problem of these reconstructive methods is that presumed phonetic values used in these methods, such as Chinese pronunciation and Sanskrit pronunciation, are also based on written materials, not audio recordings.

This means that perfect reconstructions are impossible. Consequently, scholars disagree with each other on the reconstructed phonetic values of sa-gy! and sha-gy!. I will not try to determine which reconstructions are correct. Nevertheless, reflecting on the development of sa-gy! and sha-gy! is useful for understanding the cause of the distribution of current sa-gy! and sha-gy!. 126

4.6.1 Reconstruction of Sa-gy! through Chinese characters and Sanskrit

characters (8th century ~ 1333)

The oldest documents that may be reliably used for the reconstruction of historical phonetic values are Kojiki (712),85 Nihon Shoki (720),86 and Man-y!-sh" (759).87 Since

Japanese did not have its own writing system, Japanese texts in these three documents are written in Chinese characters that were used as phonograms at least until the Heian period

(AD 794~1185).88 It is known what characters were used for each word. The reconstructed phonetic values of these characters in Chinese at that time are known as well (Mabuchi 1971).89 The phonetic values of the original Chinese characters used for the transcription of consonants of sa-gy! vary widely, such as [s], [!] (or ["]), [#], and [c]

(Miyake 2003; Mabuchi 1971; Hashimoto 1950; Arisaka 1957).90 Many scholars acknowledge uncertainty about the actual phonetic values of sa-gy! consonants in the 8th century. Hashimoto (1950) states that it is difficult to determine their phonetic values.

Arisaka (1957) claims that the distribution of the phonetic values of sa-gy! consonants is

[#] in front of a back vowel and [s] in front of a front vowel. Mabuchi (1971) reluctantly suggests the possibility that [#] was followed by a back vowel, [s] was followed by a central vowel, and ["] was followed by a front vowel. Miyake (2003), on the other hand,

85 Kojiki 古事記 is the oldest book in Japan. 86 Nihon Shoki 日本書紀 is the second oldest book of Japanese history. 87 Man-y!-sh" 万葉集 is the oldest collection of poetry in Japan. 88 These Chinese characters are called Man-y!-gana. 89 However, as discussed above, Chinese reconstructions do not accurately reflect the actual phonetic values at that time; therefore, the actual phonetic values at that time are still unknown. 90 Miyake uses ["], while Mabuchi uses [!]. Since it is almost impossible to reconstruct the subtle difference between ["] and [!], I use ["] hereafter. 127

proposes the distribution [s] in front of a non-front vowel and [s] or [!] in front of a front vowel, rejecting the idea that allophones of /s/ in the 8th century varied as widely as suggested by Mabuchi.

In the Heian period (AD 794~1185), the historical documents involve both

Chinese and Sanskrit characters. Zait!ki (the early Heian period) and Shittan Kuden (the late Heian period) were written in order to describe the pronunciation of Sanskrit. Both

Zait!ki and Shittan Kuden use the phonogram Chinese characters to transcribe the

Sanskrit pronunciations. The Sanskrit pronunciations at that time had more sounds than

Japanese and when certain Sanskrit sounds are similar to the Japanese pronunciations, they are noted. Zait!ki and Shittan Kuden also provide detailed descriptions of the authors’ introspections of places and manners of articulation (Mabuchi 1971).

Based on the data in Zait!ki, Arisaka (1957) asserts that the sounds of sa-gy! in

91 the early Heian period were ["a !i (or si) !e (or se)]. Mabuchi (1971) also assumes that the consonants of sa-gy! in the early Heian period were either ["], [#], or [!]. Shittan

Kuden, which was written in the late Heian period, says that the consonants of sa-gy! are produced by putting the right and left edges of the tongue against the upper jaw and passing air in between (Arisaka 1957; Mabuchi 1971). Shittan Kuden also reports that when pronouncing Japanese [ta], the tongue tip is put against the upper gum. These descriptions in Shittan Kuden suggest that unlike pronouncing [t$], the tongue tip did not contact the upper jaw when pronouncing [s$]. According to the descriptions above, the

91 Arisaka states that there is no data to be used for the reconstruction of /su/ or /so/ in the early Heian period (1957:159). 128

consonant of the sa-gy! in the Heian period appear to be very close to [!] (Mabuchi

92 1971:69), rather than ["] or [s]. Consequently, Mabuchi infers that sa-gy! in Heian period was pronounced as comparable to ["# "i "u "e "o]. Arisaka (1957:146) claims that the consonants of sa-gy! can be either ["] or [s].

4.6.2 Emergence of Sha-gy!

In current Japanese, palatal consonants followed by a back vowel are called y!on (lit. twisted sound). According to this definition, sha-gy! sounds should be also classified as y!on. It is often stated that the influence of Chinese loanwords contributed to the emergence of y!on (Mabuchi 1971). According to Mabuchi, sa-gy! consonants were already ‘palatalized’ before sha-gy! was established. In the following discussion by

Mabuchi, the emergence of sa-gy! /sa/ and sha-gy! /!a/ are discussed. Mabuchi (1971:70,

95) claims that when Chinese words were first borrowed, Japanese had only ["a], while

Chinese loanwords had [sa] and [si$a]. On the other hand, Japanese speakers also learned how to pronounce [s] by studying Sanskrit and Chinese. Consequently, before the emergence of sha-gy!, the distribution of voiceless coronal fricatives may have been something like shown in (2).

(2) "# "i "u "e "o (Japanese native sounds) s# su so (foreignism: Chinese and Sanskrit)

92 It is important to note that unlike [!], ["] in current sa-gy! does not require putting tongue sides so close to the middle of the palate. 129

When Japanese speakers first began borrowing Chinese words, they used [!"] to pronounce Chinese [s"], and created an innovative sound sequence that is similar to [!ii#a] in order to pronounce Chinese [si#a]. Later, sounds such as [!ii#a] came to be pronounced as [!"], and sa-gy! [!"] came to be pronounced as [s"], which is a learnèd sound sequence. Table 4.6 illustrates the historical shift of the sound [!"] between sa-gy! and sha-gy!.

Table 4.6. The historical shift of the sound [!a] between sa-gy! and sha-gy! Sa-gy! Sha-gy!

Before Chinese influence [!"] (or [$"]) ______

Early Chinese influence [!"] [!ij"]

Later Chinese influence [s"] [!"]

Table 4.6 shows that sha-gy! [!"] in the bottom row is not the ‘palatalized’ version of sa-gy! [s"]. Rather, it appears that sa-gy! [!"] in the earlier period simply shifted to sha-gy!, and the vacant spot was filled with [s"], which was introduced into

Japanese as a foreign sound sequence. Table 4.7 illustrates the change of the distribution in the complete sa-gy! and sha-gy!.

130

Table 4.7. The change of the distribution in the complete sa-gy! and sha-gy! Sa-gy! Sha-gy!

Early Chinese influence [!"] [!i] [!u] [!e] [!o] [!ij"] [!iju] [!ijo]

Later Chinese influence [s"] [!i] [su] [!e] [so] [!"] [!u] [!o]

According to Table 4.7, the sound sequences [!"], [!u], and [!o] were recruited from sa-gy! to sha-gy! and their places were filled with Chinese or Sanskrit origin sound sequences [s"], [su], and [so]. The ancestors of [s"], [s#], and [so] in current sa-gy! were created by ‘de-palatalizing’ sa-gy! [!"], [!u], and [!o] in the early period, and this ‘de- palatalization’ was promoted by the contrast between sa-gy! and sha-gy!. The remaining sa-gy! [!i] and [!e] did not have to go through the ‘de-palatalization’ process because of the lack of the sa-gy! – sha-gy! contrast.

However, other y!on did not go through the complicated process described above.

For example, the contrast between ka-gy! ([k"][ki][ku][ke][ko]) and kya-gy!

([kj"][kju][kjo]) emerged as illustrated in Table 4.8. Notice that the emergence of kya- gy! does not affect the phonetic values of ka-gy! at all.

Table 4.8. The development of ka-gy! and kya-gy! Ka-gy! Kya-gy!

Before Chinese influence [k"][ki][ku][ke][ko] ______

Early Chinese influence [k"][ki][ku][ke][ko] [kij"] [kiju] [kijo]

Later Chinese influence [k"][ki][ku][ke][ko] [kj"][kju][kjo]

131

Mabuchi’s account is intriguing and plausible. However, if all consonants of sa- gy! before the emergence of sha-gy! were [!], the phoneme of the consonants of sa-gy! is /"/, rather than /s/. It would be typologically problematic to have a phoneme /"/ without having a phoneme /s/. This issue will be discussed later in this chapter.

4.6.3 Reconstruction Through the Roman Alphabet

In the mid 16th century, Portuguese missionaries wrote and edited books on Japanese grammar and vocabulary. These documents show transcriptions of Japanese sounds in the

Roman alphabet, and by referring to the pronunciation of Portuguese at that time, reconstruction of the sounds in Japanese is possible.93

These documents demonstrate at least two significant historical facts. First, in

Kyoto, the imperial capital of Japan at that time, the phonetic values of sa-gy! sequences and sha-gy! sequences were [s# !i su !e so] and [!# !u !o] respectively. Second, in the

Kant! region, which is located about 500 km to the east of Kyoto, sa-gy! /se/ was pronounced as [se] instead of [!e]. Edo (present day Tokyo) is located in the Kant! region, and it became the de facto capital of Japan in 1603 (the Edo period: 1603~1867).

The relocation of the capital might have resulted in the spread of the Kant! region phonetic value [se] to Kyoto. Consequently, by the mid-Edo period, [!e] in sa-gy! in the

Kyoto area also changed to [se], and both eastern and western Japan had the same sa-gy!

[s# !i su se so] and sha-gy! [!# !u !o].

93 Among these documents, Vocabvlario da Lingoa de Iapam (a Japanese-Portuguese dictionary; João Rodriguez 1603,), Arte da Lingoa de Iapam (João Rodriguez 1608), Arte Breve da Lingoa Iapoa (João Rodriguez 1620), and Ars grammaticae Iaponicae linguae [Grammar of the Japanese Language] (Diego Collado 1632) have been widely used for the reconstruction of the phonetic values of Japanese in the late 16th and early 17th century (Mabuchi 1971:107). 132

4.7 Variations of Sa-gy!

We have seen that the distributions of sa-gy! were different in the Kansai region and the

Kant! region, and this difference disappeared by the mid-Edo period. Before the Edo period (1603~1867) the sa-gy! distribution was [s! "i su "e so] in the Kansai region and

[s! "i su se so] the Kant! region, and the distribution of modern standard Japanese is [s!

94 "i s# se so]. Nevertheless, this does not mean that the distribution of sa-gy! is uniform throughout Japan. Rather, the sa-gy! distributions still vary.

In certain regions, such as Ky"sh", Shikoku, T!hoku, Hokuriku, and Ch"goku, sa-gy! /se/ is still pronounced as ["e] (Okumura 1972; Sat! 2002). In particular, the distributions of sa-gy! and sha-gy! in the Ky"sh" region vary widely.95 In some areas of

Ky"sh", not only sa-gy! /se/, but also sa-gy! /sa/ in some words contains the consonant

["]. For example, Kambe (2003) studied dialects in Saga, one of the prefectures in

Ky"sh", and provided descriptions of the dialects’ grammar and basic glossaries. In the glossaries, there are words such as ["!$%#] (<[s!i%#] ‘wallet’) and ["!ka&] (<[s!k!&]

‘plasterer’). Nevertheless, the section for grammar in Kambe (2003) does not mention the existence of sa-gy! ["!] at all. Perhaps sa-gy! ["!] examples are not as consistent as sa- gy! ["e], and for this reason they were treated as exceptions and not as a general tendency. In contrast, in some dialects in Ky"sh", sha-gy! consonants are ‘depalatalized’.

In Kagoshima prefecture and in the southern part of Miyazaki prefecture, sha-gy!

94 The current phonetic value of the vowel /u/ in Kansai is different from the one in Kant!: it is rounded and, therefore, closer to [u] than [#]. 95 Ky"sh" is one of the four major islands and is located in the southwest of Japan. 133

consonants are pronounced as [s], such as in [is!do"] (<[i#!] ‘doctor’), [tesu] (<[tei#$]

‘husband’), and [sotsu] (<[#o%&$%] ‘a clear distilled liquor’) (Got! 1961:272).

In the T!hoku region (the northeastern area of Honshu), the high vowels [i] and

96 [$] are centralized to ['(] and [')], and frequently they are merged into ['(] or [')].

Consequently, some speakers of dialects of the T!hoku region do not distinguish sa-gy!

/si/ and /su/, since both of them are pronounced as either [s'(] or [s')] ("hashi 2002).

Arisaka (1957:155) is also aware of this fact, but he concludes that it is difficult to determine whether the phonetic value of the consonant of sa-gy! /si/ in the eighth century was [#] or [s], even though the existence of [s'(] in some dialects could be a potential piece of evidence that [si] was sa-gy! /si/ in the eighth century.

Kajiku (1983) indicates that before the class system was abolished, some speakers of the Shuri dialect in Okinawa used the [si] sequence instead of [*i]. In Shuri, the people in the warrior class and the common people used different phonological systems, as illustrated in (3). The system used by the warriors was “acquired by studying intentionally” (Kajiku 1983:33), although it was based on the system used by common people.97 According to Kajiku’s description, the [si] sequence used by the warriors is an innovative sound sequence. On the other hand, it is also possible that the people in the warrior class had been consciously trying to maintain the traditional [si] sequence, which had been lost in the speech of the common people.

96 ['(] and [')] are “advanced high central unrounded vowel” and “retracted high central unrounded vowel” respectively. 97 Although the warrior class speakers do not exist now, their pronunciations are still observed in traditional plays and songs in Okinawa (Kajiku 1983:33). 134

(3) Warrior Class Common People Gloss [sina] [!ina] sand [si"si"] [!i"!i"] soot [si"ju#] [!i"ju#] to attach [dza"] [d$a"] seat [midzi] [mid$i] water [ndzu] [nd$u] ditch [tsukuju#] [t!ukuju#] to make (Source: Kajiku 1983:35)

The dialectal evidence shown in (3), along with the historical evidence discussed in the previous section, suggests that the consonant of sa-gy! before the 16th century may have been either [!] or [s]. In the next section, I will discuss two extreme scenarios that can be deduced from the historical and dialectal evidence.

4.8 Two Scenarios

4.8.1 The First Extreme Scenario: Consonants of Sa-gy! Were [s]

The first scenario assumes that sa-gy! changed from [s% si su se so] to [s% &i su &e so], and finally to [s% &i s' se so]. In many languages, [s] is palatalized in front of a high front vowel; thus, the change from [si] to [&i] is consistent typologically and historically

(Ito and Mester 2007; Blevins 2004). The crucial assumption of the first extreme scenario is that sa-gy! /se/ was also ‘palatalized’ in the Heian period. As shown in Figure 4.1, Ito and Mester (2007) imply that a change from [se] - [&e] - [se] took place.

135

平安室町江戸現代 Heian Muromachi Edo Modern 900 1100 1300 1500 1700 1900 2000 so ! so !o ! !o se se se (!e) !e ( ) !e (si) !i !i !i su su !u !u sa sa !a !a

(Source: Ito and Mester 2007:676) Figure 4.1. Historical development of sibilants: time table of mergers

Notice that, strictly speaking, Ito and Mester do not support the first extreme scenario, because they do not seem to accept the change from [si] to ["i] ([!i] in Figure

4.2). However, since they maintain that [e] acts as a palatalizer, I will assess their analysis in this section.

Based on the assumption that [e], as well as [i], used to function as a palatalizer,

Ito and Mester (2007) claim that Japanese /se/ ["e] in the Pre-Edo system has been replaced by /se/ [se] in the modern system so that the overall system of contrasts can be improved. They assert that the contrast between ["i] and ["e], where both [i] and [e] act as palatalizers, is smaller than ["i] and [se], where only [i] is a palatalizer. Consequently, in order to increase the contrast, ["e] became [se] by suppressing the palatalizing function of

[e]. Ito and Mester (2007) conclude that the current system is better than the pre-Edo 136

system because the contrast between [!i] and [se] is larger than the one between [!i] and

[!e]. However, I do not agree with their claim that the current system is better than the pre-Edo system, because it implies that the current sandard Japanese system is even better than the system of the dialects that uses [!e] for sa-gy! /se/. Even if their claim is right, we still cannot find any motivation for sa-gy! /se/ to be ‘palatalized’ in the Heian period.

4.8.2 The Second Extreme Scenario: Consonants of Sa-gy! Were [!]

The other extreme scenario supposes that all sa-gy! consonants were ‘palatalized’ already and they were ‘depalatalized’ as new sound sequences, such as sha-gy!, were introduced. The second scenario assumes that sa-gy! changed from [!" !i !u !e !o] to [s"

!i su !e so], and finally to [s" !i s# se so]. As we have seen above, Mabuchi (1971) takes this position. If all sa-gy! consonants were [!], and [s] was never realized in the pre-

Heian system, there is no need to accept the existence of the phoneme /s/. Instead, we can safely assume that the phoneme for sa-gy! consonants at that time was /$/. The problem with this scenario is whether a language can have a phoneme /$/ without having a phoneme /s/. In the following section, I will further elaborate these two extreme scenarios by taking the distribution of /s/ in the Ainu language into consideration. I will also discuss articulatory characteristics of [s] and [$] in order to illustrate the cause of the difficulties involving Japanese voiceless coronal fricatives.

137

4.9 /s/ in Ainu

Ainu is spoken by the Ainu in Hokkaido, the northern island of Japan. Shibatani states that “Ainu is best described as a language-isolate” (1990:5). But the genetic relationship between Ainu and Japanese is still unclear (see Shibatani 1990 and works cited there; see also Greenberg 2000). In this dissertation, I take the position that Ainu is not genetically related to Japanese in order to avoid reaching for a simple and easy conclusion that current Ainu reflects the phonology of Old Japanese.

In Ainu, “(t)he fricative /s/ is realized either as [s] or [!]; the [!] sound consistently before [i], and in syllable-final position” (Shibatani 1990:11). Hashimoto (1950:110) states that some Ainu speakers do not distinguish [s] from [!] or ["]: they sound the same to the speakers of Ainu. The [s/!] confusion in Ainu can be observed in some words borrowed from Ainu. Some Ainu words with /s/ have been borrowed into Japanese lexicon, and Ainu /s/ often corresponds to /!/ in Japanese. For example, the loanword

["i"#mo] (

Japanese speakers. K!jien implies that the Japanese word [s#ke] or ["#ke] ‘salmon’ might originate with Ainu.98 If this assumption is correct, then it suggests that the Ainu /s/ in this word was interpreted as either /s/ or /!/ by Japanese speakers.

These pieces of evidence indicate that Ainu /s/ ranges from [s] to [!], regardless of the following vowel. In other words, actual pronunciation of Ainu /s/ does not depend on a phonological rule, such as /s/ becomes [!] in front of a high front vowel. Rather,

98 Even if [s#ke/"#ke] is not from Ainu, this word demonstrates the compatibility of [s#/"#] in modern standard Japanese. Hamada (1964) suggests that ["#ke] might reflect the pronunciation of Old Japanese. 138 pronunciation patterns vary among individual speakers. In the next section, I will review articulatory and acoustic similarities and differences between [s] and [!].

4.10 Articulatory And Acoustic Account

The confusion found in Japanese voiceless coronal fricative might be due to the fact that the articulatory gestures and acoustic signals are not in a linear relationship. The Quantal

Theory, proposed by Stevens (1972, 1989), maintains that the relationships between changes of articulatory gestures and output acoustic signals are non-linear. That is, when the shape of the vocal tract changes continuously from one phone to another, acoustic signals show a sudden abrupt change at the border of two phones. Stevens (1972, 1989) proposes that the inventory of sounds of a language is the collection of the stable acoustic regions that are divided by the unstable portions. In the case of English [s] and [!], [s] has a higher low-frequency cutoff, whereas [!] has a lower low-frequency cutoff (Nolan,

Holst, and Kühnert 1996; Perkell, Boyce, and Stevens 1979). Perkell, Boyce, and Stevens

(1979) note this distinct difference in the frequency of the lowest major spectral peak in between English [s] and [!], claiming it is due to the difference in distance between the tongue tip and the mandibular alveolar ridge. The results of their experiments demonstrate that there is a strong contact for [s], and little or no contact for [!]. They suggest that the cause of sudden change at some point in between [s] and [!] is the abrupt change of the space below the tongue blade when the tongue tip breaking away from the 139 lower teeth.99 Ladefoged and Maddieson (1996) claim that the articulation points can be either apical or laminal in English [s] and [!]. However, the illustrations drawn based on the X-ray photographs of the articulators of Peter Ladefoged clearly show that the tongue tip is used when pronouncing [s], and the tongue blade is used when pronouncing [!].

Nolan, Holst, and Kühnert (1996), on the other hand, speculate that the use of tip and blade varies across English speakers. Although tongue tip and tongue blade—along with alveolar and post-alveolar—are major articulators for [s] and [!], other articulators such as the lips and tongue body are also involved in the production of [s] and [!] in English.

Ladefoged (2001) indicates that pronouncing an English [!] requires a significant lip rounding while pronouncing [s] does not. Ladefoged and Maddieson (1986) use X-ray pictures of Peter Ladefoged, whose native language is British English, to illustrate the articulatory differences between [s] and [!]. According to the illustration, the tongue shape of [s] is “grooved,” while [!] is “domed.” Ultrasonic images created by Stone et al.

(1992) show an anterior midsagittal groove when pronouncing [s]. But when pronouncing

[!], a midsagittal groove appears in the posterior area of the tongue.

As previous studies show, articulatory gestures for pronouncing English [s] and

[!] are different in terms of lip rounding, tongue shape (domed/grooved), and articulation point of the tongue (apical/laminal). Among the studies above, we can point out some inconsistencies in the articulatory gestures. For example, the use of tongue tip/blade is

99 Nolan, Holst and Kühnert (1996), however, claim that the Quantal Theory alone is insufficient to account for deletion and blending. 140 not consistent among the studies. Nevertheless, the articulatory gestures of English [s] and [!] seem always to be distinct.

Compared with English, the articulatory gestures of Japanese [s] and ["] are more similar. In Japanese [s] and ["]: (1) neither requires lip rounding; (2) both are laminal; and (3) the tongue shapes are relatively flat. The only noticeable differences between

Japanese [s] and ["] are that [s] is alveolar, while ["] is alveolo-palatal, and significantly palatalized. But such differences in place of articulation or palatalization may be observed in other sounds in other languages, such as [k#] in key in English (Ladefoged

2001:218).

Perkell, Boyce, and Stevens (1979) report that subtle articulatory differences make an abrupt change in acoustic signals. Japanese [s] and English [s] have strong contact in alveolar ridge, and little or no contact for Japanese ["] and English [!].

According to Perkell, Boyce, and Stevens, these articulatory characteristics create the

“Quantal” effect by producing an abrupt change in acoustic signals. If the tongue tip/blade contact and the acoustic signals created by the gesture are the primary and adequate information for the distinction between Japanese [s] and ["], (or English [s] and

[!]), then why are there distinct articulatory gestures in English [s] and [!], such as lip rounding and tongue body doming, even though such gestures are apparently irrelevant to

“quantizing” phones? I hypothesize that the apparently irrelevant articulatory gestures actually facilitate “quantization,” although they do not seem to affect crucial acoustic 141 signals much.100 To support this hypothesis, I introduce Articulatory Phonology.

Articulatory Phonology, developed by Browman and Goldstein (1992), maintains that gestures, not acoustic signals, are the primary source of speech perception. Whether articulatory gestures can be the primary source of speech perception instead of acoustic signals is debatable. But if we suppose that both acoustic signals and articulatory signals can be the essential sources of speech perception, we can hypothesize that there are cases in which a lack of sufficient cues from either acoustic signals or articulatory gestures might result in perceptual confusion. If the hypothesis I propose is correct, the reason for confusion of voiceless coronal fricatives is that, because the gestural differences between

Japanese [s] and [!] are so small, they may not be adequate for the distinction, although the difference in the acoustic signals is present. In the case of English, lip rounding gestures and tongue body doming gestures might not produce the primary acoustic cues to distinguish [s] and ["]. But, if articulatory gestures create cues for speech perception, as well as acoustic signals, these “extra” articulatory gestures surely make a significant difference in the distinction of these two phones: they enhance the distinction. Lack of such “extra” articulatory gestures in Japanese [s] and [!] may result in a failure to distinguish these two phones clearly. This confusion is most evident in the sound sequences [si] and [!i].

If [s] and ["] are difficult to distinguish with the primary acoustic signals only, sa- gy! consonants in the 8th century might have been unstable. When there is no phonemic

100 However, lip rounding significantly lowers the average frequency (Johnson 2003:97). 142

distinction between /s/ and /!/, the allophones may vary from [s] to [!]. In extreme cases, some speakers may use [s] only and some may use [!] only. Sa-gy! consonants in the 8th century might have been not all [s] or all [!], but rather varied greatly depending on the words, situations, and speakers. This scenario is close to the situation of Ainu. As discussed above, Ainu speakers do not distinguish [s] and [!]: they are regarded as one phoneme /s/ since Ainu does not have a phoneme /!/. Nonetheless, some speakers use [!] exclusively and some do not. The various patterns of allophonic distribution of Ainu /s/ suggest that there could have been a similar situation in 8th century Japan. Note, however, I do not maintain that Ainu is genetically related to Japanese. I am simply using

Ainu as an example that [s/!] free variation does exist in a current language.

4.11 Why Is ["e] Prevalent While [si] Is Still a Foreignism?

We have seen the extremely unbalanced distribution of ["e] and [si] in current standard

Japanese: ["e] is widely used while [si] is rare. Not only ["e], but all the other potential innovative sound sequences except for [si] have been accepted. Why is [si] still a foreignism, whereas ["e] is not?

Traditional OT explains that there are constraints which prohibit certain sound sequences in a language. Ito and Mester (1995b) propose the constraint *!E and *SI. The constraint *!E, which is shown in (4), restricts palatal segments preceding the mid front vowel [e]. The constraint *!E, therefore, rules out the sound sequence ["e], [c"e], and

[#e].

143

(4) *!E: No sequence [alveolopalatal C] + [e] (Ito and Mester 1995b:829)

The constraint *SI, on the other hand, prohibits non-palatal coronal fricatives before [i].

The unbalanced distribution of [!e] and [si], therefore, can be explained by saying that the constraint *SI is ranked higher than *!E. This solution depicts the facts of Japanese correctly. However, I strongly doubt the existence of the constraint *!E. I argue that the issue of Japanese voiceless coronal fricatives is not grammatical, but historical.

I claim that the constraint *!E may not exist, whereas the existence of the constraint *SI is well supported by linguistic typology and historical linguistics. Ito and

Mester (1995b) regard the constraint *!E to be legitimate based on the fact that the !E sequences appear only in recently borrowed foreign items and some native exclamations that they classify as “Peripheral.” Nonetheless, these facts are not adequate to support the claim that the constraint *!E actually exists. I consider a counterexample. The sound sequence [mj"] only appears in recently borrowed foreign items (Kindaichi 1988;

Yuzawa and Matsuzaki 2004). Nonetheless, to my knowledge, no scholar has proposed a constraint that prohibits [mj"]. Why do linguists treat [mj"] and [!e] differently, while these two sound sequences show similar distributions? Native Japanese speakers know the answer to the question: the sound sequence [mj"] is included on the list of kana that is used for elementary school education (Mabuchi 1996). However, the list does not include innovative sound sequences, such as [!e], [c!e], and [#$e]. Thus, Japanese linguists do not regard [mj"] as an innovative sound sequence, not because of linguistic facts, but because it is on the list of kana. If we do not have to establish a constraint that 144

restricts [mj!], then there is no legitimate linguistic reason to set up the constraint *!E because there is no significant distributional difference between [mj!] and ["e]. The reason the sound sequence [mj!] is missing in non-foreign items is simply accidental. In contrast, the reason ["e] is found only in recently borrowed items is historical. As we have seen, in Kyoto, the change from [#e] to [se] took place some time during the Edo period—in Kant" area, this change took place even earlier. We cannot find ["e] in non- foreign items because all the non-foreign items that contain the [#e] sequence changed into words with [se] before the 18th century. Thus, lack of ["e] in non-foreign items is due to the historical change. However, this change does not necessarily mean that the speakers lost the grammatical ability to produce ["e]. That is why when ["e] was reintroduced to Japanese, it was accepted so easily. There are several pieces of evidence that support the claim that Japanese speakers kept the grammatical ability to pronounce

["e], [c"e], and [$e]. As discussed earlier in this chapter, young children often produce ["] or [c"] instead of [s] (Kamei 1984; Beckman, Yoneyama and Edwards 2003; Tsurutani

2004), and adults also palatalize [s] to mimic young children. Ito and Mester (1995b) state that sound sequences ["e] and [c"e] appear in some native words, such as [c"e%]

(swear word) and ["e&] (exclamation used by famous cartoon figure).

As we have seen, the lack of ["e] in non-foreign items is due to historical facts, and not due to changes in grammatical ability to produce ["e]. In contrast, there is no document that suggests [si] went through the change from [#i] to [si]. The sound sequence

[si] may have been absent in Japanese for a much longer period than ["e]. Moreover, the sound sequence [si] might have never appeared in the history of the Japanese language. 145

Palatalization of /s/ when adjacent to a high front vowel happens over and over in the languages of the world (Blevins 2004:142). Typologically and historically, palatalization of Japanese /s/ in front of a high front vowel is not special at all. The historical differences between [!e] and [si] may be the cause of the difference in the distribution of

[!e] and [si].

The view I have presented may not be welcomed by OT scholars who study

Japanese phonology. Ito and Mester state that if morpheme classifications such as

Yamato, Sino-Japanese, Foreign and Mimetic “were nothing more than a record of etymological history, they would be of little linguistic interest. However, as is familiar from the classical linguistic literature on the subject, they require explicit synchronic recognition if, and as far as, they continue to play a role in the grammar” (1999:63-64). In order for a phonological phenomenon to play a role in the grammar, such a phenomenon must not involve changes in input, because by definition, input must be free from any constraints (Prince and Smolensky 2004; Kager 1999). This means if a phonological phenomenon is created by a change of input, synchronic grammar (i.e. OT) cannot account for the phenomenon. Therefore, if the lack of certain sound sequence in certain vocabulary items is caused by the change of the vocabulary items (i.e. input), then it must be explained historically. Constraints or their rankings cannot affect the change of inputs because “no constraints hold at the level of underlying forms” (Kager 1999:19). On the other hand, Evolutionary Phonology, a more recent phonological theory, maintains that historical accounts should precede synchronic accounts (Blevins 2004). Studying 146 historical linguistics might help synchronic linguistics determine what in fact plays a role in synchronic grammar.

Notice, however, I do not claim that OT is wrong or inferior to Evolutionary

Phonology. I simply claim that OT may not be suitable for solving most phonological issues of Japanese innovative sound sequences, because these issues involve language change. Other issues, such as syllable structure, may be appropriate for OT analyses.

4.12 Why [si] Is Still a Foreignism

I have discussed why [!e] is more prevalent than [si]. Lack of the sequence [!e] in non- native items is due to historical facts, whereas the palatalization of [s] followed by a high front vowel is a universal typological tendency. However, this does not explain why [si] is still a foreignism. In this section, I will argue that [si] is avoided by many speakers for cultural reasons.

As discussed in the previous section, the constraint *SI may have been involved in the palatalization of sa-gy! /si/. However, current Japanese speakers—in particular younger speakers—should be able to produce both [si] and [!i]. The writing system of

Japanese has an appropriate method to transcribe the sequence [si] by combining ス

([s"]) and ィ (a small-size イ [i]). This evidence suggests that most Japanese speakers do not have much trouble producing and perceiving the sequence [si]. In other words, there is no phonological, articulatory, or perceptual constraint that prohibits [si] from occurring. 147

The sequence [si] is used when American or British culture is important in the context. Most junior high school students study English at school. They learn how to differentiate, for example, she and see. They are also familiar with Japanese popular songs that include several English words and phrases with the [si] sequence. The [si] sequence is used in many Japanese popular songs targeting youth. The audience may also sing these songs, and thereby produce the sequence [si].

People in certain positions probably use the sequence [si] more frequently than ordinary people. Some TV and radio newscasters and commentators use the [si] sequences when saying foreign items. It is important to note that newscasters are professionals in speech performance, and some TV commentators are experts in international affairs. In other words, they are prestigious speakers. Kiko, Princess

Akishino, clearly pronounced [si!di!] ‘CD’, as opposed to the normal pronunciation

rd ["i!di!], on Prince Akishino’s 43 birthday on November 30, 2008. Needless to say, the members of the royal family are perceived to be prestigious.

The sequence [si] is also used when trying to be humorous. Saying [mond#i n#ssi!$%] ‘there is no problem’ instead of [mond#i n#"i] creates a pun ([n#ssi!$%]

’nothing’ and [n#"i] ‘there is not’). The use of the sequence [si] in [sek%si! b%c"o!]

‘Sexy Manager’ (NHK 2007), a sketch in a television program, also generates comical effects. To sum up, the sequence [si] is used: (1) when the context is Western; (2) when the speakers are prestigious; or (3) when the speakers are trying to be humorous.

Finally, let us consider what would be anticipated when the sequence [si] is used in the wrong context. Since [si] is mostly used when the context is related to Western 148 culture, the speakers may feel anxiety about the loss of Japanese identity. The use of [si] by non-prestigious speakers may generate pretentiousness or unwanted ludicrousness.

Misuse of the sequence [si] contains these sociocultural risks. Japanese speakers may restrict the use of [si] in order to avoid these risks.

4.13 Conclusion

In this chapter, I investigated the distributions of the innovative sound sequences [si] and

[!e]. The sound sequence [!e] is widely accepted, whereas [si] is still a foreignism.

Nevertheless, the sound sequence [si] exhibits strong sociolinguistic effects, which suggests that Japanese speakers can at least recognize the phonetic difference between

[!i] and [si]. We still do not know exactly why [si] remains a foreignism. However, we know that the phonetic difference between [!i] and [si] plays a significant sociocultural role. The current situation of the sequence [si] poses an interesting question to phonologists, as well as phoneticians, because although speakers can recognize the phonetic difference, phonemically, the sequences [si] and [!i] are “free” variants. In order to solve the issue of the sequence [si], a collaboration of linguists from different fields is essential. 149

CHAPTER 5

JAPANESE [ti], [t!], [di], AND [d!]

5.1 Introduction

Innovative sound sequences that include Japanese alveolar stops are [ti], [t!], [di], [d!],

[t"j!], and [d"j!]. In this chapter, I will concentrate on [ti] and [t!], as well as their voiced counterparts [di] and [d!]. I will discuss the causes of the emergence of these innovative sound sequences and the causes of unbalanced distributions between [ti]/[di] and [t!]/[d!]. Modern standard Japanese has [t] and [d] in its phonemic inventory, but allows [ti], [t!], [di], and [d!] sound sequences only in foreign items. In the traditional fifty-sound display of the kana syllabary, the columns for /t/ and /d/ (henceforth ta-gy! and da-gy!, respectively) demonstrate defective distributions, as illustrated in Table

101 5.1.

Table 5.1. Distribution of ta-gy! and da-gy! sound sequences a i u e o ta-gy! t# タ c$i チ %! ツ te テ to ト da-gy! d# ダ &'i ('i) ヂ (! (z!) ヅ de デ do ド innovative sound ti ティ t! トゥ sequences di ディ d! ドゥ

101 The traditional fifty-sound display arranges vowels in rows and consonants in columns. Although gy! means ‘column’ I arrange them as ‘rows’ for illustrative purposes only. 150

The phonetic value of the consonants [c!], ["], [#$], and [%] are the following:

[c!] (voiceless alveolo-palatal affricate); ["] (voiceless alveolar affricate); [#$] (voiced alveolo-palatal affricate); and [%] (voiced alveolar affricate). As shown in the table above, instead of [ti] and [t&], [c!i] and ["&] occupy the spaces for ta -gy! /ti/ and /tu/.

Similarly, [#$i($i)], and [dz&/(z&)] fill the spaces for da -gy! /di/ and /du/, respectively.

It may appear typologically uncommon to consider three different consonants ([t], [c!], and ["]) to be in the same phonemic group, namely ta-gy!, but palatalization of [t] can also be observed in other languages, such as Brazilian Portuguese (Yoshida 2001).

Shibatani (1990) and Tsujimura (1996) regard [c!] in [c!i], as well as ["] in ["&], as allophones of /t/. Hattori (1960), however, claims that [c!i] and ["&] are phonemically different from /t/, proposing a new phoneme /c/ for allophones [c!] and ["]. In this dissertation, I avoid phonemic descriptions as much as possible. When I need to use phonemic descriptions, I refer to the gy! (columns in the traditional 50 sound display).

For example, the phonemic descriptions for [c!i] and ["&] are referred to as ta -gy! /ti/ and ta -gy! /tu/.

There are two questions I would like to address in this chapter. The first question is about the motivation for the emergence of the innovative sound sequences [ti] and [t&].

Hattori (1960) claims that the motivation for the emergence of [ti] and [t&] is to fill in the structural gaps of ta-gy! sequences. However, [ti] and [t&] seem to behave differently from other sound sequences with [i] and [&]. I will discuss the traits of [ti] and [t&], focusing on the “devoiceability”—the possibility of devoicing—of their vowels. 151

The second question is about the unbalanced distributions of [ti]/[di] and

[t!]/[d!]. The innovative sound sequences [ti]/[di] and [t!]/[d!] have common features; they both may have existed in Japanese in the past and their consonants are similar phonetically. They are historically similar as well. They were all replaced by their affricated counterparts by the mid 16th century (Mabuchi 1971) and reappeared in the late 19th century. Nevertheless, there are differences in the way they appear in non- marginal Japanese vocabulary items: [ti]/[di] are much more prevalent than [t!]/[d!].

On the other hand, in marginal items—such as song titles, brand names, and movie titles—words and phrases with [t!]/[d!] sequences are much more common than in non- marginal items. The disproportional distribution of [ti]/[di] and [t!]/[d!] cannot be explained phonetically or typologically. I argue that the disproportional distribution of

[ti]/[di] and [t!]/[d!] is due to the cultural perspectives of Japanese speakers. I also discuss how knowledge of the source words affects the distribution patterns of [ti]/[di] and [t!]/[d!].

5.2 The Diversity of Ta-gy! /Da-gy! Consonants and Confusion in the Writing

System

The ta-gy! consonants consist of three distinct consonants: [t], [c"], and [#]. This diversity caused confusion in the writing system. The consonant of ta-gy! /ti/ is [c"]

(voiceless alveolo-palatal affricate) and the consonant of /tu/ is [#] (voiceless alveolar affricate), whereas the rest of the consonants are [t] (voiceless alveolar stop). However, historical documents suggest that the consonants of ta-gy! were all [t] in the past. 152

According to Mabuchi (1971), the changes from [t] to [c!] and ["] in ta-gy! /ti/ and ta- gy! /tu/, respectively, as well as affrication of da-gy! /di/ and da-gy! /du/, took place sometime at the beginning of 16th century. These changes may have contributed to the emergence of the well-known phonological confusion called yotsugana (lit. the four kana). Yotsugana is the merger of [#i]/[z$] (ジ/ズ) and [%#i]/[dz$] (ヂ/ヅ), the voiced counterparts of [!i]/[s$] (シ/ス) and [c!i]/[ts$] (チ/ツ) respectively (Shibatani 1990;

Mabuchi 1971). The historical documents suggest that the mergers began at the beginning of the 17th century (Mabuchi 1971). Table 5.2 and Table 5.3 contrast the distributions of sequences in sa-gy! and ta-gy! and the distributions of sequences in za- gy! and da-gy!.

153

Table 5.2. Distribution of the sound sequences in sa-gy! and ta-gy! sa-gy! s! サ "i シ s# ス se セ so ソ ta-gy! t! タ c"i チ $# ツ te テ to ト innovative ti ティ t# トゥ

Table 5.3. Distribution of the sound sequences in za-gy! and da-gy! za-gy! /za/ /zi/ /zu/ /ze/ /zo/ %!(z!) &'i ('i) %# (z#) %e(ze) %o(zo) ザジズゼゾ da-gy! /da/ /di/ /du/ /de/ /do/ d! &'i ('i) %# (z#) de do ダヂヅデド innovative di ディ d# ドゥ

As demonstrated in Table 5.3, the katakana transcriptions of [&'i ('i)] and [%#

(z#)] are problematic: [&'i ('i)] is shared by za-gy! /zi/ and da-gy! /di/, and [%# (z#)] by za-gy! /zi/ and da-gy! /di/. Let us concentrate on [&'i ('i)] for now. The description

“[&'i ('i)]” means that the speakers do not discriminate between [&'i] and ['i]. The consonant [&'] is a voiced alveolo-palatal affricate, and ['] is a voiced alveolo-palatal fricative. Therefore, phonetically, [&'i] is a voiced counterpart of [c"i] (ta-gy! /ti/) チ, and

['i] is a voiced counterpart of ["i] (sa-gy! /si/) シ. Nevertheless, since current speakers of standard Japanese cannot discriminate between [&'i] and ['i], the difference between these sound sequences is treated as allophonic, rather than phonemic. It is important to 154

note that “allophonic” does not mean that [!"i] and ["i], as well as [#$] and [z$], are not in complementary distribution. Inozuka and Inozuka (2003) state that, broadly speaking, the sound sequences [!"i] and [#$] tend to appear word initially and after /%/ or as a second consonant of a gemination, while ["i] and [z$] appear elsewhere. Vance (1987) provides descriptions of more precise tendencies as follows: (1) [#] appears word initially and after the nasal, and [z] appears intervocalically, and word initially when not preceded by a pause; (2) on the other hand, ["] does not appear in careful speech and perhaps occurs intervocalically in rapid speech (Vance 1987:20, 24-25).

Before the merger, there were contrasts between [!"] and ["], and [#] and [z]. In current standard Japanese, however, [!"] and ["] have lost distinction and they are both pronounced as [!"] or ["], depending on the context, as shown in Table 5.4. The same applies to [#] and [z], which have been merged to [#]/[z].

Table 5.4. Development of yotsugana

Old Japanese 16th century 17th Century~Present

ta-gy! [ti] [tu] [c&i] ['u] [c&i] ['u]

da-gy! [di] [du] [!"i] [#u] [!"i ("i)] [#$ (z$)] za-gy! ["i] [zu] ["i] [zu]

The yotsugana mergers influenced the kana writing system when it was revised after World War II. In the pre-WWII system, ヂ (da-gy! /di/) and ヅ (da-gy! /du/) were used more often than ジ (za-gy! /zi/) and ズ (za-gy! /zu/) to transcribe [!"i ("i)] and [#$ 155

102 (z!)] in foreign items. However, after the promulgation of the Modern Kana Usage in

1946, the katakana ジ (za-gy! /zi/) and ズ (za-gy! /zu/) came to be used to transcribe ["#i

103 (#i)] and [$! (z!)], respectively (Seeley 1991). The use of the katakana ヂ and ヅ

(/di/ and /du/) in current Japanese is very limited. Two popular examples are ビルヂング

‘building’ and the brand name ブリヂストン ‘Bridgestone’. The correct katakana

104 spelling for the first example ビルヂング ‘building’ is ビルディング. However, many office buildings in Japan have names that include the word ビルヂング (birujingu), not ビルディング (birudingu). The latter case ブリヂストン ‘Bridgestone’ should be transliterated as ブリジストン (burijisuton) according to the current transliteration rules.

The use of the katakana ヅ (da-gy! /du/) is rarer.105 The confusion in the writing system might have affected the disproportional distributions of [ti]/[di] and [t!]/[d!]. This issue will be discussed in the following section.

5.3 The Recent Development of Japanese [ti], [t!], [di], and [d!]

The development of loanwords and innovative sound sequences in the 20th century has been well-documented. The Japanese government has been trying to standardize

102 Yotsugana transcription in hiragana is more complicated even in the current system; therefore, it will not be discussed. 103 However, in 1926 the Interim Committee on The Japanese Language had already proposed that ジ ["#i] should be used for the sounds that have been traditionally transliterated as ジ ["#i] and ヂ ["#i]. 104 I do not know how ビルヂング ‘building’ is actually pronounced by Japanese speakers. It is probably spelled that way in order to be pronounced as [bi%!di&'!], which is the same pronunciation as ビルディング. But I suppose that many Japanese, including myself, cannot help pronouncing it as [biɾɯɟʑiŋɡɯ] because of the katakana spelling that suggests voiced [cɕi] チ. 105 On the other hand, the hiragana counterparts ぢ ["#i] /[#i] and づ [dz!] /[z!] are used much more frequently, primarily to reflect voicing relationships. For example, はな [h(n(] ‘nose’ + ち [c)i] ‘blood’ = はなぢ [h(n("#i] ‘nosebleed’. 156 fluctuating katakana spellings of loanwords since the early 20th century. The Ministry of

Education established standard rules for transliterating loanwords in 1902, and the rules were revised several times by multiple institutes. The official rules in the early 20th century clearly intended to reflect the spellings of the source words, at least in terms of /t/ and /d/. However, subsequent rules are more focused on reflecting how most Japanese speakers actually pronounce loanwords. Of course no transliteration methods can transcribe actual pronunciation accurately. But the transitions of the katakana spelling methods and the reasons for their modifications reveal how [ti], [t!], [di], and [d!] have penetrated into Japanese phonology. Table 5.5 summarizes the various transliteration rules regarding [ti], [t!], [di], and [d!] suggested in the 20th century.

157

Table 5.5. The history of standardization of katakana spellings of loanwords INSTITUTION SUGGESTED RULES EXAMPLES a. The Ministry of Use チ [c!i], ヂ["#i], チグリス [c!i'%(is%] ‘Tigris’ Education ツ[$%], and ヅ [&%] ヂッケンズ ["#ikkendz%] ‘Dickens’ (proper names: instead of ティ [ti], デツールーズ [$%)(%)z%] ‘Toulouse’ plan) (1902) ィ[di], トゥ[t%], and カトマンヅードゥ[d%] [k*tom*n&%)] ‘Kathmandu’ b. The Historical ヂゥナン (ME) ヂュナン Society of Japan (HSJ) ["#%n*+] ‘Dunant’ (proper names: plan) (1914) c. Interim Use ジ ["#i] for sounds ラジオ [(*"#io] ‘radio’ Committee on the that have been ビルジング [bi(%"#i!'%] ‘building’ Japanese Language traditionally ジフテリア ["#i,%te(i*] ‘diphtheria’ (1926) transliterated as ジ ["#i] エジプト [e"#ip%to] ‘Egypt’ and ヂ ["#i].

Use ジュ ["#%] for ラジューム [(*"#%)m%] ‘radium’ sounds that have been イリジューム [i(i"#%)m%] ‘iridium’ traditionally transliterated as ジュ ["#%] and ヂュ ["#%].

Use チ [c!i] for sounds チップ [c!ipp%] ‘tip’ that have been ニコチン [nikoc!i+] ‘nicotine’ traditionally チーク [c!i)k%] ‘teak’ transliterated as ティ [ti] and チ [c!i].

158

INSTITUTION SUGGESTED RULES EXAMPLES d. Deliberative Use ト [to] and ド [do] ゼントルマン Council on The to transliterate the [%ento&$m'(] ‘gentleman’ National Language sounds /tu/ and /du/ in ブレーントラスト (Notation of the source words [b$&e)n to&'s$to] ‘brain trust’ Loanwords) ドライブ [do&aib$] ‘drive’ (1954) ドラマ [do&'m'] ‘drama’ Exceptions: ツーピース [*$)pi)s$] ‘two-piece’ ツリー [*$&i)] ‘tree’ ズック [%$kk$] ‘doek (Dutch)’ ズロース [%$&o)s$] ‘drawers’

It is preferred to use チチーム [c!i)m$] ‘team’ [c!i], ジ ["#i] to チンキ [c!i!ki] ‘tinctuur (Dutch)’ transliterate the sounds ラジオ [&'"#io] ‘radio’ /ti/ and /di/ ジレンマ["#i&emm'] ‘dilemma’

When there is ティー [ti)] ‘tea’ conscious awareness of ビルディング the pronunciations of [bi&$di!+$] ‘building’ the source words, using ティ [ti] and ディ [di] is allowed.

It is preferred to use チスチュワーデスュ [c!$], ジュ ["#$] to [s$c!$,')des$] ‘stewardess’ transliterate the sounds *ステュワデス /tu/ and /du/. チューブ [c!$)b$] ‘tube’ *テューブ

ヂ ["#i] and ヅ [%$] ジュース are removed from the ["#$)s$] ‘deuce’ list of katakana. ジュラルミン ["#$&'&$mi(] ‘duralumin’ Note: Kobayashi Hideo *デュラルミン mentioned that [!e] and Exception: ["#e] are pronounceable プロデューサー for the Japanese. [p$&odj$)s')] ‘producer’

159

INSTITUTION SUGGESTED RULES EXAMPLES e. Cabinet notice The katakana トゥトゥールーズ (Notation of [t!] and ドゥ [d!] [t!$%!$z!] ‘Toulouse’ Loanwords) (1991) correspond to the ハチャトゥリヤン sounds /tu/ and /du/, [h&c'&t!%i&(] ‘Khachaturian’ which are found in ヒンドゥー教 loanwords. [hind!$ kjo$] ‘Hinduism’

Generally, they can be Alternate General Transliteration: transliterated as ツツアー ["!] ズ [#!] or ト ["!&$] ‘tour’ [to] and ド [do]. ツーピース ["!$pi$s!] ‘two-piece’ ツールーズ ["!$%!$z!] ‘Toulouse’ ヒンズー教 [hin#!$ kjo$] ‘Hinduism’ ハチャトリヤン [h&c'&to%i&(] ‘Khachaturian’ ドビュッシー [dobj!''i$] ‘Debussy’ Sources: a & b) http://www.bunka.go.jp/kokugo/pdf/09_big_01.pdf (accessed August 6, 2009) 外国地名人名呼称一覧（大正三年９月）史学会 c) http://www.bunka.go.jp/kokugo/pdf/06_big_02.pdf (accessed August 6, 2009) 外国語の写し方（仮名遣い改定案補足） d) http://www.bunka.go.jp/kokugo/pdf/kokugo_series_027.pdf (accessed August 6, 2009) 外来語の表記外来語の表記について（昭和２９年３月）国語審議会 e) http://www.bunka.go.jp/kokugo/frame.asp (accessed August 6, 2009) 国語施策情報システム国語表記の基準外来語の表記 II 第2表に示す仮名に関するもの（内閣告示・内閣訓令）

The oldest document available from the web site of the Agency for Cultural

Affairs106 is the List of Foreign Proper Names by the Historical Society of Japan in 1914.

This document compares the suggestions for katakana spelling rules by the Ministry of

Education in 1902 and the suggestions by the Historical Society of Japan, in order to ask

106 http://www.bunka.go.jp/kokugo/ (accessed August 6, 2009) 160

Japanese speakers which rules are more appropriate. The rules suggested by the Ministry of Education and the Historical Society of Japan differ in many aspects. The most noticeable difference is that the suggested rules by the Historical Society of Japan are based on English, even if the source language is not English, whereas the rules by the

Ministry of Education are based on the source language. This fact suggests that the fluctuation of the katakana spellings existed at this time, and the causes of the fluctuation varied. In particular, the difference between academe and the government implies that the sounds of loanwords depended on the importer. This issue will be discussed in §5.18.

Although the two sets of suggested rules are different in many respects, they agree on using チ [c!i], ヂ ["#i], ツ [$%], and ヅ [&%] instead of ティ [ti], ディ [di], トゥ

[t%], and ドゥ [d%], as shown in Table 5.5.a and Table 5.5.b. The kana spellings ティ

[ti], ディ [di], トゥ [t%], and ドゥ [d%] were not used at that time, perhaps because such sequences were rare. The example words ヂッケンズ ["#ikken&%] ‘Dickens’ and カト

マンヅー [k'tom'n&%(] ‘Kathmandu’ suggest that the rules tried to reflect the difference in the spellings of the source words with an ‘s/z’ and a ‘d’. Since in standard

Japanese, as discussed in the previous section, the difference in pronunciation between ズ

(za-gy! /zu/) and ヅ (da-gy! /du/) had been lost by then, the purpose of the use of ヅ (da- gy! /du/) is not to reflect the pronunciations of the source words, but rather to transliterate the spelling of the source words. That is, ヅ (da-gy! /du/) does not reflect the pronunciation [d%]; however, it does indicate that the source word includes the letter

“d”—after all, the katakana ヅ belongs to da-gy!. Nonetheless, these rules might have confused Japanese speakers at that time. The largest hindrance for ordinary Japanese 161

speakers was that in order to spell the loanwords with the sequences [c!i], ["#i], [$%], and [&%] in katakana correctly, they either needed to know the spellings of the source words or needed to remember the correct katakana spelling. It became a more serious problem later when the sequences [ti] and [di] started being used in everyday speech.

In 1926 the Interim Committee on the Japanese Language published The Notation of Loanwords (Table 5.5.c). It was published in order to unify the katakana spelling rules of loanwords used in everyday life, which were “extremely inconvenient” due to their fluctuation. The rules state that ジ ["#i] should be used for the sound sequences that had been traditionally transcribed as ジ (za-gy! /zi/) and ヂ (da-gy! /di/). This suggests that there was confusion in katakana spellings because of the homophones ジ (za-gy! /zi/) and ヂ (da--gy! /di/). As discussed above, this confusion may be due to ignorance of the spellings of the source words because, according to the rules in the List of Foreign

Proper Names (the Historical Society of Japan 1914), in order to use ジ (za-gy! /zi/) and

ヂ (da-gy! /di/) properly one must know the original spellings.

In 1954 the Deliberative Council on the National Language published a newer version of The Notation of Loanwords (Table 5.5.e). The new rules encourage the use of

ト [to] and ド [do] to transliterate the sounds /tu/ and /du/ and the use of チ [c!i] and ジ

["#i] to transliterate the sounds /ti/ and /di/. On the other hand, ティ [ti] and ディ [di] are also allowed. Consequently, there are many variants. For example, candy is transliterated as キャンデー [kj'nde(], whereas dandy is ダンディー [d'ndi(]. In contrast, there are no remarks on the use of katakana トゥ [t%] and ドゥ [d%]. The council states that the causes of fluctuation are the various routes and histories of borrowing. The council 162 asserts that it respects conventional usages and ease of use by Japanese speakers. It is intriguing that one of the council members, Kobayashi Hideo, mentioned that since

Japanese speakers are capable of pronouncing [!e] and ["#e], the use of シェ [!e] and ジ

ェ ["#e] should be included in the rules. Kobayashi’s remark invites the inference that [ti],

[di], and in particular [t$], and [d$], were not easy to pronounce for most Japanese at that time.

Miyajima and Takagi (1988) investigated the katakana spellings of loanwords appearing in ninety magazines published in 1956. They found that less than 50% of the tokens followed the rules suggested by the Deliberative Council on the National

Language in 1954. They also found irregular use of ィ (small イ [i]). According to the rules, long vowels are indicated by the symbol ー. They found that in some cases ィ was used instead of ー. For example, バー [b%&] and リー ['i&] are occasionally transliterated as バァ107 and リィ. Miyajima and Takagi reported that the use of small-sized vowels instead of the symbol ー might indicate that there was an intention to signify that リィ is longer than リ ['i], but not as long as リー ['i&]. As mentioned before, [ti] is transliterated as ティ, which is the combination of [te] and small [i], and [di] is ディ. Thus, the function of the small [i] is not lengthening the vowel, but substituting the vowel [i] for [e].

However, the irregular use of small [i] in リィ ['i&] might affect the intrinsic length of テ

108 ィ [ti] and ディ[di]. There was no example of トゥ [t$] or ドゥ [d$] on the list, which suggests that these sound sequences may have been rare in the 1950s, while ティ

107 ァ indicates small ア [ɑ]. 108 The length of the ティ [ti] and ディ[di] will be discussed in more detail later in §5.9. 163

[ti] and ディ [di] were already prevalent. However, note that the word プロデューサー

[p!"od#j!$s%$] ‘producer’ was on the list, which suggests that pronouncing [d#j!] may not have been as hard as [d!] for people at that time.

The most recent set of rules suggested by the government is The Notation of

Loanwords by the Cabinet in 1991 (Table 5.5.e). The katakana トゥ [t!] and ドゥ [d!] were finally approved to transliterate the sounds /tu/ and /du/, which are found in loanwords. But it also allowed for the conventional use of ツ [&!] ズ ['!], ト [to], and

ド [do].

The history of the rules of katakana transliteration by the government does not necessarily reflect the actual pronunciations of loanwords by Japanese speakers.

Nevertheless, it suggests the following. First, since the transliteration methods for loanwords vary greatly, it is impossible to predict the form of a word even though the government has been trying to simplify the rules. Second, the causes of the variation are complex: from historical and conventional to individual. Third, the changes of the rules correspond to the synchronic distributions of ティ [ti], ディ [di], トゥ [t!], and ドゥ

[d!]—ティ [ti] and ディ [di] were approved before トゥ [t!] and ドゥ [d!] were approved.

5.3.1 The Distribution of [ti], [t!], [di], and [d!]

The sound sequences [ti] ティ and [t!] トゥ and their voiced counterparts [di] ディ and

[d!] ドゥ have similar historical backgrounds in pre-20th century Japanese: all of them were once sounds of Japanese, and their consonants were all replaced by affricates. In 164 spite of their similar historical background, [ti] and [di] appear to be more abundant than

[t!] and [d!] among non-marginal words. Table 5.6 demonstrates the unbalanced distribution of [ti] / [di] and [t!] / [d!] in the major Japanese dictionaries K!jien and

Daijirin.

Table 5.6. Distribution of [ti] / [di] and [t!] / [d!] in non-marginal Items

ti di t! d! 75 39 K!jien 761 655 (common nouns: 20) (common nouns: 6) 69 37 Daijirin 906 696 (common nouns: 21) (common nouns: 4)

More than half of the non-marginal items with [t!] or [d!] are proper nouns, such as personal names and place names. As discussed in the previous section, the

Cabinet approved the use of katakana spellings for [t!] and [d!] in order to reflect the pronunciation of a foreign proper noun as close as possible to the source language.

Accordingly, in terms of proper names, frequently there are discrepancies between katakana spellings and common pronunciation. For example, many names of historical figures in the Roman Empire are likely to be pronounced without the sequence [t!] or

[d!]. The name for Brutus is listed as ブルートゥス [b!"!#t!s!]; however, it is likely to be pronounced as [b!"!#tos!] or [b!"!#t$s!]. The latter pronunciation more closely reflects English pronunciation. Many of the non-proper nouns (i.e. common nouns) that contain innovative sound sequences are technical terms, such as musical terms borrowed 165

from Italian (e.g. [t!tti] ‘tutti’トゥッティ), and scientific terms borrowed from Latin

(e.g. [pitek"nto#op!s! e#ek!t!s!] ‘Pithecanthropus erectus’ ピテカントロプス・エ

レクトゥス). Table 5.7 is a list of non-marginal words with the innovative sound sequence [t!] or [d!] which are likely to be used in everyday conversation.

166

Table 5.7. List of non-marginal words with innovative sound sequences [t!] or [d!]

gloss K!jien Daijirin

[t!] a. [ins!t!"!ment#"!] ‘instrumental’ Yes インストゥルメンタル

b. [k#$t!$%] ‘cartoon’ Yes カートゥーン

c. [t#t!$] ‘tattoo’ Yes Yes タトゥー

d. [t!de$] ‘today’ Yes トゥデー

e. [t!n#ito] ‘tonight’ Yes トゥナイト

f. [t!mo"o$] ‘tomorrow’ Yes トゥモロー

g. [b!"!$t!$s!] ‘Bluetooth’ Yes Yes ブルートゥース

[d!] h. [#nd!$] ‘undo’ Yes Yes アンドゥー

i. [d!$ itto j!#se"!&!] ‘do-it-yourself’ Yes Yes ドゥイットユアセルフ

j. [d!$ '#pp!] ‘doo-wop’ Yes ドゥーワップ

All four items listed in K!jien were not listed in the previous edition that was published in 1998. Let me examine each word and its historical and cultural background.

Table 5.7.a [ins!t!"!ment#"!] ‘instrumental’インストゥルメンタル specifically 167 means ‘popular music without vocals’ in Japan. Thus, Japanese speakers who are interested in Western pop music or J-pop music are familiar with this word. Instead of

[ins!t!"!ment#"!], K!jien lists [ins!to"!ment#"!]. Table 5.7.b [k#$t!$%] ‘cartoon’

カートゥーン refers to comic strips by Daijirin’s definition, but most Japanese use this word to signify animations produced by American film companies, such as Disney and

Warner Brothers. This word was used by enthusiasts first, but was gradually accepted by the general population. Japanese use the native word irezumi as well as the foreign item

[t#t!$] タトゥー (Table 5.7.b) to refer to a tattoo. The former is used for traditional

Japanese style tattoos while the latter is used for fashionable tattoos or tattoos that are not in the traditional Japanese style. Traditional Japanese style tattoos have had strong associations with yakuza (Japanese criminal organization members); thus, they are regarded as symbols of antisocial groups or activities. Table 5.7.d [t!de$] ‘today’トゥデ

ー, Table 5.7.e [t!n#ito] ‘tonight’トゥナイト, and Table 5.7.f [t!mo"o$] ‘tomorrow’

トゥモロー are words that junior high school students learn in English class. Thus, many speakers know what these words mean and how they are pronounced in English. These words are rarely used in conversations, but many TV programs have a title with one of these words. Table 5.7.g [b!"!$t!$s!] ‘Bluetooth’ ブルートゥース and Table 5.7.h

[#nd!$] ‘undo’アンドゥー came from computer terminology. Table 5.7.i [d!$ itto j!#se"!&!] ‘do-it-yourself’ ドゥイットユアセルフ also consists of words that are familiar to students in junior high and above, who know at least basic English. Table 5.7.j

[d!$ '#pp!] ‘doo-wop’ ドゥーワップ is a style of vocal music popularized in the

1950s and 60s in the United States as well as in Japan. It became very popular in Japan 168 again in the early 1980s. Overall, these words are relatively new. They are based on source words known to speakers through English language education. They also have strong connections with Western culture or technology—such as music, art, or computer technology.

5.3.2 Phonological Variants with [ti]/[di]

In this section, I will discuss phonological variants involving [ti]/[di]. I define phonological variants as “two different pronunciations of a loanword.” The existence of phonological variants suggests that there was a change in speakers’ perspectives toward loanwords. First, I will discuss phonological variants with [ti] and [di]. There are two patterns of forming phonological variants with [ti] and [di]. The first pattern involves affrication: [ti]/[di] and [c!i]/["#i (#i)] phonological variants. Examples of this pattern are shown in (1).

169

(1) a. [p!"#s!c$ikk!]~[p!"#s!tikk!] ‘plastic’ プラスチック～プラスティック b. ["omanc$ikk!]~["omantikk!] ‘romantic’ ロマンチック～ロマンティック c. [e"oc$ikk!]~[e"otikk!] ‘erotic’ エロチック～エロティック d. [o%s!c$i&]~[o%s!ti&] ‘Austin’ オースチン～オースティン e. [c$iketto]~[tiketto] ‘ticket’ チケット～ティケット f. [c$i"#nodz#!"!s!]~[ti"#nodz#!"!s!] ‘Tyrannosaurus’ チラノザウルス～ティラノザウルス g. [c$i%m!]~[ti%m!] ‘team’ チーム～ティーム h. ['(i"emm#]~[di"emm#] ‘dilemma’ ジレンマ～ディレンマ

The second pattern is vowel alternation: [ti]/[di] and [te]/[de] phonological variants. Ito and Mester (2007) indicate that this is a case in which the vowel [e] is acting as a “depalatalizer” by shifting from [i] to [e]. That is, by shifting the vowel, the affrication from [ti] to [c$i] or [di] to ['(i ((i)] can be avoided. Examples of this second pattern are shown in (2).

170

(2) a. [de!"it#$%]~[di!"it#$%] ‘digital’ (Ito and Mester 2003) デジタル～ディジタル b. [deki&i']~[diki&i'] ‘Dixieland jazz’ (Ibid.) デキシー～ディキシー c. [b%$#nde']~[b%$#ndi'] ‘brandy’ (Ibid.) ブランデー～ブランディー d. [kj#nde']~[kj#ndi'] ‘candy’ キャンデー～キャンディー e. [p%$es%te'!"i]~[p%$es%ti'!"i] ‘prestige’ プレステージ～プレスティージ f. [h#ndekj#pp%]~[h#ndikj#pp%] ‘handicap’ ハンデキャップ～ハンディキャップ

The phonological variants in (1) and (2) do not indicate that loanwords with /ti/ in the source words necessarily change to [ti]. The patterns of these phonological variants appear to develop on a case-by-case basis. For example, (1f) [c&i$#nodz#%$%s%]

‘Tyrannosaurus’ is primarily replaced by [ti$#nodz#%$%s%]. The earlier editions of

K!jien list [c&i$#nodz#%$%s%] チラノザウルス, but later editions list

[ti$#nodz#%$%s%] ティラノザウルス, and [c&i$#nodz#%$%s%] チラノザウルス is not even cross referenced. On the other hand, both variants in (1e) [c&iketto]~[tiketto]

‘ticket’ チケット～ティケットand (1g) [c&i'm%]~[ti'm%] ‘team’ チーム～ティーム are accepted, but the affricated variants are much more common than the non-affricated variants. In the case of (2c) [b%$#nde']~[b%$#ndi'] ‘brandy’, [b%$#ndi'] ブランディー is probably very rare and not listed in the dictionaries even as an alternative pronunciation. However, the R&B singer Brandy (1979~) is called [b%$#ndi'] ブランデ

ィー and never [b%$#nde'] ブランデー. By contrast, in the case of (2d) 171

[kj!nde"]~[kj!ndi"] ‘candy’, [kj!ndi"] キャンディー is the dominant pronunciation while [kj!nde"] キャンデー sounds old-fashioned. The dictionaries list [kj!nde"] as an entry and [kj!ndi"] as an alternative pronunciation. But I doubt that young standard

Japanese speakers pronounce [kj!nde"] instead of [kj!ndi"]. This may be partially due to the Japanese pop vocal group [kj!ndi"z#] キャンディーズ (Candies 1972~1978), as well as the manga and anime [kj!ndi kj!ndi] キャンディ・キャンディ (Candy

Candy).109 They were extremely popular in the 1970s. The situation for the variants (2f)

[h!ndekj!pp#]~[h!ndikj!pp#] is different. [h!ndekj!pp#] (and its abbreviated version

[h!nde]) is used as golf jargon, while [h!ndikj!pp#] often refers to a physical or mental handicap or disability.110

5.3.3 Phonological Variants with [t!] and [d!]

Before I discuss the phonological variants involving [t#] and [d#], I will describe what kind of source words could be loanwords with the innovative sound sequences [t#] and

[d#]. What sound sequences do the loanwords actually have instead of [t#] and [d#]?

Japanese has had an enormous influx of loanwords from English (Loveday 1996); therefore, I will consider English as the source language in this section. There are roughly two patterns that can lead to loanwords with [t#] and [d#] sequences. One is when the source word has a [tu], [t$], [du], or [d$] sound sequence, and the other is when the source word contains a consonant cluster beginning with [t], such as [tr] and [tw].

109 Igarashi Yumiko and Mizuki Ky!ko 1975~1979 Candy Candy 110 However, Japanese dictionaries do not explicitly mention disability. [h!ndi] ハンディ, meaning ‘physical or mental handicap or disability’ doesn’t appear to be very popular. 172

Obviously, if a source word contains a sound sequence [tu], [t!], [du], or [d!], the innovative sound sequences [t"] or [d"] could be contained in the loanword. As we have seen in Table 5.7, [t#t"$] ‘tattoo’ タトゥー and [d"$ itto j"#se%"h"] ‘do-it-yourself’

ドゥイットユアセルフ are examples of this pattern. The source words of some loanwords with a [t"] sequence (such as [t"de$] ‘today’トゥデー, [t"n#ito] ‘tonight’

トゥナイト, and [t"mo%o$] ‘tomorrow’トゥモロー) do not contain [u] in their first syllables, but [!]. However, these words can have a [tu] sequence in exaggerated or very

111 careful pronunciation. The English [!] is often transformed to the Japanese [#] when borrowed. But when compound words are borrowed, loanwords tend to respect the elements in the compound words. For example, in English wiki is pronounced [w&ki] while Wikipedia is pronounced [w&k!pi%i!]. In Japanese, they are pronounced ['iki] and

['ikipedia] respectively.

It appears that source words with /tu/, /t!/, /du/ or /d!/, or with /t!/ as a part of an element of a compound word, could be loanwords with [t"] and [d"]. However, until after WWII, /tu/, /t!/, /du/ or /d!/ in source words were borrowed as [("] or [dz"]. The examples in (3) are loanwords whose source words contain [tu], [t!], [du] or [d!]. They could contain [t"] or [d"], but they have [("] or [dz"] instead.

111 Songs sometimes reveal the pronunciation of elements in compound words. For example, tonight in the song by that name from the musical West Side Story (Bernstein and Sondheim 1956) and tomorrow in the song by that name from the musical Annie (Strouse and Charnin 1982) are pronounced with the [tu] sound sequence. Thus, transcribing today as トゥデー , while not phonetically accurate, is not a grave error. 173

(3) [ts!"#] ‘tour’ ツアー [ts!#] ‘two’ ツー [ts!#$is!to] ‘tourist’ ツーリスト [ts!#$i!g!] ‘touring’ ツーリング [ts!n"] ‘tuna’ ツナ [ts!be$!k!$i%] ‘Tuberkulin (German) (tuberculin)’ ツベルクリン [ts!#$!] ‘tool’ ツール [ts!ndo$"] ‘tundra’ ツンドラ [h"!ts!#] ‘how to’ ハウツー [dz!kk!] ‘doek (Dutch) duck; canvas’ ズック [s!ts!#$!] ‘stool’ スツール

English has some frequently used basic vocabulary words with sequences [tu],

[t&], [du] or [d&], such as to, too, two, took, and do. The token frequency of these words is large. But aside from these words, [tu], [t&], [du], and [d&] appear to be rare compared with [ti], [t'], [di], and [d'].

The second transliteration pattern occurs when the source word contains a consonant cluster with [t] or [d]. Such consonant clusters may include /tw/, /dw/, /tr/, and

/dr/. For example, the English word ‘true’ contains a consonant cluster /tr/; thus it could have the [t!] sound sequence when Japanese borrows it. As we have seen in Table 5.7 above, [ins!t!$!ment"$!] インストゥルメンタル ‘instrumental’ is an example of such an item. However, [t] or [d] in source word consonant clusters are frequently altered to [(!], [to], [do], or [dz!], as shown in (4). 174

(4) [ts!is!to] ‘twist’ ツイスト [ts!i"t#"] ‘tweeter’ ツイーター [ts!i"do] ‘tweed’ ツイード [ts!i$] ‘twin’ ツイン [ts!%i"] ‘tree’ ツリー [to%#kk!] ‘truck’ トラック [to&#i%#ito] ’twilight’ トワイライト [s!to%#ik!] ‘strike’ ストライク [do%#''!] ‘drug’ ドラッグ [dz!%o"s!] ‘drawers’ ズロース

One may wonder whether [t!] or [d!] appear when the source word ends with a

[t] or [d]. This question is reasonable because Japanese cannot have a closed syllable except for certain cases—and normally a vowel [!] is epenthesized if the source word ends with a consonant. In order to transform closed syllables into open syllables in the process of borrowing, typically, Japanese adds the vowel [!] at the end of the syllable.

For example, if the syllable of the source word ends with /p/, /k/, /s/, /b/, /'/, /!/, /m/, or

/z/, that syllable will be transformed into [p!], [k!], [s!], [b!], ['!], [!'!], [m!], or

[z!] respectively. Thus, the original Japanese title of “The Ring” (Verbinski 2002), the popular Hollywood remake of a J-horror movie, is “Ringu” (Nakata 1998). This also applies to some source words with syllable-final [n] (e.g. [en!] < (the letter) ‘n’).

However, syllable final [t] or [d] has never been transformed into [t!] or [d!] in the process of borrowing: the final [t] is altered into either [(!] (e.g. [s!"(!] < suit) or [to]

(e.g. [#p#"to] < apartment), and the final [d] to (e.g. [k#"do] < card). Consequently, both

[t!] and [d!] in loanwords where the source word has a final [t] or [d] are impossible to find. 175

It is difficult to find phonological variants for [t!] and [d!] in dictionaries. The variants [ins!t!"!ment#"!]~[ins!to"!ment#"!] ‘instrumental’インストゥルメンタ

ル～インストルメンタル may be the best example of all. Daijien indicates that

[ins!to"!ment#"!] is an alternative pronunciation of [ins!t!"!ment#"!], while

K!jien lists only [ins!to"!ment#"!]. As I discussed in §5.3.1, many scientific words and proper nouns with [t!] sequence might be pronounced with [to] instead. For example,

[pitek#nto"op!s! e"ek!t!s!] ‘Pithecanthropus erectus’ ピテカントロプス・エレク

トゥス is likely to be pronounced as [pitek#nto"op!s! e"ek!tos!]. Inozuka, H. and

Inozuka, E. (2003) state that [t!dei], [t!n#ito], and [t!mo"o$] were pronounced with

[ts!] when these words were first introduced. Daijirin, as well as the Nihon Kokugo

Daijiten, indicates that [ts!dei], [ts!n#ito], and [ts!mo"o$] are alternative pronunciations. However, I have never heard these alternative pronunciations used.

Moreover, I suspect most young Japanese speakers have some difficulty recognizing the words [ts!dei], [ts!n#ito], and [ts!mo"o$] when they hear them.

As we have seen, there are two patterns of forming phonological variants: vowel shifting and affrication. Phonological variants with a [t!]~[to] alternation, such as

[ins!t!"!ment#"!]~[ins!to"!ment#"!] instrumental and [pitek#nto"op!s! e"ek!t!s!]~[pitek#nto"op!s! e"ek!tos!] ‘Pithecanthropus erectus’, are examples of a vowel shift from [!] to [o]. On the other hand, [t!dei]~[ts!dei],

[t!n#ito]~[ts!n#ito] and [t!mo"o$]~[ts!mo"o$] are examples of affrication from [t] to

[ts]. We can find more phonological variants when marginal items, such as song titles and movie titles, are considered. The [t!"i$]~[ts!"i$] ‘tree’ variation can be observed in the 176

album title [jo!"# t"$i%] The Joshua Tree (U2 1987). The [t"%]~[ts"%] ‘two’ variation is observed in the song title [t"% ob" #s"] Two of Us (the Beatles 1970). These words are pronounced with [ts"] in conversations. Interestingly, the name of the band U2 is pronounced as [j"%ts"%], with a [ts"] sequence. This fact demonstrates that the sound

112 sequence [t"] is not uniformly accepted in marginal items.

5.4 Have Innovative [ti] and [t"] Replaced Ta-gy! [c!i] and [c!"]?

In this section, I will discuss the motivation behind the emergence of [ti] and [t"] in

Japanese. I will investigate whether [ti] and [t"] have established their status as ta-gy! sound sequences by replacing [c!i] and [ts"]. I argue that [ti] and [t"] have not yet established such a status in Japanese phonology based on the fact that the vowels in [ti] and [t"] do not devoice in the devoicing environment. Although only [ti] and [t"] will be discussed in the following sections, their voiced counterparts [di] and [d"] can be assumed to share the same or similar status.

5.4.1 Akima (Gap) Theory and the Innovative Sound Sequences [ti] and [t!]

Hattori (1960) claims that the consonant phonemes of [c!i] and [ts"] do not belong to the same phoneme as the consonant in other ta-gy! sequences, namely /t/. Instead, he proposes a unique phoneme /c/ for the consonants of [c!i] and [ts"]. Consequently, he argues that ta-gy! lacks sequences /ti/ and /tu/, which creates what he calls akima (gap,

112 Interestingly, the words for the numbers twelve and twenty are always pronounced as [t"e$"b"] and [t"enti%]. They are frequently used, but marginal items. 177 lit. vacancy). Hattori claims that the motivation for the emergence of the innovative [ti] and [t!] is to fill in these gaps, as illustrated in Table 5.8.

Table 5.8. Defective distribution of ta-gy! and [ti] and [t!] as gap fillers

/ta/ [t"] akima (gap) akima (gap) /te/ [te] /to/ [to]

/ta/ [t"] /ti/ [ti] innovative /tu/ [t!] /te/ [te] /to/ [to] innovative

タティトゥテト

Hattori’s claim is consistent with the Structuralists’ view of sound change. The

Structuralists emphasize symmetry, which is one aspect of structure; therefore, they disfavor gaps in sound systems (McMahon 1994). For example, McMahon demonstrates that /#/, the voiced counterpart of /$/, appeared to fill in the gap in the pre-18th century

English fricative system, as shown in (5).

(5) Pre-18th century English fricative system (Source: McMahon 1994:28)

voiceless f " s $ h voiced v # z

The other gap in (5), the voiced counterpart of /h/, has not been filled. Instead, /h/ may be disappearing from the language: many British English accents drop /h/ now. If this happens, the emergence of /#/ and the disappearance of /h/ will eventually make the 178

English fricative system gap-free and the system will have perfect symmetry (McMahon

1994).113

The idea that the innovative sound sequences [ti] and [t!] have taken root in

Japanese phonology in order to complete an unbalanced and defective system seems logical and phonetically coherent. However, the Structuralists’ view raises a question. If language disfavors asymmetry, why did the consonants of Old Japanese ta-gy! /ti/ and

/tu/ change so that gaps were created as a consequence? Even if the innovative sound sequences appear to fill in the gaps, the filler sound sequences may not “fit” into the gaps completely. In the case of the English fricative system, /"/ surely fills the gap in the system; however, it does not behave like other voiced fricatives. The most noticeable difference is that word-initial /"/ cannot be found except for rare items, such as the word genre and the name Jacques. The sound /h/ may be disappearing in British English.

However, it is difficult to imagine that it will vanish completely from British English, much less from American English.

I find similar incompleteness in [ti] and [t!] as “fillers of the gaps”. I argue that both [ti] and [t!] have not yet achieved the status of genuine ta-gy! sequences. In addition to the fact that [ti] and [t!] appear only in recently borrowed items, they seem to behave differently from other sound sequences. One of the reasons I do not consider [ti] and [t!] as ta-gy! sequences is that unlike other similar sound sequences with a structure

113 Of course, English /h/ is often pronounced as breathy voiced [#] rather than voiceless [h]. 179

of ‘a consonant + a high vowel’, such as [c!i], [ts"], [#i], [ki], [k"] and [s"], the vowels of [ti] and [t"] do not seem to devoice in the vowel devoicing environment.

5.5 Vowel Devoicing and Innovative Sound Sequences

It is well-known that in standard Japanese high vowels (i.e. [i] and ["]) are devoiced between voiceless consonants or word-finally (Shibatani 1990; Vance 1987; Inozuka, H. and Inozuka, E. 2003; Ogasawara 2007). Some non-high vowels are devoiced as well.

For example, the first vowel of [koko$o] is likely to be devoiced (Sakuma 1929:232-232, cited in Vance 1987; Inozuka, H. and Inozuka, E. 2003). However, devoicing of non-high vowels is rare compared with devoicing of high vowels. In some areas of Japan, vowel devoicing is infrequent. For example, in the Kyoto dialect, vowel devoicing is less noticeable (Shibatani 1990). Shibatani provides the following basic conditions for vowel devoicing: A high vowel will devoice only: (1) when it is not “contiguous to a voiced sound;” (2) when it is not an initial sound; and (3) when it is not accented (Shibatani

1990:160). Vance (1987) also demonstrates that (4) vowel devoicing does not apply to a high-pitched final syllable of a word or a sentence. In addition to these basic conditions, various factors affect vowel devoicing in standard Japanese. Shibatani (1990) indicates that slow speech tempo might result in less frequent devoicing. By contrast, fast speech might cause devoicing of non-high vowels as well (Bloch 1950:103; Han 1962b:84-85, cited in Vance 1987). Han (1962b:91-92, cited in Vance 1987) argues that devoicing of vowels in two consecutive syllables is rare and that vowel devoicing in three consecutive syllables does not occur. 180

The peculiar aspect of [ti] and [t!] is that their vowels do not seem to devoice even though the devoicing conditions described above are met. Maekawa and Kikuchi (2005) analyzed vowel devoicing patterns using an excerpt from the Corpus of Spontaneous

Japanese (CSJ), a large-scale speech data base. The excerpt contains about 23 hours of segment-labeled speech with 427,973 vowel segments of 29 female and 56 male speakers. The results for the analysis of devoicing patterns of /CiC/ sequences are shown in Table 5.9. In Table 5.9, Co in /CoVcCo/ means ‘voiceless consonant’ and Vc means

‘high vowel’ (Maekawa and Kikuchi call this “close vowel”). C1 and C2 mean

‘preceding consonant’ and ‘following consonant’ respectively. /c/ is a phoneme proposed by Hattori (1960). There are two allophones of /c/: [c"] and [#]. In Table 5.9, /c/ in C1 is

[c"] and /c/ in C2 could be either [c"] and [#]. Apparently, Maekawa and Kikuchi (2005) consider ["] an allophone of /s/. Table 5.9 does not include the sound sequences whose total number of samples was fewer than 10. 181

Table 5.9. Cross-tabulation of the voicing of /i/ in the /CoVcCo/ environment by C1 and C2 VOWEL C1 C2 VOICED DEVOICED % DEVOICED

c 16 73 82.02

h 35 7 16.67

k 31 358 92.03

c p 7 44 86.27

Q 16 16 50.00

s 64 41 39.05

t 32 181 84.98

c 5 80 94.12

h 22 9 29.03

k 15 342 95.80 h Q 21 39 65.00

s 11 3 21.43

t 21 883 97.68

c 19 62 76.54 i h 167 65 28.02

k 73 476 86.70 k Q 32 51 61.45

s 144 262 64.53

t 53 791 93.71

p Q 118 9 7.09

c 7 259 97.37

h 47 14 22.95

k 50 1,102 95.66 s Q 25 92 78.63

s 259 178 40.73

t 49 6,507 99.25

k 11 0 0.00 t Q 13 0 0.00 (Source: Maekawa and Kikuchi 2005:212)

182

As Table 5.9 shows, the [ti] sequence does not devoice at all. Maekawa and

Kikuchi (2005) also made a table for sound sequences that contain the vowel /u/, but /tu/ sequences are not included in the table, presumably because of the low frequency of [t!] sequences in Japanese. As we have seen, it is atypical that vowels in [ti] and possibly

[t!] fail to devoice even in the devoicing environment. What prevents the devoicing of

[ti] and [t!]?

5.6 Reasons for Vowel Devoicing

Before discussing the cause of [ti] and [t!] evading vowel devoicing, I will describe why high vowels are subject to devoicing. The reason high vowels are devoiced in the devoicing environment may be that they are shorter than other vowels. Lehiste (1970:18-

19, cited in Vance 1987) maintains that it is probably a general tendency that high vowels are shorter than other vowels, since vowel lengths depend on the degree of opening in the vocal tract. According to Han’s study (1962a:67, cited in Vance 1987), the length differences between Japanese high vowels and low vowels are greater than other languages, such as Swedish (Elert 1964, cited in Vance 1987). These studies suggest that high vowels are shorter than the other vowels and that (usually) only high vowels are devoiced in the devoicing environment. However, it is still not clear why intrinsically shorter vowels are subject to devoicing. Nevertheless, the lengths of high vowels might be a crucial factor for vowel devoicing, as discussed below.

183

5.7 Why vowels in [ti] and [t!] do not devoice

There are several ways to explain why vowels in [ti] and [t!] do not devoice in devoicing environments. First, one could claim that this is merely an exception to the rule. But this does not explain the phenomenon at all. Second, one could claim that the acoustic differences between [t] in [ti] and [t] in [t!] are smaller than the difference between consonants in other sound sequences, such as [k] in [ki] and [k] in [k!]. The second account has to do with the retrievability of devoiced vowels. Vowel devoicing is called

‘devoicing’, not ‘deletion’, even when the whole vowel is deleted (Vance 1987), because the devoiced vowels are retrievable (Ogasawara 2007). Japanese speakers can retrieve a devoiced vowel that has no vowel-like acoustic information. It is easier to retrieve devoiced vowels when the acoustic information in the consonants is contrasting. If the acoustic information in the consonants in [ti] and [t!] is similar, then it is more difficult to retrieve devoiced or deleted vowels. Third, the vowels of [ti] and [t!] could be somewhat different from high vowels of other sound sequences. Since non-high vowels are not subject to devoicing due to relatively longer durations, the vowels of [ti] and [t!] may be intrinsically longer.

5.8 [ti] and [t!] as co-articulation

Sometimes, devoiced vowels are completely deleted (Vance 1987). Deletion of the devoiced vowel occasionally creates consecutive identical consonants, which resembles consonant gemination. However, there is a fundamental difference between consonant gemination and two consecutive consonants due to vowel devoicing and subsequent 184 vowel deletion. As discussed above, the deleted vowel caused by vowel devoicing is always retrievable.114 In contrast, in the case of consonant gemination, not only is the vowel irretrievable (because there is no vowel to retrieve), but also the first consonant of the gemination is irretrievable. This suggests that Japanese speakers cannot recognize a consonant without the following vowel. In Japanese, consonant clusters are always homorganic, and Japanese words must end in a vowel or a moraic nasal [!] (Blevins

2004). Blevins states that Japanese phonotactics allows place features only before vowels. Lack of a place feature means that the consonant is not recognized properly. This means that the existence of a following vowel is indispensable for identification of a consonant.

The deleted vowel in between two consecutive voiceless fricatives created by vowel devoicing, such as [i] in ["i"o#] ! ["i$"o#] ! [""o#] ‘obstacle’, often contains “no clear interval corresponding to the voiceless vowel” (Vance 2008:209). However, the deleted vowel is completely retrievable (Ogasawara 2007).115 On the other hand, consonant geminations, such as [j%kk%&i] ‘slowly’, are not induced by vowel devoicing or vowel deletion; thus, there is no vowel to retrieve. But the interesting fact is that when there is no vowel to retrieve after a consonant, Japanese speakers do not retrieve the

114 Inozuka, H. and Inozuka, E. (2003) claim that vowel devoicing across a morpheme boundary leads to consonant gemination (eg. [sent'k%] ‘wash’ +[k(i] ‘machine’ ! [sent'k%k(i] ! [sent'k%$k(i] ! [sent'k(k(i] ‘washer’). In the case of [sent'k(k(i] ‘washer’, the deleted vowel [%] is retrievable because of the existence of the word [sent'k%] ‘wash’. However, not all consonant geminations are triggered by vowel devoicing or vowel deletion, as we have seen in Chapter 2. The example Inozuka and Inozuka present is rather exceptional, and it applies only to Sino-Japanese items. See further arguments in Kurisu (2000). 115 The little circle under the vowel ([i]$ in ["i"$ o#]) indicates devoicing. 185 consonant either. In other words, no Japanese speaker recognizes the first [k] in

[j!kk!"i] as a consonant: it is recognized as silence. Acoustically, there is a pause in between the closure of the first [k] and the release of the second [k]. The formant trajectory signal before the closure of the first [k] does exist; nevertheless, Japanese speakers often fail to perceive the signal before the closure of the first [k]. Consequently, the first [k] is recognized as silence. Furthermore, Japanese speakers fail to perceive the first half of geminated consonants with continuant consonants, such as the first [s] in

[!ss!"#] ‘faintly’, where there is no interval of silence at all. Even though there is a clear acoustic signal that indicates the existence of [s], Japanese speakers identify the signal not as [s], but as a pause. This is consistent with the fact that all first consonants of gemination are spelled as っ (small /tsu/ in hiragana) or ッ (small /tsu/ in katakana). In addition, traditional Japanese linguists have analyzed the first consonant of gemination as an independent phoneme (Vance 1987), distinct from any syllable initial consonant.

Kawakami (1977, cited in Vance 1987) claims that when devoiced vowels are deleted, vowel coarticulation remains in the preceding consonant. But how can coarticulation be possible when there is no vowel with which to coarticulate? This situation is similar to the case where English speakers perceive an illusory [m] in [kæ$p%] camper and [!] in [æ$k%] anchor (Kenstowicz 1994:5). English lacks nasalized vowels as phonemes; therefore, the nasalized vowel [æ]$ in [kæ$p%] camper or [æ$k%] anchor is an allophone of /æ/. Vowels are nasalized when followed by a nasal. In this case, however, the vowel is nasalized without a following nasal (Kenstowicz 1994). Thus, it is not 186 inconceivable to claim that a similar process of coalescence (assimilation and then deletion) takes place when devoicing occurs in Japanese.

Nevertheless, in Japanese, I doubt that acoustic information of vowel coarticulation truly remains in the consonant whenever the following vowel is completely deleted. In the case of the nasalized vowel [æ!], the acoustic information of the nasality is clearly present. But it is debatable whether such salient acoustic information suggesting the coarticulation of the vowel is always present in the immediately preceding consonant of a deleted vowel in Japanese.

I argue that the deleted vowels are retrieved with or without vowel coarticulation.

Retrievable vowels are either [i] or ["]. If the consonants that precede [i] and the consonants that precede ["] are different, vowel coarticulation is irrelevant. In general, because of coarticulation, the acoustic information of the consonants in CV1 and CV2 is different (Liberman et al. 1967).116 Thus, it is natural that acoustic information in consonants in [Ci] and [C"] are different. Table 5.10 shows the consonants that precede

[i] or ["] in narrow transcriptions that reflect palatalization and labialization.

Table 5.10. Distribution of consonants that precede either [i] or ["] ka-gy! sa-gy! sha-gy! ta-gy! cha-gy! ha-gy! fa-gy! pa-gy!

i k# $ $ c$ c$ ç %# p# " k s $& ' c$& h/% % p

116 CV1 and CV2 indicate that these syllables share the same consonant, but their vowels are different. 187

The issue here is not whether the consonant that precedes [i] and that precedes [!] in the same gy! are different phonemes or the allophones of the same phoneme. The crucial question is whether they can uniquely determine the following devoiceable vowel when pronounced in isolation. For example, as we have seen, [s] can be followed by ["],

[!], [e], and [o], but the only devoiceable vowel that can follow [s] is [!]. Thus, in the case of [s!], it is possible to determine the deleted vowel with or without vowel coarticulation: the deleted vowel must be [!]. Similarly, [ts] can be followed by any

117 vowel, but the only possible deleted vowels that can follow them is [!]. In these vowel retrieving processes, vowel coarticulation (if it exists), plays no role. On the other hand, the consonants with rich vowel coarticulation, such as [k], sometimes allow devoicing of vowels other than [i] and [!], such as the first [o] in [koko#o] (Vance 1987). But as for the deletion of [i] and [!], vowel coarticulation does not always play a significant role in retrieving the deleted vowel. Furthermore, the preceding consonants do not always contain vowel coarticulation. Nevertheless, vowels that are deleted due to vowel devoicing are almost always retrieved successfully (Ogasawara 2007). It may be assumed that the vowels of [ti] and [t!] are not devoiced because when the vowels are completely deleted, the information in the consonants of [ti] and the information in the consonants of

[t!] are not different enough to retrieve the following vowels. The lack of a sufficient difference between the consonants in [ti] and [t!] makes it impossible to identify them as the consonants that precede either [i] or [!].

117 It appears that the vowel in [tsi] will not be devoiced. Since there are too few examples of [tsi], I will not discuss the sound sequence any further in this dissertation. 188

The other possible reason that the vowels of [ti] and [t!] do not devoice is that they are somewhat different from high vowels of other sound sequences. The vowels in [ti] and [t!] may be longer than other high vowels. As discussed in §5.8, the smaller durations of high vowels are considered one of the causes of vowel devoicing. Thus, if the vowels in [ti] and [t!] are intrinsically longer than high vowels in other sound sequences, it is expected that they would be exempt from devoicing. For example, the vowel [i] in [s"ti] ‘Erik Satie’ サティis not as long as [i#] in [ti#] ‘tea’ ティー, but not as

118 short as [i] in [c$i] ‘blood’: it is somewhere in between. There are also possibilities that the vowel in [ti] is less palatal than [i] in [ki] or [c$i], and articulation of the vowel in

[t!] involves lip protrusion. It is not entirely clear how these subtle articulatory modifications affect devoicing, but at least if the vowels in [ti] and [t!] are different from the high vowels preceded by other consonants, then logical consistency is maintained.

These hypotheses must of course be tested empirically.

If these hypotheses are correct, then Japanese has two additional vowels appearing only in [ti] and [t!], or Japanese has something like 1.5 moras appearing only in [ti] and

[t!]. Thus, the emergence of the innovative sound sequences [ti] and [t!] is not merely

“filling in the gaps,” as Hattori claims. The acceptance of these sound sequences might require some modifications of the Japanese phonological system.

The situation of [ti] and [t!] suggests that pronounceability of sound sequences does not necessarily make the sound sequences a part of an inventory. A language has its

118 Other examples are the second [i] in [h"%o# kiti] ‘Hello Kitty’ ハローキティand [kiti c$"&] ‘Kitty’ キティちゃん. 189 own phonological system, and sound sequences that maximize the function of the phonological system will be included in the inventory.

5.9 Evidence From New-type Gairaigo

This section demonstrates how the undevoiceability of [ti] and [t!] interacts with the creativity of language observed in modern slang used by young people in Japan. Akizuki

(2005) has been interested in the innovative expressions used by Japanese teenagers. He reports that since 2000, young people in Japan started using what he calls new-type gairaigo (new-type loanwords). New-type gairaigo are marginal vocabulary items used only in popular culture or on the Internet. According to Akizuki, new-type gairaigo break the traditional rules of transliterating source words into Japanese in order to make the loanwords sound more foreign, and in most cases, English. However, this does not necessarily mean that new-type gairaigo are phonetically closer to the source words than the traditional loanwords. The users of new-type gairaigo are not satisfied with the traditional transliteration method, which once connoted Western culture by being progressive, but now is hackneyed. One of the characteristics of new-type gairaigo is that when the source word ends with [t], the [t] is replaced by [ts!]—whereas traditionally, it is replaced by [to] (Akizuki 2005; Kubozono 1999). A comparison of traditional loanwords and new-type gairaigo is shown in Table 5.11.

190

Table 5.11. Examples of the comparisons between traditional loanwords and New-type gairaigo Source Words Traditional Loanwords New-type Gairaigo a. point [pointo] [poin!"] b. perfect [p#$%ek"&to] [p#$%ek"&!"&] c. pretty [p"'iti$] [p"'ic(i$]

check (it) [c(ekk"] ([itto]) [c(eki], [c(eke] d. check (it out) [c(ekk"] ([itto #"to]) [c(eki'#] (Source: Akizuki 2005:191)

The traditional transliterating method epenthesizes ["] after a word final consonant, since ["] is the shortest vowel in Japanese; therefore, it is likely to be devoiced when the devoicing environment is satisfied. However, as discussed above, word final [t] does not follow this process: [o] is epenthesized instead of ["]. Thus, point

(Table 5.11.a) becomes [pointo], and perfect (Table 5.11.b) becomes [p#$%ek"&to]. The reason why the traditional method uses [o] instead of ["] for epentheses is that traditional

Japanese does not have the [t"] sequence and applying [!"] instead of [to] requires

119 another epenthesis, namely [s] in between [t] and ["]. Another reason why [o] is used for epenthesis is that Japanese occasionally distinguishes singular and plural forms of a word whose source word ends with a [t] by assigning [to] and [!"]. For example, sweet is [s"i$to], whereas sweets is [s"i$!"&]. In spite of the advantages of the traditional method, such as one less instance of epenthesis and morphological compliance with the

119 It may be more appropriate to say that the use of [!"] instead of [to] involves complete consonant substitution, rather than epenthesis. 191

source language, new-type gairaigo use [!"]. The users of new-type gairaigo prefer

[poin!"#] and [p$%&ek"#!"#] because they reflect the pronunciations of the source words points and perfects, even though these forms include a grammatically incorrect extra morpheme -s. This method violates morphological compliance because these words are supposed to be singular. Akizuki claims that they prefer [poin!"#] and [p$%&ek"#!"#] due to the intrinsically short vowel ["], which has the potential to be devoiced. The users of new-type gairaigo sacrifice the grammatical advantages for the phonetic (at least impressionistic) similarities to the source words.

If this is correct, one may wonder, then, why the new-type gairaigo for pretty

(Table 5.11.c) is more dissimilar to the source word pronunciation than the traditional loanword. As discussed above, substituting [ti] for [c'i] is ‘old fashioned’ these days because there are a large number of marginal and non-marginal items with [ti]. Akizuki clarifies that the innovative sound sequence [ti] has been used in lieu of [c'i] since the end of WWII and does not connote a Western atmosphere anymore. For younger people, the ‘old-fashioned’ transliterating method, which uses [c'i] for [ti] in source words is more novel.120

Akizuki’s explanation of new-type gairaigo is somewhat inconsistent. For example, using [!"] for transliterating word final [t] of the source language is also ‘old-fashioned’

(cf. ['$ts"]

120 However, as discussed above, it was also common to use [te] for [ti] as an alternative repair strategy. 192 new-type gairaigo users find novelty in an ‘old-fashioned’ repair strategy. Nonetheless, whether Akizuki’s analyses are accurate or not is not the issue here. The important fact is that Japanese speakers invent new forms of existent loanwords relatively freely, yet word final [ti] and [t!] are not the repair strategies for word final [t] in the source language. If the vowels in [ti] and [t!] were devoiceable, they would be the perfect candidates for representing word final [t] in source words. Why don’t the users of new-type gairaigo adopt [ti] or [t!] for final [t]? I argue that the vowels in [ti] and [t!] are different from the ones in other sound sequences, such as [c"i], [#!], [ki], [k!], ["i], and [s!]. As discussed above, the vowels in [ti] and [t!] never devoice; therefore, it is assumed that they are more robust than high vowels in other sequences. This robustness makes the vowels in [ti] and [t!] different from the high vowels in other sequences.

5.10 Distributions of [ti]/[di] and [t!]/[d!] in Marginal Items

I remember that when I was a child in the early 1970s, Japanese people sang the song

“Happy Birthday to You” as [h$ppi% b$%s!de% #!% j!%]. Now, most young Japanese people sing this song as [h$ppi% b$%s!de% t!% j!%]. This means that the Japanese can pronounce the [t!] sequence—indeed, the replacement of [#!%] with [t!%] actually took place in singing. However, as discussed above, the replacement of [#!] with [t!] is not observed in frequently used non-marginal words, such as [#!%] ‘two’. It appears that the innovative sound sequences [t!] and [d!] are rare because the major source language

(i.e. English) does not have many potential source sequences [tu], [t&], [du], or [d&].

However, I still have the following questions. Do Japanese speakers really have few 193

chances to encounter (or use) the sound sequences [t!] and [d!]? Why are frequently used ["!] sequences, such as [ts!#] ツー ‘two’ and [ts!i$] ツイン ‘twin’, not replaced by potential phonological variants with [t!]?

To answer the first question, I examine marginal items. I investigate the use of innovative sound sequences [ti], [t!], [di], and [d!] in song titles by the British rock band the Beatles (1962~1970) in order to find out if the distribution is similar to the non- marginal items. Why the Beatles? Just as they influenced American culture, the Beatles have significantly influenced Japanese popular culture. They have been consistently popular ever since they first appeared, and new albums have been released even in this century. Consequently, different age groups are familiar with songs by the Beatles.

Japanese middle school students are familiar with the Beatles because some of their songs are in music textbooks. Many English language teachers in Japanese middle schools use Beatles’ songs as teaching material because their lyrics are written in plain

English (Terashima 2007). One may still wonder why song titles are relevant to this issue.

In Japan, Western song titles and movie titles are either transliterated into katakana, or in some cases, translated into Japanese. Thus, many Japanese are familiar with the transliterated titles, rather than the original titles. The analysis of the transliterated song titles by the Beatles in Table 5.12 demonstrates how the Japanese are frequently exposed to [t!] and [d!] sequences.

194

Table 5.12. List of the titles of songs recorded by the Beatles that contain, or potentially contain [ti], [t!], [di], or [d!] Song Titles in IPA Song Titles in Japanese Original Song Titles "Please Please Me" (1963) ["i so# h"# s!t"ndi!$! ze"] アイ・ソー・ハー・スタンディ I Saw Her Standing ング・ゼア There [%"b! mi# d!#] ラヴ・ミー・ドゥ Love Me Do [d!# j!# &onto t!# no! " ドゥ・ユー・ウォント・トゥ・ Do You Want To Know A Secret 'i#k!%etto] ノウ・ア・シークレット [(!is!to "ndo '"!to] ツイスト・アンド・シャウト Twist And Shout (Russell and Medley） "With the Beatles" (1963) [o#%! "i $otto t!# d!#] オール・アイヴ・ゴット・トゥ All I've Got To Do ・ドゥ [ti%! zea &oz! j!#] ティル・ゼア・ウォズ・ユー Till There Was You (Willson) "A Hard Day's Night" (1964) [j!# kj"nto d!# z"tto] ユー・キャント・ドゥ・ザット You Can't Do That

"Beatles For Sale" (1964) [k"ns"s! 'iti] カンサス・シティ Kansas City (Stoller and Leiber) [p"#ti# &" sonom"m"ni] パーティはそのままに I Don’t Want to Spoil the Party [ho&"tto j!# "# d!#i!$!] ホワット・ユー・アー・ドゥー What You're Doing イング "Help!" (1965) [tiketto t!# %"ido] ティケット・トゥ・ライド Ticket To Ride [j!# %"ik! mi# t!# m"cc'i] ユー・ライク・ミー・トゥ・マ You Like Me Too ッチ Much (Harrison) [di)*i# mis! %i)*i#] ディジー・ミス・リジー Dizzy Miss Lizzie (Williams) "Revolver" (1966) [%"! j!# t!] ラヴ・ユー・トゥ Love You To (Harrison) 195

[!i "!nto t# te$# j#%] アイ・ウォント・トゥ・テル・ I Want To Tell You ユー (Harrison) [&otto t# &etto j#% int# m!i ゴット・トゥ・ゲット・ユー・ Got To Get You Into My Life $!i'#] イントゥ・マイ・ライフ [t#mo$o% neb!% no#z#] トゥモロー・ネヴァー・ノウズ To morrow Never Knows "Sgt. Pepper's Lonely Hearts Club Band" (1967) [&etti!&# bet!%] ゲッティング・ベター Getting Better [hoe( !im# )ikk##ti% 'o%] ホエン・アイム・シックスティ When I'm Sixty-Four ー・フォー Magical Mystery Tour (1967) [m!*+ik!$# mis#te$i% ,#!%] マジカル・ミステリー・ツア Magical Mystery Tour ー "The Beatles" (White Album) (1968) [di! p#$#%dens#] ディア・プルーデンス Dear Prudence [ob#$!di ob#$!d!] オブ・ラ・ディ、オブ・ラ・ダ Ob -La-Di, Ob-La-Da [z! konti-j#%i!&# s#to%$i% ザ・コンティニューイング・ス The Continuing Story Of Bungalow Bill ob# b!!&!$o% bi$#] トーリー・オブ・バンガロー・ビル [m!%s! m!i di!] マーサ・マイ・ディア Martha My Dear [eb#$ibodi%z# &otto エヴリボディーズ・ゴット・サ Everybody's Got Something To Hide s m i! t h ido ムシング・トゥ・ハイド・エク ! #) &# #% ! Except Me And My セプト・ミー・アンド・マイ・ ek#sep#to mi% !ndo m!i Monkey モンキー mo!ki%] [sek#)i% sedi%] セクシー・セディー Sexy Sadie [s!boi to$!''#$#] サヴォイ・トラッフル Savoy Truffle (Harrison) "Yellow Submarine" (1969) [o%$# t#&ez!% n!#] オール・トゥゲザー・ナウ All Together Now [it,# o%$# t#% m!cc)i] イッツ・オール・トゥ・マッチ It's All Too Much (Harrison) "Abbey Road" (1969) [k!m# t#&ez!%] カム・トゥゲザー Come Together 196

[h!" m!#$es%ti"] ハー・マジェスティー Her Majesty "Let It Be" (1970) [t% ob% !s%] トゥ・オブ・アス Two Of Us [di&% ! poni"] ディグ・ア・ポニー Dig A Pony [di&% itto] ディグ・イット Dig It (Lennon - McCartney - Harrison - Starkey) [dz! 'o(& !ndo )!indi(&% ザ・ロング・アンド・ワインデ The Long and Winding Road 'o"do] ィング・ロード PAST MASTERS VOLUME ONE (1988) [h%'om% mi" t%" j%"] フロム・ミー・トゥ・ユー From Me To You PAST MASTERS VOLUME TWO (1988) [dei to'ipp!"] デイ・トリッパー Day Tripper ['edi" m!donn!] レディ・マドンナ Lady Madonna (Source: Shinko Music 1991)

Notes: All songs were composed by John Lennon and Paul McCartney except as noted below. Twist And Shout (Bert Russell and Phil Medley 1960） Till There Was You (Meredith Willson 1950, 1957) Kansas City (Mike Stoller and Jerry Leiber 1952) You Like Me Too Much (George Harrison 1965) Dizzy Miss Lizzie (Larry Williams 1958) Love You To (George Harrison 1966) I Want To Tell You (George Harrison 1966) Savoy Truffle (George Harrison 1969) It's All Too Much (George Harrison 1968, 1969) Dig It (John Lennon, Paul McCartney, George Harrison and Richard Starkey 1970)

The following song titles include words that could be transliterated with トゥ [t%]; however, they are excluded from the list because they were given Japanese titles:

I Want To Hold Your Hand (1963); I'm Happy Just To Dance With You (1964); Things We Said Today (1964); Everybody's Trying To Be My Baby (Perkins 1964); You've Got To Hide Your Love Away (1965); You're Going To Lose That Girl (1965)

I Don't Want To Spoil The Party (1964) is in Table 5.12, even though it has a Japanese title, because its translated title includes [p!"ti"] ‘party’.

197

Table 5.13. Distribution of [ti], [t!], [di], and [d!] in the titles of songs recorded by the Beatles [ti] [di] [t!] [d!] 8 10 16 5 till standing to (10) do (4) ticket dizzy too (2) doing party dear (2) tomorrow All Songs city Ob La Di two (213 songs) getting Sadie together (2) sixty dig (2) continuing lady Majesty winding

Table 5.12 and Table 5.13 show that [t!] and [d!] are not particularly rare compared with [ti] and [di]. This fact is largely due to the high frequency of the words to and do in the original titles. Although the occurrence of [t!] and [d!] is not rare, not all source words that contain /tu/ are transliterated with トゥ [t!]. For example, the source word of [ts!"#] ‘Tour’ ツアー in Magical Mystery Tour (1967) contains /tu/; nevertheless, it is not transliterated as トゥアー [t!"#]. The consonant clusters /tw/ and

/tr/ do not create a [t!] sequence as seen in [ts!is!to] (not [t!is!to]) ツイスト in

Twist And Shout (1963) and [to$"%%!$!] (not [t!$"%%!$!]) トラッフル in Savoy

Truffle (1968).121 The difference between items listed in the dictionary and items in the song titles is [ts!#] ツー and [t!] トゥ ‘two’ in Two Of Us (1970). This means that the borrowing processes of marginal words and non-marginal words are different, at least for

121 However, there are many examples of [t!] sequences in source words with consonant clusters. A phonological variant of [ts!$i#] ‘tree’ can be observed in [t!$i#] in ヨシュア・トゥリー [jo&!" t!$i#] (The Joshua Tree U2, 1987). 198

two. If [ts!"] ツー is frequently used in non-marginal situations, why is it not used for song titles? Similarly, if [t!] トゥ is pronounceable for Japanese speakers, why does

[ts!"] ツー ‘two’ not have a variant [t!] トゥ or [t!"] トゥー?

5.11 Theoretical Account

We have seen that in non-marginal items, there are differences in the distribution patterns between [ti]/[di] and [t!]/[d!]: items with [t!]/[d!] are much fewer than items with

[ti]/[di]. Items with [t!]/[d!] do not seem to have phonological variants, while a number of items with [ti]/[di] have variants. On the other hand, in marginal items, the token frequency of [t!]/[d!] is not necessarily low compared with [ti]/[di]. Phonological variants with [t!]/[d!] as well as [ti]/[di] can be found in marginal items; however, it is complicated to predict when these innovative sound sequences appear.

In this section, I will discuss whether a theoretical account for this unpredictability of [ti]/[di] and [t!]/[d!] is possible. First, as we have seen in §5.5.1, the

Structuralists’ view that [ti]/[di] and [t!]/[d!] emerged in order to fill gaps might partially explain the actuation of these sound sequences. However, it does not explain the unbalanced distribution. Second, within the framework of Optimality Theory, Ito and

Mester (1995b) propose the constraints *TI and !TSU. The constraint *TI “excludes non-palatal coronal consonants followed by the high front vowel [i]” (Ito and Mester

1995b:827) and the constraint !TSU prevents [t] followed by [!], requiring [ts!] instead

(Ito and Mester 1995b). The constraint !TSU could be applied to [d] to prevent [d!] and require [dz!]. Ito and Mester (1995b) demonstrate that the constraint !TSU is more 199 peripheral (ranked higher) than *TI, thereby creating a tendency that the sequences [ti] and [di] are more common than the sequences [t!] and [d!]. When one constraint is more peripheral than another, the former is higher ranking and more strictly obeyed than the latter.

As discussed in Chapter 4, the constraint *TI, which prevents [si], [ti], [di], and

[zi], may be typologically common. Yoshida (2001) indicates that Brazilian Portuguese speakers and female Cairene Arabic speakers palatalize and affricate /t/ and /d/ in front of a high vowel. Yoshida also demonstrates that palatalization and affrication of dental or alveolar stops appears across a word boundary (English last year, did you) and a morpheme boundary (Korean kat" + i [kat#i] ‘together’, pat" + i [pat#i] ‘plowed field +

SUBJECT’). On the other hand, affrication of /tu/ and /du/ may not be as common as affrication of /ti/ and /di/. It is possible to claim that the constraint *TI is more common than the constraint !TSU. However, this does not support the claim that !TSU is ranked higher than *TI, or that !TSU is more peripheral than *TI. If *TI is more likely to be obeyed than !TSU, then *TI should be more peripheral and ranked higher then !TSU. In that case, we should see more [t!]/[d!] than [ti]/[di].

Even if the ranking !TSU > *TI is correct, it does not explain when [ti], [di], [t!] and [d!] appear. As we have seen in the chapter on voiced geminate obstruents, the OT account of the appearance of innovative sound sequences must depend on indexing— such as, ASSIMILATED FOREIGN and UNASSIMILATED FOREIGN. However, as can be seen in words like [kj$nde%]~[kj$ndi%] ‘candy’, diachronically speaking, the latter is more assimilated than the former. It was once pronounced [kj$nde%], but now it is pronounced 200

[kj!ndi"]. Nonetheless, according to OT, the latter should be indexed as UNASSIMILATED

122 FOREIGN due to the innovative sound sequence [di]. This example demonstrates that the more a word is assimilated into Japanese, the more it is dissimilated phonologically.

That is to say, an increase in usage of a foreign word in the Japanese lexicon does not necessarily promote phonological nativization. This is due to the following historical facts. The word ‘candy’ became more common in Japan over time. At a certain point in time, people began to prefer [kj!ndi"] over [kj!nde"]. Another question is where the index comes from. Why are some words indexed as ASSIMILATED FOREIGN, while others are indexed as UNASSIMILATED FOREIGN? OT does not answer this question satisfactorily.

Thus, even though the OT analysis is correct, it does not explain the causes of the unbalanced distributions of [ti], [di], [t#], and [d#].

Both the Structuralists’ analysis and the OT analysis explain, to some extent, why

[ti], [di], [t#], and [d#] emerged in modern standard Japanese. But they do not account for the unbalanced distributions. In order to explain the unbalanced distributions of

[ti]/[di] and [t#]/[d#], we must consider extra-grammatical factors—such as culture and education.

122 Ito and Mester (1999:77) state that the input /siti/ that is realized as [$ic$i] is “roughly” indexed as NATIVE, [$iti] as ASSIMILATED FOREIGN, and [siti] as UNASSIMILATED FOREIGN. According to this classification, the input of [kj!nde"] should be indexed as NATIVE. However, since /baggu/ [b!kk#] ‘bag’ is indexed as Assimilated Foreign (Ito and Mester 1999:74), I indexed [kj!nde"] as ASSIMILATED FOREIGN and [kj!ndi"] as UNASSIMILATED FOREIGN. 201

5.12 Cultural Account

We have seen that the pattern of borrowing is complex, and for this reason, it is impossible to predict when [ti], [di], [t!], or [d!] appears in a loanword. Rule-based phonology cannot explain when rules are applied. Constraint-based phonology also cannot predict when the constraints prevent an innovative sound sequence from appearing as output. Phonetics cannot provide a sufficient account, either. The only way to account for this complex linguistic phenomenon is sociolinguistics and historical linguistics.

Language contact is an external cause of language change, and for this reason, it is often regarded as a special case of language change (Ohala 2003; Blevins 2004).

Nevertheless, studying language contact is indispensable in investigating the exact cause of the emergence of innovative sound sequences in modern standard Japanese. The innovative sound sequences [ti], [di], [t!], and [d!] emerged due to the influx of loanwords from Western languages, in particular, English. Thus, the main cause for the emergence of [ti], [di], [t!], and [d!] is language contact. McMahon (1994) emphasizes that language contact and lexical borrowing depend on bilingualism, and the degree of bilingualism required for lexical borrowing varies from “very restricted” to “far- reaching”. As we have seen, loanwords in Japanese have entered by various routes, such as English language education, computer technology, music, movies, and other products of popular culture. I will explore these routes of contact to identify the exact cause of the disproportional distributions of [ti], [di], [t!], and [d!]. 202

Phonetic variations of items with [ti], [di], [t!], and [d!], we have seen, suggest that there is no single pattern or rule that transfers the phonological forms of source words into the phonological forms of loanwords. For example, two /tu"/ is transferred into

[#!"] as a non-marginal item, whereas it appears as [t!"] in the Beatles’ song Two of Us.

However, even in marginal items, /tu/ sequences are not automatically transferred into

[t!]. The name of the band U2 (1980~) is pronounced as [j!: #!:]. The title of the

Hollywood blockbuster movie Back To The Future Part II (Zemeckis 1989) is pronounced as [b$kk! t!" z$ %j!"c&$" p$"to #!"], which demonstrates that the transfer from /tu/ to [t!] and the transfer from /tu/ to [#!] coexist within a title of a movie. These examples suggest that it is impossible to predict whether a /tu/ sequence in the source word will be transferred into [tu] or [#!] without considering the cultural context. The cultural context that affects the phonological forms of foreign items is complex. Some items are used in daily conversation by speakers of any generation, and other items may be used only by certain age groups in limited circumstances.

Before 1930, Japanese actively borrowed items from English and other Western languages. However, Japan’s occupation of Manchuria in 1931, entering into a war with

China in 1937, and entering into a war with the United States and the British Empire in

1941 led to anti-Western ideology, which repudiated Western values (Loveday 1996).

According to Loveday, as early as 1939, borrowing of English items stagnated. After

1940, attempts were made to obliterate the use of English-origin loanwords. The teaching of English was reduced in 1931, and English teaching was limited until the end of WWII. 203

Loveday (1996) states that “(t)he war years left a whole generation with no English competence, or very little” (1996:75).

After WWII, the situation changed completely, from anti-Western to westernization. The Japanese began to see knowledge of English as the “key to obtain social advantages,” “the medium through which to learn about democracy,” and “a code for liberals and internationalists” (Loveday 1996:75). Under the post-war education system, almost all Japanese middle school students study English.123 As we have seen, there is a generation gap for Japanese born in the 20th century; one generation experienced very limited English contact in their youth, and the other had a considerable amount of English education. These two generations show different attitudes toward the use of loanwords (Loveday 1996).124 The existence of two groups who have experienced different attitudes toward Western culture and language, in particular English, has affected the distribution of the innovative sound sequences.

Degrees of knowledge of the source language affect how loanwords are borrowed.

The examples in (6) demonstrate phonological variants of loanwords from Russian into

Asiatic Eskimo. The early loanwords were borrowed in the pre-Soviet period when there was little bilingualism. The later loanwords were borrowed in the later Soviet phase when

123 According to Tanaka (2002), Japanese middle school students must take one of the following foreign languages: English, French, German, Chinese, and Korean. Thus, English is not and has never been an obligatory subject in Japan. However, more than 99% of students have no choice other than taking English because language classes other than English are not offered except for special private schools. 124 To be more precise, there may be three generations because students in school before 1930 may have received more English education than students in school in 1930-1945. 204 the Asiatic Eskimos studied Russian as a second language in Russian schools (McMahon

1994).

(6) Russian Early loan Later loan [bl!utc!] [pljusa] [bljutsca] ‘saucer’ [t"aj] [saja] [t"aj] ‘tea’ [tabak] [tavaka] [tabak] ‘tobacco’ [pat"ka] [paskaq] [pat"ka] ‘bundle’ (Thomason and Kaufman 1988:33, cited in McMahon 1994:205)

As indicated in (6), the phonological forms of later loanwords are more similar to the source words. The case of loanwords from Russian to Asiatic Eskimo suggests that the degree of knowledge of the source language affects the degree of phonological adaptation of loanwords.

Several studies indicate that even minimal acquisition of L2 phonology can affect the perception and production of L1 phonology (M. Anisfeld, E. Anisfeld and Semogas

1969; Anisfeld and Gordon 1971; Flege 1987; Guion 2003). The study by M. Anisfeld, E.

Anisfeld and Semogas (1969) suggests that only a small amount of second language learning might affect the subjects’ perception and production of L1. As discussed above, those who spent their youth during WWII received little or no English education, whereas almost all children in the post-war generations studied English in middle school. The contrast in terms of English education might have affected the phonological forms of loanwords in Japan.

205

5.13 Differences in Generations

Imamura (1996) conducted research on “the difference in language use based on age,” asking questions of 150 students (age 18 to 20) at Ch!bu University in Japan. In her study, the subjects reported that older speakers, the subjects’ grandparents’ generation, cannot pronounce [ti] and [di]. Table 5.14 does not reflect actual speech of older speakers accurately, but it provides some stereotypical pronunciations of older speakers from younger speakers’ point of views.

Table 5.14. Samples of stereotypical pronunciations of [ti] and [di] by the older speakers

Standard Old Generation [ti!!" pe#p$#] [te!!" pe#p$#] ‘tissue paper’ 125 [en" ti# ti#] [en" te# te#]/[en" c!i# c!i#] ‘NTT’ [diz"ni# %$ndo] [dez"ni# %$ndo] ‘Disney Land’ [!i# di#] [!i# de#] ‘CD’ (Source: Imamura 1996:93)

Imamura (1996) claims that older speakers do not pronounce [ti] and [di] because they are not accustomed to these innovative sound sequences. Older speakers might not have had adequate opportunities to receive English language education compared with university students in the 90s, in particular in the areas of listening comprehension and speaking. Moreover, when older speakers were young, there were many fewer foreign items in daily use. The ages of the subjects in Imamura’s study were from 18 to 20;

125 Nippon Telegraph and Telephone corporation 206 therefore, her study may not reflect how older speakers actually pronounce foreign items.

However, her study clearly indicates that certain pronunciation patterns are associated with the cultural and educational background of a certain age group. In other words, if young speakers fail to use [ti] and [di] appropriately, they may be regarded as ignorant of

English or they may not be considered as truly belonging to the younger generation.

5.14 Difference in the Timing of Borrowing

By following the classification by Bloch (1950), Quackenbush and !so (1990) state that

[c!i] in [c!i"m#] ‘team’ and [$#] in [$#"] ‘two’ were entrenched as “conservative

Japanese,” whereas [ti"m#] and [t#"] indicate the diffusion of “innovative Japanese.”

However, Quackenbush and !so remark that the labels “conservative Japanese” and

“innovative Japanese” are not necessarily chronological. For example, an English word ending in -tic can be pronounced as either [c!ikk#] or [tikk#], but the timing of borrowing does not always decide which form will be adopted. Consequently,

Quackenbush and !so conclude that it is not an “absolute” rule that older loanwords adopt conservative forms and that newer words accept innovative forms, although it is a general tendency that might be helpful for teaching Japanese.

5.15 Difference in Cultural Context

Words belonging to different cultural background may undergo different sound change processes. Yaeger-Dror and Kemp (1992) and Yaeger-Dror (1996) report word-frequency effects on word-specific sound change in Quebecois French. Quebecois French 207 underwent a vowel shift; however, vowels in a particular group of words that are associated with the church, the military, and the schools did not shift. Their studies suggest that sound changes occur independent of phonological properties and dependent on the cultural background.

5.16 The Complex Structure of Cultural Values and [ti], [di], [t!], and [d!]

As we have seen, the use of [ti], [di], [t!], and [d!] in certain circumstances connotes the complex cultural background of the speaker and the word. If speakers pronounce

[t!"] instead of [#!"] meaning ‘two’ in daily conversations, many listeners, in particular older listeners, will not understand the utterance. This pronunciation may give the impression that the speakers know English, but it is more likely to give the impression that the speakers are pretentious. On the other hand, pronouncing [t!dei] may be accepted by all generations without being regarded as pretentious. The degree of acceptance of loanwords is closely connected to their cultural and historical backgrounds, and it is therefore impossible to generalize. There are only general tendencies based on cultural and historical backgrounds, as illustrated in Table 5.15.

208

Table 5.15. General tendencies of cultural values associated with [ti], [di], [t!], and [d!] [c"i] [te] [to] [#!] [do] [ti] [di] [t!] [d!] [$!] ! "

Phonetic different from the source closer to the source language Similarity to the language pronunciation pronunciation source language

Timing of old new borrowing

Generation older generations younger generations

Degree of English less English education more English education education

Cultural reference Japanese culture Western culture

Identity maintain Japanese identity loss of Japanese identity

Negative ignorance of English pretentiousness impression Intelligibility intelligible to all generations unintelligible to older generations

Table 5.15 shows the general tendencies of cultural values relevant to [ti], [di],

[t!], and [d!]. For example, pronouncing [t!] or [d!] might connote pretentiousness or

Western culture. However, it is important to note that such cultural values are relative, and each sound sequence does not carry these cultural values independent of the context.

The actual cultural values of these sound sequences are determined by how the words that contain these sound sequences are pronounced by the speaker. Pronouncing a word with a

[t!] or [d!] sequence does not necessarily convey a negative impression. Similarly, pronouncing a word with [ti] and [di] may convey a negative or positive impression 209

depending on the context. For example, saying [t!"] instead of [#!"] when meaning ‘two’ sounds pretentious not because the sound sequence [t!] has an intrinsic and absolute cultural value of pretentiousness, but because that item is normally pronounced as [#!"].

The cultural values of the pronunciations [t!] and [#!] are determined only when they are in the context. That is why pronouncing [t!dei] ‘today’, which is the default pronunciation, does not connote pretentiousness. In contrast, [#!dei] gives the impression that the speakers belong to an older generation or that they are ignorant of

English. The sequences [ti] and [di] are used daily in everyday items such as [$is! ti"]

‘ice tea’ and [ti"%$#!] ‘T-shirt’; thus, pronouncing [te"%$#!] might give an impression that the speaker is old, as Imamura’s (1996) study suggests. On the other hand, pronouncing [ti"m!] ‘team’ may give a negative impression, depending on the context, because the word is normally pronounced as [c%i"m!].

5.17 Importer’s Influence on the Phonological Forms of Loanwords

Every loanword has its own history: it is not automatically borrowed from one language to another. Some people must make a decision to import a word under certain circumstances. Thus, the importers’ cultural backgrounds and their knowledge of the source language affect their decisions on what sound sequences will be used for the loanword. Moreover, the speakers’ knowledge of the source language must be taken into consideration because the phonological form of the loanword must be accepted by the speakers in the community. Otherwise, a new loanword might emerge, but it will not diffuse. We have already seen how the intelligentsia in the early 20th century tried to 210 reflect the original spellings when transliterating foreign personal and place names into

Japanese. Such a borrowing process requires knowledge of the source language. On the other hand, speakers with little literacy in the Western languages imported foreign words as well. The phonological forms of the items borrowed by the intelligentsia and the items borrowed by the general public with limited foreign language literacy can be very different. In the next section I will discuss how the background differences of importers affect the phonological forms of loanwords.

5.18 The Intelligentsia, the General Public, and the Pseudo-intelligentsia

Yamada (2005) claims that there are three stages of borrowing English into Japanese: the first stage—from 1868 to 1921; the second stage—from 1921 to 1965; and the third stage—from 1966 to the present. In the first stage, borrowing was conducted mostly by the intelligentsia. In the second stage, both the general public and the intelligentsia borrowed foreign words. In the third stage, the general public and the pseudo- intelligentsia have been leading the borrowing. Yamada (2005) does not explicitly define who the ‘pseudo-intelligentsia’ are, but it could refer to those who engage in trading businesses, the music industry, the movie industry, the fashion industry, the sports industry, and the food service industry. These people have actively imported foreign words, as well as foreign products. We have seen that English song titles are transliterated differently from non-marginal items. This is due to the fact that the transliterations were conducted by people who worked for the music industry, and they intended to make the transliterated titles as authentic as possible for promotional purposes. 211

This promotional strategy is successful only when the users of the products appreciate it.

Those who appreciate such a transliteration method have some knowledge of English but need the assistance of the katakana transliteration. It is important to stress that if the users’ command of English is good, the katakana transliteration is unnecessary. Thus, the phonological form of a loanword is affected by the importer’s cultural background and knowledge and the user’s command of English, as well as the timing of the borrowing.

5.19 Conclusion

We have seen that the causes of phonological variants vary. The appropriateness of a phonological form of a loanword depends on the background of the word and associated cultural values. These causes are complex and cannot be explained by using one cause, for instance, old versus new. New words do not necessarily contain innovative sound sequences. The phonological form of a loanword is determined by the cultural background, the timing of the borrowing, and source language literacy of the users and importers.

CHAPTER 6

DISCUSSION

6.1 Stratification or No Stratification

I have argued that the phonology of innovative sound sequences can be explained without referring to language stratification. In contrast, most OT analysts of Japanese suppose that lexical stratification affects synchronic Japanese phonology (Ito and Mester 1995a,

1995b, 1999, 2003; Fukazawa and Kitahara 2002). The claim that synchronic Japanese phonology depends on lexical stratification is based on the assumption that the constraints *NT, *P, and *DD are applied differently to each stratum. In Chapter 2, I discussed the evidence, arguing that the issue of geminate voiced obstruents is not explainable by the constraint *DD; rather, it requires a sociolinguistic account. In Chapter

3, I questioned the legitimacy of the constraint *P, which does not seem to be innate or universal. In the next section, I argue that *NT, the constraint against post-nasal voicing, is not ranked high enough to be effective in modern standard Japanese. By refuting the effectiveness of these constraints, I propose that lexical stratification is a linguistic fact that should be explained by diachronic linguistics and sociolinguistics, but not synchronic linguistics.

6.2 The Constraint *NT and Modern Standard Japanese

The only piece of evidence that suggests *NT dominates Yamato items in modern standard Japanese is that the gerundive and past forms of a verb whose stem ends with a 213 nasal or a /b/ create a [nd] sequence.126 However, this evidence is weak because verbs that end with a /g/ also voice the following consonant of a gerundive or past morpheme

(McCawley 1968).

Ito and Mester (1995b, 1999, 2003) claim that the constraint *NT is effective for

Yamato items in modern standard Japanese for the reasons shown in (1).127

(1) a. The constraint *NT is universal. b. Yamato items follow *NT and, therefore, do not have postnasal voiceless consonants. c. The past marker /-ta/ and the gerund marker /-te/ are voiced when preceded by a nasal. d. Non-Yamato items can have postnasal voiceless consonants. e. Therefore, Yamato items are bound by the constraint *NT, but non-Yamato items are free from *NT.

Reason (1a) appears to be a well-founded claim,128 and reason (1b) is largely true, with several exceptions that I will discuss later. However, satisfying the conditions described in reasons (1a) and (1b) does not require lexical stratification because as long as the relevant faithfulness constraints are ranked higher than *NT, the correct output is obtained. The problem arises when reason (1c) is taken into account. Voicing of /-ta/ and

126 Ito and Mester (1999) state that there is another type of evidence that shows morphological alternations. They claim that the combination of the verb /fumu/ and another verb will create a [nd] sequence, such as fumu ‘to step on’ + tsukeru ‘attach’ = fun-dzukeru ‘trample on’, not *fun-tsukeru. However, this item is also used as an example of rendaku ‘sequential voicing’. See Vance (1987:137). 127 Ito and Mester (1995a,1995b) claim that the constraint *NT is also effective for Mimetic items. But this claim is refuted by Asano (2004). (See also Chapter 1.) Inkelas and Zoll (2003) indicate that a markedness reversal is required for accounting for the phonology of Mimetic items. 128 I do not have any evidence to refute the existence of the constraint *NT; therefore, I assume that it is part of UG.

214

/-te/ apparently violates a faithfulness constraint but follows *NT. This requires some repairs. Since all Yamato items follow the constraint *NT, reversing the order of the ranking to *NT > FAITH does not affect the output of Yamato items. In order to allow items in other lexical strata to evade *NT, the faithfulness constraints must be reranked to the appropriate position in other lexical strata (the cophonology, or co-grammar approach), or multiple faithfulness constraints must be indexed with strata and appropriately positioned (the index-faithfulness approach). However, if reason (1c) is proved to be invalid, there is no reason to take lexical stratification into consideration. I argue that reason (1c) may have been valid in the past, but not in modern standard

Japanese.

Ito and Mester (1999) claim that the data in (2), which involve gerundive forms and past forms of Japanese verbs, follow the constraint *NT.

(2) BASE GERUNDIVE PAST !in- !in-de !in-da ‘die’ in- in-de in-da ‘leave’ yom- yon-de yon-da ‘read’ susum- susun-de susun-da ‘progress’ hasam- hasan-de hasan-da ‘put between’

cf. mi- mi-te mi-ta ‘see’ ha!ir- ha!it-te ha!it-ta ‘run’ kaw- kat-te kat-ta ‘buy’ (Source: Ito and Mester 1999:68)

215

Judging from the data in (2), it appears that *NT is solely responsible for the alternation /t/~[d]. However, the presentation of the data in (2) is incomplete, as well as misleading. McCawley (1968) presents past tense forms differently, as shown in (3).

(3) PRESENT PAST a. taberu tabeta ‘eat’ b. miru mita ‘look’ c. !aruku !aruita ‘walk’ d. yobu yonda ‘call’ e. !oyogu !oyoida ‘swim’ f. yomu yonda ‘read’ g. dasu dasita ‘let out’ h. matu matta ‘wait’ i. sinu sinda ‘die’ j. karu katta ‘trim’ k. wara!u waratta ‘laugh’ (Source: McCawley 1968:95 with modification by author)

Example (3e) !oyogu~!oyoida alternation does not involve postnasal voicing.

McCawley states that the voicing assimilation takes place when the stem-final consonant is voiced, except for /r/.129 His rule depicts the situation better, because it covers both

130 types of /t/~[d] alternations: sinu~sinda alternation and !oyogu~!oyoida alternation.

This is a piece of evidence that the /t/~[d] alternation depends on something other than the condition that /t/ is preceded by a nasal. Therefore, it is not conclusive that the /t/~[d]

129 Of course, the Sound Pattern of English (Chomsky and Halle 1968) and post-SPE phonology can provide a more precise natural class, such as [- continuant] and [+ voice]. However, this is not the central issue. 130 I also doubt that native Japanese speakers analyze sinda as sin + ta.

216 alternation in sinu~sinda reflects the constraint *NT. Voicing of gerundive and past markers is best described as a historical exception, just like “irregular English verbs that constitute a closed system of listed allomorphs and are not part of active rule-governed phonology” (Ito and Mester 2003:123).

Now I turn to reason (1b): Yamato items do not have postnasal voiceless consonants. Traditional Yamato items certainly do not have any postnasal voiceless consonants. In modern native items, however, there are more than a handful of non- marginal words that do not follow *NT. Words in (4) are some non-marginal modern native items with a postnasal voiceless consonant, which are relatively frequently used.

(4) a. [!nt!] ‘you’ b. [i"c#iki] ‘bogus’ c. [j!"c#!] ‘naughty’ d. [e!ko] ‘sit down’ e. [$"c#i] ‘feces’ f. [p!c#i!ko] ‘pachinko’ g. [k!c#i!ko] ‘clapperboard’ h. [n!nte] ‘how’ (cf. [n!nde] ‘why’) i. [n!nk!] ‘somewhat’ j. [ot!!kon!s$] ‘nincompoop’ k. [ot!"c#i%] ‘numskull’ l. [!mpont!%] ‘nitwit’ m. [c#impi&!] ‘hooligan/punk’ n. [mompe] ‘women’s trousers for work, such as farming’

217

In particular, (4h) [n!nte] ‘how’ is a crucial piece of evidence against the claim that *NT is active in modern standard Japanese because it has a minimal pair counterpart

[n!nde] ‘why’.

There are other words whose etymological origins are not completely known, as shown in (5).

(5) ["!k!] ‘leafhopper’ (Ota 2004) [t!mpopo] ‘dandelion’ (Ibid.)

The words in (5) may be etymologically non-Yamato items. But then why do some native speakers not classify them into Sino-Japanese or Foreign items? This inconsistency is another piece of evidence suggesting that the violation of *NT does not determine judgment of lexical stratification.

Some researchers may argue that items such as anta are syncopated; therefore

*NT is not applied (Ito and Mester 2003; Kawahara 2002). However, this claim does not hold because, historically, sin+te ~ sinde was also a syncopated form of sin + ite

(Mabuchi 1971).

6.3 Historical Account

As discussed above, *NT does not apply to non-Yamato items in most cases. However, as shown in Table 6.1, some Sino-Japanese items in counter phrases show post-nasal voicing. 218

Table 6.1. Sporadic postnasal voicing in counter phrases ‘3’ ‘4’ ‘1,000’ ‘10,000’ Counter for san- yon- (-is-)sen- (iti-)man-

blocks of tofu /tyoo/ -tyoo -tyoo -tyoo -tyoo boats /soo/ -soo -soo -soo -soo books /satu/ -satu -satu -satu -satu chairs /kyaku/ -kyaku -kyaku -kyaku -kyaku chapters /syoo/ -syoo -syoo -syoo -syoo cups /hai/ -bai -hai -bai -bai drops of liquid /teki/ -teki -teki -teki -teki footwear /soku/ -zoku -soku -zoku -zoku hanging scrolls /huku/ -puku -puku -puku -puku houses /ken/ -gen -ken -ken -ken lessons /ka/ -ka -ka -ka -ka long objects /hon/ -bon -hon -bon -bon pairs of objects /tui/ -tui -tui -tui -tui pieces /ko/ -ko -ko -ko -ko pieces of clothing /tyaku/ -tyaku -tyaku -tyaku -tyaku poems /syu/ -syu -syu -syu -syu points /ten/ -ten -ten -ten -ten rolls of kimono material /tan/ -tan -tan -tan -tan shots /hatu/ -patu -patu -patu -patu small animals /hiki/ -biki -hiki -b/hiki -b/hiki times /hen/ -ben -hen -ben -ben times /kai/ -kai -kai -kai -kai volumes /kan/ -kan -kan -kan -kan years of age /sai/ -sai -sai -sai -sai (Source: Ito and Mester 2003:140)

Ito and Mester (2003) call these Sino-Japanese counter phrases with post-nasal voicing “positive exceptions” because the overapplication of *NT to non-Yamato items 219 supports the claim that the constraint *NT is universal, rather than language or stratum specific. Nonetheless, Ito and Mester do not provide specific reasons for why these exceptions occur—violating the restriction of lexical stratification they propose. In contrast, historical phonology can satisfactorily explain these “exceptions” and other *NT related phenomenon. I argue that the constraint *NT was once ranked high enough to affect Japanese phonology, but later was reranked to a position where it is no longer effective. These exceptions show that there was such a change.

Based on the data provided in Table 6.1, Ito and Mester draw some conclusions, as shown in (6).

(6) a. Postnasal voicing does not occur uniformly. b. The numeral /san/ ‘three’ causes more postnasal voicing than others. c. The numeral /yon/ ‘four’ never triggers postnasal voicing. d. The numeral /sen/ ‘thousand’ and /man/ ‘ten thousands’ are somewhere in between. e. The counter /kai/ ‘times’ does not undergo postnasal voicing, whereas the counter /kai/ ‘floors’ does undergo postnasal voicing

As for (6e), Ito and Mester assume that the reason the constraint *NT does not apply to /kai/ ‘times’ is to “keep the form distinct from san-gai ‘third floor’ (from /kai/)”

(2003:274). However, this assumption is not supported by historical facts. Historical documents suggest that the pronunciations of the counters /kai/ ‘times’ and /kai/ ‘floors’ had been different until a certain period. The counter /kai/ ‘times’ was pronounced as

[k!"i] without a numeral, whereas the counter /kai/ ‘floors’ was pronounced as [k"i] 220 without a numeral. Nevertheless, in the 17th century ‘three times’ was pronounced as

131 [s!!k"!i], and ‘three floors’ was pronounced as [s!!#!i]. Since [k"!i] ‘times’ and

[k!i] ‘floors’ were already distinct from each other, there is no need to voice [k] in ‘third floor’. Therefore, the voicing of the consonant of the counter /kai/ ‘floors’ is not motivated by the need to be distinct from the counter /kai/ ‘times’.

As shown in Table 6.1, the difference between (6b) and (6d) is so subtle compared with the difference between (6b) and (6c), that I shall now concentrate on the difference between (6b) and (6c). Ito and Mester do not provide an explanation for why the numeral /yon/ ‘four’ never triggers postnasal voicing except for providing the fact that it has “another S word” (2003:274) namely, /shi/, which is avoided by Japanese speakers because its homonymy has the meaning ‘death’. Ito and Mester (2003:139) incorrectly label the numeral /yon/ ‘four’ as Sino-Japanese. In fact, according to Nihon

Kokugo Daijiten, the numeral /yon/ ‘four’ is a Yamato item /yo/ with a word-final nasal that is derived by analogy with /san/ ‘three’, which indeed is a Sino-Japanese item. The numeral /yon/ ‘four’ had not been used widely at least until the early 17th century.132

Before /yon/ ‘four’ became prevalent, the Sino-Japanese numeral /si/ ‘four’ was predominantly used. I will not discuss why /yon/ ‘four’ has replaced /si/ ‘four’ further.

The important fact is that unlike conventional numeral-counter compounds, postnasal

131 For example, see Diego Collado's Grammar of the Japanese Language (Latin: Ars grammaticae Iaponicae linguae) (1632). 132 Collado notes that the word for ‘fourth floor’ was pronounced as [$ik!i] (1632:71). 221 voicing is not applied to relatively recently introduced NT clusters due to the replacement of /si/ with /yon/.

As for exceptional applications of *NT to Sino-Japanese items, Takayama

(2003:42) notes that post-nasal voicing was observed extensively in the past, even in

Sino-Japanese items, such as [!"#$o] ‘application form’, [kempo%] ‘constitution’, and

[z"!ki] ‘shame’, as illustrated in (7).

(7) Old Pronunciation Modern Pronunciation [!&"#'(o] [!"#$o]! 願書 ‘application form’ [kembo%] [kempo%] 憲法 ‘constitution’ [z"!!i] [z"!ki] 慚愧 ‘shame’

By contrast, Takayama also notes that except for lexically and morphologically set items, post-nasal voicing does not necessarily occur in modern Japanese. For example, ["n"t"] ‘you’ and [bok)+no+)c$i] ‘my house’ are frequently pronounced as ["nt"] and [bok)#c$i], but never as ["nd"] or [bok)#'(i], as shown in (8)

(8) ["n"t"] > ["nt"] (*["nd"])! アンタ ‘you’ [bok)+no+)c$i] > [bok)#c$i] (*[bok)#'(i]) ボクンチ ‘my house’

Close historical examination of the data provided by Ito and Mester, as well as by

Takayama, suggests that *NT was so powerful that it even affected Sino-Japanese items until a certain period in the history of Japanese. This suggestion does not contradict the claim that *NT is part of UG (Ito and Mester 2003:141). It might have been ranked high 222 enough to affect even certain Sino-Japanese items and syncopated phrases. Later, *NT was reranked into a lower position so that it does not affect Japanese phonology anymore.

Since *NT does not affect Japanese phonology, newly created words do not have to follow the constraint. Most Sino-Japanese items that used to follow *NT stopped doing so. When a postnasal consonant is created because of syncope, it always violates *NT.

6.4 Problems with Lexical Stratification

So far, I have indicated that lexical stratification does not account for synchronic phonology. In this section, I provide more detailed reasons why lexical stratification is problematic for synchronic analyses.

6.4.1 Definition of Lexical Stratification

One might assume that Japanese speakers can correctly classify any items into the appropriate lexical strata without any difficulty. However, this assumption is not correct.

Moreover, even linguists disagree with each other on the definition of stratification used in the context of synchronic phonology. The most objective definition of stratification is etymological (see Vance 2002). However, many synchronic phonologists disagree with the use of etymology in their synchronic analyses (McCawley 1968; Ito and Mester

1995b, 1999, 2003; Fukazawa, Kitahara and Ota 1998; Gelbert and Kawahara 2007;

Inkelas and Zoll 2003). Instead, they use lexical strata based on phonological facts that can be observed in synchronic language. The problem is that the methods of stratification differ greatly among phonologists. As discussed in Chapter 1, McCawley (1968) takes the position that stratification is decided solely by phonological characteristics and any 223 etymological facts should be discarded. Consequently, according to McCawley’s analyses, etymologically foreign origin items that do not have any phonological characteristics marking them as [+ Sino] or [+ Foreign]—such as [mi!"k"] ‘milk’—are automatically classified into [+ Yamato]. This system appears persuasive and phonologically consistent. However, it is in fact circular because the phonological characteristics that classify lexical strata are obtained by the observation of items that are etymologically classified. This fact suggests that stratification by phonological characteristics is impossible without help from etymology.

Another stratification strategy that is described in Ito and Mester (1999) is more widely accepted, yet is an ambiguous combination of etymology and phonology. In this method, an item is classified into the correct stratum if appropriate constraints are violated. A problem arises in this method when an item does not violate any relevant constraints. Ito and Mester state, “[i]t is of course possible for a form to already obey all the constraints holding in [Yamato]. In such cases, it has the potential to be fully admitted to the morpheme class” (1995b:830). But they do not provide any clear criteria for classifying such a form. It appears that such an item is classified according to the etymological classification. This method also has a problem that is similar to

McCawley’s system: it has to depend on both phonology and etymology.

As we have seen, stratification that is independent of etymology is impossible.

Therefore, the best solution is to treat stratification as etymology, as Vance (2002) suggests.

224

6.4.2 Accessibility to Lexical Stratification

Some phonologists claim that Japanese speakers have access to knowledge of lexical stratification (Ito and Mester 1999; Kawahara 2002). Rice (1997, 2005), on the other hand, argues that it is impossible for children to acquire knowledge of lexical stratification from the data available to them. Japanese speakers can classify many items into the correct lexical stratum. But it is also true that the speakers do not know the correct classifications of many items. For example, Ota (2004) considers [!!k"]

‘leafhopper’ a Yamato item. However, according to Daijirin and Nihon Kokugo Daijiten,

[!!k"] appears to be etymologically a Sino-Japanese item. This confusion in classification is due to the fact that speakers use learned knowledge of lexical stratification when they classify items. If their knowledge is wrong, they misclassify, and if they do not have information of the stratum of the item, either they cannot classify, or they have to hypothesize. Consequently, there are many items that are misclassified.

Nonetheless, Japanese speakers have no trouble pronouncing any misclassified items.

Take Ota’s example once again. If he classifies [!!k"] ‘leafhopper’ into the Yamato stratum, why does he not start pronouncing the item [!!#"]? This suggests: (i) either lexical stratification is unrelated to phonological knowledge; (ii) the constraint *NT does not affect the phonology of Yamato items; or (iii) both.

225

6.5 Interactions of Grammars

I have been critical of the analyses by OT researchers; however, I do not disagree with all their claims. In fact, I strongly agree with Ito and Mester’s claim that considering exceptions—underapplication and overapplication of regularities—“leads to a deeper understanding of the way phonology, as a formal system of ranked constraints, interacts with lexicon” (2003:153). I also concur with their claim that such interactions will unite

“the nonuniformities of a historically evolved stratified lexicon into a single coherent synchronic system” (2003:153). But having faithfulness constraints that index lexical strata does not seem to help in establishing “a single coherent synchronic system,” even though strict domination is maintained. It would be consistent if constraints that affect items belonging to only a certain lexical stratum simply do not exist, or such constraints are ranked so low that they are inapplicable to Japanese phonology. I interpret

“exceptions” as evidence that apparent stratum-specific grammars can merge by creating

“a single coherent synchronic system.” I use the word “apparent” because I believe that all grammatical differences observed among lexical strata are superficial and not real.

Interaction between apparent stratum-specific grammars can be observed repeatedly in the history of Japanese phonology. Old Japanese did not have syllable-final

/!/, geminate voiceless obstruents, or /Cy/ clusters, such as [kj"] and [kjo]. These sound sequences became part of Japanese phonology partly due to the interaction between the apparent grammars of the Yamato stratum and the Sino-Japanese stratum.133 Because of

133 See Vance (1987) and Hamano (1998) for more detailed arguments with emphasis on the roles of the indigenous grammar of Japanese. 226 these changes, Yamato, Mimetic, Sino-Japanese, and even Foreign items are indistinguishable in terms of the presence or absence of these sound sequences.

Consequently, most current Japanese phonologists do not consider the constraints that prohibit these sound sequences to be stratum specific.134 This fact demonstrates that apparent stratum-specific grammars interact with one another and eventually create a uniform grammar. If this is the norm of historical phonology, then underapplications and overapplications, which Ito and Mester consider “exceptions,” are in fact the norm. The case of apparently very strictly followed constraints (e.g., *P in the Yamato and Sino-

Japanese strata) is actually an exception in terms of historical phonology. And historical linguistics can provide a perfect explanation for why this exception occurred (see Chapter

3). In contrast, Ito and Mester (2003) cannot provide any satisfactory grammatical explanations for what they consider “exceptions.” They appear to imply that exceptions should be ignored (2003:149). But just because “all rules are not exceptionless” does not mean that they can evade the responsibility of providing explanations for why these exceptions occur in spite of the presence of innate grammar. It is difficult to explain why these “exceptions” occur because they are in fact the norms. On the other hand, explanation is much easier if what Ito and Mester consider the “rules”—phonological phenomena that are explained by stratum-specific markedness constraints or indexed faithfulness constraints—are merely residues of the history of Japanese. By “residues,” I mean that the interaction of stratum-specific grammars is not reflected in the distribution

134 But see Ito, Mester and Padgett (1999) and McCawley (1968). 227 of certain sound sequences and, therefore, it appears as if there are stratum-specific grammars even after the integration of grammars has been completed.

6.6 Integration of Linguistic Theories

This dissertation discussed various theories of linguistics: theoretical linguistics, sociolinguistics, historical linguistics, and acoustic and articulatory phonetics.

Traditionally, theoretical linguistics and phonetics have collaborated, because obviously phonology must follow some principles of phonetics (Jakobson, Fant, and Halle 1951;

Stevens 1989; Hayes and Steriade 2004; Kawahara 2006). Sociolinguistics and historical linguistics have a close relationship, and they have been studied together (Labov 1972,

1994, 2001). Historical linguistics also has a close connection to phonetics (Guion 1998;

Ohala 2003; Blevins 2004). However, historical phonology and sociolinguistics have had few interactions with theoretical linguistics. Of course there are quite a number of OT studies about historical linguistics and sociolinguistics. But these studies, to my knowledge, are Optimality Theoretic reinterpretations of historical and sociolinguistic studies (see Anttila and Cho 1998; Hibiya 1999; Holt ed. 2003; Ito and Mester 2007).

These studies examine and compare phonological phenomena of various grammars that differ due to social or historical factors. Such differences are often explained as differences in rankings of constraints, which are part of innate and universal linguistic knowledge—Universal Grammar (UG). It stands to reason that various grammars concur with the descriptions by constraint rankings, since such constraints and rankings are part of UG. However, a theory of UG does not explain why a constraint ranking in a language 228 changes from an old one to a new one. According to the definition of OT (Prince and

Smolensky 2004), OT simply cannot explain the cause of the change of rankings in a particular language, because it is language-specific and not universal; therefore, it is not part of UG. For example, suppose grammar A changed into grammar A1 in dialect A1 and grammar A2 in dialect A2. OT can sufficiently describe all three grammars in terms of grammaticality. But OT cannot explain ‘why’ grammar A changed into grammar A1 in dialect A1, but not into grammar A2, which is also perfectly grammatical. Since all dialects in all eras are grammatical in terms of UG, the causes of change and variation cannot be explained in terms of grammaticality. Therefore, the causes of such change and variation must be explained extra-grammatically. OT analyses of language changes and variations are collections of synchronic analyses from different times and places.135

Phonetics-based analyses of language changes share the same limitation with OT analyses. However, the researchers in phonetics-based historical linguistics are indeed aware of this limitation (Ohala 2003; Blevins 2004).

I will provide a concrete example of a case in which phonetics can initiate a language change, but fail to be the ultimate factor in that change. Among the languages on the American northwest coast, there are at least twelve languages that do not have the nasal/oral stop contrast (Kinkade 1985, cited in Blevins 2004:212). Languages without a contrast between nasal and oral stops are uncommon because of a lack of phonetically based motivation (Blevins 2004). Blevins claims that:

135 OT analyses of historical linguistics and sociolinguistics are useful for the purposes of generative linguistics. I have been critical of the analyses based on Optimality Theory. However, I have never attempted to reject or question the theoretical significance of OT. 229

[I]f nasal stops shift to highly unstable quasi-nasalized segments,136 as appears to have occurred at least once in linguistics history, then misperception and sound change due to phonetic variability will inevitably give rise to a shift of these segments to either nasal or oral stops. In the latter case, the oral/nasal contrast is lost. (Blevins 2004:213)

Blevins’s account is consistent, but raises two questions. First, why did the quasi- nasalized segments, which do not seem to be phonetically motivated, occur in the first place? Blevins’s claim that rare sound changes and sound patterns are rare due to the lack of phonetic motivation explains the rarity of such sound patterns, but does not account for why such sound changes still occur. Second, why did some languages shift the quasi- nasalized segments to nasal, while other languages shifted them to oral? Are the selections of the directions of the shift done entirely by chance? Blevins does not directly answer this question because these two groups logically share equivalent phonetic and perceptual conditions; thus, the different selections between these two speaker groups must be extra-grammatical and extra-phonetic, or completely by chance. The causes of

“irregular” sound changes, therefore, must depend on extra-grammatical factors, such as social structure and language contact, as well as phonetics. However, this is merely a constraint of phonetics-based analyses of language changes, rather than a defect. Ohala

(2003) states that phonetics and speech perception are the major factors of documented historical sound changes that are common to multiple languages, but he also acknowledges that there are “changes due to language-specific or culture-specific factors,

136 Nasals without full closure of the velum (Blevins 2004:212) 230 for example, the influence of writing, regularization of morphological paradigms, borrowing, etc.” (Ohala 2003:671). Ohala further notes that although phonetic factors lead the inception of sound change (and some aspects of the spread of sound change), the transmission of sound changes is largely influenced by social, psychological, and language- and culture-specific factors. Thus, according to Ohala, in order to understand the whole mechanism of sound changes—both the inception and the transmission—it is essential to study at least phonetics (universal factors) and cultural factors (language- specific factors).137 As we have seen, the causes of sound changes vary, and they interact in an intricate manner.

When a grammar changes, there is always a motivation. If the grammar has two grammatically equally good candidates for a new grammar, a choice must be made. UG cannot provide such motivation and cannot make choices, although all grammars in any time and place must follow UG. Phonetics, on the other hand, can be part of the motivation and can contribute toward making choices. However, ultimate decisions are often influenced by sociocultural factors and never by UG. Thus, sociocultural factors are often decisive factors of the appropriate grammar.

The unstable situation of geminate voiced obstruents (DD) discussed in Chapter 2 portrays the complexity of decision making when a new grammar is born.

Chronologically, there are roughly three stages: (1) the DD items are not accepted; (2) the

TVDDV items are accepted, but the DVDDV items are not; and (3) all DD items are

137 Here, I do not mean that phonetics is part of UG just because it is universal. I am simply indicating that phonetics is physiologically common to all language-speaking human beings. But see Blevins (2004) for her further claim that certain aspects of phonetics are UG related. 231 accepted. The situation at the time of writing this dissertation is probably between (2) and

(3). Items that were imported earlier tend to resist accepting DD. Moreover, perceptually and acoustically, DD may be more challenging than other existing sound sequences, such as TT. On the other hand, the kana writing system cannot represent [!"i] and [dz#] as voiced counterparts of [c$i] and [ts#] respectively, which may make speakers hesitate to devoice DD with [!"] and [dz]. Knowledge of the source words may affect the probability of devoicing of DD. Finally, speakers’ socioeconomic status may be influential. The current situation of DD in Japanese is influenced by all of the intricate factors above, which create many word-by-word and speaker-by-speaker circumstances. The complexity of DD suggests that the acceptance of DD depends on which of the many factors ultimately affects the decision. I interpret the situation as the grammar of DD being unstable, because various phonetic and cultural factors are sometimes preventing— and other times promoting—the establishment of a stable grammar of DD.

The interaction caused by DD does not stop there. We have seen that all non- marginal items with DD are Foreign items. However, DD can be also found in emphatic expressions and sound effects, which are not Foreign items. This fact suggests that the interaction among strata, which I discussed in the previous section, has already started even before the grammar of DD is established.

These sorts of interactions among factors based on different linguistic fields can be seen in the situations of other innovative sound sequences that were discussed in other chapters. It is important to investigate such factors in order to understand the direct causes of language change and language variation. 232

6.7 Implications for Second Language Acquisition and Teaching

This dissertation emphasizes the influence of English education in Japan, as well as other cultural factors. The English language education system in Japan has changed drastically since the beginning of the last century. The degree of knowledge of English varies among different age groups and socioeconomic groups. There is little doubt that English education affects the prevalence of innovative sound sequences. On the other hand, if the phonology of innovative sound sequences in Japanese is influenced by English so much, why do the Foreign items with innovative sound sequences not sound like the source

English words at all? The discrepancy between the loanword phonology and the source word phonology may be due to the fact that most Japanese speakers do not know the phonology of the source language very well. For example, as we have seen, many

English-origin items have geminate voiced obstruents. However, their source items do not include any geminate voiced obstruents.138 Geminate voiced obstruents in English- origin loanwords may have been developed from the Japanese kana transliteration method, which involves governmental entities, such as the Ministry of Education,

Culture, Sports, Science and Technology (see Chapter 5). Many Japanese students may apply this kana transliteration method in order to understand English taught in school.

Consequently, many Japanese may not study English phonology directly, but rather, through the kana transliteration method. This scenario suggests that English education in

Japan might contribute to the increase in the number of loanwords with innovative sound

138 Of course, loanwords whose source language allows geminate voiced obstruents are excluded from this discussion. 233 sequences, and at the same time, the kana transliteration method is preventing the students from learning authentic English phonology.

The discrepancy between the pronunciation of loanwords and the source words is oftentimes a hindrance for learners of Japanese as a second language. Native Japanese speakers may think that having many English-origin items facilitates the learning of

Japanese—in particular if the learners speak English (Quackenbush 1977; Matsui 1992).

However, this is not so, due to the fact that the source word pronunciation is difficult to reconstruct from the loanword pronunciation (Quackenbush 1977; Matsui 1992). Also, as we have seen in the previous chapters, there are quite a few phonetic variants, and it is almost impossible to know which sound sequences are used for a certain sound of the source word.139 This situation may be frustrating for many learners. Teachers of Japanese should be aware of the fact that many English-origin loanwords have variants and the causes of the variants are often extra-grammatical. Quackenbush and !so suggest that since careful learners may ask questions about these variants, it would be desirable for

Japanese language teachers to be able to answer these questions by explaining the relationship between the variants and sociolinguistic issues (1990:125). In Chapter 4 and

Chapter 5, I discussed the fact that certain sound sequences in certain loanwords have significant sociocultural meanings. For example, the use of [si] sequence may connote unwanted pretentiousness. These sociocultural meanings may be difficult to understand

139 See Quackenbush (1977) for overall descriptions of the issue of the intelligibility of English loanwords for learners of Japanese; see Smith (2006) for how borrowing processes affect the phonological forms of loanword variants. 234 for many learners. Japanese language teachers might be able to provide explicit instruction on the sociolinguistic effects of these certain sequences.

6.8 Concluding Remarks

This dissertation has discussed the following issues. First, I proposed a new method of investigating linguistic data by using electronic dictionaries. This method prevents the data from being haphazard. I described how geminate voiced obstruents show complex distributions because they involve various factors, such as aerodynamic phonetics,

English language education, socioeconomic status, and the writing system. I argued that, unlike the traditional analysis, the consonant of the Japanese /hu/ sequence is not a voiceless bilabial fricative. Accordingly, the innovative sound sequences

[!", !i, !e, !o] are not created by combining the initial consonant of /hu/ and the vowels

[", i, e, o]. I have shown that the emergence of innovative sound sequences, such as

[!", !i, !e, !o] and [#e], is associated with the historical development of Japanese phonology. I have indicated that some innovative sound sequences ([ti], [t$], and possibly [si] as well) are different from traditional sound sequences and other innovative sound sequences based on the fact that their vowels do not devoice. This dissertation addresses various issues, but lacks true empirical study. Further perceptual, production and sociolinguistic study would shed light on the issues of the consonant of /hu/, the vowels of [ti] and [t$], and geminate voiced obstruents. 235

REFERENCES

Akamatsu, Tsutomu. 1997. Japanese phonetics: Theory and practice. München and Newcastle: Lincom Europa.

Akatsuka, Fujio. 1962-1967. Osomatsu-kun. Tokyo: Sh!gakukan.

Akizuki, K!tar!. 2005. Arienai Nihongo Tokyo: Chikuma Shob!.

Aku, Y", and Shun'ichi Tokura. 1977. Nagisa no Shindobaddo. Tokyo: Victor.

Anisfeld, Moshe, Elizabeth Anisfeld, and Ruta Semogas. 1969. Cross-influences between the phonological systems of Lithuanian-English bilinguals. Journal of Verbal Learning and Verbal Behaviour 8:257-261.

Anisfeld, Moshe, and M Gordon. 1971. An effect of one German-language course on English. Lang Speech 14 (3):289-92.

Anttila, Arto. 1995. Deriving variation from grammar: A study of Finnish genitives: [ROA-63, Rutgers Optimality Archive, http://roa.rutgers.edu/].

———. 2002. Morphologically conditioned phonological alternations. Natural Language and Linguistic Theory 20:1-42.

Anttila, Arto, and Young-mee Yu Cho. 1998. Variation and change in Optimality Theory. Lingua 104:31-56.

Archangeli, Diana B., and B. Terence Langendoen. 1997. Optimality theory: an overview. Malden, MA: Blackwell.

Arisaka, Hideyo. 1957. Kokugo on'inshi no kenky" Tokyo: Sanseid!.

Asano, Makiko. 2004. On'inron hikaku. In Linguistics and Japanese Language Education edited by M. Minami and M. Asano. 117-136. Tokyo: Kuroshio.

Beckman, Mary E., Kiyoko Yoneyama, and Jan Edwards. 2003. Language-specific and language-universal aspects of lingual obstruent productions in Japanese-acquiring children. Journal of the Phonetic Society of Japan 7:18-28.

Bernstein, Leonard, and Stephen Sondheim. 1956. Tonight: MUSICADO.

Blevins, Juliette. 2004. Evolutionary phonology: The emergence of sound patterns. Cambridge: Cambridge University Press. 236

Bloch, Bernard. 1950. Studies in colloquial Japanese IV Phonemics. Language 26 (1):86- 125.

Bloch, Bernard, and Roy Andrew Miller. 1970. Bernard Bloch on Japanese. New Haven: Yale University Press.

Browman, Catherine. P. , and Louis Goldstein. 1992. Articulatory phonology: An overview. Phonetica 49:155-180.

Bunkach!. 2009. Kokugo shisaku j!h! sisutemu. Bunkach! (The Agency of Cultural Affairs) 2002 [cited August 6 2009]. Available from http://www.bunka.go.jp/kokugo/frame.asp?tm=20090806232408.

Campbell, Lyle. 1999. Historical linguistics: an introduction. 1st MIT Press ed. Cambridge, MA: MIT Press.

Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press.

Chomsky, Noam, and Morris Halle. 1968. The sound pattern of English. New York: Harper & Row.

Churchward, C. Maxwell. 1940. Rotuman grammar and dictionary. Sydney: Methodist Church of Australasia, Department of Overseas Missions.

Collado, Diego. 1632/1975. Grammar of the Japanese language [Ars grammaticae Iaponicae linguae]. Translated by R. L. Spear. Kansas: Center for East Asian Studies, University of Kansas.

Elert, C.-C. 1964. Phonologic Studies of Quantity in Swedish. Stockholm: Almqvist and Wiksell.

Flege, James. E. 1987. The production of “new” and “similar” phones in a foreign language: Evidence for the effect of equivalence classification. Journal of Phonetics 15:47-65.

Fukazawa, Haruka, and Mafuyu Kitahara. 2001. Domain-relative faithfulness and the OCP: Rendaku revisited. In Issues in Japanese phonology and morphology, edited by J. van de Weijer and T. Nishihara. 85-110. Berlin, New York: Mouton.

———. 2002. Ranking paradox in consonant voicing in Japanese. ms., Kyushu Institute of Technology.

Fukazawa, Haruka, Mafuyu Kitahara, and Mitsuhiko Ota. 1998. Lexical stratification and ranking invariance in constraint-based grammars: [ROA-267, Rutgers Optimality Archive, http://roa.rutgers.edu/]. 237

Gelbart, Ben, and Shigeto Kawahara. 2007. Lexical cues to foreignness in Japanese. Proceedings of Formal Approaches to Japanese Linguistics 4:49-60.

Goldsmith, John. 1973. Autosegmental phonology. Cambridge, MA: MIT dissertation.

Got!, Kazuhiko. 1961. Kagoshima, Miyazaki nambu In H!gengaku K!za 4, edited by Y. End!, T. Hirayama, T. "kubo and T. Shibata. 268-293. Tokyo: T!ky!d! Shuppan.

Greenberg, Joseph Harold. 2000. Indo-European and its closest relatives: The Eurasiatic family: Grammar Stanford, CA: Stanford University Press.

Guion, Susan G. 1998. The role of perception in the sound change of velar palatalization. Phonetica 55:18-52.

———. 2003. The vowel systems of Quichua-Spanish bilinguals: Age of acquisition effects on the mutual influence of the first and second languages. Phonetica 60 (2):98-128.

Hamada, Atsushi. 1955. Tsumaru on. In Kokugogaku jiten, edited by Kokugo Gakkai. Tokyo: T!ky!d!.

———. 1964. Y!-on. Kokugo Kokubun 33 (5):1-12.

Hamano, Shoko. 1998. The sound-symbolic system of Japanese. Stanford/Tokyo: CSLI/Kurosio.

Han, Mieko. S. 1962a. The feature of duration in Japanese. Onsei no Kenky# 10:65-80.

———. 1962b. Unvoicing of vowels in Japanese. Onsei no Kenky# 10:81-100.

Hansell, Mark. 2002. Functional answers to structural problems in thinking about writing. In Difficult characters, edited by M. S. Erbaugh. 124-176. Ohio: National East Asian Language Resource Center.

Hashimoto, Shinkichi. 1950. Kokugo on'in no kenky#. Tokyo: Iwanami Shoten.

Hattori, Shir!. 1960. Gengogaku no h!h!. Tokyo: Iwanami Shoten.

Hayes, Bruce P. 1996. Phonetically driven phonology: the role of optimality theory and inductive grounding. Milwaukee Conference on Formalism and Functionalism in Linguistics.

Hayes, Bruce P., and Donca Steriade. 2004. Introduction: The phonetic bases of phonological markedness. In Phonetically-based phonology, edited by B. Hayes, R. Kirchner and D. Steriade. 1-33. Cambridge: Cambridge University Press. 238

Hibiya, Junko. 1999. Variationist sociolinguistics. In The handbook of Japanese linguistics, edited by N. Tsujimura. 101-120. Oxford: Blackwell.

Hill, Patty, and Mildred J. Hill. 1893. Happy birthday to you.

Hirayama, Minami. 2005. Place asymmetry and markedness of labials in Japanese. Toronto Working Papers in Linguistics 24:125-168.

Holt, Eric D., ed. 2003. Optimality Theory and language change. Dordrecht: Kluwer Academic Publishers.

Ichikawa, Sanki. 1930. The pronunciation of English loan-words in Japanese. In A Grammatical Miscellany Offered to Otto Jespersen on his Seventieth Birthday. 179-190. Copenhagen.

Igarashi, Yumiko, and Ky!ko Mizuki. 1975-1979. Candy Candy. Tokyo: K!dansha.

Imamura, Hiromi. 1996. Gengo no nenrei-sa In Shakai gengogaku e no sh!tai, edited by H. Tanaka and S. Tanaka. 83-95. Kyoto: Mineruba Shob!.

Inkelas, Sharon, Orhan Orgun, and Cheryl Zoll. 1997. The Implications of lexical exceptions for the nature of grammar. In Derivations and constraints in phonology, edited by I. Roca. 393-418. Oxford: Oxford University Press.

Inkelas, Sharon, and Cheryl Zoll. 2003. Is grammar dependence real? In [ROA-587, Rutgers Optimality Archive, http://roa.rutgers.edu/].

Inoue, Fumio. 1989. Shiin no hatsuon no henka In K!za Nihongo to Nihongo ky!iku 2: Nihongo no onsei on'in (j!), edited by M. Sugit!. 109-134. Tokyo: Meiji Shoin.

Inozuka, Hajime, and Emiko Inozuka. 2003. Nihongo onseigaku no shikumi Edited by K. Machida, Shir"zu Nihongo no shikumi o saguru. Tokyo: Kenky#sha.

Ito, Junko, and Armin Mester. 1986. The phonology of voicing in Japanese: Theoretical consequences for morphological accessibility. Linguistic Inquiry (17):49-73.

———. 1995a. The core-periphery structure of the lexicon and constraint ranking. University of Massachusetts Occasional Papers 18:181-209.

———. 1995b. Japanese phonology. In The handbook of phonological theory, edited by J. A. Goldsmith. 817-838. Oxford: Blackwell.

———. 1998. Markedness and word structure: OCP effects in Japanese. Santa Cruz, CA: University of California, Santa Cruz, ms. 239

———. 1999. The Structure of the phonological lexicon. In The Handbook of Japanese linguistics, edited by N. Tsujimura. 62-100. Oxford: Blackwell.

———. 2003. Japanese Morphophonemics: Markedness and Word Structure, MIT Press Linguistic Inquiry Monograph Series 41. Cambridge, MA: MIT Press.

———. 2007. Systemic markedness and faithfulness. CLS 39-1: The main session:665- 689.

———. 2008. Lexical classes in phonology. In The Oxford handbook of Japanese linguistics, edited by S. Miyagawa and M. Saito. 84-106. Oxford: Oxford University Press.

Ito, Junko, Armin Mester, and Jaye Padgett. 1999. Lexical classes in Japanese: A reply to Rice Phonology at Santa Cruz 6:39-46.

Jakobson, Roman. 1971. Why 'mama' and 'papa'? In Selected Writings. 538-545. The Hague: Mouton.

Jakobson, Roman, Gunnar. M. Fant, and Morris Halle. 1951. Preliminaries to speech analysis: The distinctive features and their correlates. Cambridge: MIT Press.

Johnson, Keith. 2003. Acoustic and auditory phonetics. 2nd ed. Oxford: Blackwell.

Kager, René. 1999. Optimality theory. Cambridge: Cambridge University Press.

Kajiku, Shin'ichi. 1983. Shuri h!gen ny"mon Gekkan Gengo 12 (4):33-42.

Kambe, Hiroyasu. 2003. H!gen kiso goi sh". In Sagaken no kotoba, edited by T. Hirayama. 29-133. Tokyo: Meiji Shoin.

Kamei, Takashi. 1984. Suzume shiu shiu. In Kamei Takashi ronbunsh" 3: Nihongo no sugata to kokoro (1) On'in. 447-464. Tokyo: Yoshikawa K!bunkan.

Karimata, Shigehisa. 2002. Ry"ky" no h!gen In Asakura Nihongo K!za 10: H!gen, edited by Y. Ebata. 161-179. Tokyo: Asakura Shoten.

Kase, Tsugio. 2001. Nihongo ky!ikuno tameno onsei hy!gen. Tokyo: Gakubunsha.

Kashima, Tanomu. 2002. Nihongo ky!iku o mezasu hito no tame no kiso kara manabu onseigaku. Tokyo: Sur# $ Nettow%ku.

Katayama, Motoko. 1998. Optimality theory and Japanese loanword phonology, University of California, Santa Cruz. PhD dissertation. 240

Kawahara, Shigeto. 2002. Faithfulness among variants. On-in Kenky! [Phonological Studies] 5:47-54.

———. 2005. Voicing and geminacy in Japanese: An acoustic and perceptual study. In UMOP, edited by K. Flack and S. Kawahara.

———. 2006. A faithfulness ranking projected from a perceptibility scale: The case of [+voice] in Japanese. Language 82 (3):536-574.

Kawahara, Shigeto, Kohei Nishimura, and Hajime Ono. 2002. Unveiling the Unmarkedness Of Sino-Japanese. Japanese/Korean Linguistics 12.

Kawakami, Shin. 1977. Nihongo onsei gaisetsu. Tokyo: "f!sha.

Kenstowicz, Michael. 1994. Phonology in generative grammar. Oxford: Blackwell.

Kindaichi, Haruhiko. 1988. Nihongo Shinpan J#. Tokyo: Iwanami Shoten.

Kinkade, M. Dale. 1985. More on nasal loss on the Northwest coast. International Journal of American Linguistics 51:478-80.

Kitahara, Mafuyu. 1996. Consonant gemination and accent in Japanese loanwords. In Paper presented at Proceedings of Formal Approaches to Japanese Linguisitics 2.

Kokuritsu, Kokugo Kenky!jo Corpus of Spontaneous Japanese (CSJ) (The National Institute for Japanese Language). Kokuritsu Kokugo Kenky!jo [The National Institute for Japanese Language] [cited August 6, 2009]. Available from http://www.kokken.go.jp/en/research_projects/kotonoha/csj/.

Kubozono, Haruo. 1999. Nihongo no onsei Tokyo: Iwanami Shoten.

Kunihiro, Tetsuya. 1983. Kotoba no yure. In Tokyo daigaku k#kai k#za: Kotoba edited by R. Hirano. 81-111. Tokyo: Tokyo Daigaku Shuppankai.

Kurisu, Kazutaka. 2000. Richness of he base and root fusion in Sino-Japanese. Journal of East Asian Language 9:147-185.

Kusano, Shimpei. 1978. Kusano Shimpei Zensh!. Tokyo: Chikuma Shob#.

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

———. 1980. Locating language in time and space. New York: Academic Press.

———. 1994. Principles of Linguistic Change. Oxford: Blackwell. 241

———. 2001. Principles of Linguistic Change: Social Factors. Vol. 2. Oxford: Blackwell

Ladefoged, Peter. 2001. A course in phonetics. 4th ed. Boston, MA: Heinle & Heinle.

Ladefoged, Peter, and Ian Maddieson. 1996. The sounds of the world's languages. Oxford: Blackwell.

Leben, Will. 1973. Suprasegmental phonology. Cambridge, MA: MIT dissertation.

Lehiste, Ilse. 1970. Suprasegmentals. Cambridge, MA: MIT Press.

Lehmann, Winfred Philipp. 1992. Historical linguistics : an introduction. 3rd ed. London; New York: Routledge.

Liberman, Alvin Meyer, Franklin S. Cooper, Donald P. Shankweiler, and Michael Studdert-Kennedy. 1967. Perception of the speech code. Psychological Review 74:431-461.

Löfqvist, Anders. 2005. Lip kinematics in long and short fricative consonants. Journal of the Acoustical Society of America 117:858-78.

Loveday, Leo J. 1996. Language Contact in Japan. New York: Oxford

Lovins, Julie. 1975. Loanwords and the Phonological Structure of Japanese. Bloomington: IULC.

Mabuchi, Kazuo. 1971. Kokugo on'in-ron Tokyo: Kasama Shoin.

———. 1996. Goj!-on-zu no hanashi. Tokyo: Taish!kan.

Maddieson, Ian. 2005. Bilabial and labio-dental fricatives in Ewe. UC Berkeley Phonology Lab Annual Report:199-215.

Maekawa, Kikuo, and Hideaki Kikuchi. 2005. Corpus-based analysis of vowel devoicing in spontaneous Japanese: an interim report. In Voicing in Japanese, edited by J. van de Weijer, K. Nanjo and T. Nishihara. 205-228. Berlin, New York: Mouton de Gruyter.

Masica, Colin P. 1989. The Indo-Aryan languages. Cambridge: Cambridge University Press.

Matsui, Yoshikazu. 1992. Gaikokujin kara mita Nihongo: Tadashii Nihongo ga wakariyasui T"ky": Nihon Ky"bunsha.

Matsumura, Akira. 2006/2007. Daijirin 3rd edition/Daijirin 3rd edition [DVD-ROM]. Tokyo/Fukuoka: Sanseid"/Denshijiten. 242

McCarthy, John. 1986. OCP Effects: Gemination and antigemination. Linguistic Inquiry 17:207-263.

———. 2004. Optimality theory in phonology: a reader. Oxford: Blackwell.

McCarthy, John, and Alan Prince. 1986. Prosodic morphology. Amherst, MA, and New Brunswick, NJ: University of Massachusetts and Rutgers University.

McCawley, James D. 1968. The phonological component of a grammar of Japanese. The Hague, Paris: Mouton.

McMahon, April M. S. 1994. Understanding language change. Cambridge ; New York, NY, USA: Cambridge University Press.

Miller, Roy Andrew. 1967. The Japanese language. Chicago: University of Chicago Press.

Miyajima, Tatsuo, and Midori Takagi. 1988. Zasshi ky!jisshu shiry" no gairaigo hy"ki. In Kokugo shisaku enkaku shiry" 9 gairaigo shiry"sh!, edited by Bunkach". Tokyo: Bunkach" (the Agency of Cultural Affairs).

Miyake, Marc Hideo. 2003. Old Japanese: a phonetic reconstruction. New York, N.Y.: Routledge.

Miyazawa, Kenji. 1995. Kaze no Matasabur". In Miyazawa Kenji zensh! 9. 5-46. Tokyo: Chikuma Shob".

Morishita, Kiichi, and Makio #no. 2001. H"gen tanky!h". Tokyo: Asakura Shoten.

Nakata, Hideo. 1998. Ringu [Ring]. Japan: T"h".

Nakata, Norio. 1972. S"setsu In K"za kokugo-shi 2: On'in-shi moji-shi, edited by N. Nakata. 5-61. Tokyo: Taish!kan.

NHK. 2007. Sekus$ buch". In Sarar$man NEO. Tokyo: NHK.

Nihon Kokugo Daijiten Dainiban Hensh! Iinkai, ed. 2000-02. Nihon kokugo daijiten. 2nd ed. Tokyo: Sh"gakukan.

Nishimura, Kohei. 2006. Lyman's Law in loanwords. On-in Kenky! [Phonological Studies] (9):83-90.

Nolan, Francis, Tara Holst, and Barbara Kühnert. 1996. Modelling [s] to [ ! ] accommodation in English. Journal of Phonetics 24:113-137. 243

Odden, David. 1986. On the role of the obligatory contour principle. Language 62:353- 383.

Ogasawara, Naomi. 2007. Processing of speech variability: Vowel reduction in Japanese. Doctoral dissertation, University of Arizona.

Ohala, John J. 1983. The origin of sound patterns in vocal tract constraints. In The production of speech, edited by P. MacNeilage. 189-216. New York, NY: Springer Verlag.

———. 1992. The cost and benefit of phonological analysis. In The linguistics of literacy, edited by P. Downing, S. D. Lima and M. Noonan. 211-237. Philadelphia: John Benjamins Publishing Co.

———. 2003. Phonetics and Historical Phonology. In The handbook of historical linguistics, edited by R. D. Janda and B. D. Joseph. 669-686. Oxford: Blackwell.

!hashi, Jun'ichi. 2002. T"hoku h"gen onsei no kenky#. T"ky": !f#.

Ohno, Kazutoshi. 2005. Sei-daku: diachronic developments in the writing system. In Voicing in Japanese, edited by J. van de Weijer, K. Nanjo and T. Nishihara. 47- 69. Berlin, New York: Mouton de Gruyter.

Okumura, Mitsuo. 1972. Kodai no on-in. In K"za kokugo-shi 2: On'in-shi moji-shi, edited by N. Nakata. 63-171. Tokyo: Taish#kan.

Ota, Mitsuhiko. 2004. The learnability of the stratified phonological lexicon. Journal of Japanese Linguistics 20:19-40.

Perkell, J.S., S.E. Boyce, and Ken N. Stevens. 1979. Articulatory and acoustic correlates of the [s-$] distinction. Journal of Acoustical Society of America 65 Suppl (1):S24.

Pierrhumbert, Janet, and Mary Beckman. 1988. Japanese tone structure. Cambridge, MA: MIT Press.

Prince, Alan, and Paul Smolensky. 2004. Optimality Theory: Constraint interaction in generative grammar. In Optimality Theory in phonology: A reader, edited by J. McCarthy. 3-71. Oxford: Blackwell.

Quackenbush, Hiroko C. 1977. English loanwords in Japanese. The Journal of the Association of Teachers of Japanese 12 (2/3):149-173.

———. 1989. Gairaigono sokuonka ni tsuite: Ry#gakusei Nihongo ky"iku ni kansuru rironteki jissenteki kenky# Hiroshima University Department of Education. 244

Quackenbush, Hiroko C., and Mieko !so. 1990. Gaoraigo no keisei to sono ky"iku Tokyo: !kurash" Insatsukyoku.

Rasmussen, Jens Elmegård. 1979. Anaptyxis, gemination and syncope in Eskimo: a diachronic study, Travaux du Cercle Linguistique de Copenhague, vol. 18. Copenhagen: C.A. Reitzels Boghandel.

Rice, Keren. 1997. Japanese NC clusters and the redundancy of postnasal voicing. Linguistic Inquiry 28:541-551.

———. 2005. Sequential voicing, postnasal voicing, and Lyman's Law revisited. In Voicing in Japanese, edited by J. van de Weijer, K. Nanjo and T. Nishihara. 25- 45. Berlin, New York: Mouton de Gruyter.

———. 2006. On the patterning of voiced stops in loanwords in Japanese. Tronto Working Papers in Linguistics 26:11-22.

Rodriguez, João. 1604~1608/1955. Arte da lingoa de Iapam. Translated by T. Doi. Nagasaki/Tokyo: Sanseid".

Sait", Yoshio. 2003. Gendai nihongo no onsei In Asakura Nihongo k"za 3: Onsei on-in, edited by Z. Ueno. 1-21. Tokyo: Asakura Shoten.

Sait", Yumiko. 1990. Nihongo onsei hy"genh". Tokyo: !f#sha.

Sakuma, Kanae. 1929. Nihon onseigaku. Tokyo: Kazama Shob".

Sat", Ry"ichi. 2002. Okuni kotoba o shiru: h"gen no chizuch" Tokyo: Sh"gakukan.

Sat", Yoshimi, and Y"k" Kawamura. 1934. Guddo bai.

Seeley, Christopher. 1991. A history of writing in Japan. Leiden: E. J. Brill

Shibata, Takeshi. 1988. Nihongo ky"shi no onsei hy"gen ni tsuite [pamphlet]. In Hanashi kotoba no kiso kunren [cassette tape], edited by NHK-Nihongo-Sent$. 2-4. Tokyo: Alc.

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

Shimmura, Izuru. 2008. K"jien 6th edition/K"jien 6th edition [DVD-ROM]. Tokyo/Fukuoka: Iwanami/Denshijiten.

Shinko Music, ed. 1991. All the lyrics recorded by the Beatles. Tokyo: Shinko Music. 245

Smith, Jennifer. 2006. Loan phonology is not all perception: Evidence from Japanese loan doublets. In Japanese/Korean Linguistics 14, edited by T. J. Vance and K. A. Jones. Stanford: CSLI.

Stevens, Ken N. 1972. The quantal nature of speech: Evidence from articulatory acoustic data In Human Communication: A unified view, edited by E. E. David and P. B. Denes. 51-66. New York: McGraw-Hill.

———. 1989. On the quantal nature of speech. Journal of Phonetics 17:3-45.

Stone, Maureen, Alice Faber, Lawrence J. Raphael, and Thomas H. Shawker. 1992. Cross-sectional tongue shape and linguopalatal contact patterns in [s], [!], and [l]. Journal of Phonetics 20:253-270.

Strouse, Charles, and Martin Charnin. 1982. Tomorrow.

Takayama, Michiaki. 2003. Nihongo on'in-shi kenky! to sono kadai. Journal of the Phonetic Society of Japan 7 (1):35-46.

Tanaka, Y"ko. 2002. Nihon ni okeru eigo igaino gaikokugo ky"iku no genj" no gaikan narabini EU no gengo ky"iku no rinen to taisaku oyobi EU kameikoku no gengo ky"ku no jy"ky" no bekken. Studies in Languages and Cultures 16:95-116.

Terasima, Takayosi. 2007. In search for the starting point of English education: English teaching from a different angle. Gifu Daigaku Ky"ikugakubu Kenky! H"koku Jinbun Kagaku 56 (1):141-152.

Thomason, Sarah G., and Terrence Kaufman. 1988. Language contact, creolization and genetic linguistics. Berkeley: University of California Press.

Toyama, Eiji. 1972. Kindai no on'in In K"za Nihongoshi 2: On'inshi, mojishi, edited by N. Nakata. 173-268. Tokyo: Taish!kan

Trudgill, Peter. 1974. Sociolinguistics: an Introduction. Harmondsworth: Penguin.

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

Tsurutani, Chiharu. 2004. Acquisition of yo-on (Japanese contracted sounds) in L1 and L2 phonology. Second Language 3:27-47.

U2. 1987. The Jushua tree: Island.

Ueda, Kazutoshi. 1898. P-on k" Teikoku Bungaku 4-1. 246

Uehara, Toyoaki, and Gisaburo N. Kiyose. 1974. Fundamentals of Japanese. Bloomington, IN: Indiana University Press.

Ueyama, Ayumi. 1991. Hajimete no hito no gengogaku Tokyo: Kuroshio Shuppan.

Umegaki, Minoru, ed. 1974. Gairaigo jiten. Tokyo: Tokyod! Shuppan.

Unger, Marshall J. 2006. Ongoing sound change in standard Japanese Japanese Language and Literature 40:219-222

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

———. 2002. Another look at vocabulary stratification in Japanese. In The linguist's linguist : a collection of papers in honour of Alexis Manaster Ramer, edited by F. Cavoto. 433-435. Mèunchen: LINCOM.

———. 2008. The sounds of Japanese. Cambridge: Cambridge University Press.

Verbinski, Gore. 2002. The Ring. United States of America: DreamWorks SKG.

Woodbury, Anthony. 1987. Meaningful phonological processes: a study of Central Alaskan Yupik Eskimo prosody. Language 63:685-740.

Yaeger-Dror, Malcah. 1996. Phonetic evidence for the evolution of lexical classes: The case of a Montreal French vowel shift In Towards a social science of language; papers in honor of William Labov, edited by C. F. Gregory R. Guy, Deborah Schiffrin, and John Baugh. 263-287. Philadelphia: John Benjamins Publishing Co.

Yaeger-Dror, Malcah, and Kemp, William. 1992. Lexical classes in Montreal French: the case of (!"). Language and Speech 35 (3):251-293.

Yamada, Y"ichir!. 2005. Gairaigo no shakaigaku: Ingo-ka suru comyunik#shon. Yokohama: Shunp"sha.

Yamaguchi, Nakami. 2002. Inu wa biyo to naiteita Tokyo: K!bunsha.

Yoshida, Shohei. 2001. An element-based analysis of affrication in Japanese. In Issues in Japanese phonology and morphology, edited by J. van de Weijer and T. Nishihara. 193-214. Berlin, New York: Mouton.

Yuzawa, Tadayuki, and Hiroshi Matsuzaki. 2004. Onsei on'in tanky"h!: Nihongo onsei e no izanai, Shir$zu Nihongo tanky"h! 3 Tokyo: Asakura Shoten. 247

Zemeckis, Robert. 1989. Back to the future part II. United States of America: Universal Pictures.