Old Japanese: a phonetic reconstruction

Citation preview

OLD JAPANESE

How can sounds of a dead language be brought back to life? What is wrong with the way we pronounce Old Japanese? How can other Asian languages help us to ‘reconstruct’ a language we no longer use? Eighth-century Japanese or Old Japanese is studied by many Japanese students at school and texts are available in almost any bookstore, yet almost no one pronounces it in a way that its original speakers would recognize. Old Japanese attempts to rectify this problem. The anachronistic pronunciation of this ancient language is resolved through the book’s persuasive step-by-step guide to reconstructing the central dialect of eighthcentury Japanese. Miyake clearly presents what is already known about the history of writing in Japan, the history of Japanese phonology, and several continental Asian data sources including Sino-Korean and Sino-Vietnamese, before convincingly synthesizing this wide variety of evidence into an argument for a system of Old Japanese which utilizes thirteen consonants and seven vowels. Accessible to both specialists and non-specialists alike, this book – the first to provide an areal view of East Asian historical phonology in English – will prove a valuable resource to all those interested in linguistics, Japanese studies and the history of writing. Marc Hideo Miyake is a historical phonologist specializing in East Asian languages.

i

ii

OLD JAPANESE A phonetic reconstruction

Marc Hideo Miyake

iii

First published 2003 by RoutledgeCurzon 11 New Fetter Lane, London EC4P 4EE Simultaneously published in the USA and Canada by RoutledgeCurzon 29 West 35th Street, New York, NY 10001 This edition published in the Taylor & Francis e-Library, 2006.

“To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” RoutledgeCurzon is an imprint of the Taylor & Francis Group © 2003 Marc Hideo Miyake All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging in Publication Data Miyake, Marc Hideo Old Japanese: a phonetic reconstruction by Marc Hideo Miyake p. cm. 1. Japanese language – To 794 – Phonology. I. Title. PL540.M59 2003 495.6′15–dc21 2002154804 ISBN 0-203-51072-0 Master e-book ISBN

ISBN 0-203-34270-4 (Adobe eReader Format) ISBN 0– 415–30575–6

iv

CONTENTS

Illustrations Acknowledgments Abbreviations, symbols, and conventions

viii xii xiii

1

Introduction

1

2

Early transcriptions of Japanese

5

Sinographs, semantograms, and (ongana) phonograms 5 Transcriptions of pre-Old Japanese in Chinese histories 6 Pre-Suiko-Period transcriptions 9 Suiko Period transcriptions 12 Post-Suiko-Period transcriptions 17 The transcriptions of Nihon shoki 28 Summary 40 3

Previous research on phonograms

43

Overview 43 Premodern research 43 Modern research 49 Summary 61 4

Japanese phonology through time

66

Overview 66 Consonants 67 Vowels 77 Summary 86 v

CONTENTS

5

Old Chinese, Middle Chinese, and Sinoxenic

89

Overview 89 Internal sources for earlier Chinese 91 Sinoxenic (SX) sources for earlier Chinese 98 Other external sources for earlier Chinese 135 The reconstruction of Old Chinese 138 The reconstruction of Early Middle Chinese 143 The reconstruction of Late Middle Chinese 149 Summary 156 6

Goals and methodology

158

Goals 158 Methodology 158 7

The reconstruction of Old Japanese consonants

164

Overview 164 OJ p 164 OJ b 166 OJ m 168 OJ w 170 OJ t 171 OJ d 173 OJ n 175 OJ r 176 OJ s 177 OJ z 183 OJ y 186 OJ k 187 OJ g 191 OJ $ 194 Summary 196 8

The reconstruction of Old Japanese vowels Overview 198 OJ a 198 OJ yi 203 OJ u 207 OJ o 211 vi

198

CONTENTS

OJ iy 217 OJ ye 221 OJ ey 227 OJ wo 232 OJ i 238 OJ e 244 OJ o 250 Summary 262 9

Conclusion

265

Bibliography Index

273 285

vii

ILLUSTRATIONS

Figures 2.1 3.1 3.2 3.3 4.1 5.1 6.1 6.2 8.1

The relationship between sinographs and morphemes Origins of [i], [ye], and [wo] in Keichû’s Japanese The ‘established’ interpretation of OJ vowels Karlgren’s MC vowels Mergers of the OJ ‘vowels’ Foreign perceptions of CLMC nasals Methodology of OJ consonantal reconstruction Methodology of OJ ‘vocalic’ reconstruction The ‘peripheralization’ of the OJ B-type ‘vowels’

5 46 55 64 84 156 161 162 264

Tables 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 3.1 3.2

Suiko Period
‘branch’ rhyme phonograms for OJ -a syllables Suiko Period
‘branch’ rhyme phonograms for OJ -(y)i syllables Suiko Period
‘genitive’ rhyme phonograms for OJ -o syllables Suiko Period phonograms reflecting unpalatalized OC initials CLMC prenasalized obstruent–OJ consonant correlations in Shoki CLMC initials for OJ m- and n-initial syllables in Shoki poetry Common EMC–OJ voiced obstruent correlations in Kojiki Common CLMC ‘muddy’ initial–OJ obstruent correspondences in Shoki Syllables of earlier Japanese distinguished in Keichû’s orthography Syllables of Old Japanese distinguished in Kogen seidokukô viii

13 14 15 16 33 34 35 36 46 47

ILLUSTRATIONS

3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10 4.11 4.12 4.13 4.14 4.15 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 5.12 5.13 5.14 5.15 5.16 7.1

Orthographic distinctions in Kanazukai oku no yamamichi Possible syllables in the OJ of Kojiki: the consensus view Pierson’s (1929) reconstruction of OJ Yoshitake’s (1934) reconstruction of OJ SK and SV readings of phonograms for OJ A/B syllables in -i Japanese ‘anti-eight-vowel’ OJ A/B ‘vowel’ reconstructions American ‘anti-eight-vowel’ OJ A/B ‘vowel’ reconstructions Transcription of OJ voiced obstruents in the alpha section Various reconstructions of the OJ ‘A/B-type vowels’ Various reconstructions of OJ s The PKJ consonants of Whitman (1985) A maximal inventory of PJ consonants A minimal inventory of PJ consonants The OJ consonant inventory OJ t- and d-syllables and their later equivalents OJ s- and z-syllables and their later equivalents The PKJ vowels of Whitman (1985) The PJ vowels of Whitman (1985) The PJ vowels of Unger (1993) The PJE vowels of Hattori (1978–79) The PJ vowels of Serafim (1999a) The PKJ vowels of Serafim (1999b) OJ reflexes of PJ diphthongs Distribution of OJ ‘vowels’ The OJ ‘vowel’ inventory The consonants of pre-1945 SJ orthography The vowels of pre-1945 SJ orthography The diphthongs of pre-1945 SJ orthography SK consonants (after Kang Sinhang 1997: 119) SK vowels Written Vietnamese consonants (after Thompson 1987: 58) Written Vietnamese and PVM vowels Written OSV tones Written SV tones The ‘branch/fat/genitive/small’ rhymes in MC, SV, and SJ Phonological traits across Chinese and SX Vowel classes in Vietnamese rhyming (after Downer 1987: 140) The consonants of EMC The vowels of EMC Reconstructions of the LMC labiodentals The consonants of CLMC OJ p in Kojiki poetry ix

48 50 52 53 54 56 57 59 62 63 68 68 71 74 75 75 77 78 78 78 79 79 80 82 84 102 102 103 115 116 119 122 126 128 129 130 133 149 149 154 156 165

ILLUSTRATIONS

OJ p in Shoki poetry OJ b in Kojiki poetry OJ b in Shoki poetry OJ b in Shoki alpha section poetry OJ m in Kojiki poetry OJ m in Shoki poetry OJ w in Kojiki poetry OJ w in Shoki poetry OJ t in Kojiki poetry OJ t in Shoki poetry OJ d in Kojiki poetry OJ d in Shoki poetry OJ n in Kojiki poetry OJ n in Shoki poetry OJ r in Kojiki and Shoki poetry OJ s in Kojiki poetry Phonograms for OJ sa in Kojiki prose Suiko Period phonograms for OJ sa OJ s in Shoki poetry OJ s in Shoki alpha section poetry OJ z in Kojiki poetry OJ z in Shoki poetry OJ y in Kojiki and Shoki poetry OJ k in Kojiki poetry OJ k in Shoki poetry OJ k in Shoki alpha section poetry OJ g in Kojiki poetry OJ g in Shoki poetry OJ $ in Kojiki poetry OJ $ in Shoki poetry OJ $ in Shoki alpha section poetry The OJ consonant inventory OJ a in Kojiki poetry OJ a in Shoki poetry SX equivalents of rhyme categories for Shoki a-phonograms Common a-phonograms in Shoki poetry OJ yi in Kojiki poetry OJ yi in Shoki poetry Common yi-phonograms in Shoki poetry OJ u in Kojiki poetry OJ u in Shoki poetry SX equivalents of major rhyme categories of Shoki u-phonograms 8.11 Common u-phonograms in Shoki poetry 7.2 7.3 7.4 7.5 7.6 7.7 7.8 7.9 7.10 7.11 7.12 7.13 7.14 7.15 7.16 7.17 7.18 7.19 7.20 7.21 7.22 7.23 7.24 7.25 7.26 7.27 7.28 7.29 7.30 7.31 7.32 7.33 8.1 8.2 8.3 8.4 8.5 8.6 8.7 8.8 8.9 8.10

x

165 167 167 168 169 169 170 170 171 172 173 174 175 175 177 177 178 178 180 181 183 185 187 187 189 190 191 192 194 195 195 196 199 202 202 203 204 206 206 208 210 210 211

ILLUSTRATIONS

8.12 OJ o in Kojiki poetry 8.13 OJ o in Shoki poetry 8.14 SX equivalents of rhyme categories for Shoki o-phonograms 8.15 Common o-phonograms in Shoki poetry 8.16 The ‘primary’ vowels of OJ 8.17 OJ iy in Kojiki poetry 8.18 OJ iy in Shoki poetry 8.19 Iy-phonograms in Shoki poetry 8.20 OJ ye in Kojiki poetry 8.21 OJ ye in Shoki poetry 8.22 SX equivalents of rhyme categories for Shoki ye-phonograms 8.23 Ye-phonograms in Shoki poetry 8.24 OJ ey in Kojiki poetry 8.25 OJ ey in Shoki poetry 8.26 SX equivalents of rhyme categories for Shoki ey-phonograms 8.27 Common ey-phonograms in Shoki poetry 8.28 OJ wo in Kojiki poetry 8.29 OJ wo in Shoki poetry 8.30 SX reflexes of rhyme categories for Shoki wo-phonograms 8.31 Common wo-phonograms in Shoki poetry 8.32 The reconstructed OJ ‘vowels’ thus far 8.33 The post-OJ vowel system 8.34 A rejected EMJ vowel system 8.35 A preferred EMJ vowel system 8.36 OJ i in Kojiki poetry 8.37 OJ i in Shoki poetry 8.38 SX equivalents of rhyme categories for Shoki i-phonograms 8.39 Common i-phonograms in Shoki poetry 8.40 OJ e in Kojiki poetry 8.41 OJ e in Shoki poetry 8.42 SX equivalents of rhyme categories for Shoki e-phonograms 8.43 Common e-phonograms in Shoki poetry 8.44 OJ o in Kojiki poetry 8.45 OJ o in Shoki poetry 8.46 SX equivalents of rhyme categories for Shoki o-phonograms 8.47 Common o-phonograms in Shoki poetry 8.48 The ‘vowels’ of OJ 8.49 OJ syllables with C-type ‘vowels’ 9.1 Post-OJ SJ diphthongs 9.2 OJ period SJ vowel–glide sequences 9.3 The syllabic inventory of OJ xi

212 214 215 216 216 217 219 219 221 223 223 226 227 228 229 230 232 234 234 236 237 238 238 238 239 240 240 243 244 248 248 249 251 256 256 261 262 264 267 268 271

ACKNOWLEDGMENTS

I wish to express my deepest gratitude to the following people whose generous assistance with my doctoral dissertation made this book possible: first and foremost, Professor Alexander V. Vovin, for sharing his vast knowledge of Asian and historical-comparative linguistics with me; Professors Robert A. Blust and Anatole V. Lyovin, for teaching me historical linguistics; Professor Kenneth L. Rehg, for teaching me phonology; Professor Arthur H. Thornhill, for his willingness to read a dissertation concerning an alien side of a familiar subject; Professor Leon A. Serafim, for introducing me to Ryukyuan and to the work of Professors Edwin Pulleyblank and John Whitman; Professor Gerald B. Mathias, for teaching me the fundamentals of Japanese historical linguistics; John R. Bentley, for guiding me through the dense philological forests of Japanese texts and answering my seemingly endless questions; Mark J. Alves, for our many conversations about Chinese and Vietnamese and for sending me the invaluable gift of NgZ thiên t@ (Five Thousand Characters); last but not least, my beloved and my family, for their patience and support through my long years in school.

xii

ABBREVIATIONS, SYMBOLS, AND CONVENTIONS

Abbreviations Am BCE Cnt CE CEMC CLMC Chn CT Cz Dk EH EMC EMJ EMK Eng Freq Go Ind Jn Jpn Kan KJK LMC LMJ LMK LOC LT MC Md M. Hakka MJ

Amoghavajra Before Common Era; = BC Cantonese Common Era; = AD Chang’an EMC Chang’an LMC Chinese Colloquial Taiwanese Chaozhou DharmakÖema Eastern Han Early Middle Chinese Early Middle Japanese Early Middle Korean English frequency Go-on Indic Jñanagupta Japanese Kan-on Kojiki Late Middle Chinese Late Middle Japanese Late Middle Korean Late Old Chinese Literary Taiwanese Middle Chinese Mandarin Chinese Meixian Hakka (= Kejia) Modern Japanese xiii

ABBREVIATIONS, SYMBOLS, AND CONVENTIONS

MK ms. MSJ mss MYS NS NW OC OJ ONWC OSK OSV PG Ph PJ PJE PJI PKJ POJ PSJ PSK PSV PSX PT PVM SBBT SC SCBT SD SDBT SE SEBT SG SJ SK Skt SP STCA SV SX Shoki Tib. (tr.) Tw

Middle Korean manuscript Modern Standard Japanese manuscripts Man’yôshû Nihon shoki northwestern Old Chinese Old Japanese Old Northwest Chinese Old Sino-Korean Old Sino-Vietnamese phonogram phonogram Proto-Japonic Proto-Japonic-External Proto-Japonic-Internal Proto-Koreo-Japonic Pre-Old Japanese Pre-Sino-Japanese Pre-Sino-Korean Pre-Sino-Vietnamese Pre-Sinoxenic Proto-Tai Proto-Viet-Muong System-B-Based Transcription System C System-C-Based Transcription System D System-D-Based Transcription System E (= CLMC) System-E-Based Transcription sinograph Sino-Japanese Sino-Korean Sanskrit Sino-Paekche Sui-Tang Chang’an (Coblin 1994) Sino-Vietnamese Sinoxenic Nihon shoki Tibetan (transcription) Taiwanese

xiv

ABBREVIATIONS, SYMBOLS, AND CONVENTIONS

Ui. (tr.) Wr Viet

Uighur (transcription) Written Vietnamese

Symbols α β * – [] ‘’ //

alpha section of Shoki (see Mori 1991) beta section of Shoki (see Mori 1991) unattested form (either reconstructed or ungrammatical) morphemic boundary phonetic transcription English gloss phonemic transcription

Conventions Cantonese forms are cited in Yale romanization. Modern Chinese forms other than Cantonese, Mandarin, or Taiwanese are cited from Hanyu fangyin zihui (1962) with minor typographical modifications. h after a consonant indicates aspiration. Tones are indicated using the following superscript numerals: Level Rising Upper 1 3 Lower 2 4

Departing Entering 5 7 6 8

Italics represent conventionalized orthographic forms as opposed to phonetic or phonemic transcriptions. Japanese forms other than Sino-Japanese and Old Japanese are cited in Hepburn romanization, with the ‘mora nasal’ written consistently as n and vowel length indicated by circumflexes (™) rather than macrons (¡). Middle Korean and modern Korean forms are cited in the Yale romanization used in Martin (1992) with one alteration: the vowels [o] and [u] are consistently transcribed as wo and wu except in modern Korean names and titles. Korean toponyms are written with their conventionalized spellings in English without diacritics: e.g., ‘Koguryo’ instead of ‘KoguryÉ’ or ‘K(w)okwulye,’ ‘Seoul’ instead of ‘Sewul,’ etc. Mandarin forms are cited in pinyin romanization. Tones are unmarked on names, titles of works, and linguistic terminology.

xv

ABBREVIATIONS, SYMBOLS, AND CONVENTIONS

Middle Chinese forms are cited according to Pulleyblank’s (1984, 1991a) reconstruction unless stated otherwise with the following typographical substitutions: 1 2 3 4 5 6 7

c, chr, jr, sr, zr for tZ, tZh, d| , Z, | c, ch, j, y for t\, t\h, d9, j q for } long vowels indicated by vowel doubling: e.g., aa for a: R for # (1984) or Ü (1991a) ü and ö for y and œ Rising and Departing Tones indicated with tonal letters -q and -h

‘Tonal letters’ do not represent segmental phonemes: -q and -h are not pronounced as [}] and [h]. Since there is no agreement on the phonetic values of the Middle Chinese tones, I have chosen to represent two of the four Middle Chinese tones with abstract symbols added to the end of the syllable to avoid implying anything about the pitch and contour of the tones. For example, *kaq is [ka] with the Rising Tone (not *[ka}] ) and *kah is [ka] with the Departing Tone. Chinese reconstructions generally leave the Level Tone unmarked. Thus *ka is [ka] with the Level Tone. All syllables ending in stops (*-p, *-t, *-k) automatically have the Entering Tone. The traditional terms ‘Level,’ ‘Rising,’ ‘Departing,’ and ‘Entering’ do not necessarily describe the tones that they represent. I will propose a number of modifications to Pulleyblank’s MC reconstructions in Chapters 2 and 5. All subsequent MC reconstructions will reflect those modifications. Chang’an LMC forms are identical to other LMC forms except that all nasals have been rewritten as prenasalized stops unless followed by syllable-final *-Y: LMC *m > Chang’an LMC *mb, etc. See (2.15). Old Chinese forms are cited according to Starostin’s (1989) reconstruction unless stated otherwise with the following typographical substitutions: 1 2 3 4 5 6 7 8 9

tr, thr, dr, dhr, nr for à, àh, 1, 1h, á ts, tsh, dz, dzh for c, ch, W, Wh cr [âZ], chr [âZh], jr [¶| ], jhr [¶|®], sr [Z], zr [| ] for ö, öh, í, íh, Ö, " c [c\], ch [c\h], j [ j_9], jh [ _j 9®], ã, y, \, 9 for †, †h, £, £h, °, j, 5, ¢ cl, chl, jl, jlh for the lateral affricates §, §h, •, •h q for } long vowels indicated by vowel doubling: e.g., aa for a retroflex vowels indicated by superscript r: e.g., ar for ô Rising and Departing Tones indicated with tonal letters -q and -h

See my note on ‘tonal letters’ in Middle Chinese above. xvi

ABBREVIATIONS, SYMBOLS, AND CONVENTIONS

I will propose a number of modifications to Starostin’s OC reconstructions in Chapters 2 and 5. All subsequent OC reconstructions will reflect those modifications. Old Japanese forms in italics are cited in the romanization used in Martin (1987), but with p substituted for Martin’s f (equivalent to capital F in Miller 1967 and other works). Old Japanese reconstructions are cited according to the reconstruction proposed in Chapters 6 and 7. Old Mandarin reconstructions are cited from Pulleyblank (1991a) without tones and with *y substituted for this IPA *j. Poetic citations from Kojiki, Nihon shoki, and Man’yôshû are given in the following format: poem number, line number, and sinograph number. Thus KJK 1.1.1 represents the first sinograph in the first line of the first poem in Kojiki. Specified words in poems are cited only by poem and line number. Occurrences of sinographs in unspecified words in poems are cited by poem number, line number, and sinograph number. Textual variant sinographs are marked with lower case letters after the sinograph number. These letters are those assigned to each variant sinograph in Miyake (1999). Thus NS 2.5.3a refers to the first variant of poem 2, line 5, sinograph 3 listed in Miyake’s (1999) database of NS poetry, NS 2.5.3b to the second variant, etc. See Appendices A and D of Miyake (1999) for further details. Non-poetic citations from Kojiki and Shoki are in the following format: (Title Vol. No.: Section Title). Hence (KJK I: Age of the Gods) refers to the “Age of the Gods” section in Kojiki. The sole exception to this rule is the Preface to KJK, which will be cited as (KJK Preface). Proto-Tai (PT) forms have been reconstructed by combining the initials, finals, and tones given for each PT etymon in Li (1977). Page numbers in citations refer to the pages where Li’s (1977) reconstructions of initials and finals can be found for a given PT etymon. Li (1977) cites the initial and tone for a PT etymon (e.g., *h-C1) and/or the final and tone (e.g., *-Rï C1) for a PT etymon but never cites all three components of a PT etymon in a single reconstructed form (e.g., PT *hRï C1 ‘give’). For example, “PT *hRï C1 ‘give’ (Li 1977: 250, 289)” means that the PT initial *h- was taken from Li (1977: 250) and that the PT final and tone *-Rï C1 were taken from Li (1977: 289). Sino-Japanese forms (Go-on, Kan-on, etc.) are taken mostly from Tôdô (1978). Whenever they were unavailable in Tôdô, I have cited them from xvii

ABBREVIATIONS, SYMBOLS, AND CONVENTIONS

Morohashi (1955–60), Morohashi et al. (1981–82), or Morohashi et al. (1992). All citations are in a phonemicized transliteration of their pre-1945 kana spellings. For example,
‘armor’ is cited as SJ kahu rather than as SJ kafu or as SJ kô. Following Tôdô (1978, 1980), I have consistently rendered SJ -n < Chinese *m as -mu. However, Sino-Japanese elements in names and titles are in Hepburn romanization: e.g., jiten ‘dictionary’ instead of ziten in my bibliography. Sino-Korean forms are cited in Yale romanization from the Cenwun okphyen dictionary of c. 1796 unless stated otherwise. Taiwanese forms are cited in the Maryknoll missionary system with four modifications: (1) o for o·, (2) R for o, (3) N for n, and (4) the use of the following superscript numerals to indicate the seven tones of Taiwanese: Level Rising Upper 1 3 Lower 2

Departing Entering 5 7 6 8

Note that modern Taiwanese has no tone 4. The earlier Lower Rising tone (tone 4) merged with the Lower Departing tone (tone 6). My modified romanization and phonetic transcription of Taiwanese represent southern Taiwanese pronunciation. Tibetan forms are cited in Wylie romanization. Vietnamese forms are cited in qu3c ngw romanization. Phonetic transcriptions of Vietnamese represent idealized Hanoi dialect pronunciation.

xviii

INTRODUCTION

1 INTRODUCTION

Among the classical languages of the world, Old Japanese (OJ), the literary language of Japan up to the Nara Period (710–794 CE),1 has fared quite well. While some other classical languages are remembered only among scholars, most Japanese have been exposed to OJ and other varieties of premodern Japanese in secondary school. This situation is quite unlike that of, say, the United States, where no high school students are exposed to Old English except in translation. Untranslated Old English texts are difficult to find at non-academic bookstores, but one can easily find a copy of the OJ poetic anthology Man’yôshû (Collection of Ten Thousand Leaves) in ordinary Japanese bookstores. Although OJ is no longer a living language, it is far from forgotten. Even Japanese high school students are familiar with the rudiments of its morphology and lexicon. Yet the phonetics and phonology of OJ remain unclear. When OJ texts are taught in Japanese classrooms, they are read anachronistically in a modernized pronunciation. Similarly, OJ texts are cited in most Western-language works in a romanization reflecting that anachronistic modernized pronunciation. To demonstrate the gap between this anachronistic modernized pronunciation of OJ and the original pronunciation of OJ, I will cite the first poem in the official history Kojiki (Records of Ancient Matters; 712 CE) in two different forms below. On the left is the Hepburn romanization of the poem reflecting its pronunciation in Modern Standard Japanese. On the right is my reconstruction of the approximate pronunciation circa the early eighth century CE.2 (An English translation is in Keene (1993: 43).) (1.1) yakumo tatsu izumo yaegaki tsumagomi ni yaegaki tsukuru sono yaegaki o

*yakumo tatu *indumo yapeYgaki *tumaYgRmt ni *yapeYgaki tukuru *sRnR yapeYgaki wo

This poem is “traditionally considered to be the first Japanese poem ever composed” (Keene 1993: 43). Its composer3 would find its modernized 1

OLD JAPANESE

pronunciation quite odd. His *tumaYgRmt would correspond to modern tsumagomi. If sound is an integral part of poetry, then, if one reads OJ poetry as if it were Modern Standard Japanese, much of its essence would be lost. Is OJ merely Modern Standard Japanese with archaic words and grammar? Yet this anachronistic treatment of OJ pronunciation is partly forgivable because of (1) the lack of a scholarly consensus on OJ phonetics and phonology and (2) a complex script (man’yôgana) that obscures the original pronunciation. By using a modernized pronunciation (and romanizations based upon that modernized pronunciation), one avoids having to choose between the many competing reconstructions of OJ pronunciation based on different analyses of the man’yôgana script and other evidence. As I will demonstrate in Chapter 3, the extant reconstructions of OJ are mutually contradictory and often methodologically troublesome. This situation is hardly unique to OJ in East Asia. Other classical languages of the region such as Classical Chinese and the premodern Vietnamese of nôm texts4 are also traditionally read anachronistically in modernized pronunciations and romanized according to those modernized pronunciations. Once again, there is no consensus about the original pronunciations of these languages. Furthermore, complex scripts (Chinese characters and nôm) obscure the original pronunciations of Classical Chinese and premodern Vietnamese. However, scholars do agree on the presence of certain abstract phonological distinctions in these languages, even if they do not agree on the concrete phonetic interpretation of these distinctions. In the case of OJ, for example, most scholars agree that there were at least 87 different syllables in the language.5 One can view these 87-plus syllables of OJ as algebraic variables: X, Y, Z, etc. If so, then the problem is to find the phonetic values of these 87-plus variables. The goal of this book is to solve this problem by reconstructing OJ phonetics on the basis of an analysis of the man’yôgana orthography of the OJ poetry in the Kojiki (KJK; Records of Ancient Matters; 712 CE) and Nihon shoki (NS or Shoki; Chronicles of Japan; 720 CE). In Chapter 2, I will describe the earliest written records of Japanese and their orthographies, beginning with the Chinese transcriptions in the Wei zhi (Chronicles of Wei) and leading ultimately to the data at the heart of this book: the OJ poetry in Kojiki and Shoki written in man’yôgana. I will outline the various kinds of man’yôgana found in OJ texts and demonstrate that the poems in Kojiki and Shoki employ two distinct varieties of the ongana (phonogram)6 subtype of man’yôgana. In Chapter 3, I will deal with previous scholarship on ongana, beginning with the eighteenth-century discovery of orthographic distinctions present in ongana but absent in the later kana syllabaries and continuing on to twentieth-century attempts to reconstruct OJ phonology on the basis of 2

INTRODUCTION

ongana. I will present all extant reconstructions of OJ phonology known to me for comparison and criticism. In Chapter 4, I will sketch what is known about the phonology of the ancestors of Old Japanese from internal and external reconstruction. I will also sketch the phonological history of Japanese after the eighth century. I intend my reconstruction of OJ based on written records to be as consistent as possible with (1) the results of internal and external reconstructions of pre-Nara-Period Japanese and (2) later developments in the phonological history of Japanese. In Chapter 5, I will discuss the reconstruction of Late Old Chinese, Early Middle Chinese, and Late Middle Chinese – the three stages of the Chinese language relevant to the interpretation of ongana. This discussion will necessitate an excursion into Sinoxenic – the large strata of borrowings from Chinese in Vietnamese, Korean, and Japanese (i.e., Sino-Japanese, Sino-Korean, and Sino-Vietnamese). I will show that Sino-Korean, SinoVietnamese, and the Sino-Vietnamese-based phonograms of the premodern Vietnamese nôm writing system are important sources of data not only for Chinese reconstruction but also for the interpretation of the ongana. In Chapter 6, I will formulate my methodology for the reconstruction of eighth-century Japanese and the goals which my reconstruction must fulfill. In Chapters 7 and 8, I will reconstruct the phonetic values of the consonants and vowels of the Central dialect of OJ primarily on the basis of Chinese and Sinoxenic evidence with reference to the textual frequency and distribution of ongana and what is known about earlier and later Japanese phonology. I will then reconstruct a typologically sound phonemic system for OJ on the basis of my reconstructed phonetic values. Finally, in Chapter 9, I will summarize my findings, speculate on the implications of my Central OJ reconstruction for the reconstruction of the Nara Period pronunciation of Sino-Japanese, and indicate future directions of research.

Notes 1 The association of stages of the Japanese language with Japanese historical periods is merely convenient and not particularly accurate, especially in the case of OJ. I do not intend to imply that linguistic changes coincided with the shift of the capital to Nara in 710 CE and later to Heian (modern Kyoto) in 794 CE. Although OJ texts were compiled during the Nara Period, some or most of their content may date before the Nara Period. I cite only the periods most commonly associated with each stage in order to give a crude idea of chronology to readers unfamiliar with Japanese language history. A term such as ‘Old Japanese’ by itself tells lay readers nothing about when Old Japanese existed. 2 I will present my arguments for the reconstructions of individual OJ consonants and vowels in Chapters 7 and 8. My reconstructed pronunciation is “circa the eighth century CE” because, although Kojiki itself was completed by 712, its contents do not necessarily reflect OJ as spoken in 712. The poetry of Kojiki may

3

OLD JAPANESE

3 4

5

6

“antedate the introduction of writing” into Japan several centuries earlier (Keene 1993: 33). Furthermore, as I will argue in Chapter 2, the orthography may reflect a tradition preserving distinctions made in pre-eighth century OJ pronunciation rather than the pronunciation of the OJ of 712. Bentley (1997: 159–161) presents evidence for what he calls the “spelling convention” of OJ texts such as Kojiki. The Kojiki attributes this poem to the god Susano-o. Nôm is a Chinese character-based writing system created in Vietnam some time prior to the invention of Vietnamese romanization by European missionaries circa the seventeenth century CE. Nôm is roughly comparable to other Chinese characterbased writing systems such as Japanese man’yôgana (the subject of this book) and Korean hyangchal. See p. 129 for more on nôm. The figure of 87 refers to the number of syllables conventionally assumed for the OJ of the official history Nihon shoki (Chronicles of Japan; 720 CE) and the poetic anthology Man’yôshû (Collection of Ten Thousand Leaves; c. 759 CE); the OJ of Kojiki (Records of Ancient Matters; 712 CE) is assumed to have had 88 different syllables. However, some scholars believe that there were more than 88 distinct syllables in OJ. I will discuss this issue at greater length in Chapter 8. Although there are two types of man’yôgana phonograms, in this book, the term ‘phonogram’ will refer only to the ongana type unless noted otherwise. I will define the various types of man’yôgana in the following chapter.

4

EARLY TRANSCRIPTIONS OF JAPANESE

2 EARLY TRANSCRIPTIONS OF JAPANESE

Sinographs, semantograms, and (ongana) phonograms Sinographs (Chinese characters) were the dominant form of writing in East Asia in the first millennium CE. Although sinographs are commonly called ‘ideographs,’ they do not represent ideas without any reference to language. On the contrary, sinographs were originally intended to represent specific morphemes of a specific language – Chinese. Morphemes are minimal units of meaning associated with phonetic forms. Hence each sinograph represents both the meaning and the pronunciation of a Chinese morpheme (Figure 2.1). For example, the sinograph
represents a Chinese morpheme which means ‘horse’ and which is pronounced mf in Mandarin Chinese (Md). On the one hand,
is not a symbol for all Chinese morphemes meaning ‘horse’; one cannot write
Md wj, the ‘horse’ of the Chinese zodiac, or any other Chinese morphemes for ‘horse’ as
. On the other hand,
is not a phonetic symbol for the syllable mf in Mandarin Chinese; one cannot write the morpheme
‘number,’ also pronounced mf in Mandarin Chinese, as
. Since sinography (Chinese writing) is language-specific and, in turn, individual sinographs are morpheme-specific, scribes in early Japan faced tremendous problems in adapting sinography to Japanese. Since Japanese is a language completely unrelated to Chinese, it has no native morphemes similar to Chinese morphemes, apart from rare coincidental lookalikes sinography (the sum of all sinographs) individual sinograph Chinese morpheme meaning

sound

Figure 2.1 The relationship between sinographs and morphemes

5

OLD JAPANESE

(e.g., Modern Standard Japanese shin- ‘to die’ and Mandarin sh ‘id.’). How would one write all the OJ words which had no cognates in Chinese? For example, how would one write OJ namyi ‘wave,’ a word without any resemblance to the Chinese morpheme for ‘wave,’
EMC *pa?1 Two basic solutions were possible. One could write OJ namyi ‘wave’ with the sinograph for a Chinese morpheme of similar meaning:
EMC *pa ‘wave.’
*pa ‘wave’ would then be a semantogram for OJ namyi, serving as a translation of OJ namyi without representing its sound. Alternatively, one could write OJ namyi with two sinographs for Chinese morphemes which in combination sounded like namyi:
, with
EMC *nah ‘that’ for OJ na- and
EMC *miq ‘beautiful’ for OJ -myi.2
EMC *nah ‘that’ and
EMC *miq ‘beautiful’ would then be phonograms for OJ namyi ‘wave,’ representing only the sounds of OJ namyi ‘wave.’ These phonograms are of the dominant ongana type whose readings are derived from Chinese. Hereafter, the term ‘phonogram’ will refer solely to ongana unless specified otherwise. (See p. 25 for the less common kungana phonograms whose readings are derived from OJ.) Scribes in early Japan used both of the above methods, as well as a number of more complex but infrequent methods described on p. 26, to write Japanese. However, since the goal of this book is to reconstruct the pronunciation of OJ, I will focus hereafter on phonograms, since semantograms contain no phonetic information about OJ. In this chapter, I will outline the history of phonogram transcriptions of Japanese up to the eighth century, beginning from the earliest Chinese attempts to write Japanese words and concluding with the transcriptions of poetry in Kojiki, Nihon shoki, and Man’yôshû.

Transcriptions of pre-Old Japanese in Chinese histories “Wo ren zhuan” of the Wei zhi The earliest written record of the Japanese language may be in the “Wo ren zhuan”3 (Account of the Dwarfs) section of the Wei zhi (Chronicles of Wei), a history of the Wei state (220–265 CE) compiled by Chen Shou (233–297 CE). The Chinese author(s) of this description of third-century Japan used sinographs for their phonetic values to represent names and terms of a language which may be an early form of Japanese. For example, in the first lines of the “Wo ren zhuan,” a native title for ‘cochieftain’ is transcribed with four phonograms:
 ! (Late Old Chinese [LOC] *pye noo mRRwq le). This word is usually believed to be something like OJ pyinamori (hinamori in modernized pronunciation) ‘guard of a distant region’. The identification of
 ! as well as of many other transcribed words in the “Wo ren zhuan” is still open to question. 6

EARLY TRANSCRIPTIONS OF JAPANESE

At present, no full-scale study of the “Wo ren zhuan” transcriptions has been done using recent reconstructions of Old Chinese such as Starostin (1989) or Baxter (1992) to the best of my knowledge. Such a study would probably cast doubt on previous attempts at identification. Even if such a study were done, the “Wo ren zhuan” would still be of relatively limited value for the study of early Japanese for a number of reasons. First, the transcribed material in the “Wo ren zhuan” consists only of names and titles. There are no transcriptions for sentences, verbs, or even other types of nouns. Second, semantic glosses which would aid identification are few and far between. Names have no semantic glosses at all and the semantic glosses accompanying titles are of little or no help in determining etymologies. For example, although the “Wo ren zhuan” tells us that
 ! LOC *pye noo mRRwq le means ‘co-chieftain,’ this may not be a morpheme-for-morpheme gloss. Third, as pointed out by Omodaka (1967: 619), the accuracy of the “Wo ren zhuan” transcriptions may be open to question. Given the Chinese contempt for the early Japanese ‘dwarfs,’ there is no reason to expect much precision from the transcriptions in the Wo ren zhuan. According to Kane (1989: 100), Chinese government translators of the Ming Dynasty (1368–1644 CE) “were not well known for their competence in the [non-Chinese] languages they studied, or for their care in transcription [in the Hua-Yi yiyu (Sino-Barbarian Glossaries) ].” Although Kane was referring to a period over a millennium after the fall of the Wei in 265 CE, Chinese ethnocentrism has changed little over time. The Chinese of both the Wei and the Ming viewed non-Chinese as ‘barbarians.’ Hence it is doubtful whether anyone in the state of Wei was truly competent to transcribe early Japanese. Furthermore, the choice of characters such as
‘humble’ or
‘slave’ in the transcriptions may reflect Chinese ethnocentrism more than the actual phonetics of early Japanese. Lastly, extreme skeptics may question whether all the “Wo ren zhuan” transcriptions represent an early form of Japanese. Although some transcriptions such as the toponym
 LOC *9(h)a maarq dhRR, believed to be ‘Yamato’ (OJ yamato < pre-OJ *damatR), resemble later Japanese words,4 still others do not resemble anything in Japanese, particularly when interpreted according to modern OC reconstructions. It is at least theoretically possible that the transcriptions represent words from different dialects of the same language or even different languages which may or may not be related. The “Wo ren zhuan” does not explicitly state that all the transcriptions represent a single language. Other Chinese histories Although flawed, the “Wo ren zhuan” of the Wei zhi does supply more potential data on early Japanese than other Chinese historical works on 7

OLD JAPANESE

Japan up to the middle of the seventh century CE, which have only a handful of transcriptions often duplicating those already in the Wo ren zhuan. The sections on Japan in the Song shu (History of the Song [420–478 CE; not to be confused with the more famous Song of 960–1279 CE]; c. 500 CE) and the Nan Qi shu (History of the Southern Qi [479–501 CE]; c. sixth century CE) have only what may be transcriptions of the first syllables of the names of Japanese ‘kings.’5 Were the Japanese literate in the third century? Fortunately, the Chinese histories are not our only sources for transcriptions of pre-eighth-century Japanese; we also have sinographic sources from Japan which may have been written by the Japanese themselves. Unfortunately, as we shall soon see, it is doubtful that the Japanese were sinographically literate as early as the third century CE. Early Japan did not have a writing system of its own, despite fraudulent claims of an indigenous “script of the age of the gods” ( jindai moji). Hence all Japanese sources up to the development of the kana syllabaries during the Heian Period (794–1185 CE)6 employed sinographs. The first Japanese contact with sinography may have been as early as the first-century CE. A seal discovered on Shiga Island with the sinographs
 !" (lit. ‘Sino-Japan Na Country King’) is believed to be the seal bestowed upon a Japanese emissary by the Chinese emperor Guangwu in 57 CE. Even if the identification of this seal is erroneous, coins from firstcentury China have also been discovered among artifacts excavated from the Yayoi period (Matsumura 1972: 21). This evidence of early contact with Chinese writing does not necessarily mean that the Yayoi Period Japanese were literate in Chinese. The Japanese of the first century CE were probably not able to read the writing on the seal or on the coins; a few may have known what the writing meant without having any significant command of the Chinese script. The “Wo ren zhuan” does not provide any definite evidence for literacy among the natives of third-century Japan. It does tell us that the court of the Japanese queen ‘Himiko’7 did exchange letters with the Wei court, but this cannot be interpreted as absolute proof of the existence of native Japanese who were literate in sinography. Since the “Wo ren zhuan” (1) does not mention anything about literacy in its ethnographic description of the Japanese and (2) does not identify who was doing the reading and writing in ‘Himiko’s’ court, it is possible that the queen employed bilingual, literate Chinese as scribes for communication with the Wei court. If there was any native literacy in ‘Himiko’s’ Japan, it may have been limited at best. Although Seeley (1991: 12) notes that Chinese gave the Japanese mirrors with inscriptions, he believes that “these inscriptions do not constitute evidence of the actual use of writing in Japan at that period” and goes so far 8

EARLY TRANSCRIPTIONS OF JAPANESE

as to doubt whether the Japanese of the time even “understood the function of writing.” Owing to a lack of strong evidence for early Japanese literacy, I conclude that the transcriptions in the “Wo ren zhuan” were most likely to have been created by Chinese and were not copied from transcriptions created by Japanese.

Pre-Suiko-Period transcriptions The coming of the Paekche scholars Detailed knowledge of the sinographic system as opposed to mere exposure to sinography was probably brought to Japan circa the early fifth century CE. According to the Japanese national histories Kojiki and Nihon shoki, scholars named ‘Achiki’ and ‘Wani’8 from the Korean peninsular state of Paekche (Japanese Kudara) came to Japan during the reign of Emperor Ôjin. The exact date of their arrival is uncertain. According to Bentley’s (1998a: 59) reconstructed chronology for Nihon shoki, they arrived some time in the early fifth century CE. In any case, other evidence shows that Paekche was “closely associated politically and socially with Yamato [= Japan]” (Serafim 1995). The two states exchanged “diplomatic missions . . . on an annual, or even a semiannual, basis” and, according to the Samkwuk saki, Paekche sent “royal princes as hostages to the Japanese court” (Best 1983: 146). Buddhism came to Japan from Paekche in the sixth century. There were no missions to China at all in the sixth century (Best 1983: 146) and full-scale Japanese missions to China began only in 600 CE (Varley 1984: 21). Hence it is probable that the early Japanese initially learned most of their sinography from Paekche teachers such as ‘Achiki’ and ‘Wani’ rather than from Chinese scholars. Sino-Paekche Consequently, the Japanese may have learned Sino-Paekche pronunciations of sinographs (i.e., Chinese pronunciations adapted to Paekche phonology) rather than the readings then current in China itself. In other words, the early Japanese may have been learning a ‘second-hand’ version of Chinese pronunciation from their Paekche teachers. As I will argue on p. 110, owing to a paucity of evidence, little is known about Sino-Paekche and other early varieties of Sino-Korean, despite attempts at reconstruction. Hence it is impossible to determine whether the early Japanese were learning Sino-Paekche readings, authentic Chinese readings, or readings which were somewhere in between (i.e., Chinese readings with a ‘Paekche accent’). For convenience, from this point onward I will speak of ‘Chinese pronunciations’ and cite Starostin’s and Pulleyblank’s reconstructions of Old 9

OLD JAPANESE

Chinese (OC) and Middle Chinese (MC) sinograph readings even though the Japanese may have been learning Sino-Paekche rather than Chinese pronunciation. I will also refer to the sinograph pronunciations learned by the Japanese at a specific period as ‘systems’ to avoid having to identify the sometimes unknown sources of the pronunciations. See p. 11. Pre-Suiko inscriptions Perhaps some of the earliest native Japanese transcriptions can be found in the Inariyama sword inscription which may be from the fifth century CE. The rather ugly, sometimes almost illegible characters suggest that the inscription may have been created by Japanese unfamiliar with sinography. The transcriptions in the Inariyama inscription, the Eta Funayama sword inscription written by a
 Zhang An (fifth century CE?),9 and the Suda Hachiman mirror inscription (early sixth century CE?),10 like the transcriptions in Chinese sources, are of limited value because they are restricted to isolated proper nouns in texts which are otherwise written in Classical Chinese. Some of the phonograms in the inscriptions such as
LOC *kaar ‘add’ for ka reappear in far later texts (e.g., Kojiki, Shoki, and Man’yôshû) with the same sound values. These frequent overlaps in usage reflect a definite continuity between this early orthography and the orthography of later centuries. Seeley (1991: 23) points out that “a substantial number of phonograms” in the Inariyama inscription were “employed [as phonograms] in early Korean texts quoted in the Nihon shoki, and in sixth century Korean texts in metal and stone.” Phonograms in the other two Japanese inscriptions can also be found as phonograms in Korean texts. For example,
LOC *kaar ‘add’ of the three Japanese inscriptions appears as a phonogram for ka not only in later Japanese transcriptions but also in transcriptions of Paekche, Koguryo, and Silla toponyms in Samkwuk saki. Thus orthographic continuity existed not only between various eras in Japan, but also between Korea and Japan. Since scribes in theory could have chosen phonograms from thousands of sinographs, it would have been possible for each scribe to invent his own orthography characterized by his own idiosyncratic choice of phonograms. While one scribe could use
LOC *kaar ‘add’ to represent the pre-OJ syllable ka, others could have used any of a number of alternative sinographs all pronounced like [ka] in the Chinese of the time:
,
,
(all pronounced *kaa in LOC),
,
,
,
,
,
,
,
,
(all pronounced *kaar in LOC), etc. There was no need for Paekche scholars to write Japanese according to Paekche traditions. However, the overlaps in usage between Korean and Japanese transcriptions on the one hand and Japanese transcriptional practice through time 10

EARLY TRANSCRIPTIONS OF JAPANESE

on the other show that the Paekche scholars (and others from the continent) brought their orthographic traditions with them and that these traditions persisted in Japan, albeit with modifications, through the centuries that followed. If this were not the case, it would be unlikely that scribes of different centuries somehow independently chose the same sinographs to represent certain Japanese syllables phonetically. It is more likely that scribes were consciously following a tradition imported from Korea which evolved over time in accordance with newly imported systems of sinographic readings. Systems of sinographic readings: Systems A and B The Japanese learned not one but several different systems of sinographic readings through the centuries. The first of these systems, ‘System A,’ may have been the early LOC pronunciation current at the time of the Wei zhi. However, I have already expressed my doubts about literacy among the natives of third-century Japan. The scribes of ‘Himiko’s’ court were probably Chinese immigrants, not Japanese trained by such immigrants. Nevertheless, I believe that ‘System A’ is a useful term, for we can call the earliest transcriptions ‘System A (= LOC)-based transcriptions’ to distinguish them from later transcriptions based on newer systems of sinographic pronunciation. The second of these systems, ‘System B,’ may have been the pronunciation which early Paekche scholars taught to the Japanese. As mentioned earlier, it is unclear whether this pronunciation was LOC, Sino-Paekche, or something in between. It is possible that OJ words suspected to be early loans from Chinese such as uma ‘horse’ (<
LOC *maarq) and umey ‘plum’ (<
late LOC *mRRy) originate from the same period as System B. System-B-based transcription (SBBT) System B served as the basis for the phonetic writing in the three early inscriptions named above and was probably quite different from subsequent systems of pronunciation such as Early Middle Chinese (EMC)11 and the Late Middle Chinese (LMC) dialect of Chang’an which were the basis of later Japanese orthographies. For example, in the Inariyama inscription, one finds the sinograph
LOC *kye ‘branch’ as a phonogram for pre-OJ kye. Early Paekche scholars probably taught Japanese to pronounce ‘branch’ as something like pre-OJ kye based on LOC (or Sino-Paekche?) *kye. In the centuries that followed, the pronunciation of ‘branch’ changed in Chinese: (2.1) LOC *kye > early EMC *ciR > late EMC *ci > LMC *cri 11

OLD JAPANESE

The Japanese learned of these new pronunciations initially through Korean intermediaries and later directly from the Chinese themselves once missions to China began in the seventh century. By the eighth century, Japanese orthographic habits had adjusted to reflect the above changes in the pronunciation of ‘branch.’ ‘Branch’ never represented OJ kye because its newer Chinese pronunciations no longer had initial *k: its EMC-based Go-on and LMC-based Kan-on reading is si < OJ si (< late EMC *ci, LMC *cri), not ke < OJ kye. However, the Inariyama, Eta Funayama, and Suda Hachiman inscriptions are not sufficient to allow us to outline the orthography based on System B. Most pre-OJ syllables are not represented at all in the early inscriptions. We have only phonograms for those syllables which appeared in a handful of proper nouns.

Suiko Period transcriptions Suiko Period inscriptions The situation improves considerably from the reign of Empress Suiko (r. 592–628 CE) onward. Beginning in the late sixth century, a comparatively large body of phonogram materials appears, including the Gangôji tray inscription (596 CE), the Gangôji Jôroku ™akyamuni halo inscription (605 CE), the Hôryûji Yakushi Buddha halo inscription (607 CE), and the Tenjukoku Mandala inscription (622 CE).12 With one exception,13 the phonograms in these materials continue to appear solely in proper nouns embedded in what are otherwise Chinese or quasi-Chinese texts. Nevertheless, these phonograms represent 60 out of the 87-plus syllables of OJ, a substantial improvement over the fragmentary representation of pre-OJ in earlier inscriptions. All the higher-frequency syllables of OJ are represented in the Suiko corpus. Syllables not represented such as OJ gye tend to be those of low textual frequency in OJ rather than chance omissions.14 System C The phonograms in the Suiko materials – Ôno Tôru’s (1962: 59) “old stratum kana”15 – reflect a third system of sinograph pronunciation, ‘System C’ (SC), similar but not identical to the Go-on stratum of Sino-Japanese. System C was based on a very late variety of OC and/or an early variety of EMC which was probably filtered through Sino-Paekche and possibly other Korean peninsular varieties of Sinoxenic. It lacks some of the archaic OC features found in transcriptions based on Systems A and B (e.g., the Wo ren zhuan and the Inariyama inscription). In the next section, I will demonstrate how System C is closer in some ways to EMC than to LOC. 12

EARLY TRANSCRIPTIONS OF JAPANESE

Table 2.1 Suiko Period
‘branch’ rhyme phonograms for OJ -a syllables PG

LOC

EMC

Gloss

OJ








*ke *ge *Ye *chyeq < EH *thyeq *9e

*kiR *giR *YiR *chiRq *yiR

‘odd number’ ‘strange’ ‘fitting’ ‘extravagant’ ‘move’

ka (*ka) ga (*Yga) ga (*Yga) ta (*ta)* ya (*ya)

Note It is possible that scribes may have assumed that the uncommon sinograph
‘extravagant’ was read like the graphically similar and much more common sinograph
LOC *taa / EMC *ta ‘many,’ a high-frequency phonogram for OJ ta.

System-C-based transcription (SCBT) Ôno Tôru (1962: 60–64) describes the Suiko “old stratum kana” system (our ‘System-C-based transcription’) in terms of a rather long list of correspondences between OJ sounds and Old and Middle Chinese initial and final categories. Without going into such detail, I will outline the main features of Suiko phonograms that distinguish them from earlier and/or later phonograms, thereby demonstrating the transitory nature of the Suiko writing system. First, sinographs belonging to the rhyme
EMC *ciR < LOC *kye ‘branch’ were used in Suiko times for OJ syllables ending in -a (Table 2.1). This usage may reflect EMC *-iR. It may also suggest a transitional stage *-ya in the development of the
‘branch’ rhyme between LOC and EMC: (2.2) LOC *-(y)e > (*-ya?) > early EMC *-iR In the Inariyama inscription, the sinograph
EMC *ciR < LOC *kye ‘branch’ itself represented pre-OJ kye, not pre-OJ ka, because the System B reading of
‘branch’ was presumably closer to LOC *kye. The only Suiko Period phonogram of the EMC
‘branch’ rhyme used for an OJ syllable ending in -ye (*e) is
EMC *pyiRq < LOC *pye ‘cause’ for OJ pye (not pa). Perhaps this usage is an archaism reflecting a *pye-like System B reading of
‘cause.’ By the end of the sixth century, the pronunciation of the
‘branch’ rhyme had undergone yet another change:16 (2.3) Early EMC *-iR > Late EMC *-i Since the Suiko materials date from the late sixth century onward, some of their phonograms reflect this change (Table 2.2). Apparently, there was synchronic variation in the System C readings of sinographs of the
‘branch’ rhyme category: an *a-like rhyme (*-ya?) in an older stratum of readings 13

OLD JAPANESE

Table 2.2 Suiko Period
‘branch’ rhyme phonograms for OJ -(y)i syllables PG

Early EMC

Late EMC

Gloss

OJ










*kyiR?~*ciR *gyiR *siR *triR *triRh *ãiRq *myiR

*ci *gyi *si *tri *trih *ãiq *myi

‘branch’ ‘forked road’ ‘this’ ‘know’ ‘wisdom’ ‘you’ ‘increasingly’

kyi kyi si ti ti ni myi

(not (not (not (not (not (not (not

ka or kye) ka or kye) sa or se) ta or te) ta or te) na or ne) ma or mye)

based on late-LOC or early EMC and an *i-like rhyme in a newer stratum of readings based on late EMC. The phonograms in Table 2.1 reflect the older *a-like stratum while the phonograms in Table 2.2 reflect the newer *ilike stratum. After the Suiko Period, the *a-like readings for sinographs of the
‘branch’ rhyme become obsolete. A second feature of Suiko orthography is the use of sinographs belonging to the
OC *Ya / EMC *YtR ‘fish’ rhyme category for OJ syllables ending in -ey:
LOC *ko > EMC *ktR ‘dwell’ and
LOC *koq > EMC *ktRq ‘raise’ for OJ key. It is difficult to explain the choice of these graphs for OJ key on the basis of the reconstructed readings given here; neither LOC *-o nor EMC *-tR resembles *Ry, the reconstruction of OJ ey which I will propose on p. 231. One possibility is that
‘dwell’ and
‘raise’ had LOC (and hence Sino-Paekche and System C) readings similar to OJ key (*kRy). Despite Starostin’s reconstruction of LOC *ko and *koq for these words, the evidence for rounded vowels in their rhyme category (
OC *nga / EMC *ngtR ‘fish’) prior to the Tang Dynasty is not overwhelming. The -o in their modern Go-on and Kan-on readings (ko, kyo) probably originates from an OJ period -o (*R) and even the other Sinoxenic readings dating from the Tang Dynasty (SK ke [kR], SV cu’ [kt]) have unrounded vowels. Ting (1972: 416) reconstructed the LOC
‘fish’ rhyme as *-aò. If I adopt Ting’s reconstruction of this rhyme,
‘dwell’ and
‘raise’ would respectively be LOC *kaò and LOC *kaòq. Furthermore, if Sino-Paekche, like modern Korean, lacked a velar glide coda *-ò,
LOC *kaò ‘dwell’ and
LOC *kaòq ‘raise’ could have been borrowed as Sino-Paekche (SP) *kay. The System C readings for the sinographs
‘dwell’ and
‘raise’ which the Paekche taught to the Japanese would then be LOC *kaò and LOC *kaòq and/or Sino-Paekche *kay. Hence it would be quite reasonable for scribes to use
LOC *kaò / SP *kay ‘dwell’ and
*kaòq / SP *kay ‘raise’ as phonograms for OJ key (*kRy). Following the Suiko Period,
‘dwell’ and
‘raise’ were phased out as phonograms for OJ key (*kRy). Neither graph appears in the poetry of Kojiki, but both appear in the poetry of Shoki as phonograms for OJ ko 14

EARLY TRANSCRIPTIONS OF JAPANESE

Table 2.3 Suiko Period
‘genitive’ rhyme phonograms for OJ -o syllables PG

EH

LOC

Gloss

OJ












*qRh *kRq *sR *tyRq *nRR *pwRh *mRRq *9Rq *rRq

*qth *ktq *syt *ctq *nRRy *pRwh *mRwq *ytq *ltq

‘idea’ ‘self ’ ‘think’ ‘stop’ ‘thereupon’ ‘rich’ ‘mother’ ‘already’ ‘village’

o (*R) ko (*kR) so (*sR) to (*tR) no (*nR) po (*pR) mo (*mR) yo (*yR) ro (*rR)

(*kR), not OJ key (*kRy). This new usage reflected the shift of LOC *-aò > EMC *-tR > early LMC *-iR. A third characteristic of the Suiko transcriptions is the use of phonograms belonging to the
OC *tR / EMC ct ‘genitive’ rhyme category solely for OJ syllables ending in -o (*R) (Table 2.3). Note that neither Starostin’s high central LOC vowel *t nor Pulleyblank’s EMC *-t for the
‘genitive’ rhyme category quite matches the mid central OJ vowel -o (*R): the OJ vowel and its post-OJ reflex ( [o] ) is mid whereas the LOC vowel is high. However, Ting (1972: 415) reconstructs this rhyme with a *R as *(y)Rò for LOC. Old Sino-Vietnamese loans from EMC have -d [RR], not -e [t], in this rhyme (Pulleyblank 1984: 212). Hereafter I will reconstruct the LOC pronunciation of the
‘genitive’ rhyme as *-R(ò), which probably would have been reflected as *-R in System C.17 After the Suiko Period, LOC *-R > EMC *-t > LMC *i; hence
‘genitive’ rhyme phonograms frequently transcribe OJ syllables ending in -iy (*t) or -i (*i) rather than -o (*R). In the poetry of Kojiki, there are only three OJ -o (*R) syllables transcribed by
‘genitive’ rhyme phonograms: o (= o;18 *R), go (*YgR), and po (*pR). No
‘genitive’ rhyme phonograms ever transcribe OJ -o (*R) syllables in the poetry of Shoki. Phonogram choices reflecting unpalatalized OC consonants comprise a fourth characteristic of the Suiko transcriptions. Prior to EMC (c. 600 CE), OC dentals palatalized in ‘Type B’ syllables (Pulleyblank 1994: 79). Starostin (1989: 464) believes that this change (1) occurred during the Eastern Han period (25–220 CE) and (2) involved a medial *-y- that had developed before short vowels: (2.4) *ty > c, *thy > ch, *dy > j, *dhy > jh, *ny > ã, *nhy > *ãh OC *k also palatalized in ‘Type B’ syllables (Pulleyblank 1991a: 77). Starostin (1989: 467) believes that this change also involved a medial *-yand occurred with other OC velars: 15

OLD JAPANESE

Table 2.4 Suiko Period phonograms reflecting unpalatalized OC initials PG

EH

LOC

Early EMC

Late EMC

Gloss

OJ






*kye *thyeq *tyRq

*kye ~ *ce *che *cRq

*ciR *chiRq *ctq

*ci *chiq *ciq

‘branch’ ‘extravagant’ ‘stop’

kyi (*ki) ta (*ta) to (*tR)

(2.5) *ky > c, *gy > j, etc. However, he notes that the palatalization of velars, unlike the palatalization of dentals, “clearly did not occur in all [Eastern Han OC] dialects” and cites Middle Chinese *k~*c variation, Buddhist transcriptions, and modern Min dialect retentions of *k as evidence for the late survival of unpalatalized velars. In the Suiko transcriptions, one finds the aforementioned phonograms (Table 2.4). It is difficult to explain the usage of these phonograms without referring to their earlier unpalatalized initials. This does not necessarily mean that their initials were unpalatalized in Chinese as late as the early seventh century. I have demonstrated above that the System C readings inferable from the Suiko transcriptions contained multiple strata based on various stages of LOC and EMC. Hence the use of
‘branch,’
‘extravagant,’ and
‘stop’ for OJ kyi, ta, and to could reflect older readings of those characters with *k- and *t- rather than their later EMC-based System C readings which might have reflected palatalized initials. Since I dated the arrival of scholars from Paekche circa the early fifth century, System B must date from c. 400 CE and System C must be somewhat newer – perhaps its earlier strata date from c. 500 CE. However, one should not assume that Systems B and C were brand-new at the time of their importation into Japan. It is likely that Paekche scholars and others from the continent were teaching the Japanese Sinoxenic (Sino-Paekche, etc.) readings borrowed from literary varieties of Chinese decades or even centuries earlier and/or conservative literary pronunciations of Chinese without innovations found in colloquial pronunciation. So the early stratum of System C imported into Japan c. 500 CE may actually resemble the literary Chinese of, say, c. 300 CE. Therefore the use of
‘stop’ with an OC-like *t- for OJ to (*tR) in the Gangôji Jôroku ™akyamuni halo inscription of 605 CE does not constitute proof for an initial *t- in the word
‘stop’ in China in 605 CE. Scholars who had just arrived in Japan from the continent probably found at least some of the Suiko era readings (i.e., the older strata of System C) to be quite outdated. However, clearly not all the System C readings were old-fashioned to continental ears in the seventh century. Despite the special characteristics of 16

EARLY TRANSCRIPTIONS OF JAPANESE

Suiko orthography enumerated above, there was a large overlap between Suiko and post-Suiko phonograms. Out of the 88 Suiko Period phonograms listed in Igarashi (1969), 71 (80.7 per cent) retain their Suiko Period OJ values in post-Suiko texts. Thus the foundations of eighth-century OJ orthography were largely already in place by the early seventh century. Instead of ‘reinventing the wheel,’ post-Suiko scribes revised their orthographic habits somewhat in accordance with two new varieties of sinograph readings which I will call ‘System D’ and ‘System E.’

Post-Suiko-Period transcriptions The material written in orthographies based on these two new systems vastly surpasses all previous transcriptions of Japanese combined in terms of both quantity and quality. Most of this wealth of material dates from the Nara Period (710–794 CE) and is in the form of poetry in three texts: the national history Kojiki (Record of Ancient Matters; 712 CE) and the poetic anthology Man’yôshû (Collection of Ten Thousand Leaves; c. 759 CE), both written in System-D-based orthography, and the official national history Nihon shoki (Chronicles of Japan; 720 CE), written in System-E-based orthography. (I will list other texts containing System-D-based transcriptions later in this section.) Since the orthography of Shoki is quite different from that of other post-Suiko materials, I will postpone discussion of Shoki until p. 28 Early post-Suiko transcriptions Post-Suiko, pre-Nara, System-D-based phonogram materials from the period 629–709 CE are sparse by comparison. These include a few proper nouns in fifteen inscriptions from 650 to 707 CE,19 the more numerous proper nouns of the Taihô census (702 CE), and the proper nouns and the four poems of Jôgû Shôtoku hô-ô teisetsu (The Imperial Story of Shôtoku of the Upper Palace, King of the Law), a biography of Prince Shôtoku (573–621 CE) which may date from c. 700 CE.20 Senmyôgaki Other post-Suiko, pre-Nara phonogram materials include edicts (senmyô) and liturgies (norito). The earliest senmyô preserved in Shoku nihongi (The Continued Chronicles of Japan; 797 CE) dates from 697 CE. John Bentley (p.c.) dates the earliest norito preserved in Engishiki (Institutes of the Engi Period; 927 CE) circa 685–690 CE on the basis of their spellings of the OJ syllables mwo, mo, and po. Senmyô and norito were written in the so-called senmyôgaki (‘edict writing’) style. Senmyôgaki employed two sizes of sinographs, each with a different function. Small phonograms represented “inflections and particles” while large semantograms represented the rest 17

OLD JAPANESE

(i.e., nouns, verb stems, etc.) (Habein 1984: 19). The syntax of senmyôgaki texts is closer to Japanese than to Classical Chinese. Senmyôgaki materials are of extremely limited value for the phonetic reconstruction of OJ, since only those OJ syllables which appear in inflections and particles are written with phonograms. One could not reconstruct OJ phonetics on the basis of the limited pool of phonograms used in senmyôgaki. Kojiki A far richer corpus of System-D-based orthography is found in Kojiki (Record of Ancient Matters), the earliest surviving Japanese national history and the “oldest [surviving] Japanese book” (Keene 1993: 33) compiled by Hieda no Are and Ô no Yasumaro. One hundred and twelve poems are scattered throughout all three volumes of Kojiki. Over half of the poems are in the third volume. These poems, probably dating from the seventh century or earlier, are the most relevant for our purposes, as they constitute the first large body of OJ texts written in phonograms. Earlier phonogram materials consisted almost entirely of isolated proper nouns or, in the case of senmyôgaki, inflections and particles. The prose of Kojiki, written in a somewhat Japanized variety of Classical Chinese, is no different from previous texts written in Chinese with phonograms largely for non-Chinese (i.e., Japanese) proper nouns. The poetry of Kojiki, on the other hand, contains complete OJ utterances entirely in phonograms. For example, the first line of the first Kojiki poem cited in (1.1) in Chapter 1 is written with the following phonograms in the original text (2.6). (2.6)




*yiah kuwq maw yakumwo eight- cloud ‘Eight clouds rise.’ (KJK 1.1) EMC OJ


 ta tQ tat-u stand-FIN

The poetry of Kojiki allows us to see OJ clearly for the first time without the hindrance of semantograms. If the above line were rewritten in the senmyôgaki style, it might appear as in (2.7). (2.7)



EMC *pεrt wun OJ ya kumwo eight- cloud ‘Eight clouds rise.’


lip tat-u stand-FIN




tQ

OJ yakumo ‘eight clouds’ and the first syllable ta of the OJ verb form tat-u ‘stand-FIN’ would be represented by the large semantograms
EMC 18

EARLY TRANSCRIPTIONS OF JAPANESE

*pεrt ‘eight,’
EMC *wun ‘cloud,’ and
EMC *lip ‘stand,’ which tell us nothing about OJ pronunciation; the only phonogram would be
EMC *tQ ‘capital city’ for the second syllable tu of tat-u ‘stand-FIN.’ The line would have fared even worse in pre-Kojiki, non-senmyôgaki texts, where it would have simply been translated into Classical Chinese without any phonograms at all. Many words in the OJ lexicon which did not occur as parts of proper nouns are not attested in phonograms prior to Kojiki. For example, the OJ words ya ‘eight,’ kumo ‘cloud,’ and tat- ‘stand’ do not occur in the preSuiko Inariyama, Suda, or Eta Funayama inscriptions. The value of Kojiki for the reconstruction of OJ phonology (and for other aspects of OJ) cannot be underestimated; I consider it to be one of the three key sources for the study of the OJ language. Unfortunately, the earliest surviving text of Kojiki is the Shinpukuji text from 1371–72 CE, based on a lost archetype from 1282 CE. The eighthcentury original and, of course, the manuscripts linking the original to the Shinpukuji and other extant manuscripts have long been lost. According to John Bentley (p.c.), “Kojiki was poorly treated through history [compared to Shoki, the official national history] and most mss [manuscripts] have disappeared. There are quotes in earlier texts, but it appears that Kojiki has come down to us in a faithful text.” Over a millennium passed before kokugaku (national learning) scholars took notice of Kojiki in the eighteenth century. I will deal with their studies of its orthography in the next chapter. Fudoki Less essential for our purposes but still worthy of note are the 25 poems in the various fudoki (gazeteers) compiled after Kojiki. The fudoki “include old tales, records of places, and descriptions of the crops, mineral resources, topography, and wildlife of each region [of Japan]” and were written “mainly in Chinese, though some passages are in the mixed Sino-Japanese style of the Kojiki, and the poems are phonetically transcribed” (Keene 1993: 63). The earliest fudoki may be the Harima fudoki (714 CE). As with Kojiki, none of the original manuscripts of the fudoki has survived; the texts extant today date from the Heian Period onward (John Bentley, p.c.; Matsumura 1972: 20). Chronologically next in order is Nihon shoki (Chronicles of Japan; 720 CE), the first national history of Japan and a second major source for the OJ language. However, since Shoki, unlike all other Nara Period texts, employs a phonetic orthography based on System E rather than System D, I will discuss it in my later section on System E. Bussokuseki no uta Approximately three decades after the compilation of Shoki, the twenty-one Bussokuseki no uta (Songs of the Buddha’s Footprints; 752 CE) were “carved 19

OLD JAPANESE

into the stone representation of the footprint of the Buddha consecrated in 749 at the Yakushi-ji in Nara” (Keene 1993: 90). These poems were in the 5-7-5-7-7-7 syllable Bussokuseki-tai (Buddha’s footprint style) format, attested only once in the Man’yôshû (MYS 3884). Although newer than Shoki, these poems’ orthography was based on System D, not E. Furthermore, since these poems were carved, they are free from textual corruption. Man’yôshû Man’yôshû (Collection of Ten Thousand Leaves; c. 759 CE) is the third major source for the OJ language and the largest collection of OJ poetry, with 4,516 poems in twenty volumes. Only the poems are in OJ. The rest is in Classical Chinese. The compiler(s) and date of compilation are uncertain; see Ômori et al. (1982: 6–9) for a summary of various claims and hypotheses. The earliest Man’yôshû poem may be one attributed to Emperor Yûryaku (c. fifth century CE?); the latest dates from 759 CE. The earliest extant fragmentary manuscript of Man’yôshû, the Katsura text, dates from the middle of the Heian Period (794–1185 CE) and the earliest complete manuscript, the Nishi Honganji text, dates from the late Kamakura Period (1185– 1333 CE) (Lange 1973: 19; Ômori et al. 1982: 62). The orthography of Man’yôshû poetry is quite complex. Unlike previous poetry which was written entirely in phonograms, Man’yôshû poetry was written not only with phonograms, but also semantograms and sinographs employed in what has been called gisho ‘playful writing.’ Furthermore, the phonograms of the Man’yôshû include not only the type of phonograms I have focused on so far (i.e., ongana), but also kungana phonograms and semantograms; I will explain the distinction between ongana and kungana as well as describing gisho on p. 26. Although ongana phonograms can be found throughout the Man’yôshû, the bulk of the poetry in ongana phonograms appears in volumes 5, 14, 15, 17, 18, and 20, which together contain a total of 1,033 poems, roughly less than one-fourth of the collection. The poems in the other volumes appear in mixed script; as I will explain later, this mixed script can be difficult to interpret and is of much less value than pure ongana phonogram script for the reconstruction of OJ. But even if one disregards these poems entirely, the hundreds of poems in System-D-based ongana orthography in volumes 5, 14, 15, 17, 18, and 20 still constitute a significant amount of data for reconstruction. Other post-Suiko transcriptions The remaining materials in System-D-based orthography from the Nara and Heian Periods pale in importance next to Kojiki, Shoki, and Man’yôshû.

20

EARLY TRANSCRIPTIONS OF JAPANESE

I have already mentioned two sources of senmyôgaki materials. The first of these is Shoku nihongi (The Continued Chronicles of Japan; 797 CE), a continuation of Nihon shoki (p. 28), which contains five poems in phonograms, two poems in senmyôgaki, and one poem in a mix of semantograms and phonograms in addition to senmyôgaki edicts from 697 CE onward and the usual proper nouns in phonograms. The other is Engishiki (Institutes of the Engi Period; 927 CE) which contains senmyôgaki liturgies dating from the late seventh century onward. Other materials include the 34 poems in phonograms in the poetic handbook Kakyô hyôshiki (Models of Poetic Classics; 772 CE), Japanese glosses in phonograms in Shin’yaku Kegon-kyô ongi shiki (Private Notes on the Newly Translated Avataú5aka Sutra; 794 CE), Nihon ryôiki (Record of Japanese Miracles and Oddities; c. 820 CE), and other Buddhist works, proper nouns and a few other words in phonograms in various censuses such as the Yôrô census (721 CE), and some more proper nouns in phonograms in Kaifûsô (Fond Recollections of Poetry; 751 CE), a collection of Classical Chinese poetry composed by Japanese. For a more complete list of Nara and Heian Period phonogram materials see the index of works in Ôno Tôru (1962: 1001–1006). The decline of phonogram transcription Writing in Classical Chinese rose to prominence in the early ninth century (Habein 1984: 21). Keene (1993: 182) claims even that “for a time, indeed, there was a danger that Chinese might wholly supplant Japanese as the medium of literary expression.” Phonograms enabled the sounds of OJ to be recorded relatively unambiguously, but with Japanese poetry in decline and Classical Chinese poetry in vogue, the need for phonograms diminished. When Japanese poetry made a comeback in the early tenth century, it was written in kana syllabograms rather than phonograms. Nevertheless, Kojiki, Man’yôshû, and other works in orthographies based on System D (and System E, in the case of Shoki) were still copied in manuscript form long after phonograms went out of use. Though the originals of many Nara Period texts with phonograms are lost, some of these later manuscripts still survive. System D From this vast quantity of material, fairly definite conclusions can be drawn about System D, the pronunciation of sinographs underlying the phonograms in the majority of Nara Period texts. System D was based on EMC and/or EMC-based varieties of Sinoxenic (most likely a late stratum of SinoPaekche). It was the last system of sinograph readings imported through the

21

OLD JAPANESE

Korean peninsula; later systems came directly from China. The final stratum of System D may have been brought to Japan by scholars fleeing Paekche and Koguryo, which respectively fell to Silla in 663 and 668. Most of the readings of the Go-on strata of Sino-Japanese probably derive from System D and the later strata of System C based on Early Middle Chinese. Only a few Go-on readings such as Go-on se for
LOC *\e ‘bestow’ and Go-on ze for
LOC *j(h)eq ‘this’ appear to be derived from earlier strata. If one reads phonogram texts in System-D (SD)-based orthography with modern-day Go-on readings, one can closely approximate those texts’ modernized readings, which lack the kô-otsu (A/B) vocalic distinctions (see Chapter 3). For example, the first line of the first poem of Kojiki is written in five phonograms as: (2.8)


OJ ya eight‘Eight



 kumo tat-u cloud stand-FIN clouds rise.’ (KJK 1.1)

and is read as yakumo tatsu in modernized pronunciation. The modern Go-on readings of the five phonograms in (2.8), from left to right, are ya, ku, mou (< SD *mow? < EMC *maw ‘hair’; not Go-on *mo), ta, and tsu (< SD *tu < EMC *tQ ‘capital city’). Modern Go-on ya ku mou ta tsu (< OJ Go-on *ya ku mou ta tu < SD *ya ku mow ta tu?) is very close but not identical to the modernized readings of the line as yakumo tatsu (and to OJ yakumo tatu). Although OJ period Go-on readings, System D readings, and the OJ syllables represented by System-D-based orthography were all very similar, one should be careful to make a distinction between the three. For example, in Man’yôshû (MYS) one can find the following three phonograms for OJ ko (*kR) (Lange 1973: 162): 1 2 3


EMC *ktq ‘self’ (MYS 4209.15.5, 4408.60.2)
EMC *htRq ‘allow’ (MYS 54.5.1, 896.3.2, 1740.62.2, 1749.11.1, 4011.39.1, 4187.10.2, 4254.7.2)
EMC *gtRq ‘gigantic’ (MYS 56.5.1, 3638.1.1).

Although all three phonograms represent the same OJ syllable, they are not homophonous in Go-on:
‘self’ and
‘allow’ are Go-on ko < OJ Go-on *kR but
‘gigantic’ is Go-on go < OJ Go-on *YgR. Furthermore, if System D, like Chinese, distinguished between velar stops (e.g., k, g) and velar or glottal fricatives (e.g., x, γ or h, ®), all three phonograms for OJ ko may have had different System D readings:
SD *kR ‘self,’
SD *hR ‘allow,’
SD *gR ‘gigantic.’ 22

EARLY TRANSCRIPTIONS OF JAPANESE

System-D-based transcription (SDBT) Ôno Tôru (1962: 64) calls the post-Suiko orthography based on System D the “middle stratum kana” in opposition to the System-C-based orthography of the Suiko Period and the System-E-based orthography of Shoki (his “old” and “new stratum kana”). He describes it as “having characteristics intermediate between the old and new strata, but much more heterogenous. In general the usage of graphs was based on Go-on but it is thought that there were uses of graphs based on northern [Chinese] pronunciation of the 7th century” (1962: 65). The influence of “northern [Chinese] pronunciation of the 7th century” may seem inevitable since the Japanese “dispatched a total of four missions to the Sui during the period 600 and 614 [in the Suiko Period] and fifteen to the Tang between 630 and 838” (Varley 1984: 21), but as I will demonstrate later, this influence on the “middle stratum” was minimal at best. It was the “new stratum” orthography of Shoki which most strongly reflected the Late Middle Chinese of Chang’an, the capital of the Sui and the Tang. Ôno (1962: 65–67) characterizes the “middle stratum kana” in terms of a long list of correspondences between OJ sounds and Old and Middle Chinese initial and final categories. Since I will deal with the System-Dbased orthography of the poetry of Kojiki and its basis in Chinese pronunciation in great detail in my reconstruction of the phonetics of eighth-century Japanese in Chapters 7 and 8, I will only sketch the four most prominent features of the “middle stratum kana” below. I have already touched upon all of these features in my discussion of Suiko Period orthography above. First, the rhyme
EMC *ciR ‘branch’ generally corresponds to OJ i (*i), yi (*i) or iy (*t), rarely to OJ e (*e) or ey (*Ry), and almost never to OJ a (*a) in System-D-based orthography. This usage reflects the change (2.9) early EMC *-iR > late EMC *-i in the
‘branch’ rhyme which occurred by the end of the sixth century (Pulleyblank 1984: 148). Second, phonograms of the
EMC *YtR ‘fish’ rhyme only rarely represent OJ syllables ending in -ey (*Ry) in System-D-based orthography; instead, they generally represent OJ syllables ending in -o (*R). This shift in usage reflects the change (2.10) LOC (Ting 1972) *-aò > EMC *-tR in the
‘fish’ rhyme. Third, phonograms of the
EMC *ct ‘genitive’ rhyme category represent OJ syllables ending not only in -o (*R) but also in -iy (*t) or -i (*i)

23

OLD JAPANESE

in System-D-based orthography. The extension of
‘genitive’ rhyme phonograms to OJ -iy and -i reflects the raising of the vowel in this rhyme in Chinese between LOC and EMC: (2.11) LOC *-R > EMC *t I must emphasize once more that System-D-based orthography has a large overlap with System-C-based orthography. The two are not wholly distinct entities. Nevertheless, Kojiki, as I will demonstrate in Chapters 7 and 8, has a relatively ‘pure’ System-D-based orthography with only a few archaisms necessitated by phonetic gaps in EMC. The phonogram-heavy sections of Man’yôshû, on the other hand, have a less ‘pure’ orthography. On the one hand, some elements from earlier orthographies persist such as the use of
LOC *kye ‘branch’ for OJ kyi (*ki) and
OC *tRq ‘stop’ for OJ to (*tR) (Omodaka 1967: 892, 897; Ômori et al. 1982: 27, 36). On the other hand, one innovation stands out:
LMC *niayh (Chang’an LMC *ndiayh) ‘mud’ for OJ de (*nde) in
 OJ sode (*sonde) ‘sleeve’ (MYS 3973.32) and in other words (MYS 3926.2.5, 3957.42.2, 3969.5.2, 3979.2.7, 4008.31.5, 4155.4.4, 4189.22.2, 4253.3.2; Lange 1973: 160). As I will demonstrate on p. 32, the correspondence of LMC nasals (Chang’an LMC prenasalized obstruents) to OJ voiced obstruents is the most prominent characteristic of System E orthography. Other types of transcriptions One must keep in mind that the phonogram-dominated sections of Man’yôshû do not dominate the anthology; they constitute less than one-fourth of the whole. The remaining three-quarters or more are written in mixed script employing not only phonograms but also other types of transcription. I will describe these other types of transcription only briefly, since some of them (particularly the gisho) are not found in the majority of other texts and none of them provides strong evidence for the reconstruction of OJ. Semantograms Semantograms (p. 6) are the most prominent alternative form of transcription in Man’yôshû. Although semantograms are plentiful in other early Japanese texts as a method of writing proper nouns, the semantograms of Man’yôshû are usually combined with phonograms to form a ‘mixed script’ for the writing of thousands of OJ poems. This mixed script is often reminiscent of senmyôgaki, but without any differences in character size: phonograms represent inflections and particles while semantograms represent the rest. For example, the second poem of Man’yôshû contains the line: 24

EARLY TRANSCRIPTIONS OF JAPANESE

(2.12)




EMC *tshwRn srεrn wuwq ‘village’ ‘mountain’ ‘exist’ OJ mura -yama ar-edo many -mountain exist-GER ‘Though many mountains exist . . .’ (MYS


tRYq ‘wait’

2.2)

In this line are two semantograms:
EMC *srεrn ‘mountain’ for OJ yama (*yama) ‘mountain’ and
EMC *wuwq ‘exist’ for are-, the first two syllables of the OJ verb form aredo (*arendR) ‘exist-GER.’ The last sinograph,
EMC *tRYq ‘wait,’ is a phonogram for do (*ndR), the last syllable of aredo (*arendR) ‘exist-GER.’ The first sinograph in (2.12),
EMC tshwRn ‘village’ for OJ mura- ‘many,’ is obviously neither a semantogram nor a phonogram in the sense used here (i.e., an ongana), since
‘village’ is not a translation of OJ ‘many’ and EMC tshwRn does not phonetically resemble OJ mura-. I will deal with graphs of this type in the next section. Kungana phonograms
EMC tshwRn ‘village’ for OJ mura- ‘many’ in (2.12) is a kungana phonogram. Ongana phonograms have pronunciations derived ultimately from premodern Chinese (e.g., LOC, EMC, LMC), but kungana phonograms have pronunciations derived from OJ translations of Chinese morphemes. Unlike semantograms or ongana, reading kungana involves a two-step process. First, one must read a kungana as if it were a semantogram. Thus one would read
EMC tshwRn ‘village’ as its OJ translation equivalent, mura ‘village.’ Second, one must reinterpret this reading in terms of an OJ (near-)homophone appropriate for the context in which the sinograph is found. MYS 2 is a poem whose opening praises Mount Kagu among the mountains of Yamato; hence OJ mura ‘village’ is reinterpreted as OJ mura‘group, cluster, mass,’ a near-homophonous prefix differing only in pitch accent (Martin 1987: 488), and OJ mura-yama refers to the ‘many mountains’ of Yamato, among which Mount Kagu is the best. Although kungana are phonograms in the sense that they represent sounds, they are of limited value for our purposes because they can only tell us which morphemes (or, in some cases, morpheme sequences) were nearhomophonous in Japanese. We know that OJ mura ‘village’ was mura because it is attested elsewhere in ongana phonograms: in the Man’yôshû itself, it appears in the preface to poem 1.439 as the first part of the title
 OJ murazi (*muranzi) ‘village master’ (Martin 1987: 488). Suppose that the OJ word for ‘village’ was not attested in ongana phonograms. How would we then interpret
EMC tshwRn ‘village’ in
 OJ ?-yama ‘?-mountain’?
EMC tshwRn ‘village’ has never been used as an ongana phonogram, and 25

OLD JAPANESE

context makes it doubtful that it was a semantogram: what is a ‘villagemountain’ and why would one say that although Yamato has villagemountains, Mount Kagu is the best? By the process of elimination,
EMC tshwRn ‘village’ would probably be a kungana rather than a rare gisho graph (see below), but we could only say that the OJ morpheme represented by
EMC tshwRn ‘village’ was nearly homophonous with the (hypothetically) lost OJ word for ‘village.’ Homophony aside,
EMC tshwRn ‘village’ is as mute as a semantogram. It adds nothing to the phonetic information conveyed by the ongana spelling
 EMC *muRq ltaY for OJ mura ‘village.’ Although I cited a kungana example from Man’yôshû, kungana are not unique to that poetic anthology. Five kungana can be found in Suiko Period materials (Igarashi 1969: 160–163) and some more kungana are also found in post-Suiko materials other than the Man’yôshû. Nevertheless, kungana never became the dominant variety of phonograms in early Japanese writing. Kungana were never even devised for some OJ syllables. Although all of the poetry of Kojiki and Shoki and a considerable fraction of the poetry of Man’yôshû are written entirely in ongana phonograms, there are no examples of all-kungana writing. Owing to the limited value of kungana as phonograms and the lack of kungana in the Kojiki and Shoki poems serving as the basis of my reconstruction of OJ, I use the term ‘phonogram’ throughout this book to refer only to ongana phonograms except when referring to ‘kungana phonograms.’ Gisho (Tawamuregaki) Of even less value than kungana for reconstructive purposes are the rare transcriptive sinographs in the Man’yôshû collectively referred to as gisho or tawamuregaki (both meaning ‘playful writing’). All gisho record OJ in the most convoluted way possible. Among the several varieties of gisho, I will only cite three major categories established by Ômori et al. (1982: 41) which I call graphemograms, onomatograms, and arithmograms. GRAPHEMOGRAMS

Graphemograms are sinographs combined into short Classical Chinese phrases which describe the graphemes of a single semantogram. The most infamous example is: (2.13)



EMC *srεrn jtaYq ‘mountain’ ‘top’ Classical Chinese: ‘Atop a (MYS 1787.15)




buwk wuwq srεrn ‘again’ ‘exist’ ‘mountain’ mountain, there is yet another mountain’

26

EARLY TRANSCRIPTIONS OF JAPANESE

These five sinographs describe the sinograph
EMC *chwit ‘go out,’ which resembles two
‘mountains’ stacked upon one another.
EMC *chwit ‘go out’ in turn is a semantogram representing the OJ verb id-e ‘go out-GER.’21 ONOMATOGRAMS

Onomatograms are sinographs combined into short Classical Chinese phrases which describe the sounds being referred to. In MYS 2839.2, the second syllable of the verb suffix OJ -amu ‘tentative’ is written as
 EMC *Yuw miayY ‘cow cry,’ because it was homophonous with OJ mu ‘moo.’ ARITHMOGRAMS

Arithmograms are sinographs combined into numerals from which an OJ reading can be derived after an arithmetic operation. In MYS 3318.30, the second syllable of the retrospective verb suffix OJ -isi is written as

EMC *ãih ãih ‘two two,’ because it was homophonous with the SinoJapanese loanword for ‘four,’ si
< EMC *sih. ASSESSMENT OF GISHO

Were it not for the fact that the morphemes and words in the above examples are attested elsewhere in more easily interpretable phonograms, their gisho spellings would have rendered them unreadable. As Habein (1984: 13) writes, “this kind of writing [i.e., gisho] gives us the impression that the writers of these poems [in the Man’yôshû] were certainly enjoying their freedom in selecting characters for man’yôgana [= phonograms; see p. 6].” However, this freedom came at the expense of clarity. Although gisho are a testament to the creativity of Nara Period scribes, they tell us very little about how those scribes pronounced their language. The term ‘man’yôgana’ With the exception of semantograms, all of the remaining types of transcriptive phonograms described so far – ongana phonograms, kungana phonograms, and gisho – can be called ‘man’yôgana.’ This term, literally meaning ‘the kana [= syllabic symbols] of the Man’yô[shû],’ inevitably appears in all discussions of early Japanese writing. Unfortunately, the term man’yôgana is potentially misleading for two reasons. First, man’yôgana are not unique to the Man’yôshû; the ongana and kungana of other texts are also man’yôgana. Second, unlike kana, man’yôgana are not necessarily syllabic symbols. One ongana or kungana can represent two syllables. An example of a disyllabic ongana is
EMC *tsuawk ‘foot’ 27

OLD JAPANESE

for the first two syllables of the OJ title sukune (*sukune), spelled
 EMC *tsuawk nri in Hô-ô teisetsu. An example of a disyllabic kungana
is EMC tshwRn ‘village’ for OJ mura- ‘group, cluster, mass’ from (2.12). Since this book is devoted to the reconstruction of OJ on the basis of ongana rather than on the basis of man’yôgana as a whole (including kungana and gisho as well as ongana), I will hereafter rarely use the term man’yôgana.

The transcriptions of Nihon shoki The text of Nihon shoki As mentioned on p. 17, Nihon shoki (Chronicles of Japan; 720 CE) is one of the three major sources of Nara Period phonogram material. However, Shoki, like Kojiki and Man’yôshû, is not an all-OJ text, but a mixture of Classical Chinese prose and OJ poetry. Shoki is the first official national history of Japan and the second earliest surviving history of Japan, covering events from the creation of Japan to the end of the reign of Empress Jitô (r. 686–697). Shoki’s Classical Chinese, unlike that of Kojiki, showed little Japanese influence (Seeley 1991: 48). Bentley (1998a: 5) argues that the “beautifully written Chinese” of Shoki “guarantees that the [Japanese] court employed Korean or Chinese immigrants in various stages of the text’s compilation and creation”; as we shall later see, the analysis of phonograms in Shoki poetry by Mori (1991) lends credence to this claim. Both Shoki and Kojiki originated in the court of Emperor Tenmu (r. 673– 686) decades before the completion of both works. According to Shoki itself, Tenmu ordered the compilation of a history in 681 CE. Bentley (1998b: 15) theorizes that this resulted in a “major historiographical movement, with many people competing among themselves to compile an official history acceptable to the crown.” He further theorizes that power politics caused the compilation of both Kojiki and Shoki to span over three decades (Bentley 1998b: 19). Miner et al. (1985: 209) attribute the compilation of Shoki to Prince Toneri (d. 735), who, according to Shoku nihongi (The Continued Chronicles of Japan), submitted the final version of Shoki to the court in 720, but he may have been only “the director of the project . . . and as with all good leaders, he delegated the bulk of the nitty-gritty work to scribes and other low ranking courtiers” (Bentley 1998b: 23). For a full account of theories on the origins of Shoki, see Bentley (1998b). Unfortunately, we have no original manuscripts of Shoki, even though Shoki was revered for a long time as the Japanese history while Kojiki remained ignored for a millennium. According to Bentley (p.c.), “the earliest [surviving] fragment [of Shoki ] is from the tenth or eleventh century” and “it is difficult to tell which is the best” among the surviving manuscripts. In Chapters 7 and 8, I will demonstrate that extant manuscripts of Shoki have many probable scribal errors. 28

EARLY TRANSCRIPTIONS OF JAPANESE

The poetry of Nihon shoki Out of the thirty volumes of Shoki, volumes 1–3, 5, 7, 9–17, 19, and 22–27 contain a total of 128 OJ poems. These poems are written entirely in SystemE-based transcription (pp. 30, 32). This transcription is largely unique to Shoki and, for reasons given in the next section, cannot date before the late seventh century. Although Shoki was completed in 720 CE, the poems date from the late seventh century or earlier. The last three poems in volume 27 of Shoki are from Tenji 10 (671 CE). One should not confuse the date of composition of a poem with the date of its transcription into phonograms in Shoki. Many of these poems also occur in Kojiki, but in System-C-based transcription. Shoki allows us to see these overlapping poems from another perspective – that of System-E-based transcription. Compare the Kojiki and Shoki spellings of the first line of the first poem shared by both works: (2.14)




*yiah kuwq maw


LMC *yiah kRw mRw OJ yakum(w)o eight- cloud ‘Eight clouds rise.’ (KJK 1.1, NS EMC


ta
ta tat-u stand-FIN 1.1)


tQ
thuRh

(KJK 1.1) (NS 1.1)

The first and fourth phonograms (
and
) are identical, but the rest are different. The value of Shoki poetry for this book lies in its ability to serve as a separate witness for OJ. Seeing the OJ syllable ku in OJ kumo ‘cloud’ written as
EMC *kuwq ‘long time’ in Kojiki but as
LMC *kRw ‘hook’ in Shoki tells us that OJ ku (which I reconstruct as *ku) must have sounded like both EMC *kuwq and LMC *kRw. Although Shoki was completed in 720 CE, eight years after the completion of Kojiki, both works were compiled during the same period (the late seventh century to the early eighth century), so one cannot claim that
EMC *kuwq ‘long time’ in Kojiki represents an earlier pronunciation of OJ ku and that
LMC *kRw ‘hook’ in Shoki represents a later pronunciation of OJ ku. Shoki is not proof for radical change in OJ between 712 and 720 CE. One might argue that perhaps Hieda no Are and Ô no Yasumaro, the authors of the Kojiki, and the compilers of Shoki under Prince Toneri’s supervision had different dialects and that the differing transcriptions reflected the differences between their dialects, but both histories were products of the Japanese court, and would therefore reflect Central OJ, the dialect of the court. It would make no sense for Ô no Yasumaro or Prince Toneri to submit a national history with poems in a non-court dialect.

29

OLD JAPANESE

In fact, the Kônin Shiki (812 CE) says that Ô no Yasumaro was ordered to compile Shoki and was among those who submitted it to the court in 720 (ed. Kuroita 1932: 3–4, 191). Bentley (1998a: 25) believes that “the character usage (also known as wordprinting) of the first two books of Shoki compared with Kojiki’s preface [by Yasumaro] suggests that Yasumaro helped compile ‘The Age of the Gods’ section and nothing more.” If so, then both the Kojiki and Shoki versions of the poem in question (KJK 1 and NS 1, supposedly composed by the god Susano-o during ‘the age of the gods’) may have passed through Ô no Yasumaro’s hands. It is extremely unlikely that the Kojiki and Shoki transcriptions of this poem represent different pronunciations by the same man on different occasions. It is more likely that the Kojiki and Shoki transcriptions reflect different orthographies for Central OJ during the same time period (the late seventh century to the early eighth century). Furthermore, transcriptions of certain parts of Shoki (the ‘alpha’ sections of Mori 1991) seem to have been done by continental scribes or at least by Japanese very acquainted with Chinese, whereas Kojiki was written by Hieda no Are with Ô no Yasumaro. Thus the difference in transcriptions may reflect differences in the authors’ knowledge of Chinese. I will say more about this possibility on p. 58. System E (Chang’an Late Middle Chinese) So far, most of the previous systems of sinograph readings brought to Japan have defied precise identification. Only System A could be identified as the variety of Late Old Chinese underlying the transcriptions of Japanese in Wei zhi, and it may not have ever been learned by any Japanese if Himiko’s court scribes were Chinese. Systems B, C, and D were ultimately based on Late Old Chinese and Early Middle Chinese to varying degrees, but the degree to which they reflect Chinese directly is uncertain. I am inclined to think that all three systems were based on different strata of Sino-Paekche, but there is no definite proof of this. Although ties between early Japan and Paekche were close, not all continental immigrants to Japan were from Paekche. Also, there is no guarantee that scholars from the Korean peninsula taught Old Sino-Korean (Sino-Paekche, etc.) rather than genuine Chinese pronunciations. Perhaps transcriptions based on Systems B through D reflect syntheses of Chinese pronunciations learned from various sources. System E, on the other hand, is much easier to identify. As mentioned earlier, after sending no missions to China at all in the fifth century (Best 1983: 146), the Japanese “dispatched a total of four missions to the Sui during the period 600 and 614 [in the Suiko Period] and fifteen to the Tang between 630 and 838” (Varley 1984: 21). The Japanese no longer had to rely on Korean peninsular scholars to learn Chinese; now they could go directly to the source. Both the Sui Dynasty (581–618 CE) and the Tang Dynasty (618–907 CE) were based in Chang’an in northwestern China. By the time 30

EARLY TRANSCRIPTIONS OF JAPANESE

the short-lived Sui had reunited China in 589 CE with Chang’an as its capital, Chang’an, once the capital of the Western Han (202 BCE–8 CE), “had long sunk into provincial obscurity” (Pulleyblank 1984: 3). The Qieyun rhyme dictionary of Early Middle Chinese (EMC), though completed at Chang’an in 601 CE, probably reflected not the speech of Chang’an but what Pulleyblank (1984: 3) believes to be the “educated norms of north and south in the [preceding] period of division.” (Exactly what language, if any, is represented by Qieyun is a controversial issue which I will discuss on p. 93.) These non-Chang’an “educated norms” did not last long, and “a new standard based on the speech of the Sui-Tang capital, Chang’an” emerged during the seventh century (Pulleyblank 1991a: 2). Pulleyblank (1970–71, 1984, 1991a) calls this new standard ‘Late Middle Chinese’ (LMC). There are no records telling us exactly how LMC was brought to Japan, but it is most likely that it had been brought back by Japanese who had learned it during missions to Tang China. Continental immigrants familiar with LMC may have also contributed to Japanese knowledge of the new system of pronunciation. I will call the variety of LMC that entered Japan ‘System E’ to distinguish it from Systems A–D and from other varieties of LMC, such as the southeastern variety which I hypothesize to be the basis of Sino-Vietnamese (p. 126). System E became the basis of the new Kan-on (‘sounds of China’) stratum of Sino-Japanese as well as the orthography of Shoki. Despite our extremely minimal knowledge of LMC dialects, System E can probably be identified as the Chang’an dialect of LMC (CLMC). CLMC and other northwestern LMC dialects are notable for a feature which Coblin (1994: 58) and Tôdô (1980: 281) call “denasalization.” Northwestern Chinese transcriptions of Sanskrit (Coblin 1994) and Tibetan transcriptions of northwestern Chinese (Luo 1933a) show that EMC nasals became prenasalized obstruents: (2.15) *m > *mb, *n > *nd, *ã > *ã9, *Y > *Yg However, nasal finals sometimes blocked denasalization in certain rhymes (Tôdô 1978: 1584): (2.16)
EMC *ma ‘demon’ (cf. Go-on ma) > CLMC *mba (not *ma; cf. Kan-on ba) but:
EMC *miayY ‘bright’ (cf. Go-on myau) > CLMC *miayY (not *mbiayY; cf. Kan-on mei, not *bei). Although Coblin (1994: 58) dates denasalization in northwestern Chinese to “at least the mid-eighth century,” Tôdô (1980: 280–281) presents northwestern Chinese transcriptional data from an unpublished conference paper by Mizutani (1956) in which sinographs with earlier nasal initials were used 31

OLD JAPANESE

to represent Indic syllables with voiced stop initials as early as the mid seventh century. The feature appears in Zhitong’s transcriptions, dated 627– 653 CE, which are not among the data used in Coblin’s (1994) study of northwestern Chinese. The feature also appears in Kan-on and Shoki: for example, the aforementioned sinograph
CLMC *mba ‘demon’ is not only Kan-on ba but also a phonogram for OJ ba (*mba) in Shoki (91.2.5, 91.4.7, 91.5.5, 97.21.2). Therefore System E must have been known in Japan by the late seventh century, since the transcriptions of Shoki poetry were probably done over the decades between Tenmu’s 681 CE order for the compilation of a history and the completion of Shoki in 720 CE. I do not think it is likely that System E first came to Japan in, say, 713 CE, the year after the completion of Kojiki, and that all the Shoki poetry was rewritten in System-E-based orthography between 713 and 720 CE. In the above example, the System-E-based Kan-on and Shoki phonogram readings happened to be identical, but one should be careful not to equate the two. It is true that one can approximate the reading of texts in SystemE-based orthography by pronouncing them in Kan-on. Compare the Kan-on reading of NS 1.1 with the actual OJ reading: (2.17)



*yiah kRw ya ku yakumo eight- cloud ‘Eight clouds rise.’ (NS

CLMC Kan OJ


mbRw bou


 ta thuRh ta to tat-u stand-FIN

1.1)

Kan-on ya ku bou ta to is close, but not identical to OJ yakumo tatu:
CLMC *mbRw ‘luxuriant’ is Kan-on bou, not mo, and
‘Cuscuta chinensis’ is Kan-on to, not tu. The Kan-on reading and the System-E-based orthography reading of a phonogram share the same source (i.e., CLMC) but are not always one and the same. System-E-based transcription (SEBT) Ôno Tôru (1962: 67–69) calls the Shoki orthography based on System E the “new stratum kana” in opposition to the Suiko Period System-C-based orthography and the post-Suiko System-D-based orthography (his “old” and “middle stratum kana”). He describes these “new stratum kana” by means of a rather long list of correspondences between OJ sounds and Middle Chinese initial and final categories. Since I will describe the details of Shoki orthography in Chapters 7 and 8, I will only outline its major characteristics here. Perhaps the most obvious characteristic of Shoki transcription is its use of sinographs with Chang’an LMC prenasalized obstruents for OJ voiced 32

EARLY TRANSCRIPTIONS OF JAPANESE

Table 2.5 CLMC prenasalized obstruent–OJ consonant correlations in Shoki EMC

CLMC

OJ

Examples (all from Shoki)

*m

*mb

*n

*nd

*ã

*ãj

*Y

*Yg

b (*mb) m (*m) d (*nd) n (*n) z (*nz) n (*n) g (*Yg)










CLMC CLMC CLMC CLMC CLMC CLMC CLMC

*mba ‘demon’ for ba (91.2.5, etc.) *mba ‘demon’ for ma (24.3.1, etc.) *ndyiay ‘mud’ for de (78.6.7, etc.) *ndyiay ‘mud’ for ne (30.4.6, etc.) *ã9iq ‘ear ornament’ for zi (5.4.7, etc.) *ã9iq ‘ear ornament’ for zi (7.2.4, etc.) *Ygi ‘doubt’ for giy (31.4.6, etc.)

obstruents as well as OJ nasals (Table 2.5). In System-C- and D-based transcription, EMC *Y often transcribed OJ g (*Yg):
EMC *YiR < (preEMC *Yya?) ‘fitting’ transcribed OJ ga (*Yga) in the Suiko Period (Table 2.1) and
EMC *Yt ‘doubt’ transcribed OJ giy in KJK (25.2.6, 58.6.5, 58.8.6, 111.2.6) as well as in Shoki (31.4.6, etc.). However, apart from some exceptions in transcriptions of OJ de in Man’yôshû (p. 24), other transcriptions never used other EMC nasals (i.e., *m, *n, *ã) to represent the other OJ voiced obstruents (i.e., b, d, z). CLMC prenasalized obstruents were forced to do ‘double duty’ in Shoki transcription for two reasons. First, all initial nasals had become prenasalized obstruents unless followed by finals with nasal elements. This greatly reduced the number of nasal-initial syllables in the language. If the denasalization rule had been completely regular (e. g., no *ma or *mbiayY), then CLMC had no nasal-initial open syllables at all: i.e., no CLMC *ma, *na, etc. This posed problems for Shoki scribes because unlike CLMC, OJ did have nasal-initial open syllables. For example, to write OJ ma (*ma), scribes had to choose between open syllables with prenasalized obstruent initials like
CLMC *mba ‘demon’ or closed syllables with nasal initials like
CLMC *maY ‘weeds.’ Even if we assume that CLMC *a and OJ a (*a) were phonetically identical, neither CLMC *mba nor CLMC *maY was an exact match for OJ ma (*ma). Excluding all textual variants, OJ ma was written 28 times with
CLMC *maY ‘weeds’ and 152 times with sinographs for CLMC *mb- syllables in Shoki. In fact, nearly all OJ nasal-initial syllables were usually written in Shoki with sinographs for CLMC prenasalized obstruent-initial syllables, with the sole exception of OJ no (*nR). Table 2.6 shows how often CLMC prenasalized obstruent and nasal initials corresponded to the OJ nasals m and n in Shoki poetry, excluding textual variants. For this table, I have assumed that the denasalization rule had no exceptions (e.g., no *ma or *mbaY). The spelling of OJ no (*nR) as
CLMC *nRY ‘can’ alone accounts for 95.6 per cent (260/272) of OJ n-initial syllables spelled with CLMC *n and 81.3 per cent (260/320) of OJ nasal-initial syllables spelled with CLMC nasal initials. I suspect that this use of
CLMC *nRY ‘can’ for OJ no may have 33

OLD JAPANESE

Table 2.6 CLMC initials for OJ m- and n-initial syllables in Shoki poetry, number (%) OJ

CLMC *mb

CLMC *m (/_Y)

Total

ma (*ma) mey (*mRy) miy (*mt) mo (*mo) mu (*mu) mye (*me) myi (*mi)

152 (84.4) 38 (100) 8 (100) 154 (90.1) 66 (95.7) 18 (100) 154 (100)

28 (15.6) 0 (0) 0 (0) 17 (9.9) 3 (4.3) 0 (0) 0 (0)

180 38 8 171 69 18 154

Total

590 (92.5)

48

(7.5)

638

OJ

CLMC *nd

CLMC *n (/_Y)

Total

na (*na) ne (*ne) ni (*ni) no (*nR) nu (*nu) nwo (*no)

138 (100) 21 (100) 168 (100) 12 (4.4) 14 (53.8) 3 (100)

0 (0) 0 (0) 0 (0) 260 (95.6) 12 (46.1) 0 (0)

138 21 168 272 26 3

Total

356 (56.7)

272 (43.3)

628

Overall total

946 (74.7)

320 (25.2)

1,266

partly been a carryover from System-D-based orthography; out of 272 attestations of OJ no in Kojiki poetry, 99.3 per cent (270/272) are written as
EMC *nRY ‘can.’ If
CLMC *nRY ‘can’ were not included in my statistics, then 94.0 per cent (946/1,006) of OJ m and n syllables in Shoki poetry are spelled with CLMC prenasalized obstruents. CLMC prenasalized obstruents also had to do ‘double duty’ in Shoki transcriptions because LMC in general, unlike EMC, had no plain voiced obstruent initials. Hence CLMC had *p-, *ph-, *p®-, *mb- (except before final nasals), and *m- (only before final nasals), but no *b-. Earlier plain voiced obstruents such as EMC *b-, *d-, *g-, etc. had become “partly devoiced” as LMC *p®-, *t®-, *k®-, etc. (Pulleyblank 1984: 68). These “partly devoiced” obstruents are called
zhuo ‘muddy’ in traditional Chinese phonological terminology. Pulleyblank (1991a: 6) describes
zhuo ‘muddy’ consonants as “having voiceless onset with voiced aspiration or murmur at their release, spreading through the following syllable [sic].” I believe that Pulleyblank actually means that breathiness spread through the rest of a syllable with a
zhuo ‘muddy’ consonant: thus
LMC *p®a ‘crone,’ a Shoki phonogram for both OJ ba (*mba; NS 2.2.5, 35.15.5, 116.4.7) and OJ pa (*pa; NS 76.10.1, etc.), would be pronounced [p®&] with a breathy vowel [&]. 34

EARLY TRANSCRIPTIONS OF JAPANESE

Table 2.7 Common EMC–OJ voiced obstruent correlations in Kojiki EMC

OJ

Examples (all from Kojiki)

*b

b (*mb)


EMC *ba ‘crone’ for ba (2.11.3, etc.)

*d

d (*nd)


EMC *da ‘hill’ for da (2.16.3, etc.)

*j

z (*nz)


EMC *juwq ‘receive’ for zu (2.16.6, etc.)

*g *Y *γ/®

g (*Yg) g (*Yg) g (*Yg)

w EMC *gt ‘game of Go’ for go (1.3.3, etc.)
=EMC *Yt ‘doubt’ for giy (25.2.6, etc.)
EMC *γah/®ah ‘congratulate’ for ga (1.2.6, etc.)

CLMC ‘muddy’ consonants were less like OJ voiced obstruents (which were phonetically prenasalized, as I will demonstrate in Chapter 7) than CLMC prenasalized obstruents. Why write OJ ba (*mba) with
CLMC *p®a ‘crone’ without either a voiced stop or nasalization when one could write it with
CLMC *mba ‘demon,’ which would be a much better match? Textual variants aside, in the poetry of Shoki, OJ ba is written only three times (3/38; 7.9 per cent) with
CLMC *p®a ‘crone’ but is written 31 times (31/38; 81.6 per cent) with
CLMC *mba ‘demon’ and other sinographs with the CLMC initial *mb-. Summing up, CLMC prenasalized obstruents did ‘double duty’ because of (1) distributional gaps in CLMC (no *ma, *na, etc.) which forced scribes to use CLMC *mba, *nda, etc. for OJ ma, na, etc. and (2) their phonetic similarity to OJ prenasalized obstruents: CLMC *mb was more like OJ b (*mb) than ‘muddy’ CLMC *p®. The distribution of phonograms with ‘muddy’ initials in Chinese is the second most obvious characteristic of Shoki transcription. In other systems of transcription, LOC and EMC voiced obstruents corresponded to OJ prenasalized voiced obstruents, with the sole exception of LOC and EMC *Y as well as LOC and EMC *g and *® for OJ g (*Yg) (Table 2.7). Although EMC voiced obstruents (*b, *d, etc.) were not perfect matches for OJ prenasalized voiced obstruents (*mb, *nd, etc.), EMC did not have any prenasalized obstruents. If Sino-Paekche and other varieties of Old Sino-Korean were like Middle and Modern Korean, they did not have any prenasalized obstruents either. Thus scribes had no choice but to use EMC voiced obstruent initials to write OJ prenasalized voiced obstruents; EMC voiceless obstruents (*p, *ph) and nasals (*m, etc.) were clearly not acceptable alternatives. However, as already noted above, EMC voiced obstruents had become the partly devoiced ‘muddy’ obstruents of LMC: *b, *d, etc. > *p®, *t®, etc. ‘Muddy’ obstruents always correspond to voiceless obstruents in Kan-on but can regularly correspond to both OJ voiceless and voiced obstruents 35

OLD JAPANESE

Table 2.8 Common CLMC ‘muddy’ initial–OJ obstruent correspondences in Shoki EMC CLMC

OJ

Examples (all from Shoki)

*b

*p®

*d

*t®

*z

*s®

*g

*k®

*γ/®

*x®

p (*p) b (*b) t (*t) d (*nd) s (*s) z (*nz) k (*k) g (*Yg) k (*k) g (*Yg)


CLMC *p®a (Kan-on ha) ‘crone’ for pa (76.10.1, etc.)
CLMC *p®a (Kan-on ha) ‘crone’ for ba (2.2.5, etc.)
CLMC *t®a (Kan-on ta) ‘hill’ for ta (74.4.4b, etc.)
CLMC *t®a (Kan-on ta) ‘hill’ for da (2.5.3b, etc.)
CLMC *s®õ (Kan-on si) ‘word’ for si (7.3.1, etc.)
CLMC *s®iR (Kan-on syo) ‘preface’ for zo (50.5.5)
CLMC *k®yi (Kan-on ki) ‘forked road’ for kyi (1.2.7 etc.)
CLMC *k®üR (Kan-on ku) ‘provide’ for gu (28.6.6 etc.)
CLMC *x®uR (Kan-on ko) ‘why’ for kwo (18.1.7, etc.) (No examples of CLMC *x®- : OJ g- in Shoki)

in Shoki poetic transcription (Table 2.8). These ‘muddy’ obstruents had voiceless onsets which enabled them to transcribe OJ voiceless obstruents. Furthermore, the voiced aspiration of CLMC ‘muddy’ obstruents might have been identified with the voicing of OJ voiced obstruents, enabling them to transcribe OJ voiced obstruents as well. The choice of ‘muddy’ obstruents for OJ voiced obstruents also may have been influenced by earlier transcriptional practices and/or Go-on. For example, Shoki scribes knew that
EMC *ba / CLMC *p®a ‘crone’ was used to write OJ ba (*mba) in other texts and was read as Go-on ba < OJ Go-on *mba. Thus the three occurrences of
‘crone’ for OJ ba in Shoki (2.5.5, 35.15.5, 116.4.7) and other cases of CLMC ‘muddy’ obstruents corresponding to OJ voiced obstruents may be the result of scribes reverting to their usual habits of writing and pronunciation. One should keep in mind that System-D-based transcription was very much alive and well while Shoki was being compiled. Scribes could hardly not be influenced by the dominant orthographic tradition. In fact, I suspect that retention of traditions rather than phonetics might have been the predominant reason for the use of CLMC ‘muddy’ obstruents to write OJ voiced obstruents. CLMC *p®, *t®, etc. simply did not sound very much like OJ b (*mb), d (*nd), etc. Hence Shoki scribes preferred to use CLMC prenasalized voiced obstruents (*mb, *nd, etc.) for OJ prenasalized voiced obstruents (b = *mb, d = *nd, etc.). They also preferred to use CLMC voiceless obstruents (*p, *t, etc.) for OJ voiceless obstruents ( p, t, etc.). Excluding textual variants and ambiguous phonograms which could have been read with either ‘muddy’ or ‘clear’ (voiceless) initials in CLMC, only 16.7 per cent (348/2,079) of OJ voiceless obstruents ( p, t, s, k) and only 20.6 per cent (89/432) of OJ prenasalized voiced obstruents (b = *mb, d = *nd, z = *nz, g = *Yg) in Shoki poetry were written with CLMC ‘muddy’ initials. Still, ‘muddy’ initials were by no means rare in phonograms for OJ voiceless and voiced obstruents, with the exception of OJ g: only 1.7 per cent (3/176) of OJ syllables with initial g- (*Yg) were written with CLMC ‘muddy’ initial phonograms. 36

EARLY TRANSCRIPTIONS OF JAPANESE

The third through fifth features of Shoki transcription which I will mention here are much less obvious by comparison with the first two. Nevertheless, these remaining features also represent clear departures from earlier transcriptive practices. The third feature of Shoki transcription is the near-total absence of phonograms reflecting unpalatalized pre-MC initials. Earlier systems of transcription employed phonograms with pre-MC *ky- and *t(h)y- (Table 2.4). However, pre-MC *ky- and *t(h)y- later became affricates which were no longer appropriate for the transcription of OJ stops. Thus phonograms once read with pre-MC *ky- and *t(h)y- but read in CLMC with affricates never appear for OJ stop-initial syllables in Shoki poetry. Apparent exceptions are either sinographs with both stop and affricate initial readings or probable scribal errors (Chapter 7). The fourth feature of Shoki transcription is the use of phonograms of the
‘genitive’ rhyme solely for OJ syllables ending either in -i (*i) or -iy (*t). In System B-, C-, and D-based orthography,
‘genitive’ rhyme phonograms were used to write OJ syllables ending in -o (*R). This usage is absent from Shoki owing to the following sound changes between LOC, EMC, and CLMC in the
‘genitive’ rhyme: (2.18) LOC *R > EMC *t > CLMC *i (*[òi]? Cf. Sino-Korean -uy [ty] for this rhyme). LOC *-R was a good match for OJ o (*R), but EMC *-t was a much poorer match, and LMC *-i (*[òi]?) even poorer still. The fifth feature of Shoki transcription is somewhat troublesome and is not found throughout Shoki. Takayama Michiaki (1981, 1982, 1983, 1984) has demonstrated statistical correlations between the Middle Japanese accent recorded in the Myôgishô (1081 CE) and the Middle Chinese tones of the phonograms of poems in the parts of Shoki which Mori (1991: 138) calls the ‘alpha’ section. The Chinese Level Tone corresponds to Japanese low pitch, and the Chinese ‘oblique’ tones (Rising, Departing, and Entering) correspond to Japanese high pitch: e.g., this phrase from Shoki 109 (Mori 1991: 137):22 (2.19) LMC LMC tone OJ Myôgishô pitch ‘overgrown field’





*yia f®uh puat la Rising Departing Entering Level ya bu pa ra high high high low (NS 109.3)

From this evidence it would appear that OJ accent was not much different from that of Middle Japanese accent (Mori 1991: 138). John Bentley (1998a: 88) adds that “it is clear that the literati in late Asuka [552–645 CE] tried to 37

OLD JAPANESE

use the Tang poetic style of level and oblique tones within Japanese poetry”; an awareness of Chinese tones permeated not only the transcription but also the composition of poems. However, as Mori (1991: 138) notes, “the unsolved problems [with accentual transcription in Shoki ] are many.” Even within poems written with apparent accentual notation, there are mismatches between phonogram tones and the expected pitches of OJ syllables. For example, in Shoki 110, OJ syllables expected to have high pitch are represented by Level Tone (not oblique tone) phonograms: e.g.,
CLMC *quRLevel ‘crow’ for OJ woHigh (NS 110.1.1),
CLMC *nRYLevel ‘can’ for OJ noHigh (NS 110.1.5), and
CLMC *süRLevel ‘beard’ for OJ suHigh (NS 110.5.7). Mori claims that it would be impossible to write OJ woHigh, noHigh, and suHigh with oblique tone phonograms because there were no phonograms homophonous with
CLMC *quRLevel ‘crow,’
CLMC *nRYLevel ‘can,’ and
CLMC *süRLevel but with oblique (non-Level tones). This is not quite true; CLMC did have syllables such as
CLMC *quRqRising ‘dike’ and
CLMC *quRhDeparting ‘hate’ which apart from their oblique tones were homophonous with
CLMC *quRLevel ‘crow’ and could have been used to write OJ woHigh. In fact
CLMC *quRqRising ‘dike’ does appear 25 times in Shoki for OJ wo (NS 7.11.6, etc.), though I have not yet checked to see whether those 25 instances of OJ wo should be high or low in pitch according to Myôgishô. Nevertheless, it is true that
CLMC *nRYLevel ‘can’ and
CLMC *süRLevel lacked oblique tone near-homophones and that certain OJ syllable-pitch combinations could not be written without some degree of compromise (using a phonogram with the ‘wrong’ tone but the right segments or vice versa). Although I estimate that CLMC may have had three thousand different syllables (EMC as recorded in the dictionary Qieyun had approximately four thousand different syllables, many of which merged in LMC), it did not have every combination of tones and segments imaginable and had many distributional gaps. Mori (1991: 144) concludes that the scribe of the poems of the alpha section of Shoki did try to record OJ accent but in the end was more intent on accurately recording OJ segments rather than OJ accent. Martin (1987: 175) suggests some other reasons for the imperfect correlation between Chinese tones and OJ accent: One answer might be corruption of the orthography as the text got copied by scribes who did not realize why the particular characters were originally chosen and substituted characters they found more congenial. [Recall that no original Shoki manuscripts have survived.] Perhaps the original writers chose to mark the accents only when it struck their fancy, often preferring an overwhelming common phonogram even when it was counter-indicated by the tone [e.g., the aforementioned use of
CLMC *nRYLevel ‘can’ for OJ noHigh in 38

EARLY TRANSCRIPTIONS OF JAPANESE

NS 110.1.5;
‘can’ is used to write 95.6 per cent of all instances of no (260/272) in Shoki poetry]; scribal carelessness may have been a factor. There is also the possibility that the scribes, especially if they were non-Japanese [as Mori 1991 claims for his ‘alpha’ section of Shoki], may have miswritten some of the accents because their command of the spoken language (or of Chinese) was flawed. Martin (1987: 348–349) also brings up the problem of interpreting phonograms with both Level and oblique tone readings in Chinese. I will conclude my discussion of this issue here because I do not intend to reconstruct OJ accent in this book. However, I will note that no other transcriptions of Japanese prior to the phonogram glosses in Konkômyôsaishôô-kyô ongi (1079) are known to record Japanese pitch accent through phonograms.23 No one to my knowledge has actually tried to search for phonogram tone–pitch accent correlations in pre-1079 texts other than Shoki, so it is at least theoretically possible that this characteristic is not unique to Shoki. Still, the small number of different phonograms per OJ syllable in nonShoki texts makes it unlikely that OJ accent was recorded. In order to record OJ accent, one would need Level and oblique tone phonograms for most if not all OJ syllables, and this is the case in Shoki, where nearly all OJ syllables are written with a number of different phonograms of Level and oblique tones. But in the poetry of Kojiki, for example, OJ a is written only with a Level Tone phonogram
EMC *qa ‘hill’ (KJK 2.8.1, etc.) and OJ ya is written only with the Departing Tone (i.e., non-Level or oblique) phonogram
EMC *yiah ‘night’ (KJK 1.1.1, etc.). One could not write OJ a high (requiring an oblique tone phonogram) or OJ yalow (requiring a Level Tone phonogram) using the Kojiki orthography. Furthermore, these are not isolated cases; there are other OJ syllables written only with Level or oblique tone phonograms in Kojiki. Even if this book did try to take the accentual notation of Shoki into account, I probably could not use Kojiki as a separate ‘witness’ for OJ accent. Despite the radical differences between the System-E (CLMC)-based transcription of Shoki and other varieties of transcription, a certain degree of continuity still remains. Sinographs whose CLMC readings are similar or identical to their older EMC readings continued to be used in Shoki for the same OJ syllables.
EMC *qa / CLMC *qa ‘hill,’ the sole phonogram for OJ a in KJK, still dominates in Shoki, representing 94.3 per cent (117/124) of the instances of OJ a. Shoki scribes could have broken entirely from tradition and chosen entirely new phonograms for each OJ syllable, but they did not. Like their predecessors, the scribes of Shoki kept older practices which did not contradict their system of sinograph pronunciation (in their case, CLMC) and rejected the rest. Ultimately it appears that the Shoki orthography itself was rejected by later scribes. Although one can find isolated examples of System-E-based 39

OLD JAPANESE

orthography in the Man’yôshû (see the discussion of
CLMC *ndiayh ‘mud’ for OJ de on p. 24), this orthography is otherwise unique to Shoki. This is readily apparent from a glance at the man’yôgana usage charts of Omodaka (1967: 891–903) or Igarashi (1969: 160–163), which show many instances of phonograms found only in Shoki, particularly for OJ syllables with voiced obstruent initials. Not even the Shoku nihongi, the continuation of Shoki, used System-E-based transcription: it reverted back to System-Dbased transcriptive usage in its eight poems and elsewhere. It is unknown why System-E-based transcription was never used after Shoki. Bentley (p.c.) speculates that the Japanese court of the Emperors Tenji (r. 662–671 CE), Tenmu (r. 673–686 CE), and Jitô (r. 686–697 CE) “wanted to import the new-spanking culture of Tang, and the language came with it, but this was rejected after 730 (after the Yôrô edicts).” However, this does not explain why no other transcriptions from the same period (662–730 CE) have orthographies similar to that of Shoki. To complicate matters further, although System-E-based transcription went out of use after Shoki, the System-E-based Kan-on stratum of Sino-Japanese was a success. To this day, Kan-on readings of sinographs are generally more commonly used in Japanese than Go-on readings.

Summary Japan was most likely an illiterate land in the third century CE, despite fragmentary exposure to sinography. The earliest variety of Chinese known to the Japanese may have been Late Old Chinese (‘System A’). The earliest phonogram transcriptions of Japanese in the Wei zhi were done on the basis of Late Old Chinese (LOC). Literacy in Japan probably began circa the fifth century when Paekche scholars came to Japan with Chinese texts. Paekche scholars and perhaps others from the continent taught the Japanese ‘System B’ (≅ early SinoPaekche?), a variety of sinograph pronunciation based on LOC; the first transcriptions of Japanese done in Japan reflect System B. During the reign of Empress Suiko, a later wave of Japanese transcriptions were done using a newer system of sinograph pronunciation based on Early Middle Chinese (EMC), ‘System C’ (≅ middle Sino-Paekche?), the source of the earlier strata of Go-on. After the Suiko Period, Old Japanese (OJ) transcriptions in Kojiki, Man’yôshû, and elsewhere were done on the basis of the late EMC-based ‘System D’ (= late Sino-Paekche?), the source of the later strata of Go-on. Finally, Japanese missions to China brought back ‘System E’ (Chang’an Late Middle Chinese), the language of the capital of the Sui and Tang and the source of Kan-on. The OJ poems in Nihon shoki are written in SystemE-based transcription, which represented a radical break from earlier orthographies. 40

EARLY TRANSCRIPTIONS OF JAPANESE

Eventually, phonograms were simplified into kana and the OJ sounds that they represented were forgotten. In the next chapter, I will briefly summarize previous efforts to reconstruct the lost sounds of OJ on the basis of phonograms.

Notes 1 Unless stated otherwise, all Old and Middle Chinese reconstructions up to Chapter 5 follow Starostin (1989) and Pulleyblank (1984, 1991a) respectively with slight typographical modifications given in my list of abbreviations, symbols, and conventions. All Chinese reconstructions from Chapter 6 onward will reflect modifications presented in Chapter 5. 2 In fact, this is how OJ namyi ‘wave’ is spelled in KJK 4.9. 3 I will cite all titles of Chinese works in the pinyin romanization of their Mandarin pronunciation, following the anachronous convention of treating classical East Asian languages as if they were pronounced like their standardized modern relatives. 4 I am grateful to Alexander Vovin (p.c.) for pointing out to me that
 LOC *9(h)a maarq dhRR represents a pre-OJ form of OJ yamato (*yamatR). Most scholars interpret
 as a transcription of pre-OJ yamatai. 5 Bentley (1998a: 73–75) pointed out that the monosyllabic transcriptions of the names of the ‘Five Kings of Wa’ in Song shu are probably phonograms rather than semantograms. 6 It is impossible to designate a date for the creation of the hiragana and katakana syllabaries because they developed gradually from simplified and abbreviated forms of the Chinese characters rather than appearing as full-blown systems at a single point in time. 7 ‘Himiko’ is the conventional modern Sino-Japanese reading of the transcription
 LOC *pye mye hoo. 8 These are the modern pronunciations of the traditional Japanese renderings of the Paekche scholars’ names which are written in phonograms. 9 I am substituting the anachronous Mandarin pronunciation of
 for the unknown original (Sino-Paekche?) pronunciation. This inscription has no date. The estimate of the fifth century CE is from Matsumura (1972: 17). 10 The estimate of the early sixth century CE is from Matsumura (1972: 17). 11 Or, more precisely, EMC as filtered through Korean peninsular intermediaries. 12 See Ôno Tôru (1962: 1001) for a more nearly complete list of Suiko era materials. 13 According to Ôno Tôru (1962: 36, 50), the verb OJ pyitas.i ‘soak.GER,’ written
a
EMC *pyiq da siR in the Jôgûki (Record of the Upper Palace), is the only instance of a non-proper noun written with phonograms in the Suiko materials. 14 Some OJ syllables are barely represented at all even in the far richer materials of the Nara period. For example, the OJ syllable gye appears only once in the poetry of Kojiki and does not appear at all in the poetry of Shoki. 15 Ôno Tôru does not explicitly include the phonograms of the pre-Suiko materials in his tripartite classification of phonograms. 16 According to Pulleyblank (1991a: 15), the EMC rhyme
*-iR ‘branch’ merged into the EMC rhyme
*-i ‘fat’ “by the beginning of Tang” (i.e., the early seventh century CE). 17 If the
‘genitive’ rhyme were *-Rò in LOC, it could have been borrowed into System C (= Sino-Paekche?) as *-Ry; cf. LOC *-aò > System C *-ay. However,

41

OLD JAPANESE

18 19 20

21

22 23

this does not seem to be the case, since
‘genitive’ rhyme phonograms did not transcribe OJ or pre-OJ syllables ending in -ey (*Ry). I will demonstrate that OJ o = OJ o (*R) in Chapter 8. See Ôno Tôru (1962: 715) for a list of the names and dates of these inscriptions. Martin (1987: 78) gives “?700” for the date of Jôgû Shôtoku hô-ô teisetsu. Matsumura (1972: 20) writes that Jôgû Shôtoku hô-ô teisetsu may predate Kojiki and Shoki. According to John Bentley (p.c.), “Hô-ô teisetsu dates from the late eighth century, but clearly most of it was based on some very old records.” The orthography of the work, though generally similar to the System-D-based orthography of Kojiki (710 CE), contains archaic phonogram usages similar to those of the Suiko Period. Omodaka (1967: 890) goes so far as to include Jôgû Shôtoku hô-ô teisetsu among the Suiko Period phonogram materials. The gerund is actually OJ -eba; the -ba part is represented by a following semantogram,
EMC *ciaq ‘one who’ for OJ pa ‘topic marker.’ See
 OJ ar-edo ‘exist-GER’ of (2.12) above for another example of an orthographically split gerund. I have followed the LMC tones given in Pulleyblank (1991a) rather than those of Mori (1991: 137). Kojiki does contain sporadic notations such as
‘Rising [Tone]’ and
‘Departing [Tone]’ which may be attempts to mark OJ pitch using Chinese tone names rather than phonograms with certain Chinese tones; see Martin 1987: 245.

42

PREVIOUS RESEARCH ON PHONOGRAMS

3 PREVIOUS RESEARCH ON PHONOGRAMS

Overview This chapter points out the highlights of previous research on the reconstruction of the OJ language on the basis of phonograms. It does not attempt to provide a comprehensive outline of the vast body of research conducted on this subject. Although I will deal with methodology here, I will focus more on the results of research, reserving most of my criticism of proposed OJ reconstructions for Chapters 7 and 8. I am heavily indebted here to the discussions of the subject in Ôno Susumu (1953), Lange (1973), and Bentley (1997), which contain more details about the history of phonogram studies.

Premodern research Although the phonogram orthographies of early Japan became extinct and the OJ sounds they represented were forgotten by the eighteenth century, Japanese scholars of the time were not entirely ignorant of man’yôgana writing. The man’yôgana texts were still circulating in manuscript form. The earliest manuscripts we have of OJ works date from after the Nara period. Furthermore, these texts were still legible to a great extent owing to (1) knowledge of the relationship between man’yôgana and kana and (2) knowledge of the Sino-Japanese readings of man’yôgana. The kana syllabaries developed during the Heian period were based upon man’yôgana and were not independent inventions. For example, the katakana symbol for Japanese a,
, is based on “a slightly cursivised form” of the phonogram
EMC *qa / LMC *qa ‘hill’ for OJ a (Seeley 1991: 194). The hiragana symbol for the same Japanese syllable is
, based on a cursive form of the phonogram
EMC *qan / LMC *qan ‘peace’ for OJ a. According to John Bentley (p.c.), “this meant that later generations of scribes [after the OJ period] could still GUESS at the readings [of man’yôgana]” with what he estimates to be approximately 90 per cent accuracy. Someone who knew that the katakana
for a derived from the phonogram 43

OLD JAPANESE


EMC *qa / LMC *qa ‘hill’ for OJ a would know that
in Kojiki, Shoki, etc. was supposed to be read as a. It may seem unlikely that anyone would know that the katakana
derived from the phonogram
centuries after the creation of kana because of the lack of graphic similarity between the characters – few modern Japanese know of this relationship – but one should note that premodern kana could be written in a variety of forms, some very close or even identical to their man’yôgana prototypes. One can see the range of variation in premodern kana by consulting the appendix in Habein (1984: 208–217) or any other list of hentaigana (variant kana). System-D- and System-E-based transcription may have died out, but the Go-on and Kan-on strata of Sino-Japanese derived from System D (late Sino-Paekche?) and System E (Chang’an Late Middle Chinese) survived into later periods (and into the present day). As demonstrated on pp. 18 and 32, one can read Kojiki and Shoki poetry with a certain degree of success simply by reading their phonograms in Go-on and Kan-on, respectively. Go-on and Kan-on readings are not always identical to the readings of phonograms in System-D- and System-E-based transcription, but they are often very close. Even if a Japanese scholar did not know the derivation of the katakana symbol
for a, he or she could still read the phonogram
EMC *qa / LMC *qa ‘hill’ for OJ a as Go-on a or Kan-on a. This approach did have its limits: the phonogram
EMC *qan / LMC *qan ‘peace’ for OJ a was Go-on and Kan-on an with a final -n, not an open syllable a. Nevertheless, even armed with Sino-Japanese readings alone, later generations of Japanese could guess at the approximate readings of OJ ongana phonograms. There was probably no point in Japanese history when OJ phonograms were utterly undecipherable. I must emphasize that Japanese scholars capable of reading man’yôgana after the Nara Period did so only approximately. They read eighth-century Japanese as if it were the Japanese of their time period. There was no awareness of a difference between OJ pronunciation and then-current pronunciation. Thus, for example, if scholars (correctly) identified the phonogram
EMC *qa / LMC *qa ‘hill’ as a phonogram for OJ a and if the phoneme /a/ of the Japanese of their time was pronounced as [a], then they would pronounce the phonogram
‘hill’ as [a], even though, at least in theory, OJ a may have been *[æ] or some other vowel. In Chapter 8, I will demonstrate that OJ a indeed was a low central or back vowel *[a]. Premodern Japanese scholars were not aware of most of the OJ phonological distinctions that were no longer made in their language, but they were acutely aware of earlier orthographic distinctions that were later lost. Starting in the seventeenth century, in spite of the prevailing neo-Confucian orthodoxy, some Japanese scholars began to turn their interest toward native Japanese culture as opposed to Chinese culture. This renewed interest in things Japanese led to the investigation of OJ texts by the scholars of the kokugaku (national learning) movement. 44

PREVIOUS RESEARCH ON PHONOGRAMS

The priest Keichû (1640–1701) examined the phonogram orthography of the past and found that it was at odds with then-modern kana usage. For example, in OJ, wo (*wo) and o (*R) were distinct syllables each written with non-interchangeable sets of phonograms. After kana were developed, early kana orthography continued to use non-interchangeable symbols for wo and o such as
wo and
o. However, the syllables wo and o later merged into a single syllable [wo], perhaps as early as 883 (Tsukishima 1977: 19); this [wo] would then become [o] in the eighteenth century (Martin 1987: 79). By Keichû’s time, kana orthography did not consistently distinguish between [wo] < wo and [wo] < o; one could write [wo] < wo with kana originally meant for earlier o (i.e.,
o) and vice versa. The phonograms
EMC *®Q ‘final question particle’ for OJ wo (*wo) and
EMC *qtR ‘locative’ for OJ o (*R) were homophonous to Keichû; he would have pronounced both of them as [wo]. Nevertheless, he noticed that phonogram texts did not use
EMC *®Q ‘final question particle’ and
EMC *qtR ‘locative’ interchangeably. OJ wo (*wo) ‘accusative’ could be written with
EMC *®Q ‘final question particle’ but not with
EMC *qtR ‘locative’ or any other phonograms for OJ o. Conversely, the first syllable o (*R) of OJ omop- (*RmRp-) ‘think’ could be written with
EMC *qtR ‘locative’ but not with
EMC *®Q ‘final question particle’ or any other phonograms for OJ wo. Hence in Waji shôranshô (Corrections to Errors in Japanese Characters; 1693), a dictionary of ‘proper’ kana usage, Keichû advocated that syllables spelled in earlier texts with
EMC *®Q ‘final question particle’ and other phonograms for OJ wo should be written with kana for
wo and that syllables spelled in earlier texts with
EMC *qtR ‘locative’ and other phonograms for OJ o should be written with kana for
o. Keichû would pronounce both the o (< OJ wo) of ‘accusative’ and the o (< OJ o) of ‘think’ as [wo], but he would spell the two [wo]s differently: the first [wo] would be spelled
wo, but the second [wo] would be spelled
o. Keichû’s orthography distinguished not only between earlier wo and o but also between other earlier syllables such as i and wi, which had merged into [i], and ye (< OJ e and ye) and we, which had merged into [ye]. If taken at face value, his orthography implied the existence of 47 syllables in earlier Japanese (Table 3.1); this figure rose to 67 if one took account of the obstruent voicing which was present in both earlier Japanese and Keichû’s Japanese but which was not marked in Waji shôranshô: for instance, OJ irodo (*irRndR) ‘younger sibling with the same mother’ is listed as
 iroto (Okimori 1989: 175). In the process of restoring lost orthographic distinctions to the kana spelling of his time, Keichû discovered that at least some lost orthographic distinctions were correlated with lost phonological distinctions. According to Seeley (1979, 1991: 117), Keichû “had arrived at a clear understanding of the basis of the distinction in [the earlier] pronunciation of the pairs of
i and
wi,
e and
we, and
o and
wo.” Keichû knew that the syllables [i], [ye], 45

OLD JAPANESE

Table 3.1 Syllables of earlier Japanese distinguished in Keichû’s orthography

a i u e o

p/b

m

w

t/d

n

r

s/z

y

k/g

$

x x x x x

x x x x x

x x

x x x x x

x x x x x

x x x x x

x x x x x

x

x x x x x

x x x

x x

Keichû’s Japanese

[i]

i

x x x

[ye]

wi

ye

x

[wo]

we

o

wo

Earlier Japanese

Figure 3.1 Origins of [i], [ye], and [wo] in Keichû’s Japanese

and [wo] of his time had two sources (Figure 3.1). Unfortunately, Keichû had the earlier sound values of the kana for wo and o reversed: he thought that
had represented earlier o and that
had represented earlier wo (Seeley 1991: 117). I have corrected Keichû’s error in Figure 3.1. Keichû was not the first to notice differences between earlier phonogram usage and then-modern kana usage – that distinction may belong to the priest Jôshun, who observed such differences in his 1353 postscript to a manuscript of Man’yôshû (Seeley 1991: 116) – but Keichû went beyond mere observation to prescribe an entire system of kana spelling based on phonogram spellings. Keichû’s system was initially adopted only by kokugaku scholars but by Meiji times spread more widely and ultimately became the norm in Japan until the kana spelling reform of 1946 (Seeley 1991: 124–125). Keichû had only begun to find lost orthographic and phonological distinctions in OJ. The task was far from complete. The great kokugaku scholar Motoori Norinaga (1730–1801) noted in the preface to his Kojiki-den (1764– 1798) that: Even with the same phonetic group, there are different characters employed according to a distinct sound, and many words have set spellings. For example, the syllable is generally represented by the characters
[‘allow’] and
[‘old’]. The word ‘child (
)’ is always written
[‘old’] but never
[‘allow’] . . . For the syllable there are generally two characters used,
[‘rice’]
[‘sell’]. The word ‘woman (
)’ is always written with
[‘sell’] but never with
[‘rice’]. (Ôkubo and Ôno 1976: 9:28–29, tr. in Bentley 1997: 8) 46

PREVIOUS RESEARCH ON PHONOGRAMS

Keichû’s kana spelling system did not take account of these particular orthographic distinctions in Kojiki. In the Japanese of Keichû’s time, the OJ words for ‘eye’ (mey = *mRy) and ‘woman’ (mye = *me) were both pronounced [me]. Keichû would have written both words as
me, though in Kojiki they were spelled two different ways. Although Keichû had intended his kana spelling system to be based upon the distinctions made in the old phonogram texts, he failed to notice those distinctions which had already been lost by the Heian Period when the kana syllabaries were developing. The Heian Period verses Iroha (before 1079?) and Ametsuchi no kotoba (“Words of Heaven and Earth”; first attested in 967) were both attempts to serve as mnemonic inventories of Early Middle Japanese (EMJ) syllables. Like Keichû’s prescribed kana spelling, the two verses distinguished between
wo and
o (EMJ wo and o, which had merged into [wo] by Keichû’s time) but they did not distinguish between two different kinds of ko or me, since OJ kwo (*ko) and OJ ko (*kR) had merged into EMJ ko and OJ mye (*me) and OJ mey (*mRy) had merged into EMJ me. Motoori had gone beyond Keichû by discovering a few orthographic distinctions that had been lost in the Heian Period. Motoori’s student Ishizuka Tatsumaro would go even further. In 1796, Ishizuka wrote Kogen seidokukô (A Treatise on the Clear and Muddy in Ancient Words), which showed that OJ scribes used different phonograms to transcribe syllables with voiceless (‘clear’) and voiced (‘muddy’) initials. For instance, Ishizuka indicated that OJ ka with a voiceless initial was written with the phonograms
 !" but OJ ga with a voiced initial was written with the phonograms
 (Hashimoto 1949: 128; Koyama 1956: 283). When Motoori’s students asked him how to distinguish between syllables with voiced and voiceless initials in OJ phonogram texts, the master recommended that they read Kogen seidokukô (Bentley 1997: 5–6). Ishizuka then outdid himself by writing Kanazukai oku no yamamichi (The Mountain Road into the Secrets of Kana Usage; c. 1795). In this work, he showed that the OJ scribes of Kojiki, Shoki, and Man’yôshû made further orthographic distinctions which had nothing to do with the presence or absence of voicing in initials. He found that OJ syllables which he pronounced as e, ki, gi, ke, ko, go, so, to, do, nu, hi (h- < OJ p-), bi, he (h- < OJ p-), be, mi, Table 3.2 Syllables of Old Japanese distinguished in Kogen seidokukô

a i u e o

p

b

m

w

t

n

d

r

s

z

y

k

g

$

x x x x x

x x x x x

x x x x x

x x

x x x x x

x x x x x

x x x x x

x x x x x

x x x x x

x x x x x

x

x x x x x

x x x x x

x x x

x x

47

x x x

x

OLD JAPANESE

Table 3.3 Orthographic distinctions in Kanazukai oku no yamamichi

a i u e o

p

b

m

w

t

d

n

r

s

z

y

k

g

Ø

x • x • x

x • x • x

x • x • •K

x x

x •K x x •

x x x x •

x x • x x

x x x x •

x x x x •

x x x x x

x

x • x • •

x • x x •

x x x

x x

x • •

x

me, yo, and ro each corresponded to two non-interchangeable sets of phonograms in all three texts. He also found that the OJ syllables which he pronounced as chi (< OJ ti) and mo each corresponded to two noninterchangeable sets of phonograms only in Kojiki (Hashimoto 1949: 132– 133). Each of the above syllables was written only with one or the other set of phonograms in any given OJ word. For example, Ishizuka points out that the syllable ko in the OJ words ko ‘child,’ ko ‘small,’ otoko ‘male,’ hiko ‘prince’ (all cited here in their modernized pronunciations) was written with phonograms from a set including
 !"#$% (all pronounced with *-Q in EMC) but not with phonograms from a set including
 !"#$% (all pronounced with *-tR or *-t in EMC) (Hashimoto 1949: 129). Table 3.3 shows how Ishizuka built upon the foundation established in his Kogen seidokukô by indicating which syllables in Table 3.2 corresponded to two non-interchangeable sets of phonograms with the symbol •. The symbol •K indicates that the two non-interchangeable sets of phonograms appear only in Kojiki. Note that Ishizuka found no pairs of non-interchangeable sets of phonograms for -a syllables and found only one such pair of sets for an -u syllable (nu); this gap and near-gap will later prove to be important for the reconstruction of OJ vocalism. Although Bentley (1997: 9) writes that “Ishizuka . . . posited two different phonetic values” for each pair of non-interchangeable sets of phonograms corresponding to a single then-modern Japanese syllable, it is uncertain whether Ishizuka believed that OJ orthographic distinctions reflected OJ phonetic distinctions. On the one hand, Ishizuka claimed that in ancient times the distinctions were purely orthographic: the pairs of phonogram sets (such as
 !"#$% and
 !"#$% for ko) were homophonous and the choice of one set over another was simply set for each word: “In ancient times, the kana [phonograms] used were strictly set depending on the word even if their sounds were the same” (quoted in Hashimoto 1949: 141). On the other hand, he is quoted in a later work, Kogen betsuon shô (A Commentary on Different Sounds for Old Words; 1850) as saying that sounds which are identical today were not identical in old words and the old orthography strictly maintained these earlier distinctions. In either case, 48

PREVIOUS RESEARCH ON PHONOGRAMS

Ishizuka did not try to reconstruct the phonetics of OJ (Hashimoto 1949: 141–142). That task would not be attempted until over a century later. Until then, Ishizuka’s work was generally ignored through the nineteenth century, apart from attempts at minor revisions by Yagi Yoshiho (1800– 1854) and Kusakado Nobutaka (1818–1869). In Kanabukuro (A Bag of Kana; c. 1849),Yagi denied that chi (< OJ ti) or mo were written with two noninterchangeable sets of phonograms. In the aforementioned Kogen betsuon shô, Kusakado also omitted chi (< OJ ti) from Ishizuka’s list of syllables written with two non-interchangeable sets of phonograms, though it is unclear what he thought of mo; extant manuscripts of his work disagree on the subject (Bentley 1997: 10–11).

Modern research In 1917, the linguist Hashimoto Shinkichi (1882–1945) published an article which rediscovered Ishizuka’s Kanazukai oku no yamamichi and reinterpreted Ishizuka’s OJ orthographic distinctions as proof for OJ phonetic distinctions. From Hashimoto’s point of view, each non-interchangeable set of phonograms represented a phonetically distinct syllable in OJ. Therefore the aforementioned Kojiki phonograms
EMC *mεyq ‘rice’ and
EMC *mεryh ‘sell’ were not arbitrarily fixed spellings for one OJ syllable me; instead,
EMC *mεyq ‘rice’ represented one OJ syllable (mey in Yale OJ romanization) and
EMC *mεryh ‘sell’ represented another OJ syllable (mye in Yale OJ romanization). Hashimoto’s theory provided a more plausible motivation than arbitrary tradition for OJ phonogram spelling conventions: the ancients wrote OJ mey ‘eye’ and OJ mye ‘woman’ with different phonograms because the two words were not homophonous in OJ. In my reconstruction of OJ, ‘eye’ is *mRy and ‘woman’ is *me. Hashimoto’s 1917 article proposed an */e/ : */ye/ distinction in OJ on the basis of phonogram evidence. In a 1927 lecture at Tokyo Imperial University, he defended this proposal on the basis of Chinese dialectal readings and Sino-Korean readings of the two non-interchangeable sets of phonograms corresponding to modern Japanese e:
Md ài, SK oy ‘love,’ read without initial y- in Chinese and Sino-Korean, was a phonogram for OJ */e/, but
Md yán, SK yen ‘extend’ and other sinographs read with initial y- in Chinese and Sino-Korean were phonograms for OJ */ye/ (lecture notes by Iwabuchi Etsutarô published in Ôno 1953: 127–128). These two syllables, OJ */e/ and OJ */ye/, later merged; in Modern Standard Japanese, e is the reflex of both syllables. Hashimoto’s reconstruction of initial */y/ for the OJ syllable written with the set of phonograms including
Md yán, SK yen ‘extend’ has been unanimously accepted since. This was but the beginning of Hashimoto’s work on the subject. In 1931, Hashimoto discovered that members of pairs of OJ syllables which had merged 49

OLD JAPANESE

in later Japanese were associated with certain OJ verb classes and forms. Mostly on the basis of these associations, he called members of such pairs either ‘type A’ (kô-rui) or ‘type B’ (otsu-rui), with the exceptions of OJ */e/ and */ye/. For example, Motoori would have thought of
EMC *mεryh ‘sell’ and
EMC *mεyq ‘rice’ as phonograms for a single syllable me, but Hashimoto would classify the former as a type A me (= me A) phonogram and the latter as a type B me (= me B) phonogram following this reasoning:
EMC *mεryh ‘sell’ was a phonogram used in yodan class imperatives (meireikei); therefore it represented me A.
EMC *mεyq ‘rice’ was a phonogram used in yodan class conditionals (izenkei); therefore it represented meB. However, Hashimoto ran into difficulties when dealing with OJ syllables which (1) had merged in later Japanese but (2) which had no associations with OJ verb classes and forms. His earlier morphological criteria for classification did not work with these syllables. Which of those syllables were type A and which ones were type B? He dealt with this problem using several vaguely defined methods involving alternative readings of phonograms and the analysis of the rhymes of the phonograms’ readings (1949: 184). One need not accept any of Hashimoto’s arguments for the classification of a given syllable as type A or B since independent evidence that I will present in Chapter 8 shows that type B syllables had non-palatal, non-labial vowels unlike type A syllables. Hashimoto did not give a name to the remaining majority of OJ syllables which did not belong to either type A or type B, but in this book I will refer to these OJ syllables as ‘type C’ (hei-rui). When the type (A, B, or C) of an OJ syllable is not indicated by superscript letters, I indicate it in this book by using the Yale romanization of Martin (1987) with p for Martin’s f. See Table 3.4 for the correlations between OJ syllable types and Yale romanization. Over time, Hashimoto made various refinements to Ishizuka’s observations on phonogram orthography. He came to reject Ishizuka’s claims that the OJ

Table 3.4 Possible syllables in the OJ of Kojiki: the consensus view Yale a yi iy i u ye ey e wo o o

= = = = = = = = = = =

a(C) iA iB iC u(C) eA eB eC oA oB oC

p

b

m

w

t

n

d

r

s

z

y

k

g

$

x x x

x x x

x x x

x

x

x

x

x

x

x

x

x x x

x x x

x

x x x x

x x x

x x x

x x

x x

x x

x x

x x

x x

x

x x x

x x x

x x x

x x x

x x x

x x x

x x x

x x x

x x x

x x

x x

x x x x

x

x

x x x x

50

PREVIOUS RESEARCH ON PHONOGRAMS

syllables corresponding to later Japanese chi (< OJ ti) and nu had ever been written with two non-interchangeable sets of phonograms (Hashimoto 1949: 163). Thus he no longer thought that OJ had distinctions between ti A and ti B or nu A and nuB: OJ had only a single syllable ti and a single syllable nu. But he did discover further two-way orthographic distinctions for OJ syllables corresponding to later Japanese ge, zo, and no. Hence OJ also had distinctions between ge A and ge B, zo A and zoB, and no A and noB (Hashimoto 1949: 163; Lange 1973: 25). The majority of scholars accepted Hashimoto’s conclusions concerning the syllabic distinctions of OJ. As a result, a consensus view emerged that is still widely accepted today: the OJ of Kojiki had 88 different syllables, whereas the OJ of Shoki and Man’yôshû had 87 different syllables, owing to the lack of the mo A : moB distinction made only in Kojiki. I will deal with alternative claims for the number of syllables in OJ later in this section. Table 3.4 shows the 88 consonant-‘vowel’ combinations possible in Kojiki in modified Yale romanization, along with superscript equivalents for Yale vocalic notation. Since there are no syllables ending in -a A, -aB, -u A, or -u B, the type C notation for OJ -a and -u is redundant. Hereafter I will regard only syllables ending in -i C, -e C, or -oC as being type C. Although Table 3.4 may seem to indicate that OJ had eleven vowels, finals, or rhymes, scholars usually speak of the ‘eight vowels’ of OJ (a, yi, iy, u, ye, ey, wo, o), assuming that type C i, e, and o are equivalent to type A yi, ye, and wo. However, this assumption may not necessarily be correct, as I will demonstrate in Chapter 8. Hashimoto then tried to posit phonetic values for the type A and type B syllables largely on the basis of their phonograms’ pronunciations in modern Chinese dialects and Sino-Korean with reference to the medieval Chinese rhyme table Yunjing (see p. 96). In his early attempts, he assumed that (1) the type A/B distinction was vocalic in nature and (2) all type A syllables were distinguished from all type B syllables in the same way. According to Iwabuchi’s notes of his 1927 lectures quoted in Ôno (1953: 130), he believed that type A syllables had single vowels while type B syllables had diphthongs which may have contained a “vague vowel.” On the basis of SinoKorean pronunciations of OJ phonograms, he gave sample reconstructions of OJ kyi = kiA as *ki (cf. SK ki for
‘skill,’ a phonogram for OJ kyi) and OJ kiy = ki B as *kïi (cf. premodern SK kuy [kty] for
‘record,’ a phonogram for OJ kiy). Although he did not propose a full-scale reconstruction of OJ in 1927, Hashimoto was the first to attempt to describe the type A/B distinction in phonetic terms. By 1933, Hashimoto took account of modern Japanese dialects, SinoVietnamese, and Chinese transcriptions of Sanskrit as well as the other varieties of evidence already mentioned above. In doing so, he abandoned his original hypothesis of a single unifying distinction between all type A and all type B syllables (Morimoto 1933, quoted in Ôno 1953: 131). He 51

OLD JAPANESE

still reconstructed single vowels for type A yi and ye and diphthongs for type B iy and ey, but reconstructed single vowels for both type A wo and type B o (3.1). (3.1) OJ syllable type A B Yale OJ romanization yi iy Hashimoto 1933 *i *ïi

A ye *e

B A B ey wo o *Qi *o *ö

Nevertheless, the hypothesis of a single distinction (and in particular, a single vocalic distinction) between OJ type A and type B syllables remained popular among other scholars. Hashimoto and many of those who followed him concentrated their reconstructive efforts mostly on OJ type A and type B syllables rather than on OJ phonology as a whole. J. L. Pierson (1929) was perhaps the first to attempt to reconstruct what he believed to be all the syllables of OJ on the basis of Karlgren’s reconstruction of Middle Chinese and Ilopha (1492), a Korean textbook of Japanese. Table 3.5 is my phonemic analysis of the 70 syllables of Pierson’s OJ reconstruction. Pierson’s reconstruction was solely phonetic. Wherever his phonetic values cannot be inferred from my phonemic analysis (e.g., *[?i] for /ti/), I supply them in brackets without asterisks. Two phonetic values for a syllable (such as [φa]~[Fa] for /va/) represent alternative pronunciations of that syllable. Apart from *[φa] and *[Fa], whose environments are given in parentheses, all alternative pronunciations are in free variation. Although Pierson was familiar with Hashimoto’s work (Lange 1973: 39), he ignored the kô–otsu distinctions. Pierson’s reconstruction has too many Table 3.5 Pierson’s (1929) reconstruction of OJ v

b

m

w

t

n

d

r

s

z

y

k

g

Ø

a

x x x x [φa] (initially) [Fa] (intervocalically)

x

x

x

x

x x x [sa]~ [za]~ [tsa] [dza]

x

x

x

i

x

x

x

x

x [?i]

x

x [ëi]

x

x x x [5i]~ [¢i]~ [si] [zi]

x

x

x

u

x

x

x

x

x

x

x

x

x x x [su]~ [zu]~ [tsu] [dzu]

x

x x [ku]~ [kyu]

e x [ye]

x

x

x

x

x

x

x

x

x

x

x

x

x

o

x

x

x

x

x

x

x

x x [so]~ [tso]

x

x

x

x

x

52

PREVIOUS RESEARCH ON PHONOGRAMS

Table 3.6 Yoshitake’s (1934) reconstruction of OJ

a i = iA L = iB i~I = iC T e ≅ eA ε ≅ eB o ≅ oA Q ≅ oB ò ≅ oC

φ

b

m

ò

t

d

n

r

s

z

y

k

g

Ø

x

x x x

x x x

x x

x

x

x

x

x

x

x

x

x x

x

x x x

x

x x x

x x x

x x x

x x x

x x x

x x x

x x

x x

x x

x x

x x

x x x x x

x x x x x

x x x x x

x x x x

x x x

x x

x

x x x x

x x x x

x

Note [ò] represents a vowel halfway between *[o] and *[Q].

problems to be discussed here and has long since been forgotten, but it remains notable for two reasons: (1) he was the first to reconstruct all OJ syllables and (2) he was the first to use Chinese reconstructions in order to interpret the sound values of OJ phonograms. Yoshitake Saburô (1934) went far beyond Pierson by incorporating Hashimoto’s type A/B distinctions into his reconstruction of 86 syllables for OJ (Table 3.6). Yoshitake’s work, like Pierson’s, used Karlgren’s Chinese reconstructions, which are now long outdated. However, Yoshitake’s reconstruction is worthy of note not only because it was the first to give phonetic values for nearly all of the 87 to 88 OJ syllables,1 but also because it dealt with the problem of OJ C-type syllables. Later scholars would ignore this problem almost entirely and automatically treat OJ C-type syllables as if they were A-type syllables. Many other attempts at reconstructing OJ on the basis of phonogram evidence have been made since Hashimoto, Pierson, and Yoshitake. I will only summarize the highlights of a few of these attempts here; select end results from all attempts known to me will be listed in Tables 3.11 and 3.12 at the end of this chapter. Although Hashimoto may have been the first to examine Sino-Vietnamese (SV) as a source of evidence for phonogram interpretation, the first to present concrete examples of how SV could help in the reconstruction of OJ was the Koreanologist Kôno (1939). Kôno not only verified Hashimoto’s observation that Sino-Korean (SK) readings of the phonograms for OJ B-type kiy and giy contained a non-palatal element (Korean u [t]) lacking in the SK readings of the phonograms for OJ A-type kyi and gyi but also noted that the SV readings for OJ B-type piy, biy, and miy had unpalatalized labial initials (b [(], ph [f], m) whereas SV readings for OJ A-type pyi, byi, and myi had dental initials (t, d [z]) originating from earlier palatalized labials (t < *py, d < *my; p. 121). 53

OLD JAPANESE

Table 3.7 SK and SV readings of phonograms for OJ A/B syllables in -i Type

A

B

A

B

A

B

A

B

A

B

OJ

kyi

kiy

gyi

giy

pyi

piy

byi

biy

myi

miy

SK

ki

kuy [kty]

ki

uy [ty]

pi

pi

pi

pi

mi

mi

SV

ki

ki

ki

nghi [Yi]

ti

bi, phi [(i fi]

ti

bi, phi [(i fi]

mi

di [zi]

Note This is not an exhaustive table of correspondence patterns.

On the basis of these findings, Kôno reconstructed OJ A-type yi as *i and OJ B-type iy as a less palatal *ï. Unfortunately, this ‘Sinoxenic’ (p. 98) line of investigation was not carried out much further by Kôno or other scholars, and SK and SV evidence for OJ phonology has been ignored ever since. In Chapters 7 and 8 I will demonstrate that Hashimoto and Kôno were on the right track but had only begun to discover the significance of SK and SV for the reconstruction of OJ. Beginning from a 1948 lecture entitled “Problems in Chinese Phonology” (Ôno 1953: 153), the sinologist Tôdô Akiyasu took an interest in the reconstruction of the vowels of the OJ A/B-type syllables from a Chinese-centered perspective. Although Tôdô was aware of SK and SV (1980: 176–183), he approached the problem in terms of his reconstructions of Old Chinese (OC) and Middle Chinese (MC) rhyme categories. Unfortunately, Tôdô’s work was marred by circularity: his MC reconstruction was partly based on his OJ reconstruction and vice versa. Miller (1967: 179) writes of Tôdô’s (1957) attempt at OJ reconstruction: Assuming the “fact” that Old Japanese had an eight-vowel system and the “fact” that these eight vowels of Old Japanese are sufficiently well known to be described in terms of acoustic phonetics, he then takes these as his point of departure, following the phonetic values now commonly held by most modern Japanese scholars for the eight vowels. With this data he enters the Chinese rhyme-tables; and when he again emerges many pages later he is able to say, in effect, that the Middle Chinese materials show that these vowels not only existed in Japanese but were of the following phonetic shapes, etc., etc. – which unfortunately is the point from which his discussion originally began. Decades later, Pulleyblank (1984: 156) pointed out that “Since the reconstructed Chinese sounds are in turn the principal source available for reconstructing the older stages of Japanese, there is great danger of circularity on 54

PREVIOUS RESEARCH ON PHONOGRAMS

Front High Mid

Central

*i

Back *u

*ï *e

*ë

*ö

Low

*o

*a

Figure 3.2 The ‘established’ interpretation of OJ vowels (after Ôno 1975: 23)

both sides of the argument.” Yet Pulleyblank himself fell into that very trap by reconstructing MC partly on the basis of his OJ reconstruction and vice versa. Although I continue to believe that Pulleyblank’s MC reconstruction is the best reconstruction in print for reasons given in Chapter 5, the circularity of his work prevents me from using it as primary evidence for OJ reconstruction. Hence I will relegate it to a secondary role in my reconstruction of OJ phonetics. Many reconstructions of OJ, including Yoshitake’s above, treated the A/B-type distinctions in terms of distinct single vowels. Kiyose (1991: 204) would regard these reconstructions as varieties of “pro-eight vowel theories,” since their six A/B-type vowels (i A, i B, e A, e B, o A, and o B) and the remaining vowels a and u added up to a total of eight vowels. The earliest eight-vowel reconstruction was by Nagata (1932; Table 3.11). The most persistent variety of the eight-vowel theories regards B-type vowels and OJ a as central and all other vowels as peripheral (i.e., front or back). The centrality of the B-type vowels is marked with an umlaut. Hence the symbol *ö represents a centralized *o, not a mid front rounded vowel. Kiyose (1991: 200) calls this view (Figure 3.2) “the ‘established’ intepretation of the eight vowels.” This view dates back at least to Kindaichi (1938) and appears in many other works such as Miller (1967). The “established interpretation” may be popular, but it is also “highly improbable . . . for a natural language” owing to its “four-way distinction of mid-vowels” (Pulleyblank 1984: 97). Not all scholars believed that it was necessary to posit eight vowels for OJ. Miyake Takeo (1933) was the first to suggest that at least one set of A/B-type distinctions could be attributed to the presence or absence of a glide: he reconstructed A-type wo as *wo with *w and B-type o as *o without *w. Yasuda Kiyomon (1934) suggested that the presence or absence of *-u- could account for all A/B-type distinctions. Kikusawa Sueo (1935) came to a similar conclusion, except that he reconstructed the presence or absence of a medial glide *-w- instead of a vowel *-u-. In 1938, Arisaka Hideyo may have been the first to suggest that a medial glide *-y- instead of *-w- accounted for some of the A/B-type distinctions in OJ (Wenck 1954–57: 4.15, Lange 1973: 69). Hattori Shirô (1958, 1976) reconstructed A-type yi and ye with *y- and B-type iy and ey without *y-, 55

OLD JAPANESE

Table 3.8 Japanese ‘anti-eight-vowel’ OJ A/B ‘vowel’ reconstructions Type

A

B

A

B

A

B

OJ ‘vowel’ Miyake (1933) Yasuda (1934) Kikusawa (1935) Hattori (1958, 1976)

yi – *i *i *yi

iy – *ui *wi *i

ye – *e *e *ye

ey – *ue *we *e

wo *wo *uo *wo *o

o *o *o *o *ö

Note Dashes indicate that a scholar did not reconstruct a phonetic value for an OJ ‘vowel.’

but viewed the distinction between A-type wo and B-type o as a contrast between mid back rounded o and a centralized ö. Since all these reconstructions posit fewer than eight phonemic vowels in OJ, they represent what Kiyose (1991: 202) calls “anti-eight vowel theories” (Table 3.8). Although the anti-eight-vowel school never became dominant in Japan, Lange’s (1968) doctoral dissertation, published as Lange (1973), brought the concept to the attention of Anglophone scholars. Unlike most previous scholars, Lange aimed at a reconstruction of the entire segmental phonemic system of OJ rather than just the A/B-type distinction(s). Furthermore, Lange’s reconstruction was based on a careful examination of the phonogramheavy sections of Man’yôshû and not simply on lists of phonograms taken from various OJ texts. Only Ôno’s (1953) monograph on the phonogram orthography of Nihon shoki came close in terms of scope, though Ôno had only attempted to reconstruct the phonetic values of A-type wo and B-type o. Lange accounted for the A/B-type distinctions with not just one glide but two: *-y- and *-w-. Lange’s five-vowels-plus-glides reconstruction of the ‘eight vowels’ was a compromise between the reconstructions of Miyake, Yasuda, and Kikusawa on the one hand and the reconstruction of Hattori on the other. Although later American scholars did not accept all the details of Lange’s reconstruction, they were strongly influenced by it. Unger (1971) proposed a possible final glide *-w for A-type wo which was adopted by Ramsey and Unger (1972). Unger (1993 [1977]) proposed a slight variation on Lange’s solution which distributed the medial glides *-yand *-w- somewhat differently. Whitman (1985) reconstructed a final glide *-y for B-type iy and ey in addition to a medial glide *-y- for A-type ye. These latter reconstructions are notable not only for their treatment of OJ ‘vowels’ but also for their treatment of OJ voiced obstruents. Although Asayama Nobuya (1943) and Hamada Atsushi (1952) had proposed that OJ voiced obstruents were phonetically prenasalized, most scholars were too occupied with reconstructing A/B-type distinctions to investigate OJ voiced obstruents. Lange (1973), apparently unaware of Asayama’s and Hamada’s work, reconstructed plain voiced obstruents *b, *d, *z, *g for OJ. However, 56

PREVIOUS RESEARCH ON PHONOGRAMS

Table 3.9 American ‘anti-eight-vowel’ OJ A/B ‘vowel’ reconstructions Type

A

B

A

B

A

B

OJ ‘vowel’ Lange (1973) Ramsey & Unger (1972) Unger (1993 [1977] ) Whitman (1985)

yi *yi *i *i *i

iy *i *wi *wi *uy

ye *ye *ye *ye *ye

ey *e *e *e *ey

wo *wo *wo, *ow *wo *wo

o *o *o *o *o

Note Lange (1973: 124) reconstructs A-type pye, bye, mye as *[pye bye mye] and B-type pey, bey, mey as *[pe be me] but reconstructs A-type kye, gye as *[ke ge] (not *[kye gye] ) and B-type key, gey as *[kye gye] (not *[ke ge] ).

Ramsey and Unger (1972) and Whitman (1985) both accept Asayama’s and Hamada’s arguments and reconstruct prenasalized voiced obstruents (*mb, *nd, *nz, *Yg in my notation) in place of Lange’s plain voiced obstruents (*b, *d, *z, *g). This view of OJ voiced obstruents has become widely accepted in the United States but remains unpopular in Japan itself. A textbook of Japanese language history (Okimori 1989) does not mention anything about the prenasalization of OJ voiced obstruents. Even Mori’s (1991) landmark study of Nihon shoki phonograms reconstructs the plain voiced obstruents *b, *d, *dz (before *a and *R), *z (before *u and *o), and *j (before *i and *e) for OJ. (But Mori 1991 does reconstruct a prenasalized voiced velar obstruent *Yg.) Mori (1991) is notable for a number of other reasons. Just as Lange (1968, 1973) attempted to reconstruct OJ on the basis of the phonograms of Man’yôshû poetry, Mori attempted to reconstruct OJ on the basis of the phonograms of Nihon shoki poetry. Although Mori, like Lange and all others before him, was hampered by an outdated Karlgrenian reconstruction of Middle Chinese, he did use statistically based arguments which have been used in very few other works on the subject. Prior scholars paid little or no attention to how often a phonogram was used or even to where it was used: any phonogram was equally valid, whether it was used a hundred times or just once. Mori, on the other hand, examined which phonograms were most common and where those phonograms were found among the volumes of Shoki. Furthermore, he argued that there were two different writing systems in Shoki instead of a single System-E-based orthography. Phonograms in what he called the “beta section” of Shoki (volumes 1–13, 22, 23, 28, and 29) reflected what Arisaka (1955: 191) called “Japanese sounds” (wa-on); i.e., OJ period Sino-Japanese pronunciations rather than authentic Middle Chinese (MC). On the other hand, phonograms in the rest of Shoki – his “alpha section” – directly reflected MC. Mori posited three characteristics that distinguished alpha section phonograms from their beta counterparts: 57

OLD JAPANESE

(1) MC fricative (e.g., *h, *®) initial phonograms for OJ k-initial syllables are absent from the alpha section, though they are common in the beta section (Mori 1991: 122). (2) MC aspirate initial phonograms are generally avoided in the alpha section, though they are frequent in the beta section (Mori 1991: 18–21). (3) OJ -a syllables in the alpha section are usually written with MC phonograms of the MC
‘song’ rhyme, reconstructed by Mori with a low back vowel *-q, rather than phonograms of the MC
‘hemp’ rhyme, reconstructed by Mori with a low front vowel *-a. Mori explains all three characteristics by claiming that the alpha section was written by Chinese who “directly transcribed” Japanese according to “the northern Chinese dialect of the time” (1991: 163). These Chinese would have no reason to write OJ k with their MC *h or *®. They avoided using phonograms with aspirated initials because OJ probably did not have aspirated obstruents. Lastly, they avoided using phonograms of the MC
‘hemp’ rhyme to write OJ -a syllables because OJ a was closer to the low back *q of their
‘song’ rhyme rather than to the low front *-a of their
‘hemp’ rhyme. I am hesitant to accept Mori’s claims completely. Although it is true that his alpha section completely lacks fricative-initial phonograms for OJ k-syllables, Mori’s other two characterstics are hardly absolute. In fact, Mori’s own chart of phonograms for OJ -a syllables shows that phonograms of the MC
‘hemp’ rhyme are in the minority throughout Shoki. The pattern of distribution is nearly identical in either section, with MC
‘hemp’ rhyme phonograms appearing for OJ ma and ya in both sections and for five instances of sa and only one instance of ka in the beta section (Mori 1991: 55). If the Chinese scribes of the alpha section were as phonetically astute as Mori supposes, then why did they have difficulty with consistently transcribing OJ voiced obstruents (b, d, z, g)? Both the alpha and beta sections use a mix of MC ‘muddy’ obstruents (*p®, *t®, etc. in Pulleyblank’s LMC reconstruction) and nasals (Chang’an LMC [CLMC] prenasalized voiced obstruents) to write OJ voiced obstruents. Even if we look at Mori’s own statistics (1991: 273–276), the majority of OJ voiced obstruents in the alpha section are transcribed with CLMC prenasalized obstruents, but only 21.8 per cent are written with ‘muddy’ obstruents. This figure of 21.8 per cent rises to 33.3 per cent (60/180) if one excludes phonograms for OJ g, which overwhelmingly (97.2 per cent; 106/109) begin with the CLMC prenasalized obstruent initial *Yg-. The alpha section scribes could not even transcribe OJ s consistently. Mori (1991: 126) reconstructs three environmentally conditioned allophones of OJ s, *ts, *s, and *c [t6 ]: (3.2) OJ Mori

su *su

so *so

sa so *tsq *tsR 58

si se *ci *ce

PREVIOUS RESEARCH ON PHONOGRAMS

Table 3.10 Transcription of OJ voiced obstruents in the alpha section, % (number) OJ initial

CLMC prenasalized obstruents

CLMC ‘muddy’ obstruents

CLMC ‘clear’ obstruents

b d z g Total

65.2 55.1 57.1 97.2 75.4

31.8 (21/66) 34.2 (27/79) 34.3 (12/35) 2.7 (3/109) 21.8 (63/289)

3.0 3.8 8.6 0 . 2.8

(43/66) (49/79) (20/35) (106/109) (218/289)

(2/66) (3/79) (3/35) (0/109) (8/289)

Note ‘Clear’ obstruents include aspirates.

According to Mori’s MC reconstruction and most other MC reconstructions (with the exception of Pulleyblank’s LMC),2 MC speakers had nine distinct s-like phonemes: *s, *ts, *tsh, *Z, *cr [âZ], *chr [âZh], *\, *c [c\], and *ch [c\h]. Assuming that Mori’s reconstructions of OJ and MC are correct,3 then we should expect the alpha section scribes to transcribe OJ s with their *ts, *s, and *c in the appropriate environments (3.2). But this is not always the case. Let us look at the spellings of OJ si in the alpha section. Chinese speakers would not confuse
MC *ciÜL ‘genitive,’
MC *siÅ ‘this,’ and
MC *\i ‘begin’. All MC reconstructions agree that these three sinographs were read with three distinct initials and this is true even when they are read in modern Mandarin as zhc, sc, and shh. Yet why do we find these three sinographs to be the first, third, and fourth most common phonograms for OJ si in the alpha section (Mori 1991: 274)? (The second most common phonogram is
MC *ciÜLh ‘intention,’ nearly homophonous with
MC *ciÜL ‘genitive’ apart from its tone.) Mori’s own statistics show that in the alpha section: (1) 50.8 per cent (59/116) of OJ si are written with MC *c-; (2) 25 per cent (29/116) are written with MC *\-; (3) 24.1 per cent (28/116) are written with MC *s-. We would expect the Chinese scribes of the alpha section to write OJ si (Mori’s *ci) with MC *c- much more than 50 per cent of the time; chance errors aside, they would have no reason to write OJ si (Mori’s *ci) with MC *\- or MC *s-, since Mori does not reconstruct *\i or *si as alternative pronunciations of OJ si (his *ci). English speakers are capable of distinguishing between [t6i], [6i], and [si]. If English speakers were to transcribe OJ si (Mori’s *ci [t6i] ), would they write it as chi [t6i] 50.8 per cent of the time, shi [6i] 25 per cent of the time, and si [si] 24.1 per cent of the time? This unexpected inconsistency in transcription is not unique to OJ si: (1) Although Mori reconstructs OJ se as *ce [t6e], in the alpha section, only 21.7 per cent (5/23) of OJ se are written with *c-, while the majority of OJ se are written with what Mori would consider to be the ‘wrong’ initials: 43.5 per cent (10/23) are written with MC *s- and 34.8 per cent (8/23) are 59

OLD JAPANESE

written with MC *\- (1991: 274). (2) 36 per cent (9/25) of OJ sa in the alpha section are written with MC *s-, not the MC *ts- that one would expect from Mori’s reconstruction of OJ sa as *tsa (1991: 274). It is true that more inconsistency can be found in the beta section (Mori 1991: 278), but, if Mori were correct, we should expect far less inconsistency in the alpha section. Mistakes in transcription are inevitable, but it is difficult to believe that Chinese scribes capable of distinguishing between *c, *\, and *s in their own language would write 78.3 per cent (18/23) of OJ se (Mori’s *ce [t6e] ) in the alpha section with the ‘wrong’ initial. I believe that there is a simpler explanation for the inconsistency of transcriptions for OJ s in the alpha section. Sino-Japanese today (and presumably Sino-Japanese in the past as well) reflects the nine MC initials *s, *ts, *tsh, *Z, *cr, *chr, *\, *c, and *ch as s. Sino-Korean today reflects those same nine MC initials as s, c, and ch. Presumably earlier varieties of SinoKorean also did not make a nine-way distinction (Yu 1991). If some alpha section scribes were non-Chinese incapable of distinguishing between all of those initials, then such ‘mistakes’ would be inevitable and frequent – far more frequent than chance errors made by Chinese who could distinguish between all those initials of their native language. I disagree with Mori’s absolute division between the alpha and beta sections of Shoki. Both sections reflect the traits of System-E-based transcription as outlined on p. 32, apart from the pitch accent orthography unique to the alpha section. It can be difficult to tell whether a poem is from the alpha or the beta section of Shoki by glancing at its orthography, but it is far easier to tell whether a poem is in the System-E-based transcription of Shoki or the System-D-based transcription of Kojiki or Man’yôshû. Both sections appear to reflect the influence of non-Chinese scribes recycling traits of earlier transcriptive systems, though such influence is clearly more apparent in the beta section. The difference between the two sections is probably a matter of degree. I hypothesize that the alpha section may have been written by second-language speakers of Chinese, and some of those second-language speakers were incapable of making all the phonemic distinctions of Chinese and/or still occasionally employing earlier transcriptive methods. (It is highly unlikely that Japanese and immigrant scribes accustomed to System-D-based transcription could completely break away from their old habits.) The beta section, on the other hand, may have been written by more conservative scribes whose knowledge of Chinese lagged behind that of the alpha section scribes. I do find Mori’s alpha/beta distinction a useful one, and will refer to it in Chapters 7 and 8, but I will also continue to look at the orthography of Shoki as a whole. As shown above, OJ s was transcribed with a variety of MC initials in both sections of Shoki. The problem of the reconstruction of OJ s (and, to a much lesser extent, its voiced counterpart OJ z) has fascinated scholars for a long time. The number of conflicting reconstructions of OJ s rivals the 60

PREVIOUS RESEARCH ON PHONOGRAMS

number of conflicting reconstructions of the A/B-type distinctions, as can be seen from Table 3.12 at the end of this chapter. Reconstructions of OJ s fall into two categories: “pro-s” and “anti-s” (Miyake 1995: 3–5). The pro-s school represented by Sandness (1985) and Miyake (1995) believes that OJ s was phonemically */s/ and was phonetically *[s]. The anti-s school has been far more popular than the pro-s school, spawning a wide variety of factions which agree only on positing anything other than *[s] for some or all OJ s-initial syllables. Pierson (1929) may have been the first to reconstruct *[ts] as well as *[s] as possible pronunications of OJ s, but the two most influential proponents of the idea were Arisaka (1955) and Mabuchi (1959). Arisaka advocated *[s] or *[6] before i and e and *[ts] elsewhere whereas Mabuchi advocated *[6] before all vowels. Since Sandness (1985) and Miyake (1995) provide arguments against these reconstructions and those of many other scholars, I will end my discussion of this subject here and return to the problem of the reconstruction of OJ s and z on pp. 177 and 183. Nearly all of the reconstructions examined above apart from Pierson’s worked within the consensus view of 87 to 88 syllables for OJ. While a few scholars such as Yoshitake and Lange reconstructed fewer than 87 to 88 syllables for OJ (86 in Yoshitake’s case and 82 in Lange’s), some scholars thought that there were even more distinct syllables in OJ. Following the nearly simultaneous confirmation of two types of mo in Kojiki by Ikegami Teizô and Arisaka Hideyo, Nagata Kichitarô (1934) proposed that there were two types of o, si, and po in OJ on the basis of the orthography in Kojiki. This hypothesis was later popularized by Mabuchi (1957). Both Ôno Tôru’s (1962) comprehensive guide to OJ phonogram usage and Igarashi’s (1969) dictionary of OJ phonogram usage rejected the Nagata–Mabuchi hypothesis of two types of o and si but acknowledged two types of po. Inukai (1978) and Bentley (1997) criticized the proposal of A/B-type distinctions for o and si but cited evidence in favor of such a distinction for po. I agree with Inukai and Bentley, but I will reserve my arguments until I discuss the C-type vowels i and o in Chapter 8. The final variety of phonogram-based OJ reconstruction which I will mention here deals with the study of apparent pitch accent orthography in the alpha section of Shoki and to a far lesser extent in Kojiki. Since I have already discussed this subject at some length on pp. 37–9, I will not return to it here.

Summary Knowledge of phonograms never truly died out in Japan. However, the original pronunciations of the phonograms and the orthographic distinctions based on these pronunciations were lost. Beginning in the late seventeenth century, kokugaku (national learning) scholars such as Keichû, Motoori 61

OLD JAPANESE

Norinaga, and Ishizuka Tatsumaro examined OJ phonogram texts and rediscovered earlier orthographic distinctions and at least some earlier phonological distinctions. In the twentieth century, Hashimoto Shin’kichi correlated all of these orthographic distinctions with phonetic distinctions and proposed that certain syllables had ‘A/B-type’ distinctions. The reconstruction of these distinctions has been attempted over and over again, with no consensus in sight (Table 3.11). Almost as hotly debated is the reconstruction of the phonetic value(s) of OJ s. Once again, there is little agreement among scholars Table 3.11 Various reconstructions of the OJ ‘A/B-type vowels’a Type

A

B

A

B

A

B

OJ ‘vowel’ Nagata (1932) Hashimoto (1933)b Yoshitake (1934) Yasuda (1934) Kikusawa (1935) Hashimoto (1938) Kindaichi (1938) Hattori (1948) Tôdô (1948)c Hashimoto (1950) Arisaka (1955) Hattori (1958, 1976) Miller (1967) Ramsey & Unger (1972) Lange (1973)d Unger (1977, 1993) Hattori (1978–79) Tôdô (1980) Ôno (1982) Mabuchi (1983) Pulleyblank (1984) Whitman (1985) Kiyose (1991) Mori (1991) Miyake (1995) This book

yi *i *i *i *i *i *i *i *i *i *i *i *yi *i *i *yi *i *i *i *i *i *yi *i *i *i *(y)i *i

iy *y *ïi *L *ui *wi *ïi *ï *ïi *R/*£ *ïi *ïi *i *ï *wi *i *wi *ti *ï *ï *üR *i *uy *t *L *t *t

ye *e *e *e *e *e *e *e *e *ε *e *e *ye *e *ye *ye *ye *e *e *e *üe *ye *ye *e *e *(y)e *e

ey *ε *Qi *ε *ue *we *Ri/*Re *ë *ai *£/*R *Ri/*æ *ée/*Rü *e *ë *e *e *e *Re *ë *ε *ε *e *ey *R *Rû *ε *Ry

wo *o *o *o *uo *wo *o *o *o *o *o *o *o *o *wo, *ow *wo *wo *o *o *o *uo *o *wo *u *o *(w)o *o

o *ø *ö *Q *o *o *ö *ö *ö *R/*å/*Q *ö *R *ö *ö *o *o *o *ö *ö *Θ *o *R *o *o *R *R *R

Notes a This table is an expanded version of the tables in Lange (1973: 69) and Kiyose (1991: 215). It is not an exhaustive listing of reconstructions. Umlauts signify centralization. b This reconstruction, unpublished to the best of my knowledge, is cited in Morimoto (1933) and quoted in Ôno (1953a: 131). c From a lecture, “Problems in Chinese Phonology,” cited in Ôno (1953a: 154–155). d Lange (1973: 124) reconstructs A-type pye, bye, mye as *[pye bye mye] and B-type pey, bey, mey as *[pe be me] but reverses these reconstructions for velar initials: A-type kye, gye are *[ke ge] and B-type key, gey are *[kye gye].

62

PREVIOUS RESEARCH ON PHONOGRAMS

Table 3.12 Various reconstructions of OJ s Pierson (1929) Yoshitake (1934) Arisaka (1955) Mabuchi (1959) Miller (1967) Osada (1972) Lange (1973) Tôdô (1980) Maruyama (1981) Kobayashi (1981) Yamada (1983) Pulleyblank (1984) Sandness (1985) Whitman (1985) Kiyose (1991) Mori (1991) Unger (1993) Miyake (1995) This book

*s or *ts? *s *ts / _a, u, o, o; *s~6 / _i, e OJ /s/ < merger of pre-OJ */s/, */ts/ *6 *ts / _a, u, o, o; *s / _i, e *6 or *s or *c [t6]? *s or *6? *s *ts (for 17th-century Japanese) *c [t6] / _a, u, o; or *6 *ts *ts / _a, o; *6 / _i, e; *s / _u, o *s possibly with allophonic variants *ts or *6? *c [t6] or *6? *c ([ts] or [t6]?) *ts / _a, o; *c [t6] / _i, e; *s / _u, o *s *s / _a, u, o, o; *\ / _i, e *s / _a, u, o, o; *s or *\? / _i, e

(Table 3.12). Other issues in OJ reconstruction such as the nature of OJ voiced obstruents, the possibility of A/B-type distinctions beyond those proposed by Hashimoto, or OJ pitch accent have received far less attention. Most of the above reconstructions of OJ based on phonogram evidence suffered from one or more of the following five problems: (1) With the exception of Pulleyblank (1984), no published phonological reconstruction of OJ known to me has taken into account recent advances in Chinese reconstruction represented by Pulleyblank (1984, 1991a), Starostin (1989), or Baxter (1992). The vast majority of OJ reconstructions rely either upon Karlgren’s long-outdated reconstruction which has changed little since the 1920s or upon various other outdated reconstructions often similar to Karlgren’s to interpret the Middle Chinese (MC) pronunciations of phonograms. The use of ‘Karlgrenoid’ and modern reconstructions of Chinese can lead to very different results. For example, there is a phonogram
‘allow’ used to write OJ ko in Kojiki and the beta section of Nihon shoki. According to Karlgren (1957: 36), this ongana was read *xüwo (ignoring its tone) in MC. This would lead one to conclude that the OJ syllable ko represented by
‘allow’ was perhaps something like *ko with a rounded vowel *o, substituting OJ *k for MC *x and ignoring the *-üw- of the original Chinese reading. (It is known that OJ probably borrwed MC *x as OJ *k. Furthermore, no attested variety of Japanese known to me permits medial glide sequences such as *-yw-.) However, if one looks at a modern reconstruction of 63

OLD JAPANESE


‘allow’ such as Pulleyblank (1991), one does not find a rounded vowel, or even a medial glide sequence such as *-üw-: EMC *htRq. This would lead one to conclude that OJ ko was more like *kR than *ko. The central unrounded vowels in the Middle-Chinese-based Sino-Korean (SK) and SinoVietnamese (SV) readings of
‘allow’ support this conclusion: SK he [hR] and SV hEa [htR] are not unlike my proposed *kR. (2) Using MC reconstructions can invite the danger of circularity, since some MC reconstructions are based upon OJ reconstructions. Two examples of this problem are Tôdô (1957) and Pulleyblank (1984). (3) With the exception of Mori (1991) and Miyake (1995), statistical analyses of OJ phonograms have rarely been done. Many scholars seem to believe that they can reconstruct OJ phonology by looking at inventories of phonograms in a given text (or, worse yet, a body of heterogenous texts) and at the (usually outdated) reconstructed MC readings of those phonograms. Aside from the problems posed by MC reconstructions, this approach is faulty because it ignores the frequencies of individual phonograms. ‘Averaging’ the reconstructed MC readings for all the phonograms for an OJ syllable will not do. Such a technique gives the mainstream and the marginal equal weight. (4) Previous attempts at OJ phonological reconstruction disregarded phonological typology. Earlier OJ reconstructions and the Karlgrenoid MC reconstructions upon which they were based often have marked, ‘unnatural’ consonant and vowel systems. Some of the reconstructions of A/B-type distinctions in Table 3.11 above appear to be nearly random collections of phonetic symbols. How many attested vowel systems have the diphthong *Rû which Mori (1991) reconstructs for OJ B-type ey, or, as Pulleyblank noted earlier, four mid vowels (*e, *ë, *ö, *o) but only three high vowels (*i, *ï, *u) and one low vowel (*a)? A similar complaint could be made about the extremely complex vowel systems of Karlgrenoid MC reconstructions. Karlgren reconstructed 15 vowels for MC (Figure 3.3). These vowels are distributed unevenly within the system and also among the roughly four Central

Front High

*i, *A

*u *C

*e,*B

Mid

*o

*ä *K

*D

Low

Back

*å *ââ, *â

*aa, *a

Figure 3.3 Karlgren’s MC vowels (after Ramsey 1987: 131) Note: I use doubling to indicate vowel length for Karlgren’s *a and *â.

64

PREVIOUS RESEARCH ON PHONOGRAMS

thousand syllables of MC. There are eight front vowels representing five degrees of height (high, upper mid, mid, lower mid, and low) while there are only two central vowels and five back vowels. Many of these vowels occur only in diphthongs or triphthongs. Most subsequent reconstructions of MC continued to posit similar vowel ‘systems.’ Some reconstructions such as Mineya (1972) were even more intricate than Karlgren’s, substituting even more diphthongs and triphthongs for the simple vowels which were already in the minority in Karlgren’s original reconstruction. It is therefore not surprising that previous studies relying upon Karlgrenoid MC reconstructions of OJ phonogram readings concluded that OJ also had a convoluted vowel ‘system’ containing odd diphthongs such as Mori’s (1991) *Rû. It is, however, unfortunate that those studies did not try to reinterpret their results in more typologically plausible terms. (5) Previous studies disregarded not only typology but also internal and comparative reconstructions of earlier Japanese. As I will show in the next chapter, internal reconstruction has shown that some of the ‘eight vowels’ of OJ have diphthongal origins. For example, most instances of OJ B-type ey originate from pre-OJ *a + *i. Although Lange cited this very sound change (1973: 56) when reviewing the work of Arisaka (1934) and Ôno (1953b), he ignored it later when he reconstructed OJ ey after velars as *ye and as *e elsewhere (1973: 124). If OJ B-type ey had come from a pre-OJ *ai and if Lange’s OJ reconstruction is correct, then why did pre-OJ *ai become OJ *ye only after velars? It is difficult for me to believe that pre-OJ *ai could become OJ *ye in any environment without a series of intermediate sound changes: *ai > *æ > *e > *ye. Few if any reconstructions of OJ have examined comparative evidence. A rare counterexample is Miller’s (1967) attempt to study the OJ A/B-type distinctions by examining vowel correspondences between Tokyo Japanese and the Satsuma dialect, but this has since been discredited by Aoki (1974). I believe that the ideal reconstruction of OJ should be compatible with evidence from all three types of reconstruction: internal reconstruction, external reconstruction, and reconstruction based upon written records. In the following chapter, I will summarize what is known about pre-OJ on the basis of internal and external reconstruction and what is known about postOJ on the basis of written records and modern dialects.

Notes 1 Unlike Hashimoto,Yoshitake did not reconstruct the low-frequency syllable OJ zoA which appears in Man’yôshû, but not in the poetry of Kojiki or Nihon shoki. 2 Pulleyblank reconstructs only six such phonemes for LMC because he believes the retroflexes and palatals merged into a single retroflex series. I will argue against this view in Chapter 5. 3 All ‘MC’ forms are cited according to Mori’s reconstruction with typographical modifications to match the orthography I use for other Chinese reconstructions.

65

OLD JAPANESE

4 JAPANESE PHONOLOGY THROUGH TIME

Overview This chapter briefly outlines the history of OJ segmental phonemes. It is neither a history of attempts at the internal or external reconstruction of Japanese nor is it a substitute for more detailed accounts of Japanese language history such as Martin (1987). The viewpoint presented here is admittedly rather mainstream. I will discuss only consonants and vowels here, completely ignoring pitch accent. These limitations reflect this book’s goal: the reconstruction of OJ segmental phonetics based on written records. I consider the evidence below to be secondary evidence for my reconstruction. I will discuss the sinograph readings that are at the heart of my reconstruction in the following chapter. Before beginning to discuss the history of Japanese segments, I should define some names for the various stages of the Japanese language, in chronological order. Proto-Koreo-Japonic (PKJ), a term used by Serafim (1994), is a hypothetical ancestral language of Korean and Japanese reconstructed by Whitman (1985). (Whitman’s own term for this ancestral language is ‘proto-JapaneseKorean.’) Since (1) the very notion of PKJ has been questioned even by an advocate of a genetic relationship between Korean and Japanese (Vovin 1996a)1 and (2) this book does not concern a genetic relationship between those two languages, I will deal with PKJ only minimally in this book. Japonic refers to all the languages of ‘mainland’ Japan and the Ryukyus. Japonic is generally divided into two branches, Japanese and Ryukyuan. Proto-Japonic (PJ) is the hypothetical ancestor of all Japonic languages. Pre-Old Japanese (POJ) is a generic term for any variety of Japanese prior to roughly the seventh century. I consider the Suiko Period (592–628) transcriptions to be OJ rather than POJ. Proto-Japonic-Internal (PJI) is reconstructed primarily on the basis of the internal reconstruction of the ‘mainland’ (i.e., non-Ryukyuan) branch of Japonic with limited or no reference to the Ryukyuan branches. Examples of PJI reconstructions are in Whitman (1985), Martin (1987), and Unger (1993). 66

JAPANESE PHONOLOGY THROUGH TIME

Proto-Japonic-External (PJE) is reconstructed on the basis of the comparison of both the mainland and the Ryukyuan branches. Examples of PJE reconstructions are in Hattori (1978–79) and Serafim (1999a). Both PJI and PJE are ancestral to OJ. The different terms reflect their origins in different types of reconstruction (largely internal for PJI and wholly comparative for PJE). Old Japanese (OJ) is, strictly speaking, the language of the Nara Period (710–794); characterized by the so-called ‘eight vowels.’ However, since (1) the Japanese of the Suiko Period transcriptions appears to exhibit the same ‘eight vowels’ as OJ proper2 and (2) much of the content of eighth-century Nara Period texts actually dates from the seventh century or even earlier, I tentatively also consider the Suiko Period transcriptions to represent OJ rather than late POJ. I may reconsider this classification of Suiko Period Japanese after examining System-C-based orthography more closely in the future. Most of the remaining terms are unfortunately correlated with political periods, implying that changes in language accompanied changes in rulers. Although of course this was not the case, these terms are sometimes useful. I generally follow the definitions for these terms given in Martin (1987: 77). Early Middle Japanese (EMJ) is the language of the Heian Period (794– 1185). Late Middle Japanese (LMJ) is the language of the Kamakura Period (1185–1379), the Muromachi Period (1392–1573), and the (Azuchi)Momoyama Period (1573–1603). Modern Japanese (MJ) is the language of the Edo Period (1603–1867). Modern Standard Japanese (MSJ) is the standard language of modern Japan, supposedly based on “a variety of the Tokyo dialect typical of educated, middle-class natives of the Yamanote (‘uptown’) region” (Vance 1987: 1).

Consonants Pre-OJ consonants Whitman (1985) has proposed the PKJ consonants in Table 4.1. This inventory is similar to the maximal PJ consonant inventory in Table 4.2, which incorporates the PJ proposals of Whitman (1985), Martin (1987), and Unger (1993). If PKJ existed and if Whitman’s PKJ reconstruction is valid, most of the PKJ consonants remained unchanged in early Japonic, with the following exceptions. (1) PKJ *s, *\, and *l merged into PJ *s. One cannot reconstruct a threeway distinction between *s, *\, and *l purely on the basis of Japonic evidence. (2) PKJ *j may have become either PJ *d or *z. If Unger’s alternative reconstruction of PJ *z as *j is correct, then PKJ *j may have remained 67

OLD JAPANESE

Table 4.1 The PKJ consonants of Whitman (1985)a

Voiceless obstruents Voiced obstruents Nasals Liquid and glide

Labial

Dental

*p *b *m

*t *d *n *-r-, *-l-b

*s *z

Palatal

Velar

*c *j

*k *g

*\

*y

Notes a Since Whitman (1985) does not explicitly list all of his PKJ consonants, I have based this table on Martin’s (1991: 273) list of Whitman’s (1985) PKJ consonants. b Leon Serafim (p.c.) speculates that PKJ *l may have been palatalized: *[ly]. Whitman (1985: 185) compares his PKJ *l to Proto-Altaic *l2.

Table 4.2 A maximal inventory of PJ consonants

Voiceless obstruents Voiced obstruents Nasals Liquid and glide

Labial

Dental

Palatal

Velar

*p *b *m

*t *s *d *z *n *r

*c (*j)

*k *g

*y

Note Unger (1993: 46) reconstructs either *z or *j.

intact as PJ *j. I have not been able to find an explicit statement in Whitman (1985) concerning the PJ (Whitman’s ‘pre-OJ’) reflex(es) of PKJ *j. (3) Medial and final PKJ *r was sometimes lost in PJ. Since PKJ *r otherwise continued unchanged in PJ, these losses did not affect the structure of the PJ consonantal inventory. I will not elaborate on the environments of *r loss here; see Whitman (1985: 189–208). Whitman (1985) has also proposed medial and final *m loss, but this proposal is more problematic (Alexander Vovin, p.c.) and will not be dealt with here. Now I will discuss the PJ consonants of Table 4.2, focusing on (1) the validity of the reconstruction of each consonant and (2) the OJ reflex(es) of each consonant. I will begin with the more controversial consonants first, eliminating those which I believe are dubious in order to create the minimal inventory of PJ consonants listed in Table 4.5. (1) All scholars agree that PJ had a phoneme *s, but disagree about *z (or *j) and *c. Furthermore, they disagree about the OJ reflexes of PJ *s. Whitman (1985) reconstructs PJ *s with an OJ reflex s and PJ *z with the OJ reflexes y, s, and z depending on environment. Whitman’s proposal of PJ *z is similar to that of Ramsey and Unger (1972), with one difference: Ramsey and Unger’s PJ *z became OJ zero in all environments. Unfortunately, Ramsey and Unger (1972: 290) and Unger (1993: 35) posit PJ *z on the basis of phonological symmetry: if PJ *p and *t had voiced counterparts 68

JAPANESE PHONOLOGY THROUGH TIME

*b and *d, then PJ *s might have had a voiced counterpart *z as well. Ramsey and Unger (1972: 290) also claim that it is necessary to posit *z to explain the monophthongizations between PJ and OJ which resulted in at least some of the A/B-type ‘vowels’ of OJ (p. 80). In the previous chapter, for example, I mentioned that some or all OJ B-type ey originate from an earlier combination of *a and *i: (4.1) pre-OJ *ai > OJ ey Ramsey and Unger’s proposal of PJ *z would allow them to reformulate (4.1) as follows: (4.2) PJ *a(z)i > *ai > OJ ey But this *z cannot be reconstructed through the comparative method (the reflex of their PJ *z throughout Japonic is zero) or through internal reconstruction (there are no alternations suggesting their PJ *z involving the OJ A/B-type ‘vowels’ or their modern reflexes). Hence I reject the Ramsey and Unger proposal of PJ *z and the similar PJ *z (or *j) of Unger (1993). Martin, like Ramsey and Unger (1972) and Whitman (1985), also proposes PJ *s and *z. Martin reconstructs *z as the initial consonant of a few OJ morphemes which appear in two forms: a free form beginning with a vowel and a bound form beginning with s. For example, OJ amey ‘rain’ appears in the compound OJ parusamey ‘spring rain’ as samey with s- (MYS 3969.49). Martin (1987: 381) reconstructs ‘rain’ in PJ as *[z]ama-Ci with an optional *[z] to explain the ‘disappearing s.’ However, Martin (1987: 36) himself notes that Unger (no citation given) rejects the notion that the proto phoneme in question (if it existed) was voiced. He points to the fact that when there is an alternation it is the -s- form that occurs between vowels, an environment that offers no phonetic motivation for the devoicing that is later evidenced: why should speakers turn off the voice in the middle of haru-same– ‘spring rain’? Unger (1993: 41) proposes that ‘disappearing s’ comes from PJ *s, not *z, and that PJ *c, a “palatal stop,” is the source of OJ s: (4.3) PJ *c > OJ s (in all environments) PJ *s > OJ $ (initially) OJ s (intervocalically) But Unger also admits that “this hypothesis has little to support it.” Yamaguchi (1974) has found only four morphemes with ‘disappearing s’ in 69

OLD JAPANESE

OJ, but there are far many more morphemes with OJ ‘non-disappearing s’ (< Unger’s PJ *c) listed in Miyake’s (1995) database. This would imply that PJ *c was far more common than PJ *s, though I know of no language in which a palatal stop is extremely frequent and s is extremely marginal. Unger does not give any evidence whatsoever for the palatal articulation of *c. He reconstructs OJ s as *s (1993: 22) and I do not understand why he did not simply project this *s back into PJ. Why reconstruct a less marked *s for only four morphemes and reconstruct a more marked palatal stop *c for all others? There may be other explanations for ‘disappearing s’ that do not involve positing a low-frequency phoneme. Martin (1987: 36) wonders whether “the -s- could be the relic of a genitive marker, like the Middle Korean -s-, perhaps a variant of the Old Japanese particle (-)tu.” Unger (1993: 26) makes a similar speculation. Miller (1967: 194) does not merely speculate: he derives ‘spring rain’ from *Faru tu amë ‘spring-GEN-rain.’ But Alexander Vovin (p.c.) points out that since “OJ *t never shifts to *s,” it is unlikely that OJ -s- is etymologically related to the OJ genitive marker tu. In any case, I reject Martin’s PJ *z and Unger’s PJ *c for the reasons given above, preferring to reconstruct only PJ *s with an OJ reflex s. I will not attempt to solve the problem of ‘disappearing s’ here because this book is primarily concerned with the synchronic reconstruction of OJ segmental phonetics. (2) Ramsey and Unger (1972) and Unger (1993) propose a PJ *g using arguments similar to those for their PJ *z. Martin (1987: 20) also proposes a PJ *g on the basis of comparisons with Korean. All these scholars agree that PJ *g is reflected in OJ as zero. However, as Whitman (1985: 19) writes, “there is no systematic internal Japanese evidence” for PJ *g > OJ $. One cannot reconstruct PJ *g without reference to Korean. (Recall that Whitman does reconstruct a Proto-Korean-Japanese *g.) I have already explained why I find Ramsey and Unger’s phonological symmetry and supposed explanatory value arguments to be unconvincing. Hence I reject PJ *g as well as all the above proposals for a PJ *z. (3) I reject a PJ-level distinction between *y and *d. To the best of my knowledge, one can only tell whether a given OJ y came from an earlier *y or *d via comparison with Korean. This suggests to me that *y and *d merged at the PJ level into a single phoneme, unless *y and *d merged independently throughout Japonic. Following Martin (1987) and Unger (1993), I will reconstruct the PJ source of OJ y as *d. Although I would rather reconstruct PJ glides *y and *w instead of PJ voiced stops *d and *b, the evidence for either view is controversial. In any case, there is no evidence for a phonemic distinction between glides and voiced stops in PJ. Apart from *b and *d, all of the PJ consonants in Table 4.3 are relatively uncontroversial. No one has proposed eliminating, say, PJ *p, *t, or *k. Nonetheless, there is some disagreement over the details of their 70

JAPANESE PHONOLOGY THROUGH TIME

Table 4.3 A minimal inventory of PJ consonants

Voiceless obstruents Voiced obstruents Nasals Liquid

Labial

Dental

Velar

*p *b *m

*t *s *d *n *r

*k

development. What follows is a concise account of changes in consonants from PJ to OJ. (1) PJ *p has an OJ reflex which I romanize as p. Many scholars, beginning with Hashimoto (1928), believe that a pre-OJ voiceless bilabial stop *p lenited into a voiceless bilabial fricative *φ in OJ (my p). I will argue against the lenition of PJ *p in Chapter 8. (2) Whitman (1985: 27) claims that: (4.4) PJ *t PJ *d, *n, *r, (*z)3

> OJ s > OJ y

/ _i, y / _i, y

One might expect OJ to have no syllables such as ti, di, ni, or ri, yet OJ did have such syllables (Table 3.4). This means that OJ ti etc. cannot be from PJ *ti etc. The PJ ancestors of OJ ti etc. must not have been subject to the rules in (4.4). I will deal with the origins of these OJ C-type syllables when I discuss the origins of OJ vowels on p. 79. (3) Although Whitman (1985) believes that the lenition of PJ *d was conditioned by *i or *y (4.4), Ramsey and Unger (1972), Martin (1987), and Unger (1993) believe that both PJ *b and *d lenited in all environments, resulting in OJ w and y. Ramsey and Unger (1972) date this change as late as the seventh century CE, after the importation of Go-on and before the importation of Kan-on. However, I agree with Whitman (1985: 14) that “this hypothesis raises more complications than it resolves.” First, if Ramsey and Unger are correct, then we should expect the following pattern of correspondences: (4.5) Pre-seventh-century Japan

*b, *d, (*g, *z)4

Post-seventh-century Japan w, y, ($, $)

transcribed with Chinese *b, *d, etc. transcribed with Chinese *w, *y, etc.

Words transcribed with Chinese *b and *d in earlier orthographies should have been transcribed with *w and *y in the System-D-based orthography of the eighth century, reflecting the shift of PJ *b, *d, (*g, *z) > OJ w, y, 71

OLD JAPANESE

($, $). Since Ramsey and Unger do not reconstruct any glides (*w, *y) for pre-OJ, we should not expect to see any pre-eighth-century Japanese words transcribed with Chinese *w or *y. Let us now test Ramsey and Unger’s theory. According to Igarashi (1969: 154), the OJ name element wakey (< PJ *baka-Ci?) appears in a gloss in the first volume of Kojiki as
 EMC *®wa khtyh (modern Go-on wa ke) with EMC *®w- for OJ w. This word should have been pronounced with *b- in pre-eighth-century Japanese. Therefore, it should appear in earlier transcriptions with a LOC or EMC *b-. However, in the Inariyama inscription (2.4.3), we find the fifth-century equivalent of wakey spelled with
 LOC *wheerk kaò with LOC *wh-, not LOC *b-. Unless the identification of

LOC *wheerk kaò as wakey is incorrect, it appears that earlier Japanese *b had already lenited to w by the fifth-century. What about earlier Japanese *d? Is there any evidence that it had not lenited to y prior to the seventh century? In Kojiki, we find OJ to yo (< PJ *tRdR; Martin 1987: 551) ‘rich’ spelled as
 EMC *tRY ytRq with EMC *y for OJ y. In the Gangôji tray inscription (596 CE), we find the late sixthcentury equivalent of OJ to yo spelled as
 LOC *tRRYq yRq / EMC *tRYq ytq with LOC/EMC *y-, not LOC/EMC *d-. This is not an isolated instance; Igarashi (1969: 99–100) lists at least three more Suiko Period attestations of to yo with phonograms read with LOC/EMC *y-. In my examination of the System-B- and System-C-based transcriptions from the fifth to the seventh century CE, I have never found any cases of Chinese voiced obstruents (*b, *d, etc.) used to write OJ words attested in the eighth century with w or y. We would also expect the early transcriptions to reflect the pre-OJ *z and *g proposed by Ramsey and Unger, but I have never found any cases of Chinese *z- or *g-like voiced obstruents corresponding to zero in OJ words attested in the eighth century. I conclude that there is no philological evidence for Ramsey and Unger’s claim. If pre-OJ *b and *d did lenite to OJ w and y, they must have done so before the fifth century. Even without looking at pre-eighth-century transcriptions, Ramsey and Unger’s dating of the lenition of pre-OJ *b and *d is still suspect. Let us suppose that Go-on was borrowed into Japanese before the lenition of pre-OJ voiced obstruents. Thus
EMC *ba ‘crone,’
EMC *da ‘hill,’
EMC *zia ‘evil,’ and EMC
*®a ‘what’ (which are all phonograms in Kojiki) would have been borrowed as *ba, *da, *za, and *ga. Then in the seventh century, *b became w, *d became y, and *z and *g became zero. Thereafter, the Go-on readings for these four sinographs should have been wa < *ba, ya < *da, a < *za, and a < *ga. However, in modern Go-on, they are still read with voiced obstruents as ba, da, za, and ga. Ramsey and Unger (1972: 292) try to explain this by claiming: The simplest explanation [is] that Go-on morphs formed a distinct lexical stratum, were therefore not necessarily susceptible to sound 72

JAPANESE PHONOLOGY THROUGH TIME

changes taking place in native morphemes, and, moreover, were especially resistant to surface alteration . . . their users attached great importance to preserving them in as uncorrupted a form as possible. Thus Ramsey and Unger (1972) believe that pre-OJ voiced obstruents did lenite in the seventh century in all Japanese words except for Go-on loans. However, I believe that there is a simpler explanation: Go-on was borrowed after (1) the pre-OJ voiced obstruents lenited and (2) a new series of voiced obstruents developed from nasal-obstruent clusters (see below). The Japanese borrowed EMC voiced obstruents (via Systems C and D) with this new series of native voiced obstruents. (4) Medial clusters of PJ nasals (*m, *n) with PJ voiceless obstruents (*p, *t, *s, *k) resulted in the voiced obstruents of OJ (Ramsey and Unger 1972: 278): (4.6) PJ *-mp-, *-np*-mt-, *-nt*-ms-, *-ns*-mk-, *-nk-

> > > >

OJ b d z g

These PJ clusters originated via “syncope and progressive voicing assimilation” from earlier nasal–vowel–obstruent sequences defined by Ramsey and Unger (1972: 278) as: (4.7) {*m, n} + V + {p, t, k, s} For example, the -z- in the OJ title
 murazi (MYS 1439 preface) originated from the sequence *ns < *n-u-s in *mura-nusi ‘village-master’ (Martin 1987: 488); the *-u- dropped out and the remaining *ns became OJ z. Notice that the formulation in (4.6) allows only for PJ medial clusters as sources of OJ voiced obstruents. OJ had only medial voiced obstruents, apart from initial voiced obstruents in Chinese loans (Whitman 1985: 7). According to Ramsey and Unger (1972), Martin (1987), and Unger (1993), none of the OJ voiced obstruents originate from PJ voiced obstruents: OJ b is not from PJ *b, etc. Although Whitman (1985: 187) believes that PJ *d (and PJ *z and *g, which I have rejected above) “merged with the corresponding voiced obstruents derived from nasal-obstruent clusters at the pre-OJ stage,” this is a minority view. (5) Apart from those environments described in (2) and (4) above, PJ consonants except for *b and *d (and possibly *p) remained intact in OJ. The main difference between the PJ and OJ consonant systems lies in the appearance of the new voiced obstruent series discussed in paragraph (4) above. 73

OLD JAPANESE

Table 4.4 The OJ consonant inventory

Voiceless obstruents Voiced obstruents Nasals Liquid and glides

Labial

Dental

p (φ ?) b m w

t s d z n r

Palatal

Velar k g

y

At this point I will leave open the possibilities that (1) OJ p was *φ and (2) OJ s and z may have been palatal rather than dental and/or affricates rather than fricatives. I will resolve the problems involving the phonetic reconstruction of these phonemes in Chapter 7. Post-OJ consonants Owing to the extreme complexity of phonological change in post-Nara Japanese – particularly those changes called onbin (translated by Erickson 1998 as ‘sound convenience’) – I will mostly deal with major, regular changes here. For more detailed discussion of onbin see Martin (1987), Frellesvig (1995), and Erickson (1998). (1) OJ p is widely believed to have become the bilabial fricative EMJ φ in the Heian Period. Alternatively, if one believes that OJ had φ instead of p, this φ remained unchanged in the Heian Period. Kiyose (1991), on the other hand, believes that OJ p remained unchanged in the Heian Period. All scholars, Kiyose included, agree that the phoneme in question had become a fricative by the Muromachi Period. Foreign sources provide extensive written evidence for a fricative pronunciation. The Portuguese missionaries of the late sixteenth and early seventeenth centuries romanized the phoneme as f and the Korean textbook of Japanese, Chephay sine (Rapid Understanding of a New Language; 1676) wrote it with the hankul letter ph [ph] or the hankul letter combination hw since Korean lacked a bilabial fricative. Later, word-initial φ became the MSJ fricative generally romanized as h with a number of allophones: [ç] before i, [φ] (written as f in Hepburn romanization) before u [T], and [h] elsewhere. MSJ also has a voiceless bilabial stop p, mostly in foreign loans (e.g., paatii ‘party’) and in what Martin (1987: 11) describes as native “mimetic and marginal words” (e.g., pachin ‘click!’); this MSJ p is almost never a reflex of OJ p. Medial OJ p generally became -w- by the Heian Period (Matsumura 1972: 95, Martin 1987: 11). Today, MSJ reflects medial OJ p as: (a) -w- before a (MSJ kawa ‘river’ < OJ kapa ‘id.’). (b) zero before other vowels (MSJ koi ‘love’ < OJ kopiy ‘id.’) 74

JAPANESE PHONOLOGY THROUGH TIME

Table 4.5 OJ t- and d-syllables and their later equivalents OJ LMJ MSJ

ta ta [ta] ta [ta]

ti chi [c\i] chi [c\i]

tu tçu [tsu] tsu [tsT]

te te [te] te [te]

two, to to [to] to [to]

da da [da] da [da]

di gi [ _j 9i] ji [ _j 9i]

du dzu, zzu [dzu] zu [dzT]

de de [de] de [de]

dwo, do do [do] do [do]

zu zu [zu] ji [dzT]

ze je [9e] zu [dze]

zwo, zo zo [zo] zezo [dzo]

Table 4.6 OJ s- and z-syllables and their later equivalents OJ LMJ MSJ

sa sa [sa] sa [sa]

si xi [\i] shi [\i]

su su [su] su [sT]

se xe [\e] se [se]

swo, so so [so] so [so]

za za [za] [dza]

zi ji [9i] za [©9i]

Note The [dz] allophone of MSJ z can also be pronounced as [z].

I will discuss w-loss before all vowels other than a in (5) below. Exceptions to the above generalizations involve reduplicative forms, compounds, Chinese loans, and a handful of other lexical items (see Martin 1987: 10 for examples). (2) OJ t and d became affricates before i and u by the late Muromachi Period. Compare the OJ t- and d- syllables with their LMJ equivalents as spelled by the Portuguese missionaries (Matsumura 1972: 147) and their MSJ equivalents today in Table 4.5. (3) s and z were palatalized after i and e in the LMJ recorded by the Portuguese missionaries. In Table 4.6 I reproduce the Portuguese spellings of LMJ s- and z- syllables cited in Matsumura (1972: 147). In MSJ, palatal varieties of s and z mostly occur before i in native words, but palatalized s and z before e in native words can still be found in other modern Japanese dialects (see Martin 1987: 18 for examples). Note that the LMJ reflexes of OJ di, zi, du, zu were still distinct: [ _j9i], [9i], [dzu], [zu]. In MSJ, these four syllables (the ‘yotsugana’ or ‘four kana’; Martin 1987: 16’s ‘the kana Gang of Four’) have merged into two: ji [©9i] and zu [dzu]. (4) MSJ reflects the OJ voiced obstruents as plain voiced obstruents: OJ b, d, z, g correspond to MSJ b, d, z, g. Note though that MSJ g may be pronounced as a nasal [Y] in medial position: MSJ daga [daga] or [daYa] ‘but.’ However, written evidence for LMJ suggests that its reflexes of OJ voiced obstruents were prenasalized. For example, the aforementioned Chephay sine spells LMJ b, d, z, g with the hankul letter sequences -mp-, -nt-, -nz-, -ngk- (Martin 1987: 22–23). Some modern Japanese and Ryukyuan dialects still have prenasalized voiced obstruents or traces of them: OJ b, d, z, g 75

OLD JAPANESE

correspond to Akita mb, nd, nz, Yg (Vovin 1996c). For an extensive listing of all evidence for prenasalized voiced obstruents in Japonic from the fifteenth century to the present see Martin (1987: 20–25). (5) OJ distinguished the following seven syllables: (4.8) i e o

: : :

wi we : ye wo

According to the chronology presented in Martin (1987: 79), these distinctions were carried over into early EMJ, but by late EMJ they began to break down. Early EMJ e and ye merged into later EMJ ye and early EMJ o and wo merged into later EMJ wo: (4.9) i

:

wi we : wo

ye

w- was then lost before the front vowels i and e: wi, we > i, e. This ‘new’ e from earlier we and ye merged into very late EMJ ye:5 (4.10) i ye wo The EMJ evidence for these changes comes from the kana ‘spelling mistakes’ which Keichû (p. 45) tried to ‘correct’ centuries later. These ‘mistakes’ arose because kana which were once non-homophonous (e.g., o and wo) had become homophonous (e.g., both o and wo as [wo] ) in later periods. We also have foreign written evidence from LMJ onward for these changes. The Portuguese missionaries transcribed the syllables in (4.10) as i, ye, and uo or vo (Matsumura 1972: 143, 146), confirming that they were pronounced as [i ye wo] in LMJ. Ye and wo later lost their initial glides: LMJ ye, wo > MSJ e, o. (4.11) i e o (6) After OJ, two new phonemes developed which have often been written /N/ and /Q/.6 Vance (1987) calls them the ‘mora nasal’ and the ‘mora obstruent.’ Their phonetic realization in MSJ is quite complex. Before a pause, /N/ is a uvular nasal [n]; it assimilates to the place of articulation of a following stop, affricate, or nasal (e.g., /Nk/ > [Yk] ) and is pronounced 76

JAPANESE PHONOLOGY THROUGH TIME

as a nasalized vowel before vowels, glides, and fricatives (e.g. /oNo/ > [oZo] ). /Q/ becomes a phonetic copy of a following obstruent (e.g., /Qk/ > [kk] ). See Vance for further details (1987: 34–44). Many of these ‘mora consonants’ arise from earlier consonant–vowel sequences (Vance 1987: 57, 58): (4.12) OJ yom-yite mat-ite

MSJ yonde /yoNde/ ‘read-GER’ matte /maQte/ ‘wait-GER’

In (4.12), OJ myi corresponds to MSJ /N/ and OJ ti corresponds to MSJ /Q/. The /N/ and /Q/ in (4.12) respectively result from changes called sokuonbin and hatsuonbin; two other types of onbin, i-onbin and u-onbin, are so called because they resulted in the loss of various consonants before the vowels i and u. Onbin generally occur only in certain morphological environments and hence are not regular sound changes. The various onbin and the ‘mora consonants’ all date from EMJ (Martin 1987: 79), but since they will contribute little to my arguments for the reconstruction of OJ, I will end my discussion of them here. See the references mentioned at the beginning of this section for more on these subjects.

Vowels Pre-OJ vowels Whitman (1985: 129) has proposed the inventory of PKJ vowels in Table 4.7. PKJ had palatal (or ‘front-back’) vowel harmony: three of the four front vowels (*ü, *ö, *e) were in opposition to the three back vowels (*u, *o, *a). The high front vowel *i was neutral. Whitman (1985: 60) believes that these seven vowels were reduced to four in PJ (Table 4.8). This vowel system is similar to the four-vowel system of Proto-Austronesian, though the similarity is probably only coincidental. Unger (1993: 46), on the other hand, posits five vowels for PJ. He assumes that PJ *o was a back vowel and believes that a four-vowel system with three two back vowels and only one front vowel would be unbalanced (1993: 30). Of course, this imbalance would be remedied if *o were a central vowel

Table 4.7 The PKJ vowels of Whitman (1985), after Serafim (1994: 185)

High Mid Low

Front unrounded

Front rounded

*i (neutral) *e

*ü *ö

Back unrounded

Back rounded *u *o

*a

77

OLD JAPANESE

Table 4.8 The PJ vowels of Whitman (1985) Front High Mid Low

Central

Back

*i

*u *o (= *[R]; Martin 1987’s o) *a

Note Whitman (1985: 60) says that “*o and *a are redundantly specified as [+back].” However, since he reconstructs *o as a central vowel *[R] and his chart places *o (= *[R] ) and *a in the center rather than with the [+back] vowel *u, I have treated *o and *a as central vowels.

Table 4.9 The PJ vowels of Unger (1993) Front High Mid Low

Central

*i *e

Back *u *o

*a

Table 4.10 The PJE vowels of Hattori (1978–79) Front High Mid Low

Central

*ü, *i(i) *e(e)

*R(R) *a(a)

Back *u(u) *o(o)

Note This chart is based on that given in Martin (1987: 68). The notation *V(V) indicates that Hattori reconstructed both a short and a long vowel; hence *a(a) is shorthand for *a and *aa.

as in Whitman’s PJ vowel system (Table 4.8), but Unger (1993: 30) claims that “the grounds for such an argument are lacking” without any elaboration. He reconstructs PJ *e as the front counterpart of back PJ *o to restore balance to the system and provides arguments mostly based on internal reconstruction in favor of this *e (1993: 30–33). Hattori (1978–79: 21.98, 22.108) posits seven vowels (disregarding vowel length) for PJE (Proto-Japonic-External) on the basis of correspondences between OJ and ‘Period A Shuri,’ his reconstruction of the Ryukyuan Shuri dialect circa 1400 (Table 4.10). These vowels combined to form various diphthongs: *ai, *Ri, *ui, *üi, *ia, *au. Note that the reconstructions in Tables 4.8 and 4.9, unlike Hattori’s in Table 4.10, were based primarily on Japanese-internal reconstruction and not on comparative reconstruction. Strictly speaking, Whitman’s and Unger’s systems are PJI (Proto-Japonic Internal) in contrast with the PJE (ProtoJaponic-External) system of Hattori. 78

JAPANESE PHONOLOGY THROUGH TIME

Table 4.11 The PJ vowels of Serafim (1999a) Front High Mid Low

Central

*i *e

Back *u *o

*ö (= [R] ) *a

Table 4.12 The PKJ vowels of Serafim (1999b) Front

High Mid Low

Central

Back

tense

lax

lax

tense

*i *é

*è

*ù *ò

*ú *ó

*a

Note Acute accent = tense, grave accent = lax.

Serafim’s (1999a) PJE reconstruction contains six vowels (Table 4.11). Which of these views is closest to the truth? Three of the four reconstructions (Tables 4.8, 4.9, and 4.11) point toward a simple pre-OJ system. I have already discussed this issue at length in Miyake (forthcoming), where I conclude that Serafim’s system is the best one currently available. Serafim (1999b) derives the six-vowel PJ system of Serafim (1999a) from an eight-vowel PKJ system (Table 4.12). For a detailed accounts of the stages bridging Serafim’s PKJ and PJ systems, see Serafim (1999b: 4–10). Since both internal and external reconstructions (with the exception of Hattori’s) indicate a small PJ vowel system, hereafter I will not distinguish between PJI and PJE and will refer only to PJ without reference to methodology. How did the OJ ‘eight vowels’ develop from the simpler six or so vowels of PJ? Two of the OJ ‘eight vowels’ directly reflect the original four PJ vowels: (4.13) PJ *a *R

> >

OJ a o (rewritten according to Yale romanization conventions)

Mid non-central vowels merged with their high counterparts. (4.14) PJ *i, *e > *u, *o >

OJ yi (rewritten according to Yale romanization conventions) u 79

OLD JAPANESE

Table 4.13 OJ reflexes of PJ diphthongs *-a *a*i*u*R-

*i(C)a > OJ ye *u(C)a > OJ wo

*-i

*-R

*a(C)i > OJ ey

*i(C)R > OJ ye *u(C)R > OJ wo

*u(C)i > OJ iy *R(C)i > OJ iy

I do not intend to imply here that the phonetic values of the PJ vowels were also preserved in OJ. In theory, OJ o could have been *[o] or *[ö] rather than *[R]. The other four OJ ‘vowels’ (iy, ye, ey, wo) derived from PJ diphthongs (Yamaguchi 1971, Whitman 1985: 60–64, Martin 1987: 57–64, Unger 1993: 48). These combinations of PJ vowels may have been original and/or the results of consonantal loss (e.g., *VCV > *VV). I hesitate to say that PJ vowel sequences ‘monophthongized’ into OJ ‘vowels’ since we cannot be sure at this point if all of the so-called OJ ‘eight vowels’ were really vowels. Some of the OJ ‘eight vowels’ may actually turn out to be diphthongs (possibly preserved from PJ) or combinations of vowels and glides. Recall that the Yale romanization digraphs for these ‘vowels’ are merely conventional and may not bear any resemblance to their actual pronunciations: e.g., wo might have been *[o], *[ow], *[u], etc., rather than *[wo]. I list the least controversial derivations proposed for OJ iy, ye, ey, and wo in Table 4.13. The evidence for monophthongization comes from the internal reconstruction of OJ. A number of alternations in noun forms are explainable if one posits a suffix -(C)i added to free forms and absent in bound forms (Martin 1987: 62–64, 435, 517, 554): (4.15) OJ ey < *a(C)i ‘rice wine’ OJ sakey OJ saka-

(free form) (bound form)

<
> >

B-type B-type B-type B-type 82

*tiy *diy *niy *riy

> > > >

OJ C-type OJ C-type OJ C-type OJ C-type

ti di ni ri

JAPANESE PHONOLOGY THROUGH TIME

Thus OJ C-type ti, di, ni, ri were at one time syllables with the B-type ‘vowel’ iy. If PJ *ia had become *ya before becoming OJ ye, then according to (4.27), PJ *tia, *dia, *nia, *ria would become *tya, *dya, *nya, *rya and result in OJ se, ye, ye, ye. Thus OJ te, de, ne, re must have come from syllables exempt from rule (4.27): (4.29) PJ *ta(C)i PJ *nta(C)i PJ *na(C)i PJ *ra(C)i

> > > >

B-type B-type B-type B-type

*tey > *dey > *ney > *rey >

OJ OJ OJ OJ

C-type C-type C-type C-type

te de ne re

Does all of the above mean that we should treat the ‘vowels’ in ti, te, di, de, ni, ne, ri, and re as B-type in OJ? And what about other OJ syllables with C-type i and e? Even if we know that a given C-type ‘vowel’ descends from an earlier A- or B-type ‘vowel’ via internal reconstruction, I do not think that we should jump to any conclusions about that ‘vowel’ in OJ, since the diachronic origin of a ‘vowel’ may not necessarily help us determine the synchronic status of a ‘vowel’: in theory, a pre-OJ A-type ‘vowel’ could have become an OJ B-type ‘vowel,’ or vice versa. Perhaps OJ C-type i and e could be phonemically identified with other OJ A- or B-type ‘vowels’ (yi, iy, ye, ey), enabling us to avoid positing separate phonemes for i and e, but whether i and e were phonetically identical to any of those four vowels (yi, iy, ye, ey) would still be open to question. We do not even know whether all syllables with C-type i or C-type e really had the same ‘vowel.’ What if ti C had B-type iy but si C had A-type yi? And what if teC had B-type ey but seC had A-type ye? On pages 238 and 244, I will examine the OJ phonogram data to determine whether a C-type ‘vowel’ i or e in any given syllable is phonetically like or unlike another OJ ‘vowel.’ (4) Ignoring the controversy over an A/B-type distinction for OJ po, OJ A-type wo and OJ B-type o occur after all initials except b, w, and zero. (5) Again ignoring the controversy over an A/B-type distinction for OJ po, OJ C-type o is in complementary distribution with A-type wo and B-type o, occurring only wherever they do not occur: after b, w, and zero. One is tempted to analyze phonemically the C-type o in OJ bo, wo, and o as either an A-type wo or a B-type o so that C-type o can be eliminated entirely. But despite such a temptation, one must wait until I examine the OJ phonogram data on p. 250 to determine whether such reanalyses are possible. As with C-type i and e above, even if one knew the historical origins of the C-type ‘vowel’ in OJ bo, wo, and o, one cannot rely exclusively on that knowledge to classify that vowel synchronically. Also, even if a C-type o in any of those three syllables could be identified phonemically with A-type wo or B-type o, there is no guarantee that it was phonetically identical to either of those vowels. 83

OLD JAPANESE

Table 4.15 The OJ ‘vowel’ inventory Front?

Central?

*yi *iy *ye *ey

High? Mid? Low?

*o *a

Back?

Unknown

*u *wo

*i *e

*o

Putting aside the messy C-type ‘vowels’ i, e, and o which might or might not be phonetically if not phonemically distinct from all others, I conclude that OJ had eight ‘vowels’ or, strictly speaking, ‘finals,’ a term which encompasses all of their possible manifestations (monophthong, diphthong, glide-vowel, vowel-glide, etc.). Notice the question marks in Table 4.15. The placement of the ‘vowels’ in this chart is based purely on the phonetic attributes (right or wrong) suggested by the Yale romanization of OJ and has no real value. I cannot overemphasize the fact that the Yale romanization is conventionalized. yi may look like *[yi], but in theory it could be a mid *[e] or even a diphthong *[ai]. These ‘vowel’ symbols are more like algebraic symbols than any sort of phonemic or phonetic notation. One could even replace these symbols with more neutral symbols (e.g., 䊊, 䊐, etc.) not suggesting any phonetic values at all, but this would be too awkward. I will supply the phonetic values for these symbols by analyzing OJ phonogram transcriptions in Chapter 8. Post-OJ vowels The ‘vowel’ distinctions of OJ in Table 4.15 did not last long. EMJ of the Heian Period had only five vowels derived from mergers of the OJ ‘vowels’ (Figure 4.1). This figure is potentially misleading, as it may seem to imply that (a) the A- and B-type ‘vowels’ merged into the C-type ‘vowels’ and (b) OJ a, i, u, e, o survived intact into EMJ. Again I must stress that the OJ ‘vowel’ notations are algebraic in nature. The OJ ‘vowels’ symbolized by a, i, u, e, o in Yale romanization may or may not have been phonetically identical to EMJ /a i u e o/. Without looking at other evidence for OJ and EMJ vowels, I cannot rule out the possibility of phonetic changes in individual ‘Vowel’ type OJ

EMJ

A B C

A B C

A B C

a

yi iy i

u

ye ey e

wo o o

/a/

/i/

/u/

/e/

/o/

Figure 4.1 Mergers of the OJ ‘vowels’

84

JAPANESE PHONOLOGY THROUGH TIME

vowels or even a vowel shift between the two stages, though it would be simplest to assume some degree of continuity. Unfortunately, there is no definite evidence for the pronunciation of the five EMJ vowels, but foreign transcriptions do provide evidence for the pronunciation of their LMJ reflexes. In the Portuguese missionaries’ romanization, the five basic LMJ vowels appear as follows (Matsumura 1972: 143–145): (4.30) LMJ Romanization

/a/ a

/i/ /u/ i u

/e/ /o/ e o

In the Korean textbook Chephay sine (1676), we find the following hankul spellings of the five basic early seventeenth-century12 Japanese vowels (Martin 1987: 22–23): (4.31) LMJ Hankul transcription Pronunciation of hankul transcription

/a/ a [a]

/i/ /u/ /e/ /o/ i u ye(y) wo [i] [u] [yR(y) ] [o]

Spellings of LMJ /e/ with -y- in Chephay sine (and in the hankul transcriptions of LMJ in the Ilopha of 1492; Matsumura 1972: 143–144) may lead one to suspect that perhaps LMJ /e/ was phonetically [ye]. However, since the Korean of the time lacked a [e], I think that the hankul spellings reflect attempts to render LMJ [e] with Korean ye [yR] or yey [yRy]. (The [e] of modern Korean did not yet exist.) The -y- of the hankul spellings may also reflect palatalized allophones of consonants preceding LMJ /e/: Chephay sine transcriptions such as tye [tyR] may represent LMJ /te/ [tye], etc. If LMJ /e/ were really [ye] with a glide [y], one would expect to see a -y- after e in the Portuguese romanizations. However, only LMJ ye is attested; there are no spellings with C-y clusters such as pye, tye, kye, etc. In any case, the five basic LMJ vowels were similar to the five basic MSJ vowels: (4.32) LMJ MSJ

/a/ /i/ /u/ [a] [i] [u] /a/ /i/ /u/ [a] [i] [T]

/e/ [e] ( [ye]?) /e/ [e]

/o/ [o] /o/ [o]

The basic Japanese vowels have remained stable since EMJ. I use the term ‘basic’ to refer to short vowels. LMJ and MSJ have long vowels which developed from EMJ vowel sequences in native and Sino-Japanese words. Since these secondary long vowels are irrelevant to OJ reconstruction, I will cite only a few example derivations below from Martin’s exhaustive listing of long vowel sources (1987: 45–47). 85

OLD JAPANESE

(4.33) MSJ aa [aa]13 MSJ ii [ii] MSJ û [uu] MSJ ei [ee] MSJ ô [oo] MSJ ô [oo]

< < < < <
OJ w, y) and PJ nasal-obstruent clusters had become OJ voiced obstruents (PJ *np, *nt, etc. > OJ b, d, etc.). Foreign transcriptions show that LMJ reflexes of OJ voiced obstruents were prenasalized. Some modern dialects today still have prenasalized reflexes. Two of the ‘eight vowels’ of OJ (a, o) were carried over from PJ (*a, *R). Two more resulted from PJ mergers of mid and high vowels (OJ i, u < PJ *i, *e, *u, *o). The other four ‘vowels’ of OJ (iy, ye, ey, wo) originated from PJ diphthongs (*VV) which in turn may have originated from the loss of intervocalic consonants (*VCV > *VV). OJ also had three indeterminate C-type ‘vowels’ (i, e, o) which may or may not turn out to be phonemically and/or phonetically identical to any of the other OJ ‘vowels.’ The eight (or more?) ‘vowels’ of OJ collapsed into a simpler five-vowel system in EMJ. MSJ today still has the five basic vowels of EMJ. Owing to the elusive nature of the OJ ‘vowels,’ nothing can be said about OJ syllable structure apart from the following vague formulation: (4.34) *C + ‘vowel’ The OJ ‘vowels’ may not necessarily be single short vowels. Some OJ ‘vowels’ may actually be long vowels, diphthongs, and/or combinations of vowels and glides. Thus in theory OJ could have had any of the following syllable structures: (4.35) *CV, *CV(V), *C(G)V, *CV(G), *C(G)V(G), etc. (G = a glide; i.e., OJ w or y) 86

JAPANESE PHONOLOGY THROUGH TIME

I will propose my own formula for OJ syllable structure in Chapter 9 on the basis on my analysis of OJ phonogram transcriptions in Chapters 7 and 8. I have devoted this chapter exclusively to (a) internal and comparative evidence for pre-OJ segmental phonology and (b) philological and modern evidence for post-OJ segmental phonology. In this chapter, I have struggled to keep my assumptions about OJ phonology to a minimum in order not to ‘contaminate’ the above evidence. Projecting my assumptions about OJ into internal and comparative reconstructions of pre-OJ and then using such reconstructions as evidence for my reconstruction of OJ would be circular. Although I have cited my own reconstruction of OJ in Chapters 1–3, such citations will not result in circularity, since I will not use them as evidence when I reconstruct OJ in future chapters. Chapters 1–3 were introductory in intent and were peripheral to my argument. Even if one eliminated all of my OJ reconstructions from previous chapters, one would find that this would not affect the validity of the claims in those chapters. For example, it is simply a fact that the scribes of the alpha section of Shoki transcribed OJ b, d, z, and g with various kinds of Chinese obstruents, regardless of how one reconstructs OJ b, d, z, and g (Table 3.10). Note that I formulated all previous claims made about OJ using the theory-neutral Yale romanization with my own reconstructions occasionally added only for reference. I will continue to maintain a neutral stance about OJ phonology by using the Yale romanization symbols for OJ consonants and ‘vowels’ as algebraic variables as opposed to phonemic or phonetic symbols. I will attempt to reconstruct the values of these variables on the basis of Sino-Korean, SinoVietnamese, and reconstructed Chinese readings of OJ phonograms in Chapters 7 and 8. In the next chapter, I will discuss this non-Japanese evidence for OJ phonology in depth.

Notes 1 Vovin (1996a) hypothesizes that Korean and Japonic do not form a subgroup within Altaic owing to a lack of shared innovations. 2 Miller (1967: 191) claims that Suiko Period Japanese had a six-vowel system distinct from the eight-vowel system which he reconstructs for Nara Period Japanese (= OJ), but does not explain how he reconstructed this Suiko Period system or its Pre-Suiko predecessor with only five vowels. I will ignore these proposals of Miller’s, for I find the evidence for smaller vowel systems to be ambiguous in the System-B- and C-based orthographies. 3 PJ *z is in Whitman’s original formulation of this rule. I have placed it in parentheses since I have rejected the reconstruction of PJ *z. 4 Ramsey and Unger (1972) propose pre-OJ *z and *g. I have placed these protophonemes in parentheses since I have already argued against them. 5 Martin (1987: 79) notes that “attestations indicate we > ye directly,” but I agree with him that “it seems more natural to assume that we > e at some point before e > ye, even if the transition period were short (i.e., the labial onset [w] was lost and then the palatal glide [y] developed).”

87

OLD JAPANESE

6 Erickson (1998) argues against positing such abstract phonemes. 7 The -ne- in toneri ‘palace servant’ is a C-type syllable in OJ, but its etymology shows that at one time it may have been a B-type syllable *-ney- < *-no-i-. 8 I will argue in Chapter 8 that the OJ C-type syllable o with an ‘indeterminate’ (i.e., non-A/B-type) vowel was actually a B-type syllable o. 9 This figure includes textual variants. 10 However, Unger (1993: 24) reconstructs OJ /wu/. I will test this reconstruction on pp. 194–6. 11 PJ *Z is in Whitman’s original formulation of this rule. I have placed it in parentheses since I have rejected the reconstruction of PJ *z for reasons given earlier. I will ignore this *z in my discussion of (4.27). 12 Chephay sine was written in the early seventeenth century by Kang Wusen but was published posthumously in 1676 (Martin 1987: 22). 13 Note that the Hepburn romanization used in this book for MSJ does not consistently transcribe phonetic long vowels. Most long vowels are written with macrons (replaced with circumflexes in this book), but some long vowels are written as digraphs: for example, [baai] ‘circumstances,’ [ii] ‘good,’ [eigo] ‘English’ are written as baai, ii, and eigo, not *bâi, *î, and *êgo.

88

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

5 OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

Overview As shown in Chapter 2, the most reliable written evidence for OJ consists of transcriptions in sinographs used for their phonetic values (i.e., phonograms). In order to interpret the phonograms of an OJ text properly, one cannot simply read the phonograms using modern Chinese, Sino-Japanese, or SinoKorean pronunciations, though I have shown on pp. 18 and 32 that one might be able to crudely approximate the pronunciation of a poem by reading its phonograms in Go-on or Kan-on. One needs to know which system of pronunciations was the basis for that text’s orthography. In Chapter 2, I demonstrated that the Japanese were exposed to at least four (if not more) varieties of Chinese pronunciation (Systems B through E) up to the eighth century. The poetry of Kojiki is written in an orthography based mostly on System D, which apparently was Early Middle Chinese (EMC) filtered through some variety of Sino-Korean (probably late SinoPaekche). The poetry of Shoki, on the other hand, was written in an orthography based mostly on System E, which was Chang’an Late Middle Chinese (CLMC). To complicate matters further, a few phonograms in Kojiki appear to be carryovers from an earlier orthography based on System C, which apparently was Late Old Chinese (LOC) filtered through some variety of Sino-Korean (probably middle Sino-Paekche). Shoki also has its share of orthographic retentions from the older System-D-based orthography. Ideally, we would have to know the LOC, EMC, CLMC, late SinoPaekche, and middle Sino-Paekche readings of phonograms to reconstruct OJ. Unfortunately, we do not have indisputable evidence for those readings. We must rely on reconstructions of readings based on more indirect evidence. I write these reconstructions with asterisks to distinguish them from attested forms. As we shall see later in this chapter, the extant reconstructions of premodern Chinese and Sino-Paekche are problematic to various degrees. Furthermore, even the best reconstructions are always subject to revision when new evidence is discovered or old evidence is reanalyzed. 89

OLD JAPANESE

I believe that it is dangerous to rely entirely on reconstructions because of their secondary and transient nature. Hence I prefer to rely more on hard, attested evidence: the Sino-Korean and Sino-Vietnamese readings of the phonograms and the readings of phonograms in the premodern Vietnamese nôm writing system. Although Vietnamese may seem completely irrelevant to the problem of OJ reconstruction, I will show on p. 117 that Vietnamese is far from irrelevant. This chapter will briefly (1) discuss extant sources and reconstructions of OC, EMC, and CLMC and (2) describe the attested and lost varieties of Sinoxenic. I will cover only the major points necessary to understand my arguments for the reconstruction of OJ in Chapters 7 and 8. Old Chinese (OC) ‘Old Chinese’ (OC), also called ‘Archaic Chinese’ by Karlgren and earlier generations of scholars, is an extremely vague term. In the broadest sense it can refer to any variety of Chinese up to c. 600 CE. In this book, I will generally avoid referring to OC as a whole and will refer to specific stages using Starostin’s (1989) terminology. ‘Western Han’ (WH) and ‘Eastern Han’ (EH) are named after the Western and Eastern Han Dynasties (202 BCE–8 CE and 25–220 CE). ‘Late Old Chinese’ (LOC) or ‘Postclassical Old Chinese’ is the language of the third through the fifth centuries CE. Starostin (1989: 476–502) divides LOC into three stages: Early LOC (ELOC, c. the third century CE), Middle LOC (MLOC, c. the fourth century CE), and Late LOC (LLOC, c. the fifth century CE). Unlike other scholars who draw the line between OC and Middle Chinese at c. 600 CE, he considers early sixth-century CE Chinese to be ‘Early Middle Chinese’ (EMC), not OC. Early Middle Chinese (EMC) Most scholars speak of ‘Middle Chinese’ (MC), not ‘Early Middle Chinese’ (EMC) or ‘Late Middle Chinese’ (LMC). (Karlgren and earlier scholars used the term ‘Ancient Chinese’ rather than ‘Middle Chinese.’) MC in the broadest sense refers to the Chinese of the Sui Dynasty (581–618), the Tang Dynasty (618–907), and the Five Dynasties period (907–960). MC in the narrowest sense refers to the ‘language’ recorded in the Qieyun (Cut Rhymes) dictionary of 601 CE (p. 93). In this latter sense, MC is equivalent to one sense of the term EMC as used by Pulleyblank (1984, 1991a). In this book, I will use the term MC to refer to EMC and LMC ( p. 91) as a whole. Otherwise, I will use the more specific terms EMC and LMC. I have already mentioned that Starostin (1989) considers pre-Sui Chinese of the early sixth century to be EMC, not LOC. Pulleyblank (1991a: 2) would seem to agree with Starostin; he defines EMC as “a single, coherent, form of 90

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

the Chinese language, namely the elite standard common to educated speakers from both north and south in the period of division that came to an end with the Sui reconquest of the south in 589.” Following their example, I too will regard pre-Sui sixth-century Chinese as EMC, though I have certain reservations concerning Pulleyblank’s precise definition which I will express on p. 93. Starostin (1989) calls the ‘language’ of Qieyun of 601 ‘MC’ rather than ‘EMC,’ but Pulleyblank (1991a: 2) prefers to call it EMC, and I will follow Pulleyblank’s usage. Late Middle Chinese (LMC) The term ‘Late Middle Chinese’ (LMC) is almost unique to the works of Pulleyblank (1970 –71, 1984, 1991a), though a few other scholars such as Baxter (1992) have also used it. Pulleyblank (1991a: 2) defines LMC as “a new standard based on the speech of the Sui-Tang capital, Chang’an.” He believes that the sound tables of the Yunjing (Mirror of Rhymes) were based on LMC. Since I have some qualms about these views (p. 95), I will define LMC somewhat more broadly as the language of the Tang Dynasty (618– 907) and the Five Dynasties period (907–960). Although Pulleyblank refers to the Chang’an dialect of LMC simply as ‘LMC,’ I will use the term ‘Chang’an LMC’ (CLMC) to distinguish it from other dialects of LMC which lacked ‘denasalization’ ( p. 31).

Internal sources for earlier Chinese Internal sources for Old Chinese Although Chinese had a “fully developed writing system” as early as c. 1300 BCE (Norman 1988: 58), one cannot simply look at early inscriptions and determine how early OC was pronounced. Even though the vast majority of sinographs have been phonetic from the very beginning, they convey phonetic information in a somewhat oblique manner. For example,
Mandarin hfi [xay] ‘sea’ is a phonetic compound consisting of a semantic indicator ‘water’ on the left and a phonetic indicator
mgi ‘every’ on the right. The similarity of the two graphs tells us that the two morphemes must have been pronounced similarly in early OC, despite the highly divergent modern pronunciations. A number of scholars beginning with Bernhard Karlgren have attempted to reconcile these present-day ‘mismatches’ through internal reconstruction, resulting in closer ‘matches’ in OC: e.g., Karlgren (1957: 249) reconstructs
‘sea’ as OC *xmRg and
‘every’ as OC *mwRg. The trouble is that without external evidence ( pp. 98 and 135), there is no way to confirm these attempts at reconciliation. If some non-Chinese language had borrowed these 91

OLD JAPANESE

OC words as, say, xamag ‘sea’ and mag ‘every,’ we would know that
‘sea’ had an OC *x- and that
‘every’ did not, but otherwise we cannot tell exactly how
‘sea’ differed from
‘every’ in OC. Unlike Karlgren, Starostin (1989: 549) reconstructs
‘sea’ as OC *smRRq with OC *s- rather than OC *x-; Starostin’s OC *sm- > EMC *h- (equivalent to Karlgren’s EMC *x-). Unless we can find external evidence, we can test whether Karlgren or Starostin is right only on grounds such as simplicity and typological probability. Every reconstruction of OC deals with the ‘phonetic mismatch’ problem in somewhat different ways. Apart from an agreement that OC had consonant clusters lacking in later stages of the language, there is no consensus view on OJ consonants. Each reconstruction has its own inventory of consonants and consonant clusters. The situation with OC rhymes is somewhat better. Qing Dynasty (1644– 1911) scholars rediscovered the rhyme categories of the OC poetic anthology Shijing (Classic of Poetry) which differed considerably from those of MC poetry. Modern linguists have accepted these Shijing rhyme categories with some modifications in their OC reconstructions. The OC
‘genitive’ rhyme repeatedly mentioned in Chapter 2 is one of those rhyme categories. This rhyme has been variously reconstructed as *-Rg (Karlgren 1954, Li Fang-kuei 1971), *-R (Starostin 1989), *-Rò (Pulleyblank 1991b; see p. 15), and *-t (Baxter 1992). Notice the lack of agreement in these reconstructions about whether the vowel was high (Baxter) or mid (the others) and whether there was a final consonant (Karlgren, Li, Pulleyblank) or not (Starostin, Baxter). Without examining external evidence (pp. 98 and 135), it is difficult to test which of these proposed reconstructions is correct. The disagreement over OC rhymes is as great as or even greater than the disagreement over OC consonants. At one extreme, Karlgren (1954) reconstructed a 14-vowel ‘system’ for OC as complicated as his 15-vowel ‘system’ for MC in Figure 3.3. At the other extreme, Pulleyblank (1963) advocated a two-vowel system (*a, *R) for OC. Most efforts toward OC reconstruction have focused on the ‘language’ of the Shijing. I write ‘language’ in quotation marks because it is doubtful that the 305 poems in Shijing compiled circa the sixth century BCE represent a single variety of Chinese. The poems span approximately five centuries (Norman 1988: 42) and their anonymous authors could have spoken different dialects. Although it is true that all the poems generally adhere to the same system of rhyming, one finds many ‘inexact’ rhymes if one assumes that all the poems used precisely the same system of rhyming. And even if all the poems did use one system of rhyming, this still does not guarantee linguistic homogeneity. Different dialects could have different phonetic values for a given rhyme class. Fortunately, we do have post-Shijing OC poetry the authors of which are known, and we can use the rhyming patterns of individual authors to 92

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

reconstruct later stages of OC. Ting (1972) and Starostin (1989) reconstructed LOC rhymes on the basis of their analyses of rhyming in LOC poetry. Although a number of dictionaries of LOC were compiled during the Six Dynasties period (Pulleyblank 1984: 133), none of these dictionaries has survived, so we must infer LOC rhyme categories from LOC poetry. Internal sources for Early Middle Chinese I have already mentioned that EMC in the narrowest sense of the term refers to the ‘language’ of Qieyun (Cut Rhymes) dictionary, compiled by Lu Fayan (also known as Lu Ci) in 601 CE. Although the original Qieyun is lost, Qieyun fragments and copies of later recensions have survived. The most important of these recensions is the Guangyun (Expanded Rhymes) of 1008. Despite its date, Guangyun is a dictionary of EMC rather than the language of the early Song Dynasty (960–1279) and has been used as a substitute for the lost original Qieyun. I will cite rhymes and readings from Guangyun rather than Qieyun in this book. What Qieyun was intended to represent is open to question. I have already mentioned that Pulleyblank (1991a: 2) believes that Lu Fayan intended to codify “a single, coherent, form of the Chinese language.” Karlgren (1954: 212) identifies this language with the Chang’an dialect of Ancient Chinese (our EMC), but later scholars no longer hold this view. Pulleyblank, for example, does not think that Qieyun represented Chang’an EMC because Chang’an was “a cultural backwater during the time of division [i.e., before the reunification of China by the Sui]”; he claims that the language of Qieyun “went back to the dialect of Luoyang in the second and third centuries” (1991a: 2). While Karlgren and Pulleyblank thought that Qieyun represent a single EMC language, many other scholars such as Norman and Coblin (1995) disagree. They regard the ‘language’ of Qieyun as a “maximally differentiated compromise between the reading traditions of all three areas, further complicated by the inclusion of distinctions attested in various earlier rime dictionaries” (1995: 578). They deny that the Qieyun ‘language’ is “a language in any common sense of the term” (1995: 580). What did Lu Fayan himself have to say about what his dictionary was meant to represent? In the preface to Qieyun, he explains that the dictionary resulted from his discussions of phonology with other scholars at his home in the late sixth century CE. They found what they considered to be inconsistencies in previous LOC rhyme dictionaries (see above) and the past and present rhyming practices in the north and south. Lu hoped to resolve these inconsistencies with Qieyun. It is clear from Lu’s own testimony that the ‘language’ of Qieyun is an artificial construct. Does this mean that Qieyun is of no value for EMC reconstruction? Norman and Coblin (1995) take a very dim view of Qieyun, but Pulleyblank (1984: 134) does not think that the Qieyun ‘language’ was wholly artificial: 93

OLD JAPANESE

One can draw a good analogy to the situation Lu Fayan faced in compiling the Qieyun from modern English, in which there are also several mutually intelligible competing norms . . . The Qieyun gives a maximal diasystem, and few people may have preserved all its distinctions in their speech at the time it was published. If Pulleyblank is correct about Qieyun, then its ‘language’ is as artificial as the language of a dictionary of English or any other standard language. No one really speaks the languages in dictionaries, but that does not mean that those dictionaries have no basis in reality. Lu Fayan did not invent any phonological distinctions. He only recorded existing distinctions known to him and his friends. Can Pulleyblank’s claims about the distinctions of Qieyun be substantiated? Using the comparative method, Karlgren (1954) has demonstrated that the vast majority of distinctions in Qieyun can be reconstructed on the basis of sinograph readings from modern Chinese languages and Sinoxenic. Conversely, modern Chinese languages and Sinoxenic make very few distinctions which are not reflected in Qieyun. The history of Chinese phonology on the whole consists of a long series of mergers. Most of the non-Qieyun distinctions are in the conservative Min dialects, which apparently split off from the rest of Chinese before the composition of Qieyun. If the ‘language’ in Qieyun were an arbitrary collection of old and new pronunciations from various regions, one would not expect this to be the case. One would expect to find several distinctions which cannot be verified via the comparative method, not just one or two. And one might expect to find some distinctions attested in modern Chinese languages and Sinoxenic to be missing. There are clear, regular correspondences between the initials and rhymes recorded in Qieyun and the initials and rhymes of modern Chinese languages and Sinoxenic. Such clarity and regularity in correspondences would be impossible if Lu Fayan had listed one graph under its thirdcentury rhyme in the south, another graph under its sixth-century rhyme in the north, and so forth. I conclude that we can use the ‘language’ of Qieyun only as an approximation of EMC. Unlike Karlgren and Pulleyblank, I do not think that Qieyun reflected any specific variety of EMC. Lu Fayan never said that his intent was to record the Chang’an EMC assumed by Karlgren (1954: 212) or Pulleyblank’s (1991a: 2) “elite standard common to educated speakers from both north and south.” Nevertheless, the ‘language’ of Qieyun is obviously not entirely artificial. It must have been based on real varieties of EMC. How was this ‘language’ recorded in Qieyun? Qieyun (and Guangyun) was organized by rhymes. Within each rhyme, graphs were organized into homophone groups. Each of these groups in turn was supplied with a fanqie (‘reverse cutting’) spelling. For instance, the homophone group containing
‘know’ had the fanqie
. This means that
‘know’ and its 94

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

homophones were pronounced with the ‘initial’ of
‘ascend’ and the ‘final’ of
‘separate.’ ‘Initials’ are generally initial consonants, though sometimes they can be initial consonant–glide combinations. ‘Finals’ are the remainder of the syllable: medial glides, vowels, and codas. Pulleyblank (1991a) reconstructs
‘ascend’ as *trik and
‘separate’ as *liR, so
‘know’ is pronounced as *tr- + -iR = *triR [âiR].1 By examining the graphs used as ‘initial spellers’ and ‘final spellers’ in the fanqie of Qieyun, one can determine how many ‘initials’ and ‘finals’ were present in EMC. However, as Norman (1988: 28) writes, “Nothing found in the Qieyun or its descendants [such as Guangyun] tells us anything about the phonetic values of these categories. They represent form without substance; the substance must be supplied from elsewhere.” Karlgren and others have turned to modern Chinese languages and Sinoxenic to supply that substance. Their EMC reconstructions combined the comparative method with the phonological categories of Qieyun. Qieyun and its Song recension Guangyun are not the only sources of EMC, though they definitely overshadow all others. We can also confirm the rhyme categories of Qieyun by examining EMC poetry. Further EMC fanqie spellings are found in other sixth- and seventh-century works such as the fragments of the Yupian dictionary (Jade Book; 544), Lu Deming’s glosses for the Chinese classics in Jingdian shiwen (Explanations of the Classics; c. early seventh century), and Xuanying’s glosses for the Buddhist canon in Yiqiejing yinyi (Sounds and Meanings in the Buddhist Canon; c. 650 CE). Internal sources for Late Middle Chinese Pulleyblank (1991a: 3) claims that “we have even better evidence for its [LMC’s] phonological categories than we have for EMC.” However, even Pulleyblank (1991a: 3) admits that none of the LMC dictionaries which were compiled such as Yuan Tingjian’s Yunying (Flower of Rhymes; c. 750) has survived. However, although we lack an LMC equivalent of Qieyun or Guangyun which lists pronunciations (albeit in fanqie form) of an entire ‘language,’ we can rely on four types of Chinese evidence for LMC reconstruction. The first type of evidence comes from Chinese languages. Many ‘literary’ readings in Chinese languages are ‘inner Sinoxenic’ loans from Chang’an LMC or other prestigious varieties of LMC. In theory, one could try to reconstruct LMC on the basis of the comparative method using these readings alone (or in combination with LMC-based Kan-on, Sino-Korean, and Sino-Vietnamese), but this has never been attempted. Karlgren (1954) did use the comparative method on these LMC-based readings, but he also used data from EMCbased Go-on and Min readings, resulting in a ‘contaminated’ reconstruction which he believed to be the ‘Ancient Chinese’ (our EMC) of Qieyun. The remaining types of evidence are philological. The second type of evidence consists of fanqie in works such as Huilin’s Yiqiejing yinyi (Sounds 95

OLD JAPANESE

and Meanings in the Buddhist Canon, not to be confused with Xuanying’s earlier work of the same title; c. 800). The third type of evidence comes from the rhyming of Tang poets such as Li He (791–817) and Bai Juyi (772–846). The fourth type of evidence is unique to LMC and needs to be explained in some detail. There are a number of works called ‘rhyme tables’ which date from the mid-Song Dynasty (960–1279). The most important of these is Yunjing (Mirror of Rhymes), whose original author and date are unknown; the earliest known text has prefaces from 1161 and 1203 (Pulleyblank 1984: 257). Pulleyblank believes that Yunjing provides a blueprint for LMC phonology, but I am skeptical of this claim for reasons I will soon give below. Yunjing appears to be a syllabary of some sort of Chinese. Like Qieyun, it divides syllables into initials and finals. It classifies initials into seven classes according to place of articulation which are then subdivided into four types: 1 2 3

4

‘clear’ (qing; voiceless unaspirated) ‘second clear’ (ciqing; voiceless aspirated) ‘muddy’ (zhuo; usually assumed to be voiced, but voiced aspirated or murmured according to Pulleyblank 1984: 67 or with “voiceless onset with voiced aspiration or murmur at their release” according to Pulleyblank 1991a: 6) ‘second muddy’ (cizhuo; sonorants)

There are two types of ‘lip sounds,’ ‘heavy’ (zhong) and ‘light’ (qing). ‘Heavy lip sounds’ are bilabials (*p, *ph, *p®, *m) and ‘light lip sounds’ are labiodentals (*f, *f®, *v). Thus we find the following classification for ‘[back] tooth sounds’ (velars) in Yunjing: (5.1) Clear: Second Clear: Muddy: Second Muddy:

*k *kh *k® *Y

(voiceless unaspirated velar stop) (voiceless aspirated velar stop) (velar stop with voiceless onset and voiced aspiration) (velar nasal)

The Yunjing does not tell us exactly what is meant by, say, a ‘clear [back] tooth sound’. These terms are opaque without reference to concrete phonetic data outside Yunjing. For example, we know that ‘clear sounds’ are voiceless aspirates because graphs listed in Yunjing with ‘clear sounds’ are read with voiceless aspirates in modern Chinese languages. Yunjing classifies finals in a number of ways: (1) The rhymes of the EMC (!) dictionary Qieyun. (2) ‘Inner [turning]’ (nei zhuan) and ‘outer [turning]’ (wai zhuan). These terms are somewhat controversial. Luo Changpei (1933b) has interpreted 96

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

‘inner’ as having a high vowel and ‘outer’ as having a low vowel. Pulleyblank (1984) accepts this interpretation, though there are some cases that appear to be exceptional. Thus the LMC final *-i with a high vowel is ‘inner’ and the LMC final *-aa with a low vowel is ‘outer.’ (3) ‘Open [mouth]’ (kaikou) and ‘closed [mouth]’ (hekou). These terms are less controversial. Generally speaking, ‘closed’ rhymes contain labial elements (*-u- and/or *-w- depending on reconstruction) but ‘open’ rhymes lack them. Hence the LMC final *-i without *-w- is kaikou and the LMC final *-wi with a *-w- is hekou. There are a few problems with this definition, but I will not deal with them here. (4) The ‘Four Grades’ (si deng). This term does not appear in Yunjing itself, but Yunjing does make the four-way Grade distinction made in later rhyme tables. The interpretation of the Grades is extremely controversial. Pulleyblank (1984: 75) reconstructs the Four Grades in terms of degrees of palatalization:2 (5.2) Grade I: Grade II: Grade III: Grade IV:

no *i or *y *-y*-i*-yi-







LMC LMC LMC LMC

*kaw *kyaaw *kiaw *kyiaw

‘high’ ‘associate’ ‘tender’ ‘pour water’

Although the phonological reality of the Four Grades of Yunjing can be demonstrated with modern language data, other aspects of Yunjing are dubious. One can confirm the existence of the initials of Yunjing through the comparative method (Karlgren 1954: 218–230). However, the finals of Yunjing are another matter. As noted earlier, Yunjing organizes syllables by the EMC rhymes of Qieyun. The first preface of Yunjing is from 1161. This does not mean that the body of Yunjing itself is from 1161. On the other hand, close examination of the list of initials in the Yunjing’s preface shows that it could not possibly be from the same period as Qieyun (i.e., c. 601 CE). Yunjing’s list of initials contains ‘light lip sounds’ (labiodentals) lacking in Qieyun, the pre-LMC-based layers of Old Sino-Vietnamese (OSV), and the pre-LMC layers of southern Chinese languages (Pulleyblank 1984: 149). Coblin (1994: 60) believes that labiodentals developed in what he calls “SuiTang Chang’an,” the Chang’an dialect circa the seventh century CE. Yan Shigu (581–645 CE) distinguished between ‘light’ and ‘heavy’ labial initials in the fanqie glosses of his commentary to the Han shu (History of the Han), completed in 641 CE (Coblin 1994: 34–35). Perhaps labiodentals date as far back as the sixth century. Although scholars disagree about the details of the development of labiodentals, all agree that they are innovations and not retentions from pre-LMC. I conclude that the Qieyun rhymes in Yunjing do not reflect the rhymes of the language of Yunjing’s creator. Recall that Guangyun, the expanded redaction of Qieyun (p. 93), appeared in 1008, roughly 150 years before the 97

OLD JAPANESE

first preface of the oldest known edition of Yunjing. Guangyun preserved the rhymes of Qieyun only with minor changes. This is not surprising because EMC-based rhyme sets explicitly listed in Guangyun were still in use in during the Song Dynasty. Yunjing, like Guangyun, simply maintained rhyme categories that were culturally prestigious but phonologically obsolete by the eleventh century. The author of Yunjing probably intended to present the ‘language’ of Qieyun in table form, but he may not have realized that Qieyun did not distinguish between ‘heavy’ and ‘light’ labials. Thus he unintentionally created a composite non-language that mixed EMC and LMC elements. Although Yunjing’s Four Grades, 36 initials, and finals do correspond to distinctions in modern sinograph readings, it is improbable that Yunjing was a record of any real language containing all of its distinctions. There is no single stratum of readings reflecting the distinctions made in Yunjing. Some of Yunjing’s Qieyun-based rhyme distinctions are found only in older EMC-based layers of modern readings whereas the ‘heavy/light’ distinction is found only in newer LMC-based layers. Pulleyblank (1984: 62) considers Yunjing to represent the “Tang standard language,” e.g., the Chang’an dialect of LMC, but, as Norman and Coblin (1995: 580) point out: The temporal and regional origins of the Yunnjinq [Yunjing] are obscure in the extreme. We have no “Yunnjinq Preface” to tell us who wrote the text or why. No contemporary or later Chinese source has ever associated the Yunnjinq with either Charng’an [Chang’an] or the early northwest dialects in general. The dialectal foundation of this text, if indeed such a thing can ever be determined, is a topic for future research rather than an established fact on which assumptions about the dialect of the Tarng [Tang] capital can be based. Although I believe that Yunjing is a valuable resource for the phonological categories of earlier Chinese in general, it is not the key to LMC that Pulleyblank (1984, 1991a, 1998) believes it is. Even if Yunjing really did reflect LMC, it is only a syllabary in table form and not a dictionary. We would still have no resource for LMC analogous in scale to Qieyun, which listed fanqie pronunciations of thousands of characters. I consider the newer strata of Chinese languages and Sinoxenic to be the best evidence for LMC.

Sinoxenic (SX) sources for earlier Chinese Samuel E. Martin (1953: 4) coined the term ‘Sino-Xenic’ to refer to SinoJapanese (SJ), Sino-Korean (SK), and Sino-Vietnamese (SV), which he describes as “vast bodies of borrowed language forms.” In this book, I use 98

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

the spelling ‘Sinoxenic’ (SX) and refer to SJ, SK, SV as ‘branches’ of X. I will present a more detailed definition of SX below. SX refers to a special class of borrowings from Chinese into non-Chinese languages. This class of borrowings is characterized by two traits: (1) Scale: SX consists not of sporadic borrowings of a few words, but the importation of the entire Chinese lexicon. (2) Accuracy and consistency: SX borrowings are not random attempts to imitate Chinese pronunciations. They represent conscious attempts to imitate Chinese pronunciation as much as possible. As a result of this, they may not be exact copies of Chinese, but they do have regular correspondences with Chinese. Each Chinese initial, tone, and final was consistently rendered with the same foreign phoneme or sequence of phonemes. This regularity was inevitable because of the high status of Chinese culture. In ancient East Asia, ‘civilization’ was synonymous with ‘China.’ SX was only part of a much larger process of borrowing. The Vietnamese, Koreans, and Japanese did not merely borrow sinographs and their readings. They also imported Chinese language and culture as a whole. Until modern times, Vietnamese, Korean, and Japanese scholars were all fluent readers and writers of Classical Chinese with a profound knowledge of Classical Chinese literature. The Japanese even borrowed the layout as well as the language of Chang’an. Their capital of Nara was modeled after the capital of the Sui and the Tang. In such a sinophilic atmosphere, the non-Chinese elite would not accept poor imitations of Chinese pronunciation. Thus each branch of SX was an attempt to imitate the sounds of Chinese as closely as possible. Such attempts tested the limits of non-Chinese phonological systems and even expanded them, particularly in the case of Japanese. The non-Chinese elite did not mind straining their tongues in order to pronounce sinographs ‘properly.’ Perhaps they could not perfectly replicate Chinese, but nonetheless they probably tried their best. SX originated as Chinese with a foreign accent. When the Japanese began to learn System B (i.e., LOC and /or LOCbased Sino-Paekche) from Paekche scholars circa the fifth century, they did not intend to create ‘Sino-Japanese,’ an entity distinct from System B. Like modern students of English in Japan who do not regard their pronunciation of English as ‘Anglo-Japanese,’ the ancient Japanese elite regarded their accented pronunciation of System B as ‘Chinese,’ not as ‘Sino-Japanese.’ Some modern linguists regard the branches of SX as ‘virtual dialects’ of Chinese. Karlgren (1915–26) treated SV, SK, and SJ as if they were Chinese dialects and compared them with genuine Chinese dialects to reconstruct his ‘Ancient Chinese’ (our EMC). Several others including Pulleyblank (1984) and Starostin (1989) have also used this methodology in their reconstructions of EMC and LMC. The section entitled “Dialect Evidence” in Martin’s (1953: 4) The Phonemes of Ancient Chinese surprisingly begins with SX ‘dialects’ rather than authentic Chinese dialects. 99

OLD JAPANESE

The extreme regularity of sound correspondences between authentic and ‘virtual’ Chinese dialects justifies the use of SX in the study of Chinese historical phonology. Each stratum of a branch of SX could be considered to have been a near-twin of its Chinese source at the time of borrowing. Unfortunately, those who have used SX data for Chinese reconstruction have paid relatively little attention to the histories of the Vietnamese, Korean, and Japanese languages. Although each stratum of an SX branch began as a second-language-speaker version of Chinese, these strata have not been preserved intact over the centuries. They were subject to changes within Vietnamese, Korean, and Japanese that took place after they were borrowed. One cannot simply look up SX readings in a dictionary and use them as evidence for Chinese reconstruction; one must understand the historical phonology of Vietnamese, Korean, and Japanese so that one can distinguish between SX features that reflect earlier Chinese and SX features that reflect later changes within those non-Chinese languages. In order to prevent the reader from taking the SX evidence for OJ phonology in Chapters 7 and 8 at face value, I will summarize aspects of Japanese, Korean, and Vietnamese historical phonology which are relevant to the development of SJ, SK, and SV on pp. 100 –35. These sections will also briefly deal with the origins and characteristics of the various strata of the branches of SX and with writing systems partly based on SX with a focus on nôm (p. 129). Since I have already discussed man’yôgana, the origins of SJ, and Japanese historical phonology in Chapters 2 and 4, I will repeat only the key points of those chapters on p. 104. Sino-Japanese (SJ) Sino-Japanese (SJ) is unique among the three branches of SX because its EMC strata (collectively referred to as Go-on) have been preserved almost as well as its LMC-based stratum (Kan-on). Every sinograph has a Kan-on and at least one Go-on reading. Some even have Tô-Sô-on readings (p. 107). All of these readings are in use today to various degrees. Although all branches of SX are multilayered and SV has more surviving layers than either SJ or SK, we have very few SK and SV readings based on stages of Chinese earlier or later than LMC with the exception of OSV words (p. 124). In theory, SJ offers twice as much evidence for earlier Chinese than SK or SV do. Even if one excludes ‘pseudo-Go-on’ and ‘pseudo-Kan-on’ (pp. 104 and 107), one has enough genuine Go-on and Kan-on readings to see how the Japanese rendered all the initials and finals of both EMC and LMC. The same cannot be said of SK or SV. Although the newer strata of those two branches of SX reflect all the initials and finals of LMC, OSV reflects many but not all of the initials and finals of pre-Tang Chinese, and the older strata of SK (p. 109) give us insights only on a handful of SK finals. 100

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

SJ may surpass the other SX branches in terms of quantity, but in terms of quality it is somewhat lacking. The sound correspondences of SJ are just as regular as those of SK, SV, or genuine Chinese languages but they tend to involve multiple initials or finals corresponding to a single Japanese initial or final. The most infamous instance involves Kan-on s, which corresponds to all LMC dental, retroflex, and palatal fricatives and affricates. Kan-on final correspondences are not much better. For instance, the MSJ Kan-on final [oo] corresponds to several different LMC finals: *-Rw, *-aw, *-aaw, *-aY, *-aaY, *-aayY, *-aawY, *-RwY, *-ap, *-aap. In this book, I use a transliteration of the pre-1945 kana spelling of SJ forms to maintain distinctions lost by subsequent sound changes ( pp. 74 and 84). In post-1945 kana orthography, MSJ Kan-on [oo] is written only with ou, but in pre-1945 kana orthography, it is written with three different spellings: ou, au, and ahu. The pre-1945 spellings have a historical basis. In the case of the spellings for MSJ Kan-on [oo], we do know for sure that ou and a(h)u were once pronounced differently. The Portuguese missionaries of the late sixteenth century romanized ou as ô [oo] but romanized a(h)u as i [QQ]. But even if I use pre-1945 spellings rather than post-1945 spellings, I cannot deny that Kan-on did not make as many distinctions as SK, SV, or Chinese. The pre-1945 spelling au still corresponds to six different LMC finals: *-aw, *-aaw, *-aY, *-aaY, *-aayY, *-aawY. The situation with Go-on is only slightly better. Go-on s and z correspond to all Chinese dental, retroflex, and palatal fricatives and affricates and Go-on makes only a few more final distinctions than Kan-on. One can understand why SJ makes so few distinctions when one looks at the consonant and vowel inventories of pre-1945 SJ spelling. SJ has only 13 or 14 initial consonants, depending on whether one includes zero, whereas the fanqie spellings of Qieyun indicate 39 initial consonants for EMC (Pulleyblank 1984: 63, Starostin 1989: 12–13) and Yunjing’s preface lists 36 initials for its ‘language.’ Not all scholars accept these figures, but all agree that there were roughly three dozen initial consonants in either stage of MC. There is no doubt that Chinese consonants far outnumbered Japanese consonants when Go-on and Kan-on were borrowed. The SJ consonant inventory (Table 5.1) is identical to the OJ consonant inventory in Table 4.4 (p. 74) except for two differences: (1) I transcribe the voiceless labial obstruent of pre-1945 SJ spelling as h. (2) I have listed zero ($ ) among the SJ consonants because it corresponds to the LOC and MC glottal stop phoneme *q-. However, I have enclosed zero in parentheses since I ignore zero in my transcriptions of OJ. All of the consonants in Table 5.1 can occur in syllable-initial position in SJ. Their distribution in other positions is more restricted. The pre-1945 SJ orthography also indicates two medial glides: (1) -w- only after velars (k, g) and before non-labial vowels (a, i, e):
Go-on kwau ‘wide,’
Go-on kwi ‘ghost,’
Go-on kwe ‘melon.’ 101

OLD JAPANESE

Table 5.1 The consonants of pre-1945 SJ orthography

Voiceless obstruents Voiced obstruents Nasals Liquid and glides

Labials

Dental

h Go-on si ‘stop’ for OJ to were still in use as late as the eighth century (Ôno Tôru 1962: 141). Such archaic phonograms are completely absent from Shoki. (4) The use of phonograms of the
‘genitive’ rhyme solely for OJ syllables ending either in -i or -iy. Go-on and SDBT readings for
‘genitive’ rhyme characters end either in i-like vowels or in o-like vowels, whereas Kan-on and SEBT readings of
‘genitive’ rhyme characters end only in i-like vowels. I will conclude this section with a note on ‘pseudo-Kan-on.’ Although Kan-on is the dominant variety of SJ today, there are some sinographs whose Kan-on readings (a) were forgotten or (b) were never transmitted to Japan. Thus lexicographers have supplied these sinographs with Kan-on ‘readings’ generated on the basis of fanqie (cf. the pseudo-Go-on on p. 104). Some lexicographers even altered attested Kan-on readings to fit preconceived notions about Kan-on: e.g., changing
Kan-on mei ‘bright’ to pseudo-Kan-on bei so that there are no Kan-on readings with nasal initials (Miller 1967: 106). Fortunately, my sources for SJ (Tôdô 1978, Morohashi 1955–60, Morohashi et al. 1981–82, and Morohashi et al. 1992) do not alter attested readings. Nevertheless, SJ data from dictionaries can be used only with great caution. Other strata The importation of Chinese readings into Japan did not stop with the Kan-on of the late seventh and early eighth centuries. The Japanese court continued to send missions to Tang China until 838. Even after the court severed its official relations with the deteriorating Tang in the late ninth century (Varley 1984: 54, 75), Japanese continued to go to China. They learned late Tang and Northern Song (960–1127 CE) readings which Tôdô (1980: 172) calls “New Kan-on” (Shin Kan-on). But this New Kan-on never caught on in Japan. For example, New Kan-on
 hosa < Northern Song *pho saq ‘bodhisattva’ did not replace the established Go-on word bosatsu. Practically no New Kan-on pronunciations survive in MSJ. One cannot even find genuine or fake New Kan-on pronunciations in dictionaries. After New Kan-on came various strata collectively known as ‘Tô-Sô-on’ (lit. ‘sounds of Tang and Song’). The term ‘Tô-Sô-on’ is somewhat misleading since the Tang Dynasty had long since fallen by the time Sôtô and Rinzai Zen monks brought back the readings that they had learned in the Southern Song (1127–1279) capital of Hangzhou (Tôdô 1978: 1587). ‘Tô-Sô-on’ can refer to any SJ readings later than Go-on and Kan-on, though in the strictest 107

OLD JAPANESE

sense it refers to borrowings from late Northern Song and early Yuan (Tôdô 1980: 174). Even though Tô-Sô-on never seriously threatened Kan-on or Go-on, a number of Tô-Sô-on words such as
 isu (cf. Go-on and Kan-on *isi) are still common in MSJ. Note that Tô-Sô-on -u in words such as
su ‘child; nominal suffix’ corresponds to -i in Kan-on and Go-on. This word belonged to the EMC
*-t ‘genitive’ rhyme. Late EMC had merged the EMC
*-iR ‘branch,’
*-i ‘fat,’ and
*-t ‘genitive’ rhymes into a single rhyme *-i which persisted into early LMC. Hence the later stratum of Go-on and Kan-on as a whole render these rhymes with -i. Early LMC *-i then became some sort of non-palatal vowel after sibilants (*s, *s®, *ts, *tsh, *ts®, *Z, *Z®) in late LMC. Pulleyblank reconstructs this vowel as *rC before retroflex sibilants (*Z, *Z®) and as *zC before dental sibilants (*s, *s®, *ts, *tsh, *ts®). (5.4) *i > *i >

*rC / *zC /

C_# C_#

C = {*Z, *Z®} C = {*s, *s®, *ts, *tsh, *ts®}

I prefer to reconstruct this vowel as *t in both environments: (5.5) *i >

*t

/

C_#

C = {*s, *s®, *ts, *tsh, *ts®, *Z, *Z®}

Regardless of which reconstruction is correct, all evidence points toward a non-palatal vowel for prestige dialects of late LMC. SK, SV, and the literary layers of nearly all modern Chinese languages have non-palatal vowels for words of the early LMC *-i rhyme with retroflex and dental sibilants: (5.6)
‘child’

Early Late SK LMC LMC *tsi *tst co [ts£]

SV

Md

Literary Taiwanese tO [ttt] zh [ts)] chu3 [tsu]

Colloquial readings of characters in southern Chinese languages still preserve the earlier *-i:
Literary Taiwanese su 5 < LMC *st ‘four’ corresponds to Colloquial Taiwanese si 5 < pre-LMC *si. OSV also preserves pre-LMC *i:
SV tO [ttt] < LMC *tst ‘child’ corresponds to OSV tí [tii] (Wang 1958: 380). Although this discussion may seem completely irrelevant, I must point out this phenomenon for two reasons: (1) Non-palatal vowels for words of the early LMC *-i rhyme with retroflex and dental sibilants are late Tang innovations. Knowing this will help us to date SK and SV in the next two sections (pp. 109 and 117). (2) In Chapter 8, we will see many cases of SK -o [£] and SV -e [tt] corresponding to OJ -i. Knowing that the rhymes SK -o [£] and SV -e [tt] reflect an innovation (*i > *t) which occurred after the borrowing of Go-on 108

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

and Kan-on will help us to avoid misinterpreting them as evidence for nonpalatal pronunciations of OJ i (e.g., *[£] or *[t] ). Sino-Korean (SK) Unlike Japan, the Korean peninsula is directly adjacent to China. Contact between the non-Chinese tribes of the peninsula and the Chinese of the mainland was inevitable. Weiman (Korean Wiman) of the northeastern Chinese state of Yan took over the northern part of the peninsula circa the third century CE. The Western Han invaded this territory and founded four colonies in 108 BCE. The natives of the north may have been exposed to Western and Eastern Han Dynasty Chinese for several centuries. The Chinese never did dominate the entire peninsula. The king of the tribal state of Koguryo (Jpn Kôkuri) was crowned by the Chinese in the first century CE. Koguryo destroyed the last of the Chinese colonies in 313 CE. To the southwest and southeast of Koguryo, Paekche (Jpn Kudara) and Silla (Jpn Shiragi) emerged as states in the third and fourth centuries CE. The enigmatic tribal confederation of Kaya (Jpn Mimana) was formed on the southern coast between these two states. The peninsula was entirely under native rule by the fourth century CE (Lewin 1973: 22, Ki-baik Lee 1980: 507, Haewoo Lee 1994: 208–209). Each of these states probably had its own language. Although independent from China, the ‘Three Kingdoms’ of the peninsula – Koguryo, Paekche, and Silla – imported Chinese culture and sent missions to the various Chinese states during the Six Dynasties Period (Yu 1991). I believe that LOC and EMC dialects of this period formed the basis of ‘Old Sino-Korean’ (OSK): Sino-Koguryo, Sino-Paekche, and Sino-Silla. (If Sino-Kaya existed, it too would have been based on LOC and EMC.) Best (1983: 146) claims that Paekche was recognized as “the most culturally sophisticated of the Three Kingdoms.” Even the Chinese “were speaking respectfully of the accomplishments of Paekche’s [Confucian] scholars.” Paekche men of learning such as ‘Wani’ and ‘Achiki’ spread their knowledge of Chinese language and culture and Buddhism to their Japanese allies in the southeast. Perhaps the early Systems underlying Go-on and the bulk of Japanese transcriptions apart from Shoki were based on various strata of their Sino-Paekche pronunciations rather than on LOC and EMC dialects. Paekche was ultimately doomed. Its ally Japan was no match for the new Tang Dynasty. Intent on expanding into the peninsula, the Tang forged an alliance with Silla, which had already conquered Kaya in the sixth century. With the help of the Tang, Silla defeated Paekche in 663 CE and Koguryo in 668 CE, thereby unifying the peninsula for the first time. Paekche and Koguryo refugees “played a vital role as scribes, craftsmen, and artists in the advancement of culture and civilization in Japan” (Varley 1984: 22). Their languages died out, but the final stages of their branches of SX may live on in the form of Go-on. Since students from all of the Three 109

OLD JAPANESE

Kingdoms had gone to the Tang capital of Chang’an in the seventh century (Haewoo Lee 1994: 220), it is also possible that these Paekche and Koguryo refugees taught CLMC (System E) to the Japanese. Unified Silla (668–935) borrowed a new system of character readings from northern LMC which formed the core of Sino-Korean proper. Old Korean, the Silla language, was related to the ancestor of the Early Middle Korean (EMK) of Koryo (935–1392), the Late Middle Korean (LMK) of the fifteenth century, and modern Korean (Vovin 1996b). This section will briefly outline what little is known about the varieties of OSK, sketch the phonology of SK, and attempt to determine the origins of SK. Since I am only using SK evidence for this book, I will not discuss aspects of Korean historical phonology which are not relevant to SK. Old Sino-Korean (OSK) ‘Old Sino-Korean’ refers to all of the branches of SX of the Korean peninsula prior to its unification in 668 CE: Sino-Koguryo, Sino-Paekche, Sino-Silla, and possibly Sino-Kaya. Practically nothing is known about Sino-Kaya at this point, so I will not mention it again. Our knowledge of the other three varieties of OSK is not much better. Like the Japanese prior to the eighth century, the peoples of the Three Kingdoms used phonograms to record proper nouns. The degree of overlap between Paekche and Japanese phonograms is striking and confirms the Paekche origins of Japanese writing (Bentley 2001). Toh (1977, 1981, 1984, 1985, 1987), Lyu (1983), Yu (1991), Eom (1991, 1994), Ryu (1994), Bentley (1998b, 2001) are some of the many studies on Three Kingdoms phonograms and OSK. Unfortunately, most of these studies with the exception of Bentley (1998b, 2001) suffer from four problems. (1) Most scholars seem to assume that OSK readings and the readings of phonograms in transcriptions were identical. But this was not the case in Japan, and it may not have been the case in Korea. The aims of SX and sinographic transcription necessitate different readings. SX is meant to mimic Chinese, whereas sinographic transcription is meant to record non-Chinese sounds with phonograms representing the nearest Chinese equivalents. Unless the syllabic inventories of a variety of Chinese (or its SX incarnation) and a non-Chinese language perfectly match, compromise is inevitable in transcription. (2) To know how much compromise is involved, we must know something about the phonotactics of Chinese and the languages of the Three Kingdoms. Unfortunately, Koguryo and Paekche are extinct, so we cannot draw upon any modern data from them. Silla is the ancestor of Old Korean (OK; see below), but our knowledge of OK phonology is shaky. Thus we must look at LOC and EMC phonotactics to understand what choices were available for the scribes of the Three Kingdoms. 110

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

Yet no one has ever analyzed the transcriptions of the Three Kingdoms with an understanding of the phonotactics of earlier Chinese. It seems that scholars assume that OC and MC had more than enough syllables to render nonChinese languages, even though there are gaps in the syllabic inventory of Chinese that prevent exact matches between Chinese and non-Chinese syllables. (3) If we want to understand the phonotactics of earlier Chinese, we need to work with the best available Chinese reconstructions. But apart from Bentley (1998b, 2001), all scholars in this field use outdated Karlgrenian reconstructions. (4) Some scholars superimpose assumptions upon the actual evidence. Believing that all Altaic languages have vowel harmony,3 Kiyose (1991) reconstructs vowel harmony for the languages of the Three Kingdoms, even though the phonograms’ Chinese readings do not support his reconstructions. For example, he reconstructs Koguryo
 LOC *koq sye ‘gem’ as *kösi with a front rounded vowel *ö absent in the LOC reading of the phonogram
LOC *koq ‘old’ (1991: 262). One must value evidence over one’s own assumptions. Even if one did avoid all four of the problems listed above, one still would be far from reconstructing all but the vaguest picture of the various branches of OSK. Since not all initial and final classes are represented in the transcriptions, we would have no idea how the branches of OSK rendered certain LOC and EMC initials and finals. Furthermore, we do not know enough about phonological differences between LOC dialects to determine the Chinese dialectal origins of the peninsular branches of SX. Yu (1991: 66) theorizes that Sino-Koguryo is based on the Chinese of the northern kingdom of Wei while Eom (1991: 11) theorizes that Sino-Paekche is based on southern Chinese, but these theories cannot be verified at the present time. Since so little is known about OSK, I will not use OSK evidence when I reconstruct OJ in Chapters 7 and 8. Sino-Korean proper (SK) After Silla unified the Korean peninsula in 668 CE, Sino-Silla became the dominant and eventually the only variety of SX in the region. Late SinoKoguryo and late Sino-Paekche withered away along with the Koguryo and Paekche languages. The descendants of their speakers adopted the Silla language. Yi Kimun (1961: 22) refers to the language of Unified Silla as ‘Old Korean’ (OK). The Sino-Silla of OK was probably a multistratal mix of readings derived from LOC, EMC, and LMC. Traces of the earlier strata may be found in three early types of Korean orthography: hyangchal (‘local letters’) itwu (‘clerk readings’), and kwukyel (‘oral formulae’).4 These extinct writing systems are beyond the scope of this book. 111

OLD JAPANESE

The newest stratum of Sino-Silla has survived up to the present day. Haewoo Lee (1994: 210) calls this stratum ‘Late Sino-Silla.’ I will call it ‘SK proper’ or simply ‘SK.’ Its dating is somewhat controversial. Haewoo Lee (1994: 207) has summarized six proposals for the origins of SK (Maspero 1920, Karlgren 1926, Arisaka 1936,5 Kôno 1964–67, Pak 1971, and Sin 1982). I will add two more views to Lee’s list, including Lee’s own view. (7) SK derives “from the northern Late Middle Chinese in about 7–8 centuries [sic], particularly from the new standard language of Changan [sic] in the period of Late Middle Chinese” (Haewoo Lee 1994: 207). (8) SK derives from four stages of Chinese: OC (before the second century CE), MC (circa the seventh century), Old Mandarin (circa the fourteenth century), and modern Mandarin (Eom 1998: 326). I will summarize my own view below. Modern SK is multistratal. A few readings are OC-based relics:
SK kay ‘general object classifier’ < Western Han *kaayh (cf. OSV cái [kaay] ‘id.’). Still others are from Old Mandarin (OM):
‘defeat’ has two SK readings, (1) pwuk < LMC *puRk and (2) pay < OM *pjy (Eom 1998: 332). There are a handful of readings of sinographs based on modern Mandarin (Eom 1998: 333) which I regard as ordinary borrowings rather than SK since they are extremely rare and unsystematic. Nonetheless the majority of SK readings – perhaps up to 95 per cent – are from LMC as claimed by Haewoo Lee and not from the EMC ‘language’ of Qieyun as claimed by Karlgren (1926), Pak (1971), or Sin (1982). I believe that SK represents a variety of northern Chinese somewhat newer than the variety reflected by Kan-on (p. 105). If Kan-on dates from the seventh century, then perhaps SK dates from the eighth century. Thus I am largely in agreement with Haewoo Lee (1994). Haewoo Lee (1994) gives four arguments for the Chang’an LMC origin of SK: (1) The z of fifteenth-century SK corresponds to Chang’an LMC *r which appears in Tibetan transcriptions as zh [9] (Luo 1933a). Lee assumes that the LMK letter ‘triangle’ transcribed as z in Yale romanization was actually a voiced fricative *z. However, this view has been challenged by Vovin (1993), who reconstructs z as a palatal nasal *ã mostly on the basis of Korean-internal evidence. Moreover, SK has nasals corresponding to Chang’an LMC prenasalized stops: SK m and n correspond to CLMC *mb and *nd. In this respect, SK resembles non-northwestern Chinese dialects which never developed Chang’an-style prenasalized stops. (2) SK has a medial glide -y- in Grade II words corresponding to Pulleyblank’s reconstruction of a medial glide *-y- in LMC Grade II words. However, a medial -y- is not found in CLMC-based Kan-on or in the Tibetan transcriptions of northwestern LMC. This implies that SK cannot derive from CLMC. (3) SK does not distinguish finals which were distinct in EMC. For example, SK, like Kan-on, generally has one equivalent (SK i) of the EMC 112

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC


*-iR ‘branch,’
*-i ‘fat,’ and
*-t ‘genitive’ rhymes. (SK -o [£] for these three rhymes corresponds to LMC *t. See (5.6).) Thus SK must be based on an LMC dialect which had merged these rhymes. For other examples of mergers reflected in SK, see Haewoo Lee (1994: 216–217). (4) The SK coda -l corresponds to the *-r of northwestern LMC rather than to the *-t of EMC and of southern LMC. The origin of SK -l is rather controversial. For a summary of various views, see Haewoo Lee (1994: 217– 220). Like Haewoo Lee (1994), Arisaka Hideyo (1957: 305), and Yi Kimun (1961: 56), I believe that SK -l reflects northern LMC *-r < EMC *-t rather than a lenition of southern LMC *-t. SK -l corresponds to -r in the Tibetan transcriptions of northwestern LMC and to -t(V) in Kan-on, SV, and southern Chinese languages: (5.7) SK -l

Tib. tr. of Kan-on S. LMC SV, Hakka, NW LMC Cnt, Taiwan -r -ti/-tu *-t -t

Note that Kan-on has -ti/-tu with -t- rather than -r-. Assuming that Kan-on -t- was *t in the seventh century, I believe that this -t- reflects early CLMC *-t which later became the later CLMC *-r attested in the Tibetan transcriptions. I have already eliminated a northwestern LMC origin for SK on the basis of the different treatments of nasals and Grade II in CLMC and SK. Moreover, I can eliminate a southern LMC origin for SK on the basis of the different treatments of *-t in the south and SK. Only northeastern LMC remains as a possible source for SK. However, Alexander Vovin (p.c.) has pointed out that it would make no sense for SK to be borrowed from the northeast, which was less culturally prestigious than the Tang capital of Chang’an in the northwest. At present I cannot reconcile this sociolinguistic observation with the isogloss evidence. Even if Haewoo Lee (1994: 210–211) is correct in assuming that the name reforms of King Kyengtek were based on SK proper, we have very little evidence for SK readings from the Unified Silla period up to the early Yi Dynasty in the fifteenth century. The situation improves immeasurably after the promulgation of the Korean hankul alphabet in 1446. We are blessed not only with the first alphabetic transcriptions of SK readings but also the first SK dictionary, the Tongkwuk cengwun (Correct Rhymes of the Eastern Nation, hereafter TKCW; 1447) which lists readings in hankul for thousands of characters arranged by their rhymes. But as its title implies, TKCW was intended to codify the ‘right’ readings for characters. When writing sinograph readings in hankul, one was supposed to use the readings in TKCW, not the actual readings that had been imported from China centuries ago during the Unified Silla period. The TKCW readings are an interesting blend of elements from the Chinese rhyme dictionaries and tables (pp. 93 and 96) on the one hand and actual SK 113

OLD JAPANESE

readings on the other. Samuel E. Martin (1997: 264) views the TKCW system of readings as “a reconstruction of Middle Chinese phonology in terms of the Korean sound system as represented by the hankul symbols.” Until recently, there has been a tendency to “dismiss the readings given in that dictionary as wholly artificial and prescriptive” but Kang (1997: 124– 125) sees TKCW as “an extremely valuable work not only for interpreting the Korean readings of Chinese characters, but also for understanding Middle Korean and the Korean alphabet in a much broader sense.” Kang (1997: 125–128) proves his point by demonstrating the close correlations between TKCW finals and the finals of the authentic SK preserved in fifteenthcentury spelling ‘mistakes’ and in the non-TKCW-based SK orthography of later centuries. But although I also think the true worth of TKCW has been grossly underestimated, no one can claim that the TKCW readings are genuine SK readings. Thus I will generally avoid citing TKCW readings in this book. Instead, I will use the SK readings taken from the Cenwun okphyen (Jewel Book of All Rhymes, hereafter CWOP) dictionary of c. 1796 which was modeled after the Chinese Kangxi zidian dictionary of 1716 CE (John Bentley, p.c.). Although CWOP was published over three hundred years after the first attested real SK readings in hankul, its orthography is extremely faithful to the post-TKCW SK orthography of the sixteenth century. I call this orthography the ‘premodern SK orthography’ as opposed to the prescriptive fifteenth-century TKCW SK orthography or the modern postwar SK orthography. All ‘SK’ forms in this book are cited in Yale romanizations of premodern SK spellings with premodern pronunciations. The premodern SK orthography originated as a phonetic transcription of actual SK pronunciations in the late fifteenth century without the artificial distinctions of TKCW or the pitch marks that were a feature of all early hankul writing. Despite changes in SK pronunciation, this orthography remained fairly stable during the Yi Dynasty and persisted into the Japanese colonial period. Many of the spellings in Choy Namsen’s Sin cacen (New Sinographic Dictionary) of 1915 are identical to those of earlier centuries, though Sin cacen also lists alternative spellings reflecting twentieth-century pronunciations. One can see numerous examples of the conservatism of early twentieth-century SK orthography by comparing the Sin cacen readings with those of CWOP and earlier works in Nam (1973). There is only one major difference between early and late premodern SK orthography. According to Nam (1973: 8, 183), the hankul letter z (LMK *[ ã] or *[z] ) used in the fifteenth-century TKCW orthography continued to be used in non-prescriptive SK orthography to the sixteenth century. Thereafter it was replaced by the hankul letter for a zero initial. Since CWOP was compiled in the eighteenth century, it does not distinguish between zero-initial and z-initial readings. Hence I will use TKCW and the pre-seventeenth-century SK readings collected by Nam (1973) in order to 114

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

Table 5.4 SK consonants (after Kang Sinhang 1997: 119) Labial

Dental and palatal

Velar

Glottal

Unaspirated

p

t

c [ts]

k

(q) [}]

Aspirated

ph [ph]

th [th]

ch [tsh]

kh

h [kh]

‘Muddy’

( pp)

(tt)

(ss)

(cc)

(kk)

(hh)

Nasal

m

n

{z} {[z] or [ ã] }

Liquid or glide

w

l [r]/[l]

s

y

ng [Y] (zero)

(-lq)

determine whether characters listed in CWOP with zero initials were once read with initial z. I will now sketch the phonology of the SK recorded in premodern orthography with reference to the phonology of the SK of TKCW. Table 5.4 lists all of the consonants in TKCW and premodern SK orthography in Martin’s (1992) Yale romanization for LMK. Consonants found only in TKCW are given in parentheses. The z found in early premodern SK orthography is given in curly brackets ({}) to distinguish it from the prescriptive letters unique to TKCW. Since premodern SK orthography was based on the actual pronunciation of fifteenth-century SK, the SK sound values given here and elsewhere are those of the fifteenth century. All of the consonants in Table 5.4 are potential SK initials with the exception of -lq which occurs only as a coda (see below). The following peculiarities of the TKCW initials require explanation. (1) TKCW had a series of ‘muddy’ obstruents ( pp, tt, ss, cc, kk, hh) which corresponded to the ‘muddy’ obstruents listed in Yunjing and distinguished in fanqie from ‘clear’ and ‘second clear’ obstruents. The ‘muddy’ stops and affricates ( pp, tt, cc, kk) of TKCW corresponded to non-aspirates ( p, t, c, s, k) and aspirates ( ph, th, ch) in SK. The ‘muddy’ fricatives of TKCW (ss, hh) corresponded to the ordinary SK fricatives s and h. I will not speculate about how LMK speakers pronounced the TKCW ‘muddy’ initials here, though Kang (1997: 119) assumes that they were like the tense unaspirated obstruents of modern Korean ( pp, tt, ss, cc, kk). There are two modern SK readings with ss- and one with kk-, but these rare initials have nothing to do with TKCW or Chinese (Martin 1992: 112). (2) TKCW aspirates regularly correspond to Chinese aspirates. However, aspiration in authentic SK is unpredictable. (3) I have already noted that z ( [z] or [ ã] ) was not an artificial initial, since it could be found in real SK readings up to the sixteenth century. 115

OLD JAPANESE

Although it is absent from the orthography of CWOP, it can be restored on the basis of pre-seventeenth-century readings. (4) TKCW allowed ng in initial position. This ng- was meant to correspond to the Chinese initial
*Y ‘doubt,’ but TKCW often confuses ng- (GD ) with the zero initial (G) that is obligatory before vowels and glides without preceding consonants. Many cases of ng- in TKCW are unetymological. I suspect that the graphic similarity between the ng- and zero initial letters may have caused this confusion. I also doubt that many actually tried to pronounce the prescriptive initial ng- of TKCW. In premodern SK orthography, the letters for the TKCW initial ng and the zero initial became interchangeable symbols for phonetic zero in initial position. Thus the partly etymological ng-zero initial distinction of TKCW is completely absent from SK. (5) TKCW distinguishes between a glottal stop initial q- and the zero initial. Premodern SK orthography dispenses with this distinction entirely and replaces q- with the zero initial. TKCW and SK allow the following codas: (5.8) TKCW: -p SK: -p

-m -w -lq -n -m -$ -l -n

-y -k -y -k

-ng -ng

Both the SK of TKCW and the SK of the premodern orthography have a total of seven vowels whose phonetic interpretation has been extremely controversial (Table 5.5). The phonetic values given in Table 5.5 are identical to those of modern Korean, with the exception of o [£], which no longer exists; it has merged with a in SK readings. Note however that premodern diphthongs have been monophthongized in modern Korean: e.g., ay [ay] is now pronounced [ε]. Kim Wancin (1963) has proposed that the fifteenth-century LMK vowels were phonetically quite different from the modern Korean vowels. This view is quite popular among Korean linguists, though some prefer to date the shift before LMK. However, I do not believe any such shift existed. Various types of evidence have been cited against the shift: internal (Martin 1998), comparative (Martin 1998), transcriptional (Vovin 1998), Sinoxenic (Miyake 1998), and theoretical (Oh 1998). I will not repeat the arguments made in the aforementioned works here. I will simply assume that Korean vowels Table 5.5 SK vowels

High Mid Low

Front

Central

Back

i [i]

u [t] e [R] a [a]

wu [u] wo [o] o [£]

116

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

have remained stable at least since the time of the borrowing of SK proper in Unified Silla. Many earlier SK glide –vowel and vowel–glide combinations have monophthongized into single vowels in modern SK pronunciation (e.g., ay [ay] > [ε] ), but I will ignore these monophthongizations and cite the earlier pronunciations. The TKCW orthography even marked LMK pitch equivalents of MC tones with dots. However, the premodern SK orthography ignores pitch entirely. This will not be a problem for me, since I will not be reconstructing OJ pitch accent. In the modern standard pronunciation of SK, there is a correlation between the MC Rising and Departing tones and vowel length. However, Martin (1992: 112) notes that this correlation has many exceptions. The segments described above combine to form SK readings with the structure: (5.9) C(G)V(G)(C) C = {all consonants including zero and excluding ng} G = {w, y} V = {all vowels} G = {y} C = { p, m, l, n, k, ng, y) I have excluded pitch and vowel length from the above formulation. Sino-Vietnamese (SV) Contrary to popular belief, the Vietnamese language, like Japanese and Korean, is completely unrelated to Chinese. Vietnamese belongs to the Vietic branch of the Mon-Khmer language family of Southeast Asia. It is Sinoxenic that makes the Japanese, Korean, and Vietnamese languages look ‘Chinese.’ But unlike the Koreans and the Japanese, the Vietnamese were ruled directly by the Chinese for over a millennium. What we know today as northern Vietnam was the Chinese province of Jiaozhi which passed through a number of dynastic hands from 111 BCE to 939 CE. Thus the Vietnamese were exposed to multiple stages of Chinese ranging from the Western Han stage of OC to the final stage of LMC during the Five Dynasties period (907–960 CE). They did not have to rely on a few immigrants and returnees from missions to distant Chinese capitals for their knowledge of Chinese, since they could hear the language firsthand from native speakers. Even the illiterate majority were profoundly affected by the tongue of their foreign rulers. The Sinitic influence went far beyond the adoption of many everyday words such as the generic classifier OSV cái <
Western Han *kaayh ‘id.’ or the development of multiple strata of Sino-Vietnamese 117

OLD JAPANESE

(pp. 124 and 126). Long exposure to Chinese had Sinified the language not only lexically but also phonologically. Once Vietnam broke free from the Chinese in the tenth century, the language was altered almost unrecognizably by the changes that had occurred over the last millennium. Although the syntactic and lexical core of Vietnamese remained very un-Chinese – neither its modified–modifier pattern nor its basic vocabulary was lost – the language looked superficially like Chinese. The newly independent country itself was modeled after its former oppressor. The Chinese rulers were gone, but Chinese culture and language remained. The Vietnamese established a Chinese-style system of civil service examinations that lasted until the beginning of the twentieth century. Scholars learned the Confucian classics in the original Chinese in hopes of passing the exams. In Vietnam, Classical Chinese was “the medium of written communication among the educated people (like Latin in medieval Europe) and the vehicle of literary creations either in verse or in prose” (NguyWn 1997: 2). But like the peoples of the Korean peninsula and Japan before them, the Vietnamese devised ways to write their own language with sinographs. However, the chw nôm [ctt nom] (possibly from
 LOC *dzRq nRRm ‘characters southern’) script (p. 129) would turn out to be far more complex than any of its northern counterparts. Despite its difficulty, the script would continue to be used as late as the early twentieth century. In the seventeenth century, European missionaries developed systems of romanization to spread their Catholic faith among the Vietnamese. One such French missionary, Alexandre de Rhodes, came to Vietnam in 1624, mastered the language, and codified a synthesis of earlier romanizations in his Dictionarium annamiticum, lusitanum et latinum (Annamite–Portuguese– Latin Dictionary; 1651). His dictionary’s romanization, which later came to be called qu3c ngw [kwok Yt] ‘national language’ <
 LMC *kuRk YüR ‘id.’ under uncertain circumstances (DeFrancis 1978, Thompson 1987: 55), initially did not spread beyond the missionaries and their converts. However, the French later promoted the script and today it is the only medium in which Vietnamese is written. Although sinographs and chw nôm have never been entirely forgotten in Vietnam, their status is akin to man’yôgana in Japan. They are now only the objects of scholarly study, not scripts for everyday use. This radical change in writing systems has not diminished the roles of Sino-Vietnamese proper (SV) and Old Sino-Vietnamese (OSV) in the language. The old sinographic and chw nôm spellings for SV and OSV may be gone – but SV and OSV persist not only in old words but also in the neologisms necessitated by csng-skn [kRw§ saan] (<
 LMC *k®üwYh Zaanh) ‘Communism.’ Like SJ and SK, SV is indispensable for sophisticated discourse. SV and especially OSV have penetrated the core lexicon to an extent unparalleled by SJ or SK: e.g., the ordinary Vietnamese word for ‘cat’ is OSV mèo [mεw] <
EMC *mεrw ‘id.’ 118

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

Table 5.6 Written Vietnamese consonants (after Thompson 1987: 58) Labial

Dental

Retroflex

Palatal

Velar

Glottal

Unaspirated 17th-c. Hanoi

p

t

tr *[â] [c]

ch *[c] [c]

c/k/qu

( [}] )

Aspirated 17th-c. Hanoi

ph *[ph] [f ]

th *[th]

s *[Z] [s]

x *[\] [s]

kh *[kh] [x]

h

Implosive

b [(]

$ [ß ]

‘Softened’ 17th-c. Hanoi

(%) *[β] [v]

d *[dy ] [z]

gi *[©9] [z]

g/gh *[g] [γ]

Liquids/glides North

v/o/u [v]/[w]

l [l]

Nasals

m

n

nh [ ã]

ng/ngh [Y]

r [z]

Since at least the nineteenth century, dictionaries listing sinographs and chw nôm characters with their readings in qu3c ngw have been compiled. For a survey of dictionaries and other sources of Vietnamese character readings, see Mineya (1972: 21–33). NguyWn (1992: 410–411) surveys more recent nôm dictionaries. I have taken the SV forms in my book from Mineya’s (1972) appendix. I use a 1997 reprint of NgZ thiên t@ (earliest known edition from 1909), as my source of chw nôm readings. Before discussing SV proper, OSV, and nôm in the following three sections, I would like to describe the qu3c ngw orthography and outline aspects of Vietnamese diachronic and synchronic phonology which are relevant to SV as a whole. Written Vietnamese has the consonants in Table 5.6. The letter % is no longer used in modern qu3c ngw. Although all of the letters in Table 5.6 represented distinct consonants in seventeenth-century Vietnamese, no modern Vietnamese dialect makes all the distinctions indicated in spelling. I have cited reconstructed seventeenth-century phonetic values from Thompson (1976: 1120) whenever they differ from phonetic values in modern dialects with two exceptions. First, I follow Gage (1985) and reconstruct the seventeenth-century value of s as a retroflex fricative *Z because (1) Gage (1985: 506) has shown that s < *Cr-, (2) s is still a retroflex [Z] in central and southern dialects, and (3) s corresponds to OC *r-clusters and MC retroflex initials. Second, I reconstruct g as *g since de Rhodes described g as being similar to a European g (Gregerson 1969: 165, 182).6 Phonetic values for 119

OLD JAPANESE

modern Hanoi are taken from Thompson (1987: 98) except when they can be inferred from qu3c ngw: e.g., p is [p], etc. I cite modern Hanoi phonetic values of qu3c ngw forms unless stated otherwise. Qu3c ngw consonants correspond to Vietnamese consonantal phonemes with exceptions explained in Thompson (1987: 62–64). Some of the phonetic values for consonants are quite unexpected. Orthographic d is a voiced fricative [z] in the north and a palatal glide [y] elsewhere but never [d], though in the seventeenth century it was presumably a palatalized dental stop [dy ]. These mismatches resulted from massive phonological changes that can be confirmed by comparative and philological evidence. I will list the most relevant changes for our purposes below. (1) The original Vietnamese voiced obstruents were devoiced: (5.10) *b, *d, *j, *g, etc. > *p, *t, *c, *k, etc. One can determine the presence or absence of earlier voicing by looking at the tones. Syllables with proto-voiceless initials have ngang, s8c, or hni tones while syllables with proto-voiced initials have huyVn, n'ng, or ngk tones. I will discuss tones further on pp. 126 and 128. (2) Vietnamese initial unaspirated labial and dental stops became voiced implosives: (5.11) *p > (

*t > ß

Thus Vietnamese b [(] and d [ß ] come from both voiceless and voiced sources (Thompson 1976: 1128, Gage 1985: 503):7 (5.12) *p, *b > *p > *(

*t, *d > *t > *ß

(3) Earlier Vietnamese voiced implosives (*(, *ß ) became nasals (Thompson 1976: 1,134–1,135, Gage 1985: 505, 506). (5.13) *( > m

*ß > n

I suspect the following ‘drag chain’ shifts took place: (5.14) p > ( > m

t>ß>n

(4) The shift in (5.14) may have been even more complex than this. Earlier Vietnamese *s became t (Gage 1985: 303). This shift can be integrated into my earlier chain of changes as (5.15): (5.15) s > t > ß > n

120

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

Sino-Vietnamese t has several non-*s sources. One is Chinese *ts. The Vietnamese may have borrowed *ts as *s (cf. English pronunciations of tsunami as [sunami] ). Others are LMC *py and *p®y which appear in Grade IV SV readings (p. 97): (5.16) *py, *p®y > *p\i > *psi > *ts > *s > t (cf. 5.22). These changes left Vietnamese without any [s]. Notice that the seventeenthcentury consonant system in Table 5.6 has no [s]. In all modern dialects of Vietnamese, the palatal fricative x *[\] has become [s] (Thompson 1987: 98). Thus x has filled the hole left behind by the *s which had become t: (5.17) \ > s > t > ß > n (5) The aspirates ( ph, th, kh) generally correspond to OC and MC aspirates (*ph, *th, *kh). kh always corresponds to Chinese *kh. ph has two sources, OC and EMC *ph and two of the LMC ‘light lip sounds’ (LMC *f, *f®; p. 96). th has several sources besides OC and MC *th: (a) OC and MC *tsh (Mineya 1972: 84): (5.18) *tsh > (*sh = aspirated s?) > th (b) OC and MC palatal fricatives and affricates (*\, *\®, *9, *j) (Mineya 1972: 91): (5.19) *\, *\®, *9, *j > (*sh = aspirated s?) > th I would rather not posit an exotic aspirated s in (5.18) and (5.19), though such a sound does exist in Burmese. Mineya (1972: 113) does not posit any such intermediate stage. Note that Chinese palatal fricatives are also occasionally borrowed as x (Mineya 1972: 91). (c) LMC *phy in Grade IV words (p. 97): (5.20) *phy > *ph\i > *phsi > *tsh > *sh > th (cf. 5.16) (6) So far, I have explained that Vietnamese lost its original voiced obstruents (*b, *d, *j, *g, etc.) (5.10). But where did the so-called ‘softened’ voiced obstruents % *[β], d *[dy ], gi *[ j], and g *[g] of seventeenth-century Vietnamese in Table 5.6 come from? (6a) Seventeenth-century % *[β] derives from prefixed labials (*C-p, *C-b) (Haudricourt 1965, Pulleyblank 1981, Gage 1985: 504). This % *[β] later merged with v *[v].

121

OLD JAPANESE

(5.21) *C-p, *C-b > % *[β] > v [v] (6b) d *[dy ] corresponds to OC and MC *y and LMC *my. It also derives from prefixed dentals (*C-t, *C-d) (Gage 1985: 504). (5.22) *y, *my, *C-t, *C-d > d *[dy ] > d (northern [z], southern [y] ) (6c) gi *[ _j9] corresponds to LMC *ky. It also derives from prefixed palatals (*C-c, *C-j) (Pulleyblank 1981: 279, Gage 1985: 504). (5.23) *ky, *C-c, *C-j > gi *[ j_9] > gi (northern [z], southern [y] ) (6d) g corresponds to prefixed velars (*C-k, *C-g) (Haudricourt 1965, Pulleyblank 1981, Gage 1985: 504): (5.24) *C-k, *C-g > g *[g] > g [γ] The vowels of Vietnamese have apparently undergone far fewer changes than the consonants. Downer (1987: 141) points out that Barker and Barker (1970) and Thompson (1976) both reconstructed the fourteen vowel nuclei (eleven vowels and three diphthongs) found in seventeenth-century and modern Vietnamese for Proto-Viet-Muong (PVM). I list these fourteen nuclei in Table 5.7. According to Thompson (1976: 1149), As might be expected, (Viet-Muong) vowels present a number of problems that are difficult to deal with. However, these problems are of relatively small importance – to a great extent the system of PVM seems to be very similar to that of the modern languages [i.e., Vietnamese and Muong; italics mine].

Table 5.7 Written Vietnamese and PVM vowels Front Higha High + Midb Mid Low

i, y ia, iê ê e

Central [i] [iR] [e] [ε]

â #

e ea, [R] [a]

[t] ed [tR] d [RR] a [aa]; [a] / _c, nh

Back u [u] ua, uô [uR] ô [o] o [Q]

Notes a The vowel [i] is sometimes written as y at the end of a syllable. See Thompson (1987: 64) for more details on when [i] is spelled as i or y. b The diphthongs [iR tR uR] are written as ia, ea, ua at the end of a syllable and as iê, ed, uô before a coda.

122

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

I will list the most relevant vowel changes between PVM and Vietnamese for our purposes below. (1) PVM *R, *RR raised to *t before velars (*k, *Y) (Barker and Barker 1970). (5.25) *R(R) >

/ _k, Y

e [t]

(2) PVM *ε dissimilated to a [a] before palatals (*c, *ã) (Barker and Barker 1970). (5.26) *ε

>

a [a] / _c, ã

(3) PVM *o, *Q respectively ‘broke’ to [Rw] and [aw] before velars (*k, *Y) (Thompson 1976: 1151–1152). Some of the labiality of the vowels ‘moved’ to the final velars which gained labial co-articulation: (5.27) *ok > ôc [Rw¶] *oY > ông [Rw§] *Qk > oc [aw¶] *QY > ong [aw§] Henderson (1985: 21) considers this an example of ‘feature shuffling.’ Note that *u did not break to the expected [tw] before velars, though velars after *u did gain labial co-articulation: (5.28) *uk > uc [uw¶]

*uY > ung [uw§]

Gage (1985: 514–516) has found that Vietnamese vowels correspond very closely to written Khmer vowels. Since (1) Khmer and Vietnamese are very distantly related within Mon-Khmer and (2) Khmer and Vietnamese relations have been negative for centuries, such correspondences are probably the result of independent retentions rather than shared innovations or the result of language contact. We cannot say that Vietnamese vowels are identical to PVM vowels. Nor can we say that PVM vowels are identical to Proto-Mon-Khmer vowels. Nevertheless, when Gage (1985: 514–516) worked out the correspondences between Vietnamese vowels and Shorto’s (1976) Proto-Mon-Khmer vowels, he found many near-identities. There is no evidence for a Vietnamese ‘Great Vowel Shift.’ Thus I will take SV and OSV vowels at face value, with exceptions due to rules (5.25–28). The above vowels can be followed by any of ten codas: (5.29) -p -m -o/-u [w]

-t -n

-ch -nh [c] [ ã]

123

-i/-y -c [y] [k]/[¶]

-ng [Y]/[§]

OLD JAPANESE

SV and OSV syllables have the following structure: (5.30) C(G)V(C) + Tone C = {all consonants except p} G = {w} V = {all vowels; however, #, â ≠ V/ _#} C = {p, m, w, t, n, c, ã, y, k, Y} All vowels are phonetically long in syllable-final position owing to a bimoraic syllable structure constraint. Since the vowels # [a] and â [R] are inherently short, they cannot stand at the end of a syllable: syllables such as c# [ka] or câ [kR] are impossible. I will discuss tones in the next two sections. Old Sino-Vietnamese (OSV) Old Sino-Vietnamese (OSV) refers to all the strata of Chinese loanwords in Vietnamese predating Sino-Vietnamese proper, which I will demonstrate to be of late Tang origin in the next section. Thus OSV words can be borrowed from any variety of Chinese ranging from Western Han OC of the second century BCE to early Tang EMC of the seventh century. Our knowledge of premodern Chinese dialects is too minimal to pinpoint the exact geographic origin of the source dialect(s) of OSV. Presumably OSV is based on southeastern Chinese rather than on the distant dialects of the Western Han capital of Chang’an or the Wu, Eastern Jin, and Six Dynasties Song capital of Jiankang,8 though perhaps those dialects exerted some influence over the speech of the ‘extreme south.’ So far, we only know a few hundred OSV words. Most of them are not associated with sinographs except by scholars who know their etymologies. In traditional Vietnam, OSV words were often written with nôm characters rather than with the sinographs for their source etyma. OSV words have certain phonological characteristics that distinguish them from LMC-based SV proper. Wang Li (1958) and Pulleyblank (1981, 1984) give extensive lists of sound correspondences between OSV and EMC with copious examples. I do not intend to repeat their work here, but I must acknowledge my debt to them because they have established the OSV equivalents of EMC initials, rhymes, and tones. Since only a few hundred OSV words survive, we have no idea what the OSV readings of thousands of sinographs were. But Wang’s and Pulleyblank’s work allows us to make reasonable guesses about those lost readings. I will cite the correspondences established by Wang (1958) and Pulleyblank (1981, 1984) whenever necessary in Chapters 7 and 8. The only two types of correspondences that I will mention here are initial and tonal.

124

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

(1) OSV often has ‘softened’ initials (v < %, d, gi, g) corresponding to OC and EMC stops and affricates (Pulleyblank 1981): (5.31) OC, EMC 17th c. OSV OSV

*p, *b *t *ts, *dz, *c, *j *k, *g, *® % *[β] d [dy ] gi [ _j9] g [g] v [v] d [z] gi [z] g [γ]

The above pattern is extremely common; see Wang (1958), Ray (1979), and Pulleyblank (1981) for dozens of examples. These ‘softened’ initials originated from earlier clusters (5.21–24). Yet we have no evidence from the Qieyun tradition nor any other evidence to suggest that LOC or EMC still had the initial consonant clusters that characterized stages of Chinese prior to LOC. I have wondered whether colloquial Chinese of the Six Dynasties still preserved some clusters absent in the refined pronunciation reflected in the fanqie of Qieyun. Norman (1973, 1974) has theorized that Proto-Min had a series of ‘softened’ initials which were somehow distinguished from ordinary stops and affricates. Norman (1973: 237) suspects that these ‘softened’ initials “originated from the reduction of clusters of some type.” However, Cheng and Miyake (1995) found only one correlation between Norman’s ProtoMin ‘softened’ initials and the ‘softened’ initials of OSV. I doubt that there is any relationship between the ‘softened’ initials in the two languages. Starostin (1989: 52–57) goes even further and denies the very existence of ‘softened’ initials in Min. I will not argue for or against him here. I suspect that the ‘softened’ initials in Vietnamese may be remnants of native (i.e., non-Chinese) affixes. Mon-Khmer languages in general are rich in prefixes and infixes. Ruc, a Vietic language, still has affixes such as pa-/-a‘causative,’ -n-/-r- ‘nominalizer,’ mu- ‘quantifier,’ kV- ‘animal,’ k- ‘natural phenomena,’ and tR- ‘stative verb’ (Alves 1998: 14). These affixes have either been lost in Vietnamese or have resulted in ‘softened’ initials: (5.32) Ruc words with k- ‘natural phenomena’ Viet ku-mRa 2 ‘rain’ : *k-m- >mea [mtR] ‘id.’ ka-yQ 3 ‘wind’ : *k-y- > gió [zQQ] ‘id.’ Perhaps the early Vietnamese who still had such morphology applied it to OSV words in the spoken language. Unfortunately, this hypothesis cannot be tested at present, because we do not know much about early Vietnamese morphology. We cannot predict what prefixes would have been attached to which OSV loans. In any case, it appears that OSV ‘softened’ initials are Vietnamese innovations. Therefore we cannot use ‘softened’ initials as evidence for similar initials in OJ. It is unlikely that EMC-based System D (Sino-Paekche?) readings and Chang’an LMC – the foundations for the transcription systems

125

OLD JAPANESE

Table 5.8 Written OSV tones OC/EMC

Level (*-$)

Rising (*-q) and Entering (*-C)

Departing (*-h)

Clear Muddy

ngang huyVn

s8c n'ng

hni (rarely ngang) ngã (rarely huyVn)

of Kojiki and Shoki – had such initials. Neither the fanqie of the EMC ‘language’ of Qieyun nor the Tibetan transcriptions of northwestern LMC suggest the existence of ‘softened’ initials. Although I will not be using Chinese tones or their SX counterparts to reconstruct OJ pitch accent, I will note that OSV–Chinese tonal correspondences are different from SV–Chinese tonal correspondences. When OSV and SV words are segmentally identical, the suprasegmentals (tones) give away their ages: the Chinese Rising Tone (marked with the tonal letter *-q) of
EMC *tQq / LMC *tuRq ‘gamble’ appears as the s8c tone in OSV $3 ‘challenge’ but appears as the hni tone in SV $7 ‘gamble’ (Wang 1958: 380). One can see the Rising Tone: s8c tone and other OC/EMC: OSV tonal correspondences in Table 5.8. Each of the Vietnamese tone names contains and describes the tone it represents: 1 2 3 4 5 6

ngang ‘level’ has the unmarked mid level ngang tone. huyVn ‘dark’ has the low falling breathy huyVn tone, marked by (`). s8c ‘sharp’ has the high rising s8c tone marked by (´). n'ng ‘heavy’ has the low creaky n'ng tone marked by a subscript dot. hni ‘question’ has the low rising hni tone marked by a superscript question mark. ngã ‘fall’ has the high creaky ngã tone marked by a tilde.

I am mentioning OSV tone correspondences here only because I use them as well as the segmental correspondences listed in Wang (1958) and Pulleyblank (1981, 1984) to identify OSV words. Sino-Vietnamese proper (SV) I have already discussed Vietnamese historical phonology as a whole, so I will focus here only on arguments for the approximate date and location of the Chinese source of SV proper. There is a general consensus that SV readings are quite late relative to other branches of SX. Maspero (1912: 14) thought that SV was based on ninth- and tenth-century Chinese. Mineya (1972: 164) more or less agrees with Maspero’s view, though he does not accept all of the details. Wang (1958: 300) attributes the origin of SV to 126

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

Tang Dynasty Chang’an Chinese. Hashimoto thinks SV originated from a southeastern Chinese koine. Although I have rejected the details of Hashimoto’s theory (Miyake 1999: 317–320), I suspect that SV did originate from the south. SV and many Yue dialects share an unusual innovation: *i in all closed syllables has centralized and lowered to a. As far as I know, this development of *-i- is unique to Yue. SV â [R] seems to reflect a Yue dialect with *-a- or *-R- corresponding to *-i- elsewhere: (5.33)

LMC
=‘heart’ *sim

Cnt sàm [sam]

SV tâm [tRm]

Md xcn [\in]

SK Kan-on sim simu

Northwestern LMC did not shift earlier *-i- to *-a-:
‘heart’ was spelled in Tibetan script as sim, with -i-, not *sam with *-a- (Luo 1933a: 210). There is no evidence for a Vietnamese-internal shift (*-i- > -R-) that took place after SV was borrowed. Furthermore, Vietnamese does have OSV forms with i corresponding to â [R] in SV and to i elsewhere: OSV tim <
EMC *sim ‘heart.’ Like Pulleyblank (1984: 205), I suspect that the change -i- > *-R- took place in the “Cantonese-like” Chinese dialect that was the source of SV. OSV tim ‘heart’ with -i- was borrowed before the change and SV tâm ‘heart’ with -â- was borrowed after the change. Canton was the capital of the Tang Dynasty’s Lingnan Province, which encompassed Jiaozhou (northern Vietnam). Although it is shaky to rely purely on one piece of evidence, I wonder whether SV originated as late Tang Dynasty Cantonese with a Vietnamese accent. SV cannot be based on a northwestern dialect such as Chang’an LMC for three reasons: (1) SV nasals correspond to Chang’an LMC prenasalized stops. This is the weakest of the three anti-Chang’an arguments because earlier prenasalized stops in Vietnamese merged with nasals (Gage 1985: 505–506): (5.34) *mp > m

*nt > n

*ãj > nh [ ã]

*Yg > ng [Y]

So in theory the Vietnamese who went to Chang’an could have brought back CLMC *mb, *nd, etc. and these prenasalized stops later became *m, *n, etc. (2) SV is anomalous in that it reflects a *-y- in Grade II absent in modern southern dialects and in the Tibetan transcriptions of northwestern LMC: e.g., the Grade II word
‘associate’ is SV giao [zaaw] < *kyaaw but is transcribed in Tibetan as ka’u or ke’u, not *kya’u or *kye’u (Luo 1933a: 210, 218). (3) Unlike SK, SV has the southern coda t instead of the northern LMC coda *r recorded in Tibetan transcriptions. According to Gage (1985: 510– 511), earlier *r generally became [y] after non-front vowels and zero after front vowels in Vietnamese: 127

OLD JAPANESE

(5.35) *-r > -y /V [-pal]_

*-r > -$ /V [+pal]

If SV were borrowed from late Chang’an before this change, then we would expect the following correspondence: (5.36) SV *-r > -y/-$

S. Chinese -t

:

SK -l

:

If SV were borrowed after the change *-r > -y/-$, the Vietnamese would have to render Chang’an *-r with some other consonant. Since Siamese, a language whose phonology is typologically similar to Vietnamese, renders foreign -r as -n, perhaps the Vietnamese would have done the same, resulting in these correspondences: (5.37) SV -n

:

S. Chinese -t

:

SK -l

But neither (5.36) nor (5.37) matches the actual correspondences: (5.38) SV -t

:

S. Chinese -t

:

SK -l

SV conforms to the southern rather than the northern pattern in this respect but not others (such as *-y- in Grade II). SV has one feature that is neither northern nor southern. The SV ngang tone corresponds to the MC Level Tone after second muddy as well as clear initials (Table 5.9). According to Hashimoto (1978: 16), this is “a characteristic found in no known dialect of modern Chinese.” It is also not found in the OSV equivalents of the Level Tone (Table 5.8). Furthermore, if one compares Table 5.9 with Table 5.8, one finds that the Vietnamese counterparts of the Rising and Departing Tones have ‘flip-flopped.’ I will not present my speculations about the reasons for these oddities here. I will simply note once more that one can determine whether a ‘Sinoid’ word in Vietnamese is OSV or SV by checking the tonal correspondences. So far, I have accumulated conflicting evidence about the areal origins of SV. What about SV’s temporal origins? I have three major reasons to assume that SV is LMC-based:

Table 5.9 Written SV tones LMC

Level (*-$)

Rising (*-q)

Departing (*-h) and Entering (*-C)

(2nd) Clear 2nd Muddy Muddy

ngang ngang huyVn

hni ngã ngã/n'ng

s8c n'ng n'ng

128

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

Table 5.10 The ‘branch/fat/genitive/small’ rhymes in MC, SV, and SJ Rhyme


‘branch’


‘fat’


=‘genitive’


‘small’

EMC OSV Go-on LMC Kan-on SV

*-iR -ia [iR] -e, -i

*-i -i [ii] -i

*-t -d [RR] -o, -i

*-ty -di [RRy] -e, -i

*-i -i -e [tt] (after certain LMC sibilants; see p. 108), -i [ii] (elsewhere)

(1) Merged finals: EMC finals have one-to-one correspondences with OSV, whereas multiple EMC finals correspond to a single SV final. In Table 5.10, EMC and OSV make a four-way distinction, modern Go-on makes a partial three-way distinction, and LMC, Kan-on, and SV have merged all four finals into a single final class. (2) Labiodentals: EMC bilabials became labiodentals in certain LMC Grade III rhymes. SV reflects these LMC labiodentals: (5.39) LMC *f *f® SV ph [f ] ph [f ]

*v v [v]

Note that SV ph was probably [ph] in the seventeenth century (Table 5.6). If SV ph were a fricative in de Rhodes’s time, he or the missionaries before him would have spelled Viet [f ] as f. *ph was the closest native Vietnamese equivalent to LMC *f(®) at the time of borrowing. (3) SV reflects the LMC depalatalization of *i after dental and retroflex sibilants (5.4 or 5.5). See (5.6) for an example. Summing up the above discussion, SV appears to be an LMC-based amalgam of northern and southern features. In terms of the traits in Table 5.11, SV is closest to Mandarin and Cantonese, less similar to CLMC, SK, and Kan-on, and very different not only from Go-on and EMC but also from OSV. I have yet to find a Chinese language which has the same combination of features as SV. Geography would seem to dictate a southern origin for SV, but I cannot speculate much further. In any case, it is clear that SV, like SK, is not based on the Chang’an dialect of LMC. Chw Nôm Chw nôm [ctt nom] (possibly from
 LOC *dzRq nRRm ‘characters southern’) is the Vietnamese counterpart to Japanese man’yôgana and Korean hyangchal. Although it too has fallen out of use, it was perhaps more successful than either of its northern cousins. Nôm was still in use in the early twentieth century, long after man’yôgana and hyangchal had passed 129

OLD JAPANESE

Table 5.11 Phonological traits across Chinese and SX Trait

CLMC

Kan

Md

SK

SV

Cnt

OSV

Go

EMC

a b c d e f g h i j

+ – – – –?/+ + + + + 3~4

+ – – – – + ? – – 5~6

– + + – ? + + + + 2~3

– ? + – + + – – + 4~5

– – + + – + + + + 0

– – – + – + + – +/– 2~3

– – – – – – – – – 6

– – – – – – – – – 6

– – – – – – – – – 6

Key to traits a Partial denasalization of nasals b Partial denasalization of *ã only c *-y- in Grade II d Shift of *i to R or a e *-t > -r. Note that early CLMC probably had *-t f Merger of EMC finals g LMC labiodentals h *m > *v (> Md w) i Depalatalization of *i after sibilants (pp. 108–9). Cantonese once had this trait, but lost it (Ramsey 1987: 101) j Number of differences from SV.

into obsolescence. Modern scholars still know how to read its thousands of characters, albeit in modernized pronunciation. The original pronunciations of nôm characters – like the original pronunciations of man’yôgana phonograms – were very different from those used today. Although Vietnamese vowels have apparently been quite stable over the time, we have seen that the Vietnamese consonant system has changed considerably in the last four hundred years. The finals of nôm character readings today are probably not much different from the finals of the characters in centuries past, but many of the initials must have been altered almost unrecognizably since the time of the script’s creation. All agree that nôm reflects pre-de-Rhodes Vietnamese phonology, but the age and origins of the script are obscure. NguyWn (1992: 384–397) summarizes the competing claims of various scholars concerning the invention of nôm, ranging from the nationalistic and mythological to the more reasonable and scholarly. Most of those claims are highly speculative and not worthy of mention here. I will briefly state my own views below. The Chinese rulers of Jiaozhi (later Jiaozhou) may have been the first to attempt to transcribe Vietnamese. Although I doubt that the Chinese had any great motivation to learn the language of their subjects, surely there were local names and words that were written with characters. See NguyWn 130

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

(1992: 387–388, 392–393) for two early attempts at Vietnamese transcription. These sporadic attempts did not yet constitute a system. After ten centuries of ‘northern rule,’ Vietnamese had been remolded into a language well suited for quasisinography. Formerly disyllabic and sesquisyllabic morphemes had been reduced to monosyllables which could each be rendered with one character. However, Vietnamese had more syllables than Chinese had. This made a man’yôgana-style Chinese-based syllabary impossible. Modern Vietnamese, which no longer has any of the initial consonant clusters that had characterized its ancestor, has 3,525 syllables, whereas its SV component has only 1,310 syllables (NguyWn Ngic San 1987, quoted in NguyWn 1992: 406). Vietnamese must have had more than 3,525 syllables before the loss of all consonant clusters. Downer (1987: 142–143) points out that modern Vietnamese has 71 rhymes but its SV component has only 33 or 34. Presumably earlier Vietnamese might have had a few more native rhymes. If the Vietnamese of today used only preexisting sinographs as phonograms with their official SV readings, they could not write two-thirds of the syllables in their language. This figure of two-thirds was probably more like three-quarters a thousand years ago. Sinography had to be drastically modified and expanded if it was to be a viable writing system for Vietnamese as a whole rather than just a means to record a word here or a name there. The Vietnamese were well acquainted with the workings of sinography and found many of the solutions to their problem within the preexisting Chinese tradition of writing. Following the Chinese model, the Vietnamese gradually decided to use sinographs for their approximate phonetic values either as loan characters or as building blocks for new phonetic compounds. These two types of characters formed the core of what would become the nôm script. An example of a nôm loan character is the aforementioned
SV hu [hii] ‘sorrow’/ SV hí [hii] ‘cry,’ a phonogram for an unrelated word which also happened to be of Chinese origin, OSV hdi [hRRy] ‘breath’ <
Western or Eastern Han *khRyh ‘id.’
‘sorrow/cry’ did not have an SV reading homophonous with hdi ‘breath’ but its sound was close enough. (Or was it? Although hdi was not in the SV syllabic inventory, why didn’t the Vietnamese choose a phonetic element such as
SV hki [haay] with an di [RRy]-like final instead? I will soon deal with this issue.) Some nôm ‘loan’ characters were simply the original characters for OSV words. OSV gan [gaan] <
EMC *C-kan ‘liver’ was written with
SV can [kaan] ‘id.’ (Gan was not in the SV syllabic inventory.) However, this principle was not used consistently. One did not write OSV hdi [hRRy] ‘breath’ with
SV khí [xii] ‘id.,’ even though OSV hdi was borrowed from the Western or Eastern Han word *khRyh ‘breath’ represented by
. One simply had to know which OSV words were written etymologically (e.g., gan ‘liver’) and which ones were not (e.g., hdi ‘breath’). 131

OLD JAPANESE

Both types of nôm loan characters were identical in appearance to ordinary sinographs. No one literate in sinography but ignorant of nôm would ever guess that
‘sorrow/cry’ stood for an indigenous Vietnamese word meaning
‘breath.’ An example of a nôm phonetic compound is c for n#m [nam] ‘five.’ This character did not exist in Chinese at all and would be utterly unintelligible to anyone without a knowledge of both sinography and Vietnamese. The left side of character is the phonetic indicator
SV nam [naam] ‘south.’ (N#m was not in the SV syllabic inventory.) The right side of the character is the semantic indicator
SV ngZ [Yuu] ‘five.’ The structure of the character c tells the reader that it stands for a word that sounds like
SV nam but means
‘five.’ Phonetic compounds are but one type of ‘vietographs’: graphs unique to nôm. Vietnamese also used a number of non-Chinese methods to create further types of nôm characters. I will mention only some of them below. Diacritics were added to phonetic elements: ( LLOC *-RRy (Starostin 1989: 497) (5.46) OC *-t > MC -oy (Baxter 1992: 578–579). I agree with Pulleyblank’s (1993: 364) observation: “spontaneous generation of a palatal glide by a central or back vowel, [R] or [£], seems completely unmotivated.” Pulleyblank (1993: 365) proposes that the MC palatal glide originated from an earlier OC velar glide: (5.47) LOC *-Rò A > EMC *-Ry (Baxter’s -oy) He does not specify what happened to the final *-ò in Type B syllables of the OC
‘genitive’ rhyme, but I think it simply monophthongized with the preceding vowel: (5.48) LOC *-Rò B > *-R > EMC *-t Third, OSV and nôm evidence seem to indicate a final *-ò in Type B
‘genitive’ rhyme. The Vietnamese seem to have borrowed this *-ò as native *-y. OSV sometimes renders the Type B
‘genitive’ rhyme as -di [RRy]: e.g.,
OSV khNi [xRRy] ‘arise.’ Type B
‘genitive’ rhyme graphs such as
SV kì/kz ‘his/her/its’ are bases for vietographs with a final [y] such as
cày [kay] ‘kind of insect.’ (2) Although Starostin (1989: 485) reconstructed the LOC Type B
‘fish’ rhyme as *-o (< Eastern Han *-a), this rhyme regularly corresponds to OSV ea [tR] (Wang 1958: 367–368, Pulleyblank 1984: 209), not to OSV ô [o] or o [Q]. Nôm bases of the LOC Type B
‘fish’ rhyme are not used in vietographs for words with labial vowels (u, ua, ô, o). Hence I will reconstruct the vowel of the LOC Type B
‘fish’ rhyme as nonlabial *a instead of labial *o. I do agree with Starostin’s reconstruction of the LOC Type A
‘fish’ rhyme as *-oo (< Eastern Han *-aa), though I do not think the vowel length needs to be specified (see below). This rhyme regularly corresponds to a [aa] in the older layer of OSV and to o [QQ] in the newer layer. Nôm bases of this rhyme consistently appear in vietographs for words with labial vowels (u, ua, ô, o). Apparently the earlier OC velar glide coda *-ò was lost in this rhyme by the LOC period.

141

OLD JAPANESE

Thus I revise Starostin’s reconstructions of the LOC Type A and B
‘fish’ and
‘genitive’ rhymes as follows: (5.49) Rhyme
‘fish’
‘fish’
‘genitive’
‘genitive’

Type A B A B

Starostin’s LOC Miyake’s LOC *-oo *-oA *-o *-aò B *-tt *-Rò A *-t *-Rò B

I will explain other differences between my reconstruction and his shortly. (3) In (5.49) I have emended Starostin’s LOC *t(t) to *Rò. OSV shows that OC *R in the Type B
‘genitive’ rhyme may have persisted into LOC and perhaps even into EMC. According to Starostin’s system of reconstruction, the Type B
‘genitive’ rhyme word
‘time’ should appear in various stages of OC and EMC as: (5.50)
‘time’

Classical OC Western Han *d(h)R *d(h)R ELOC EMC *j(h)t *j(h)t

Eastern Han *d(h)yR > *j(h)R

Starostin (1989: 485) believes that *R raised to *t between Eastern Han OC and Early LOC of the third century. Therefore OSV words of the Type B
‘genitive’ rhyme from Eastern Han should have the final -d [RR] (the short vowel -â [R] is not permitted as a Vietnamese final) and newer OSV words should have the final -e [tt]. But we rarely find -e [tt] as an OSV counterpart of the OC Type B ‘genitive’ rhyme. We find -di [RRy] or -d [RR]: e.g.,
OSV thIi [t h RRy] ‘time’ and
OSV cI [kRR] ‘flag’ (Pulleyblank 1984: 212). Notice the [y] in OSV thIi ‘time’ which looks like a trace of the LOC final velar glide *-ò. Pulleyblank 1984 and Starostin 1989 agree that EMC did not have any *-ò in the Type B
‘genitive’ rhyme, so the absence of a trace of LOC *-ò would be a sign of late origin. Hence I believe that OSV -di [RRy] words such as
thIi ‘time’ reflect LOC *-Rò B (5.48) and that OSV -d words such as
cI ‘flag’ reflect late LOC *-R. Therefore I would revise (5.50) as in (5.51). (5.51)

Classical OC
‘time’ *d(h)Rò B ELOC *j(h)Rò B

Western Han *d(h)Rò B EMC *j(h)t

Eastern Han *d(h)yRò B > *j(h)Rò B

(4) As mentioned earlier, Starostin (1989) believes that the OC Type A/B distinction involved vowel length. He came to this conclusion largely on the basis of correlations between Types A and B in OC and vowel length in the 142

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

Tibeto-Burman Lushei language. OC Type A corresponded to Lushei long vowels and OC Type B corresponded to Lushei short vowels. But Starostin (1989: 328) himself admits that out of the OC–Lushei cognates presented in chapter 3 of his book, OC Type A corresponded to Lushei short vowels in 42 cases and to Lushei long vowels in only 32 cases. As Pulleyblank (1994: 91) writes, “even if one accepts all Starostin’s [sic] cognate word pairs (and many of them seem to me to be doubtful) the statistical correlation he claims to find between Lushei long vowels and Type A is hardly impressive.” Starostin found 72 cases of OC Type B corresponding to Lushei short vowels. He does not give the exact number of exceptions, but notes only that there are not more than ten of them. There very well may be a correlation between the OC Types and Lushei vowel length partly obscured by secondary lengthening in Lushei and/or shifts within Chinese. Nevertheless I remain agnostic and unwilling to commit to Starostin’s interpretation of the OC Type A/B distinction. Pulleyblank once advocated a similar hypothesis (1962) but now views the OC Type A/B distinction as a “prosodic opposition between prominence on the first and prominence on the second mora of the syllable” (1994: 94). He has found such an opposition in Sizang Chin and cites four Sizang Chin numerals “whose relation to Chinese is uncontroversial” to show that the position of the ‘syllabic peak’ in Sizang Chin corresponds with the OC Types (1994: 92). However, he admits that the syllabic peaks of two Sizang Chin numerals do not match the types of their OC cognates. Pulleyblank’s hypothesis requires much more testing. I currently think that Norman’s (1994) hypothesis of a pharyngeal origin for the OC Type A /B distinction may be the most likely, but I am hesitant to commit to it in its present form. For now, I will ignore Starostin’s OC vowel length and indicate the type of an OC syllable with a superscript ‘A’ or ‘B’ at the end of a rhyme. (5.52)

OC (Starostin 1989) OC (Miyake) Type A *VV *V A Type B *V *V B

Apart from the revisions explained above and a number of typographical modifications listed in my initial list of abbreviations, symbols, and conventions, I will quote all OC forms according to Starostin (1989).

The reconstruction of Early Middle Chinese Karlgren’s ‘Ancient Chinese’ (my EMC) reconstruction, first presented in (1915–26) and published in final form in (1957), though extremely outdated, remains influential today. Karlgren was a pioneer in drawing data not only from dialects throughout China but also from all three branches of Sinoxenic. 143

OLD JAPANESE

Many other reconstructions came after his, but most of them did not represent significant advances. Too many Karlgrenoid reconstructions imitated his complex fifteen-vowel system and, worse yet, added further quirks to it. Many EMC reconstructions bear little resemblance to any variety of Chinese or Sinoxenic in terms of phonological typology. However, the realism of the consonant systems of Karlgrenoid reconstructions has rarely been seriously questioned. One can easily and independently verify much of Karlgren’s original EMC consonants through the comparative method without any knowledge of the ‘36 initials’ listed in Yunjing. Even post-Karlgrenian reconstructions such as Pulleyblank (1984), Starostin (1989), Baxter (1992), and Coblin (1994) preserve most of Karlgren’s original sound values for EMC consonants. Since there is so much consensus on the initials, I decided to choose Pulleyblank’s (1984, 1991a) EMC reconstruction mainly because I believe that his reconstruction of the finals is the best available. I will enumerate my reasons for this belief in the next section. Post-Karlgrenian EMC reconstructions There are at least four published post-Karlgrenian reconstructions of EMC which lack the convoluted vowel systems of earlier efforts: Pulleyblank (1984, 1991a), Starostin (1989), Baxter (1992), and Coblin (1994). Coblin (1994) is the least useful for us because it is a reconstruction of ‘Old Northwestern Chinese.’ I doubt that the EMC basis of the orthography of Kojiki poetry originated from the northwest. The Three Kingdoms of the Korean peninsula probably got their sinographic readings from the east and south of China rather than the distant, backward northwest. Prior to the unification of China by the Sui, peninsular peoples would have little motivation to journey to the former Han capital of Chang’an and bring back its rustic pronunciations. If the term Go-on (‘sounds of Wu’; i.e., sounds of the south) is truly meaningful, then it is further counterevidence against a northwestern EMC origin for System D, the basis of Go-on and the phonogram orthography of the poetry of Kojiki. I will not use Coblin’s Old Northwestern Chinese, but I will review and reject his northwestern LMC reconstruction in the following section. The remaining three EMC reconstructions attempt to express the abstract phonological distinctions of the ‘language’ of the Qieyun dictionary of 601 CE (p. 93). Pulleyblank, Starostin, and Baxter all agree that the distinctions of Qieyun were real, but their interpretations of those distinctions differ considerably. Pulleyblank’s (1984, 1991a) interpretation stands out among the three because it is the only one which uses abundant and appropriate evidence. Besides philological sources such as Qieyun, he uses three types of modern sources: (1) the colloquial strata of Min dialects, (2) Go-on and man’yôgana (which potentially entails circularity; see Chapters 3 and 6), and (3) OSV. 144

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

No other reconstruction of OC or MC has ever tapped the rich mines of OSV. Furthermore, his reconstruction has a largely typologically plausible vowel system not unlike that of Vietnamese and can neatly account for Go-on, OSV, and many of the Min forms. Starostin (1989) tried to reconstruct ‘Middle Chinese’ (EMC)12 by combining the initials already reconstructed by other scholars (primarily Karlgren 1915–26 and Li Rong 1952) with finals derived via the comparative method. His choice of sources for comparison was extremely unfortunate. He restricted himself only to Sinoxenic, selecting EMC-based Go-on and LMCbased Sino-Vietnamese and Sino-Korean. Since the latter two are much newer than Go-on, they lack many of the distinctions of Qieyun. When one looks at Starostin’s (1989: 23–27) table of correspondences between SV, SK, and Go-on, one is struck by the number of Qieyun finals which are not distinguished in all three. Forty-six sets of SV, SK, and Go-on finals corresponded to fifty-eight Qieyun finals. A strict application of the comparative method would require Starostin to reconstruct only forty-six ‘Middle Chinese’ finals. Yet he manages to fill in the holes by plucking pieces of others’ reconstructions and picking fragments from modern Chinese dialects. The end result is not as bad as one might expect – in fact, it is not too different from Pulleyblank’s solution – but it does have some major flaws. Starostin reconstructs a lax–tense distinction for EMC vowels (underlining signifies tenseness): (5.53) lax *a, *ä, *u, *e, *o

vs.

tense *a, *ä, *u, *e, *o

and assigns the tense vowels to Grade III syllables without any explanation. I do not know how he came up with this unusual interpretation of Grade III, since the SX branches he worked with do not have a lax–tense distinction. He also takes advantage of his knowledge of previous OC reconstructions to posit retroflex vowels which correspond to OC r-vowel clusters but are completely absent in modern languages. Like Karlgren, Starostin reconstructs *-ie- in Grade IV, even though such an EMC diphthong was long ago discredited by Maspero (1920) and is absent in many Karlgrenian as well as post-Karlgrenian reconstructions which reconstruct a simple *-e- or *-ε-. Although Starostin’s OC is based on his EMC reconstruction, the above problems and others do not greatly mar his OC. Since his OC reconstructions can be verified by looking at pre-MC archaisms in the colloquial strata of Chinese and pre-MC loans in Vietnamese and Tai, it seems that he came to the right conclusions about OC in spite of the errors in his EMC reconstruction. Thus I have chosen to use his version of OC but not his version of EMC. As for Baxter’s (1992: 27) version of EMC, Baxter himself denies that it is a reconstruction: “The notation I introduce here is not intended as a reconstruction; rather it is a convenient transcription which adequately represents 145

OLD JAPANESE

all the phonological distinctions of Middle Chinese while leaving controversial questions open.” In spite of this declaration, Baxter does use his ‘transcription’ as a de facto reconstruction upon which he builds his OC reconstruction. If I were also to use his ‘transcription’ as a de facto reconstruction, I would have to make a number of typographical changes, since his transcription is inconsistent and confusing (e.g., *[y] is represented by both *y and *j, and the latter can also represent [ò] before a back vowel; Baxter 1992: 55, 579–580). But such changes would not alter the fact that his EMC cannot always account for Go-on or OSV: (5.54) Rhyme
‘genitive’
‘anxious’

Pulleyblank Baxter Go-on OSV *-t *-i o, i d(i) [RR(y) ] *-uR *-ju [òu] u ua [uR]

Older Go-on -o and d(i) [RR(y) ] make no sense if the
‘genitive’ rhyme were Baxter’s high front i. Pulleyblank’s *t is not a perfect match, but at least it is a non-front vowel. And where would the schwa in OSV ua [uR] come from if the
‘anxious’ rhyme were Baxter’s *-ju [òu]? I do not wish to give the impression that Pulleyblank’s reconstruction is the last word on the subject. There are a number of aspects of his EMC reconstruction which I find questionable. In some cases I favor Starostin’s or Baxter’s solutions and in others I favor my own. But since (1) these questionable aspects are almost entirely in the finals and (2) I will use Pulleyblank’s finals as convenient labels for EMC rhymes rather than as primary evidence for OJ vocalism, I will list only a few emendations below. Proposed revisions to Pulleyblank (1984, 1991a) Pulleyblank has published two slightly different reconstructions of EMC (1984, 1991a). The reconstruction used in this book is based on his (1991a) reconstruction with the following alterations: (1) Pulleyblank (1991a: 12) has changed his (1984) reconstructions of the EMC Grade II vowels *ar and *er to *at and *Rt. I find these new diphthongs to be dubious. Pulleyblank is correct in noting that Tai does have diphthongs ending in *-t. However, I know of no Tai language that allows such a diphthong in medial position: i.e., syllables such as *CatC are probably impossible in Tai languages. Yet Pulleyblank’s (1991a) EMC is full of syllables such as
EMC *kRtp ‘first Heavenly Stem.’ Moreover, I have never seen any forms in Chinese or in other languages which suggest *Vt diphthongs in Grade II. Colloquial Taiwanese e and OSV e [ε] regularly correspond to Pulleyblank’s EMC *at: compare
EMC *γ atq ‘summer’ with Colloquial Taiwanese he6 and OSV hè [hεε] ‘id.’ If Pulleyblank were correct, we might expect forms such as Colloquial Taiwanese *hai 6 and OSV *hày [hay]. I will abandon Pulleyblank’s (1984) and (1991a) Grade II vowels 146

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

in favor of Baxter’s (1992: 61) *æ and *ε, which account for the data without resorting to marked retroflex vowels or exotic diphthongs (1992: 61). (2) If I follow Baxter in reconstructing *ε in Grade II, this would conflict with Pulleyblank’s (1984, 1991a) reconstruction of the EMC Grade IV vowel *ε. Pulleyblank (1984: 198–199) himself gives the solution when he points out that the usual OSV rendition of this vowel is ê [e]. OSV e [ε] is a rarer rendition and is otherwise associated with Grade II (see above), not Grade IV. Since Pulleyblank’s (1984, 1991a) EMC has no *e whatsoever, I will substitute EMC *e for his EMC *ε. EMC *e would account for the OSV data better. (3) Pulleyblank (1984, 1991a) reconstructs the vowel *Q for the Grade I
‘model’ rhyme on the basis of its OSV counterpart o [Q]. This vowel does not contrast with an upper-mid vowel *o. Pulleyblank (1991a: 7) suggests that “one could substitute o for Q without ambiguity for practical purposes.” I will adopt his suggestion. (4) Other than the diphthongs (*at, *Rt) discussed in (1) above, Pulleyblank (1984, 1991a) reconstructs three diphthongs (*iR, *tR, *uR) in syllable-final position which are in near-complementary distribution with three diphthongs (*ia, *ta, *ua) in syllable-medial position: i.e., CVR and CVaC but not *CVa or *CVRC. There are only two exceptions to the above generalization: the common rhyme *-ia and the very rare rhyme *-ta. I believe that these diphthongs are in need of reinterpretation. Let us ignore the rhymes *-ia and *-ta for the moment and look at the Go-on and OSV renderings of (*iR, *tR, *uR) and (*ia, *ta, *ua): (5.55) Final Medial Final Medial Final Medial

EMC *-iR *-ia*-tR *-ta*-uR *-ua-

Go-on -e -e-o -o-u-u-

OSV -ia -iê-ea -ed-ua -uô-

[iR] -[iR][tR] -[tR][uR] -[uR]-

Pulleyblank’s differentiation between three *-VR and three *-Va- diphthongs is unnecessary, since it appears that the Japanese and the Vietnamese perceived medial *-Va- and *-VR similarly. I will simplify (5.55) by positing only three *VR diphthongs: (5.56) Final Medial Final Medial Final Medial

EMC *-iR *-iR*-tR *-tR*-uR *-uR-

Go-on -e -e-o -o-u-u-

OSV -ia -iê-ea -ed-ua -uô147

[iR] -[iR][tR] -[tR][uR] -[uR]-

OLD JAPANESE

This would eliminate the need for a diphthong *ua. But how can I eliminate the two remaining diphthongs ending in *-a (*-ia and *-ta)? Go-on and OSV render EMC *-ia with (y)a, not -e or -ia. If Pulleyblank’s identification of medial *-ia- (our *-iR-) with final *-ia were correct, then why was *ia not rendered the same way in both medial and final position (5.57)? (5.57)

EMC Go-on OSV Final *-ia -ya -a [aa] Medial *-ia- (my *-iR-) -e-iê- [iR]

I could eliminate *-ia by rewriting it as *-ya, which does not otherwise occur in Pulleyblank’s reconstruction. (Pulleyblank reconstructs medial *-yonly before *i.) Compare the Go-on and OSV renditions of EMC *-ya (Pulleyblank’s *-ia) and EMC *-a (5.58). (5.58)

EMC Go-on OSV Final *-ya (Pulleyblank’s *-ia) -ya -a [aa] Medial *-a -a -a [aa]

Now we only have four EMC diphthongs left (*iR, *tR, *uR, *ta). *-ta is of such low textual frequency that one is tempted to overlook it, but it does appear among the OJ phonograms, so we cannot ignore it. *-ta appears only after velars and is restricted almost entirely to phonograms for Indic velar-initial syllables. According to Pulleyblank (1965, 1984: 167–168), EMC velars had back allophones before back vowels. EMC */ka/ may have been *[qa]. Therefore EMC had no native *[ka]. When EMC speakers had to transcribe Indic syllables such as *[ka], they created special phonograms such as
EMC *kta ‘transcriptional character for Indic ka and ka’. I would like to reconstruct the rhyme *-ta with a velar glide *-ò- as *-òa, which would be parallel to my revision of *-ia with a palatal glide *-y- as *-ya. (5) Pulleyblank (1991a) reconstructed the voiceless and voiced ‘throat sounds’ of EMC as velars (*x and *γ). Norman (1988: 31) argued that perhaps we should take the term ‘throat’ literally and reconstruct these sounds as glottals (*h and *®). I will do precisely that, though they may have been velars. The evidence is ambiguous. Pulleyblank (1962) himself once reconstructed them as glottals and as late as 1984 (163–164) believed that they were glottals in southern EMC. (6) See my initial list of abbreviations, symbols, and conventions for minor typographical substitutions such as *y for *j. I list the consonants and vowels of our modified version of Pulleyblank’s (1984, 1991a) reconstruction in Table 5.13. *h and *® may have been *[x γ ], so I have added these possibilities in parentheses. 148

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

Table 5.13 The consonants of EMC

Voiceless stops/affr. Voiceless fricatives Aspirated stops/affr. Voiced stops/affr. Voiced fricatives Nasals Glides

Labial

Dental

Retroflex

Palatal

Velar

Glottal

*p

*t

*tr

*th *d

*m *w

*n *l

*k (*x) *kh *g (*γ) *Y *ò

*q *h

*ph *b

*c *\ *ch *j *9 *ã *y

*ts *s *tsh *dz *z

*thr *dr

*cr *Z *chr *jr *|

*nr

*®

Table 5.14 The vowels of EMC

High High + Central Upper mid Lower mid Low

Front

Central

Back

*i *iR *e *ε *æ

*t *tR *R

*u *uR *o *a

The distinction between *ε and *æ may have been an archaism in Qieyun. Rhyme classes with those vowels are poorly distinguished in OSV and Go-on and have merged in modern Chinese languages and the rest of SX. The syllable structure of my modified EMC is: (5.59) C(G1)V(C) + Tone C = {all consonants in Table 5.13 except *ò} (G) = {*w, *y, *ò} V = {all vowels in Table 5.14} (C) = {*p, *m, *w, *t, *n, *y, *yk, *yY, *k, *Y, *wk, *wY} Tone = {zero = Level, *-q = Rising, *-h = Departing, *-stop = Entering} *yk and *yY are palatalized velars and *wk and *wY are labialized velars. Starostin (1989) also reconstructs these codas for EMC. I use Pulleyblank’s (1984, 1991a) unusual notation for these four sounds. They could otherwise be written as *ky, *Yy, *kw, *Yw.

The reconstruction of Late Middle Chinese (LMC) LMC has not received the attention that has been showered upon EMC, largely because of the paucity of materials. There is no LMC equivalent to Qieyun. We do have the list of 36 initials in Yunjing and other rhyme tables, but the rhymes in those tables are taken from Qieyun and do not represent 149

OLD JAPANESE

the rhymes of LMC. I know of only two post-Karlgrenian reconstructions of LMC and will briefly review them below. Post-Karlgrenian LMC reconstructions After revolutionizing the OC field in (1962), Edwin Pulleyblank published the first draft of his Late Middle Chinese reconstruction in (1970–71). Pulleyblank’s premises were radically different from those of his predecessors. He viewed rhyme tables as evidence for northwestern Chang’an LMC rather than for southern EMC. He also viewed the Four Grades of the tables in a new way. Previous scholars had tried to make sense of the Four Grades via EMC with little success. But having freed himself from the need to shove EMC into the supposedly LMC framework of Yunjing, Pulleyblank declared that the Grades of LMC were grades of palatalization, ranging from zero to maximal. Nearly fifteen years later, Pulleyblank presented a revised reconstruction of LMC in his (1984) monograph. This reconstruction reappeared with slight modifications in his (1991a) Lexicon. Meanwhile, W. South Coblin had been pursuing the problem of LMC (and of northwestern MC in general) from a completely different perspective. Instead of relying upon rhyme tables which he rejected as artificial, he sought to reconstruct ancient northwestern speech on the basis of Chinese transcriptions of foreign languages and foreign transcriptions of Chinese. Luo (1933a) had barely dared to venture beyond Karlgren’s EMC reconstruction, but Coblin (1994) devised reconstructions of northwestern LOC, EMC, and LMC that were intended to be a radical departure from those of his predecessors. Both of these takes on Chang’an LMC would seem to be indispensable for the interpretation of the CLMC-based Shoki orthography. If two researchers could arrive at the same results independently using different methodologies, we could safely assume that they had approximated the truth. Unfortunately, this is far from the case. Pulleyblank’s LMC reconstruction is an example of theory overriding data. Although Pulleyblank is always to be commended for collecting data neglected by his peers – particularly Sinoxenic data – his ultimate (1984, 1991a) version of LMC relies heavily on his assumptions. For example, he reconstructed a medial *-y- in Grade II for Chang’an LMC, even though such a yod is exceedingly rare in the extensive transcriptive materials available for northwestern Chinese, infrequent in SK (which may not even be based on CLMC; see p. 112), and absent from Kan-on. He believed that CLMC was the standard language of Tang and spread to the south, yet southern Chinese has no trace of this yod. This is all very ironic since Pulleyblank has protested (rightfully, I believe) against the Karlgrenian reconstruction of yod for Grade III in EMC and Type B syllables of OC. 150

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

The problems with Pulleyblank’s reconstruction are not restricted to Grade II. Unlike his EMC reconstruction which accounts nicely for Go-on and OSV, his LMC reconstruction cannot account for any of the later strata of SX. Why do Vietnamese long vowels (SV cao [kaaw] ) correspond to LMC short vowels (
LMC *kaw ‘high’)? Why would the Vietnamese, who had a perfectly good final -ua [uR], borrow
LMC *muR ‘model’ as mô [moo] and
LMC *YüR ‘anxious’ as ngu [Yuu]? Why not simply reconstruct LMC *mo and LMC *Yu? There are many other cases of mismatches between attested forms and his reconstruction which I will not list here. Occasionally Pulleyblank tried to reconcile mismatches with dubious results. According to his ‘grades of palatalization’ theory (5.2), Grade III was the second most palatalized Grade and Grade IV is the most palatalized Grade. Grade III was characterized by the vowel *i and Grade IV was characterized by the glide-vowel combination *yi. SK violates this theory because SK has u [t] in Grade III and i [i] in Grade IV. In order to make SK fit his ‘grades of palatalization’ theory, he reinterpreted Late Middle Korean u as [i] and i as [yi] (1984: 100–104). This would imply an otherwise unjustifiable chain shift (*yi > *i > t) between LMK and modern Korean. But Pulleyblank neither pursued the consequences of his claims for Korean historical phonology nor cited Korean-internal evidence to justify his claims. Coblin (1994) is the antithesis to Pulleyblank’s theory-heavy approach. His Compendium of Phonetics is almost all data and no theory. Surely Coblin, grounded more firmly in the real world, would be more likely to produce realistic results. Regrettably, the very nature of Coblin’s methodology bars him from obtaining such results. Coblin principally draws upon two types of data: Tibetan transcriptions of Chinese and Chinese transcriptions of Indic languages. The Tibetan script of the time had only six vowel letters. It probably could not accurately record all of the phonemic distinctions of LMC. It is roughly analogous to katakana transcriptions of English in phrase books for Japanese tourists. Using those transcriptions, one could not reconstruct English sounds such as [æ R θ 2 W] which have no Japanese equivalents. The Chinese Buddhist transcriptions pose a different problem. Ignoring vowel length and syllabic liquids (è, ù, ê), Sanskrit has only a five-vowel system (a, i, u, e, o). Middle Indic languages also have a five-vowel system. The Chinese would naturally tend to use characters with vowels similar to the five Indic vowels in their transcriptions. Coblin (1994: 79) finds himself unable to learn much about the early northwestern sound value of the
‘fish’ rhyme because it is rarely used to transcribe Indic. Obviously, the
‘fish’ rhyme was among the rhymes that did not resemble anything in Indic. How much could we determine about English phonology based on English transcriptions of Indic words? We could not learn anything about English sounds not used in Indic transcription. We would never know that English had a /z/ or /W/ on the basis of transcriptions of Sanskrit, which has no /z/ or /W/. 151

OLD JAPANESE

Pulleyblank (1998: 201) also points out that Coblin is not entirely liberated from the philological tradition that he rejects. Coblin’s ‘Qieyun System’ is but a minor revision of Karlgren’s ‘Ancient Chinese.’ Coblin is more conservative than he realizes. Coblin (1994: 49–50) reconstructs an “intrinsically front” *i in Grade III and an “anti-palatal” *t in Grade IV. Since the Indic data he discusses for these vowel do not have a distinction between /i/ and /t/, he apparently falls back on the Karlgrenian assumption that Grade III was more palatal than Grade IV. This contradicts everything else known about those two Grades. Compare Coblin’s (1994: 49, 217, 226, 241) Sui-Tang Chang’an (STCA) reconstructions with (1) Coblin’s ‘Qieyun System’ (QYS), (2) the Tibetan transcriptions listed in Coblin (1994), (3) SK, and (4) SV in (5.60). (5.60)
‘sad’
‘compare’
‘machine’
‘discard’

Grade III IV III IV

STCA *pi *bt *ki *kht

QYS *pyi *bi *kyei *khi

Tib. pyi bt, ’byi, ’byt kt (none)

SK SV pi bi [(ii] pi tì [tii] kuy [kty] cd [kRR] ki khí [xii]

Notice that STCA *i correlates with the presence of a *y in QYS and that STCA *t correlates with the absence of a *y in QYS. So far, so good. However, the Tibetan transcriptions only confuse us: Tibetan yi and t correspond to STCA *i, and Tibetan t, yi, and yt correspond to STCA *t. The SK and SV forms, on the other hand, have a pattern. SK distinguishes Grades III and IV only after velars and SV distinguishes those Grades only after labials. SK uy [ty] corresponds to STCA *i and SK i corresponds to STCA *t. This is precisely the reverse of what we would expect given Coblin’s STCA reconstruction:
SK *ki ‘machine’ and
SK *kuy [kty] ‘discard.’ If Mineya (1972) et al. are correct in deriving SV t from a palatalized labial, then why is SV t in Grade IV (
SV tì ‘compare’) if Grade IV had Coblin’s “anti-palatal” vowel *t? And why is SV t not in Grade III (
SV bi ‘sad’), which Coblin reconstructs with an “intrinsically front” *i? The vowel of
SV cd [kRR] ‘machine’ (Grade III) also does not suggest an “intrinsically front” *i. It is true that
SV cd’ ‘machine’ is a ‘carryover’ reading from pre-Tang OSV; the expected LMC-based SV reading would be *ki. Nonetheless Coblin’s (1994: 241) fifth-century ‘Old Northwestern Chinese’ reconstruction *ki for
‘machine’ does not explain the non-front vowel in SV cd. On pp. 112 and 127, I have demonstrated that SK and SV have different origins. If they both reflect Grade III as non-palatal and Grade IV as palatal, these manifestations of the Grades are most likely to be traits shared by their source LMC dialects rather than coincidental Korean and Vietnameseinternal developments. It is theoretically possible that Chang’an LMC could have reversed the pattern found elsewhere, but it would be simpler for us to 152

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

assume otherwise. If we ignore the hypothetical Karlgrenian QYS and look at the hard data, we see that Indic and Tibetan tell us nothing about Grades III and IV but that SK and SV do tell us something quite different from Coblin’s account. This is not much ado about nothing, for all four of the characters in (5.60) appear in Shoki.
‘sad’ and
‘machine’ are phonograms for OJ piy and kiy with B-type iy (NS 123.4.5a, 7.2.3) and
‘compare’ and
‘discard’ are phonograms for OJ pyi and kyi with A-type yi (NS 6.2.1, 96.21.3) We will see in Chapter 8 that OJ B-type iy correlates with Grade III and that OJ A-type yi correlates with Grade IV. The Grade III/IV distinction is crucial for the reconstruction of these two OJ ‘vowels.’ Neither Pulleyblank’s nor Coblin’s LMC reconstruction is satisfactory. The ideal solution would be to use the comparative method to create reconstructions untainted by either Karlgrenian or post-Karlgrenian preconceptions. But such a task is far beyond the scope of this book, so I shall not attempt it here. In the next section, I will explain how I will approximate LMC without using either Pulleyblank’s or Coblin’s reconstructions. The initials of Chang’an LMC Since no LMC equivalent of Qieyun exists, no lists of LMC finals compiled by LMC speakers exist. The post-LMC rhyme tables such as Yunjing still listed the long obsolete finals of Qieyun. Instead of trying to reconstruct LMC finals, I will use SK and SV as my main witnesses for LMC finals. As we have seen, the Tibetan transcriptions of northwestern LMC finals are inconsistent and unreliable. Even though SK and SV probably do not descend from Chang’an LMC, the similarities between their finals and those of modern Kan-on suggest that there were no radical differences between the finals of their source LMC dialects and the Chang’an LMC source of Kan-on. However, SK and SV cannot tell us much about the initials of Chang’an LMC for two reasons. First, they were apparently based on LMC dialects which had not undergone the denasalization that was peculiar to CLMC. Second, neither currently reflects the original consonantal distinctions of LMC. SK aspiration is not correlated with aspiration in Chinese or in SV. Many consonant mergers and shifts have taken place in SV. Thus I will recycle the consonants of modified EMC (Table 5.13) with modifications to serve as a makeshift substitute for the initials of CLMC. But it is not clear whether all the EMC consonantal distinctions survived into CLMC. Pulleyblank (1984, 1991a) reconstructs retroflexes but no palatals because he believes that EMC palatals had merged with EMC retroflexes in LMC. Although late LMC-based SV still maintains a distinction between the two series, we cannot be certain that such a distinction existed in the northwest as well as in the south. Coblin (1994) reconstructs separate series 153

OLD JAPANESE

of palatals and retroflexes for Sui-Tang Chang’an, which would have been the basis of Kan-on and the orthography of Shoki. I believe it is best to err in the direction of overdifferentiation rather than underdifferentiation, so I will keep LMC palatals separate from LMC retroflexes. I will make only three changes to the EMC consonants of Table 5.13 to reflect the major characteristics of CLMC. (1) Pulleyblank (1991a: 6) believes that EMC voiced obstruents had become LMC ‘muddy’ obstruents “having voiceless onset with voiced aspiration or murmur at their release.” Even without referring to the voiceless modern Kan-on counterparts of LMC ‘muddy’ obstruents, we can confirm this by looking at the Tibetan transcriptions (Luo 1933a): (5.61) LMC *k *kh *k® Tibetan transcription k, g kh k, kh, g Similar patterns of correspondence exist for unaspirated, aspirated, and ‘muddy’ obstruents at other points of articulation: (5.62) LMC Tibetan transcription

*p p, b

*ph *p® ph ph, b

Apparently the Tibetans perceived ‘muddy’ obstruents as either nonaspirates (k, g, b) or aspirates (kh, ph). Pulleyblank’s *C® reconstructions account for both patterns of transcription. Therefore I will rewrite all EMC voiced obstruents (*b, *d,*g, etc.) as *C® clusters (*p®, *t®, *k®, etc.). (2) Unlike EMC, LMC distinguished between ‘heavy lip sounds’ (bilabials) and ‘light lip sounds’ (labiodentals). The ‘heavy/light’ distinction must have been known to the scribes of Shoki, since it is attested in the fanqie for Yan Shigu’s Han shu commentary of 641. The ‘light lip sounds’ have been reconstructed in a number of different ways (Table 5.15). Table 5.15 Reconstructions of the LMC labiodentals Yunjing initial name


‘not’ Unaspirated


‘spread’ Aspirated


‘present’ ‘Muddy’


‘small’ Nasal

Tib. translit. (Luo 1933a) Tôdô (1978, 1980) Pulleyblank (1984, 1991a) Coblin (1994) Miyake

ph *f *f *pf *f

ph *fh *f *phf *fh

ph, b, ’b *v *f® *bv *f®

m, b, ’b *w *v *m(v) *mv

Note Coblin (1994: 58) reconstructs *m for seventh-century Sui-Tang Chang’an and *mv for eighthcentury Mid-Tang Chang’an.

154

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

My reconstruction is a compromise between those of Tôdô (1978, 1980), Pulleyblank (1984, 1991a), and Coblin (1994). Although both Pulleyblank and I doubt the existence of a distinction between *f and *fh – there would be no other pair of fricatives distinguished by aspiration in LMC – I will err on the side of overdifferentiation and keep *f and *fh separate. For consistency with other ‘muddy’ obstruents, I will preserve Pulleyblank’s reconstruction of the ‘muddy’ labiodental as *f®. I will revise Pulleyblank’s *v as *mv to reflect the nasality evident in the Tibetan transcriptions m and ’b (probably *[mb] ). Following Coblin (1994: 34, 59), I will rewrite EMC bilabial initials (*p, *ph, *b, *m) followed by the vowel *u as CLMC labiodental initials: (5.63) EMC *p, *ph, *b / _u > CLMC *f, *fh, *f® (5.64) EMC *m / _u (except / _uw) > CLMC *mv A separate rule (5.64) is necessary for *m because EMC *m does not shift to CLMC *mv if it is followed by a *w: EMC *muw, *muwY, *muwk all have initial *m in CLMC (Pulleyblank 1984: 90). (3) *mv in (5.64) above is one of six prenasalized obstruent initials (*mb, *mv, *nd, *ndr, *ã9, *Yg) in LMC that developed from EMC nasal initials in all but certain environments (see below). All six correspond to modern Kan-on voiced obstruents and voiced obstruents prefixed with a.chung (symbolized here as an apostrophe) in Tibetan transcription (also see 5.42). Uighur transcriptions (Barat n.d.) also show denasalization in five of the six initials (*ndr is not represented in the data): (5.65) EMC LMC Kan-on Tib. translit. Uighur

*m *mb b ’b p

*m *mv b ’b v

*n *nd d ’d l

*nr *ndr d ’d –

*ã *ã9 z 9 z

*Y *Yg g ’g k

The exact circumstances under which denasalization was blocked are unclear. Coblin (1994: 58) thinks the process “occurred primarily in syllables whose finals lacked a nasal element of any kind . . . in some varieties of northwest Chinese it may have begun to spread beyond its original phonetic environment to syllables having final nasal elements.” For now, I will regard nasals before finals without nasal elements as denasalized: *m, *n > *mb, *nd, etc. I will regard nasals before finals with nasal elements as potentially denasalized: *m(b), *n(d), etc. In any case, denasalization was probably subphonemic in CLMC. Nothing in Kan-on or the transcriptions suggests the existence of minimal pairs such as *ma vs. *mba or *maY vs. *baY. However, what may have been 155

OLD JAPANESE

CLMC allophone *[m] – perceived by non-Chinese as */m/? CLMC phoneme */m/ CLMC allophone *[mb] – perceived by non-Chinese as */mb/?

Figure 5.1 Foreign perceptions of CLMC nasals

Table 5.16 The consonants of CLMC Labial Voiceless stops/affr. *p Voiceless fricatives *f(h) Aspirated stops/affr. *ph ‘Muddy’ stops/affr. *p® ‘Muddy’ fricatives Prenasalized obst. *mb *mv Nasals *m Glides *w

Dental

Retroflex

*t

*ts *tr *s *th *tsh *thr *t® *ts® *t®r *s® *nd *ndr *n *nr *l

*cr *Z *chr *c®r *Z®

Palatal

Velar

Glottal

*c *\ *ch *c® *\® *ã9 *ã *y

*k (*x) *kh *k® (*x®) *Yg *Y *ò

*q *h *h®

subphonemic for a Chinese may have been quite distinct for a Korean or Japanese scribe (Figure 5.1), so I will distinguish between denasalized and (potentially) non-denasalized allophones of nasal phonemes. Using the above proposals, I derive the CLMC consonants in Table 5.16 from the EMC consonants of Table 5.13.

Summary In this chapter, I have surveyed all the types of evidence available for Chinese reconstruction as well as the latest Chinese reconstructions themselves. I adopted Starostin’s (1989) LOC reconstruction and Pulleyblank’s (1984, 1991a) EMC reconstruction because they draw upon and account for the widest variety of data. I rejected the LMC reconstructions of Pulleyblank (1984, 1991a) and Coblin (1994) and proposed a set of LMC initials. In the next chapter, I will outline my methodology for OJ reconstruction and explain how I will use the sources discussed up to this point.

Notes 1 EMC *tr represents a single consonant *[â]. Pulleyblank does not reconstruct any consonant clusters for EMC apart from consonant–glide clusters such as *tw-. 2 Pulleyblank’s (1984: 75) examples are from examples of the Four Grades in a manuscript fragment from Dunhuang. These examples are similar to four of the entries in table 25 of Tôdô and Kobayashi’s (1971: 74) edition of Yunjing.

156

OLD CHINESE, MIDDLE CHINESE, AND SINOXENIC

3 Kiyose (1991) uses the term ‘Altaic’ to denote a group of typologically similar languages rather than a family of genetically related languages. 4 I have adopted Kim-Renaud’s (1997: 2) translations of these terms. 5 I have not been able to identify Arisaka (1936). It is absent from Haewoo Lee’s bibliography. 6 Gregerson (1969: 165) reconstructs g as either *g or *γ, but I favor *g since (1) de Rhodes says “g is used just as it is with us” (Gregerson 1969: 182) and (2) *γ would have been written with a special letter or digraph instead of g. 7 According to Gage (1985), earlier *c, *j > *c and *k, *g > *k but *p, *b and *t, *d change directly into voiced implosives without the intermediate stages *p and *t that I propose in (5.12). 8 After the fall of the Eastern Han, the Wu state, one of China’s Three Kingdoms, ruled Jiaozhou (northern Vietnam) from 222 to 280. Jiaozhou was then controlled by various other states such as Eastern Jin and Song during the rest of the Six Dynasties period before falling under Sui and finally Tang control. 9 I have excluded characters which retain their SV readings and meanings when used in nôm: e.g.,
SV cô ‘father’s sister.’ 10 Some scholars including Downer treat Vietnamese e [t] and d [RR] as back vowels (i.e., as /T/ and /[/). However, notice that Downer’s ‘back’ (my ‘labial’) vowel class consists only of rounded back vowels and excludes these two unrounded central or back vowels. 11 I use the non-Wylie symbol t to transcribe the ‘inverse gi-gu (letter i)’ of the Tibetan transcriptions. The phonetic value of this vowel is uncertain (Che 1994: 75), though I suspect it may have been [t]. 12 Starostin (1989) uses the term ‘EMC’ to refer to sixth-century Chinese rather than the Chinese of Qieyun (601 CE).

157

OLD JAPANESE

6 GOALS AND METHODOLOGY

This chapter will be quite short and relatively simple compared to earlier chapters. My aims and techniques are not as complex as the background knowledge and tools required for OJ reconstruction.

Goals I intend to reconstruct the segmental phonetics and phonology of the Old Japanese (OJ) language of the poetry of Kojiki and Nihon shoki. This reconstruction must be: (1) Statistically based: I will not draw conclusions about OJ on the basis of the readings of a few phonograms. Neither will I treat all phonograms for an OJ syllable equally. I will examine phonogram frequency to avoid being misled by nonce oddities. I am interested in finding dominant patterns of phonogram usage throughout the entire corpus of Kojiki and Shoki poetry. (2) Typologically sound: I want to reconstruct a plausible phonological system for OJ that (a) has precedents in natural human languages and (b) is as unmarked as possible. Extraordinary solutions will require extraordinary evidence. (3) Consistent with Japonic-internal evidence: I want my solution to fit what is already known about stages of Japanese after the eighth century.To avoid circularity, my solution will not be based on claims about pre-OJ made in Chapter 4. Instead, I will test such claims during the course of the following two chapters. There is no guarantee that philological reconstruction will confirm the results of internal and comparative reconstruction.

Methodology I have examined all poems from Kojiki and Nihon shoki in their original phonogram orthography, including all textual variants listed in the Shintô taikei (Outline of Shintô) edition of Kojiki and the Nihon koten bungaku taikei (Outline of Classical Japanese Literature) edition of Shoki. I have also examined textual variations between three modern printed editions of Kojiki 158

GOALS AND METHODOLOGY

and Shoki poetry: (1) the Shintô taikei edition of Kojiki, (2) the Nihon koten bungaku taikei edition of Shoki, and (3) the Nihon koten bungaku taikei anthology of non-Man’yôshû OJ poetry, Kodai kayô shû (Collection of Ancient Songs). I will quote OJ poetry in modified Yale romanizations for each phonogram based on (1) the kana transcriptions of the poems given in Kodai kayô shû and (2) the A/B-type classification of phonograms given in Igarashi’s (1969) dictionary of OJ orthography. I have chosen the poetry of Kojiki and Shoki to be my primary evidence for OJ phonetics and phonology for four reasons: (1) They constitute a large body of connected language. The poetry of Kojiki and Shoki contain a total of roughly 12,800 phonograms, including textual variants. It would be dangerous to reconstruct OJ on the basis of transcriptions of isolated proper nouns because they may contain archaic and/or aesthetic phonogram usages. One could not easily reconstruct the sounds of English on the basis of English personal and place names, which often have spelling irregularities lacking in the orthography for the rest of the language. Using two complete bodies of poetry (with the exception of a single poem; see below), I can reconstruct practically all of the syllables of the language rather than just those which happen to occur in names. I will also be able to reconstruct phonetic and phonological forms for many of the words in the language, since the bulk of the known OJ lexicon appears in poetry. (2) The System-D-based transcription (p. 18) of Kojiki and the System-Ebased transcription (p. 32) of Shoki are quite different, yet the poems being transcribed are often the same or very similar. Thus these two orthographies will give us two different perspectives on the phonetics and phonology of the poems shared by both works and of OJ in general. Since Kojiki and Shoki are contemporaneous, I hope to integrate their separate perspectives into a single coherent picture of the OJ sound system. (3) Kojiki and Shoki are both national histories produced for the Japanese court and presumably reflect the prestigious Central dialect of OJ. Man’yôshû, on the other hand, contains poems in the Azuma (Eastern) dialect of OJ as well as the Central dialect. The Azuma dialect is quite different from the Central dialect and is less well understood. It deserves separate treatment and I hope to examine its orthography, phonetics, and phonology elsewhere. I believe that the reconstruction of the Central dialect in this book is a prerequisite for reconstruction of the Azuma dialect. Hereafter, I will use the term OJ to refer only to the Central OJ dialect. (4) Unlike Man’yôshû, Kojiki and Shoki employ pure phonogram orthography without any ambiguous semantograms or tawamuregaki ( p. 26). The difficulties of interpretation that arise when dealing with the complex writing system of Man’yôshû poetry are largely absent from the simpler writing systems of Kojiki and Shoki poetry. As a general rule, I will exclude cases of variation including variant phonograms and variant interpretations of phonograms from my study. 159

OLD JAPANESE

Although I am frequently sure that some variant phonograms are scribal errors, I do not wish to explain away the statistical noise they can generate. Nor do I wish to deal with determining the correct readings of those rare phonograms which my sources interpret differently. But on the other hand, I do not wish to hide the warts of the extant texts. I am first and foremost interested in the big picture and will cite variants only when they can make major contributions to my study. For the most part I will treat Kojiki and Shoki as single texts. I will not take Mori’s (1991) division of Shoki into alpha and beta sections as a given. In Chapter 3, we have seen that the distinction between the two sections was not entirely clear-cut. I will further test Mori’s hypothesis in the next two chapters when I check to see whether certain traits of Shoki transcription are unique to one or the other section. Following the precedents set by Ôno’s (1953a) and Mori’s (1991) studies of Shoki orthography, I will exclude Shoki poem 122 from my analysis since it has not yet been adequately deciphered. However, I will use all other Kojiki and Shoki poems. From here onward, I will use the terms ‘Kojiki’ and ‘Shoki’ to refer only to the poetry of those two works and not to their prose unless otherwise specified. Now that I have explained my textual policies, I will now describe the three-stage process by which I will reconstruct the language of these texts. Chapter 7 constitutes the first stage. I will reconstruct the initial consonants of OJ on the basis of the initials of the Early Middle Chinese (EMC) readings of Kojiki phonograms and the Chang’an Late Middle Chinese (CLMC) readings of Shoki phonograms. My reconstructions of EMC initials will follow those of Pulleyblank (1984, 1991a) with the modifications given on p. 146. My reconstructions of CLMC initials will be based on Pulleyblank’s (1984, 1991a) reconstructions of EMC initials with the modifications given on p. 154. These modifications will ‘update’ and ‘localize’ the EMC initials to reflect the characteristics of the Chang’an dialect during the early Tang Dynasty. I have chosen to rely primarily on the reconstructions of Middle Chinese (MC) initials rather than on attested Sinoxenic (SX) initials because MC initials are relatively uncontroversial and are poorly preserved in SX. Authentic Sino-Korean (SK), as opposed to the prescriptive variety of the Tongkwuk cengwun dictionary, consistently reflects neither the ‘muddiness’ nor the aspiration of MC: unaspirated and aspirated SK initials can correspond to ‘clear’ (unaspirated), ‘second clear’ (aspirated), or ‘muddy’ LMC consonants (p. 115). Vietnamese has undergone radical consonant shifts since the borrowing of Sino-Vietnamese (SV) during the late Tang (p. 120). Since the Tibetan transcriptions definitely reflect the denasalization of northwestern LMC but are limited in quality and quantity (p. 135), I will also cite SK initials from Cenwun okphyen (c. 1796), SV initials from Mineya 160

GOALS AND METHODOLOGY

(1972), and the initials found in alphabetic transcriptions of northwestern Chinese (Luo 1933a, Barat n.d.) as secondary and tertiary evidence. Although Pulleyblank (1984: 96–100, 154–159) did employ man’yôgana evidence for the reconstruction of the finals of his two stages of MC, his arguments for his reconstructions of EMC and LMC initials did not rely upon man’yôgana. Hence it is not circular to use his reconstructions of initials to reconstruct OJ initials. In each section of Chapter 7, I will determine which EMC initials most frequently appear in the readings of the phonograms for syllables with the same OJ consonant in Kojiki and reconstruct that OJ consonant on the basis of those initials. Then I will repeat this process with LMC initials in Shoki. Finally, I will reconstruct the OJ consonant on the basis of my consonantal reconstructions for Kojiki and Shoki. I will also look for correlations between MC consonants and OJ ‘vowels.’ For example, I may find that OJ s is predominantly written with MC *s before OJ u and wo, MC *ts before OJ a and o, and MC *\ before OJ i and e. In such instances, I will reconstruct allophones for each environment. The process of consonantal reconstruction is diagramed in Figure 6.1. Chapter 8 constitutes the second stage: the reconstruction of the ‘vowels’ of OJ. Since I am not using any reconstruction of LMC finals, I will look at which EMC final categories are most frequently correlated with each of the ‘vowels’ in the LMC-based orthography of Shoki as well as the EMC-based

OJ consonant X

EMC initials of phonograms for X-initial syllables in Kojiki

LMC initials of phonograms for X-initial syllables in Shoki

Most frequent EMC initials

Most frequent LMC initials

(EMC-based SX evidence)

(LMC-based SX and transcriptions)

EMC-based reconstruction of OJ X

LMC-based reconstruction of OJ X

Final reconstruction of OJ X

Environmentally conditioned allophones

Figure 6.1 Methodology of OJ consonantal reconstruction

161

OLD JAPANESE

orthography of Kojiki. Although somewhat anachronistic, this practice of viewing later stages of Chinese through an EMC prism has a long history going back to the rhyme table tradition. Modern examples include the Hashimotos’ studies of Cantonese (1972) and Hakka (1973). Besides, the gap between EMC and LMC is hardly as dramatic as that between EMC and modern Cantonese and Hakka. I do not intend to imply that all the EMC final categories survived into LMC. The comparative and transcriptive evidence clearly shows many mergers between the two stages. I use EMC final categories rather than the reconstructed shapes of EMC finals as primary evidence for the OJ ‘vowels’ because I reject Pulleyblank’s (1984) man’yôgana-based EMC final reconstructions as potentially circular. In each section of Chapter 8, I will use SX and nôm to interpret phonetically the EMC final categories most commonly associated with a single OJ ‘vowel.’ Since few Old Sino-Vietnamese (OSV) words correspond to the phonograms in Kojiki, I will look at the regular OSV counterparts of their EMC finals whenever necessary. I will also look at the readings of vietographs from the 1997 edition of Ngw thiên t@ (Five Thousand Characters) whose phonetic elements (‘bases’) belong to these EMC final categories. I will reconstruct the phonetic shape of each OJ ‘vowel’ in Kojiki on the basis of this OSV and nôm data. Using the finals of the SK and SV readings of Shoki phonograms as substitutes for reconstructions of the LMC reflexes of OJ ‘vowel’ X

EMC final categories of phonograms for X-‘vowel’ syllables in Kojiki

LMC final categories of phonograms for X-‘vowel’ syllables in Shoki

Most frequent EMC final categories

Most frequent EMC final categories

EMC-based OSV counterparts of the above finals and finals of readings of vietographs with bases with the above finals

LMC-based SK and SV finals of phonograms with the above finals

EMC-based reconstruction of OJ X

LMC-based reconstruction of OJ X

Final reconstruction of OJ X

Environmentally conditioned allophones

Figure 6.2 Methodology of OJ ‘vocalic’ reconstruction

162

GOALS AND METHODOLOGY

EMC rhyme categories, I will then reconstruct the phonetic shape of each OJ ‘vowel’ in Shoki. Finally, I will reconstruct each OJ ‘vowel’ on the basis of my vocalic reconstructions for Kojiki and Shoki. I will also look for correlations between EMC final categories and OJ ‘vowels.’ For example, I may find that OJ C-type i is transcribed with EMC *-i except after OJ w, where it is transcribed as EMC *-uy. In such instances, I will reconstruct allophones for each environment or even reinterpret the A/B/C-type classification of the ‘vowel’ if necessary. We may discover that some or all ‘C-type’ ‘vowels’ are really A-type or B-type ‘vowels.’ Chapter 9 constitutes the third and last stage: the reconstruction of the structure and inventory of OJ syllables. At this point I will have complete inventories of both the consonants and the ‘vowels’ of OJ. I will formulate rules to combine these elements into syllables. Using these rules, I will generate a complete list of reconstructed OJ syllables. I will conclude with an attempt to resurrect the eighth-century pronunciation of an OJ poem attributed to the mythological hero Yamato Takeru.

163

OLD JAPANESE

7 THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

Overview In this chapter, I will reconstruct OJ consonants in an order determined by the phonetic classes implied by their Yale romanizations. I have grouped OJ consonants into sets according to their apparent places of articulation, with the zero initial ordered last. Within each of the above sets, I have grouped OJ consonants according to their apparent manners of articulation. I must emphasize the distinction between the romanization and the reconstruction of an OJ consonant. The former is an algebraic notation. Only the latter has phonological substance. The Yale symbol p does not necessarily represent an OJ bilabial stop *p. It may have been an OJ bilabial fricative *φ or even an OJ glottal fricative *h. I have grouped OJ s and z with the dentals, though they may actually be palatals (*c and *j). The orthographic zero initial ($ ) could be a glottal stop (*}). My arrangement of the OJ consonants may not match their reconstructions’ places and manners of articulation. During the course of these next two chapters, we will see how closely my OJ reconstruction corresponds to Yale romanization. Before I begin my reconstructions, I should note that I use the terms ‘Kojiki’ and ‘Shoki’ to refer only to the poetry of those two works unless otherwise specified.

OJ p Three EMC consonants (*p, *ph, and *b) correspond to OJ p in Kojiki (Table 7.1). 65.6 per cent (339/517) of OJ p in Kojiki were without a doubt written with EMC *p. 24.6 per cent (127/517) of OJ p in Kojiki in theory may have been written with either EMC *b or EMC *p. All of these cases involve the OJ syllable pyi. 99.2 per cent (127/128) of OJ pyi in Kojiki were written with the phonogram
, which had three EMC readings with *b and two EMC readings with *p. I will assume that
was selected for its *p readings. It would make no sense to reconstruct allophonic variation (*p~*b) only 164

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

Table 7.1 OJ p in Kojiki poetry OJ \ EMC

*p

*ph

*p/*b

*b

a yi iy u ey ye o Subtotals Total

204 0 0 54 18 0 63 339 517

0 0 10 0 0 0 0 10

0 127 0 0 0 0 0 127

4 1 3 0 0 31 2 41

Table 7.2 OJ p in Shoki poetry OJ \ CLMC

*p

*p/*ph

*ph

*p/*p®

*p®

*f

*fh

*f®

a yi iy u ye ey o Subtotals Total

134 18 4 6 7 6 15 190 475

0 0 0 0 0 1 0 1

52 8 0 0 0 0 0 60

0 64 0 0 0 0 0 64

16 13 0 0 15 21 41 106

0 0 0 17 0 0 0 17

2 0 0 17 0 0 0 19

0 0 0 17 0 0 1 18

before the OJ ‘vowel’ yi. If we regard
as an EMC *p-initial phonogram, then 90.1 per cent (466/517) of OJ p corresponded to EMC *p. OJ pye is the only p-initial syllable in Kojiki which was consistently written with EMC *b. Nonetheless I am reluctant to reconstruct an allophone *b for OJ p only before ye (or ye and yi). Thus I reconstruct OJ p in Kojiki as *p for pye as well as for pyi. Although OJ piy is usually written with EMC *ph, I would rather not reconstruct *ph (or a labial fricative in OJ but absent in EMC) before iy. Hence I will reconstruct OJ p in Kojiki as *p before all ‘vowels.’ Six CLMC consonants correspond to OJ p in Shoki (Table 7.2). 88.6 per cent (421/475) of OJ p in Shoki were written with CLMC stops (*p, *ph, *p®). The most common of these stops is *p. There is only one instance of CLMC *ph in the entire alpha section:
for OJ pyi (75.19.7). The 59 other instances of unambiguous *ph are in the beta section. If Mori is correct in claiming that the alpha section is phonetically more accurate, then the near-absence of CLMC *ph in the alpha section and the low frequency of CLMC *ph throughout Shoki prevents us from reconstructing OJ p as an aspirated *ph. 165

OLD JAPANESE

There are two phonograms which have CLMC readings with and without simple *p:
*p-/*p®- for OJ pyi and
*p-/*p®- for OJ pey,1 which account for all 65 cases of ambiguous *p in Shoki. If I add these 65 ambiguous cases to the total of 190 unambiguous cases of *p, then the total for CLMC *p in Shoki rises to 53.7 per cent (255/475). Only 11.3 per cent (54/475) of OJ p in Shoki correspond to CLMC fricatives (*f, *fh, and *f®). All of these fricatives are found in the readings of phonograms for the OJ syllables pa, pu, and po. Only 1.0 per cent (2/204) of OJ pa and 1.8 per cent (1/57) of OJ po were written with an CLMC fricative, whereas 89.5 per cent (51/57) of OJ pu was written with an CLMC fricative. OJ pu was always written with an CLMC fricative (*f or *f®) in the alpha section. Should I reconstruct a fricative allophone (*φ or *f) before OJ u and a stop allophone (*p) elsewhere? I do not think a fricative allophone before OJ u is necessary for two reasons. First, CLMC probably had no syllables such as *pu, *phu, or *p®u. Pulleyblank (1984: 86) and Coblin (1994: 34) agree that EMC labials became CLMC labiodentals before the vowel *u: *pu > *fu, etc. Assuming that (1) this change occurred without any exceptions, (2) OJ p was *p, and (3) OJ u was *u (i.e., a vowel similar to EMC *u), scribes faced with the problem of writing OJ *pu would have to compromise. They would have to select (a) phonograms with inappropriate initials and appropriate vowels (e.g.,
=CLMC *f-, perhaps *fuh; cf. SK pwu [pu] and SV phú [fuu] ) or (b) phonograms with appropriate initials and inappropriate vowels (
CLMC *p-, perhaps *poh; cf. SK phwo [pho] and SV b3 [(oo] ). They apparently preferred the former solution and wrote OJ *pu as CLMC *fu rather than as CLMC *po. Second, even if one disregards the above assumptions, one cannot deny that EMC *ph, the closest EMC approximation of a labial fricative (*φ or *f ), never appears in Kojiki spellings for OJ pu. Now that I have rejected *ph and *φ or *f as possible reconstructions of OJ p in Shoki, there are only two candidates left: *p and *p®. Since *p clearly outnumbers *p® and is less exotic than *p®, I reconstruct OJ p as a voiceless unaspirated bilabial stop *p in Shoki as well as in Kojiki.

OJ b EMC *p and *b correspond to OJ p in Kojiki (Table 7.3). There are 20 ambiguous cases involving phonograms which could be read with either EMC *p or *b. If we ignore these cases, 96.9 per cent (95/98 [= 118 − 20] ) of OJ b in Kojiki were written with EMC *b. Perhaps I should reconstruct OJ b in Kojiki as *b. However, before I come to that conclusion, we should examine the data from Shoki (Table 7.4). Let us suppose that OJ b was *b for Shoki as well. Since CLMC has no *b, the best substitute available for *b would be *mb (cf. its Tibetan 166

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

Table 7.3 OJ b in Kojiki poetry OJ \ EMC

*p

*b

*p/*b

a yi iy u ey ye o Subtotals Total

3 0 0 0 0 0 0 3 118

0 1 0 19 0 0 0 20

63 15 4 0 6 1 6 95

Table 7.4 OJ b in Shoki poetry OJ \ CLMC

*p

*p/*p®

*p®

*mb

*mv

a yi iy u ye ey o Subtotals Total

4 0 0 0 0 0 0 4 84

0 0 0 3 0 0 0 3

3 0 2 2 0 9 10 26

31 16 1 0 2 0 0 50

0 0 0 1 0 0 0 1

transliteration as ’b *[mb]?), even though the CLMC voiced stop has prenasalization absent in OJ *b. *p® also might be a viable substitute, since the Tibetans transcribed it as b or even ’b as well as ph. Thus it is unsurprising that 90.5 per cent (76/84) of OJ b in Shoki were written either with *mb or *p®. Nevertheless, I am hesitant to reconstruct a simple OJ *b which would not explain prenasalization in later reflexes of OJ b ( p. 75). The sound change *b > mb is rather bizarre. I would rather reconstruct *mb for OJ and assume that LMJ mb had inherited this prenasalization. But how would I explain the fact that 31 per cent (26/84) of OJ b were written with *p®? Mori believed that the alpha section was written by Chinese. If both CLMC and OJ had prenasalized obstruents, then we would expect CLMC native speakers to identify their own *mb with OJ *mb. So we might expect the alpha section to have few or no instances of *p® for OJ b. This is not quite the case, for we find that only two-thirds (22/33) of OJ b in the alpha section were written with CLMC *mb (Table 7.5). Why would CLMC speakers write OJ *mb as *p or *p®? Furthermore, OJ bu and bo were always written with *p® in the alpha section. All six instances of OJ bo in the beta section were written with *p®. 167

OLD JAPANESE

Table 7.5 OJ b in Shoki alpha section poetry OJ \ CLMC

*mb

*p

*p®

a yi u o Subtotals Total

11 11 0 0 22 33

5 0 0 0 5

1 0 1 4 6

All nine instances of OJ bey are in the beta section and were written with *p®. Yet it would make no sense to reconstruct a non-nasal allophone *b preceding certain random ‘vowels’ (OJ ey, o, etc.) and a nasal allophone *mb preceding all other ‘vowels’ (OJ a, yi, etc.). I will refine my earlier statement to account for the above phenomena by adopting Ramsey and Unger’s (1972) proposal of nasalized vowels preceding OJ voiced obstruents: 1 2 3

OJ b was generally a prenasalized voiced unaspirated bilabial stop *mb. ‘Vowels’ preceding *mb may have been nasalized (*~). *mb may have been reduced to *b if a preceding ‘vowel’ was nasalized (*~). Thus OJ *mb may have had the following allophones in free variation:

(7.1) (1) *[mb]

(2) *[~mb] (3) *[~b]

Since all instances of OJ *mb in Shoki are intervocalic, any OJ *mb could potentially be reduced to OJ *~b. There was no *~b in CLMC, so scribes wrote this non-nasal voiced stop allophone of OJ b as CLMC *mb (a prenasalized voiced stop), *p (a non-nasal voiceless stop), or *p® (a nonnasal stop with voiceless onset and voiced aspiration). One could write my reconstruction of OJ b as *~b or even as *b or *np without specifying nasality or voicing, but I will continue to write it as *mb.

OJ m OJ m in Kojiki corresponds only to EMC *m (Table 7.6). Therefore I will reconstruct OJ m in Kojiki as *m. The picture is more complicated in Shoki, where CLMC *m, *m(b), and *mv correspond to OJ m (Table 7.7). CLMC *mb is the most frequent initial (82.3 per cent = 525/638). Were OJ m and b homophonous? Should I

168

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

Table 7.6 OJ m in Kojiki poetry OJ \ EMC

*m

a yi iy u ey ye wo o Total

198 180 15 72 29 24 37 135 690

Table 7.7 OJ m in Shoki poetry OJ \ CLMC

*m(b)

*mb

*mv

a yi iy u ye ey o Subtotals Total

28 0 0 3 3 0 17 51 634

151 151 0 12 15 38 154 521

0 0 8 54 0 0 0 62

reconstruct a post-OJ phonemic split? I do not think such a drastic measure is necessary. The key to the solution lies in the phonotactics of CLMC. CLMC allowed initial nasals (as opposed to prenasalized obstruents such as *mb or *mv) only in syllables with final nasal elements: CLMC *mVN was possible, but CLMC *mV was impossible, since earlier *mV syllables had become *mbV and *mvV syllables. If we assume that OJ m was *m, scribes had to choose between CLMC *mbV and *mvV phonograms or CLMC *mVN phonograms to write OJ m-initial syllables. Apparently, they favored using open syllable phonograms even though those phonograms’ prenasalized obstruents were not perfect matches for OJ *m. We should not project the lack of *mV syllables in CLMC onto OJ. If OJ m were *mb, we might expect to see at least a few cases of EMC *b for OJ m in Kojiki, but we find only EMC *m (Table 7.6). Furthermore, we know that OJ m corresponds to m in all later stages of the language. Therefore I reconstruct OJ m as a bilabial nasal *m for both Kojiki and Shoki.

169

OLD JAPANESE

Table 7.8 OJ w in Kojiki poetry OJ \ EMC

*®w

*w

a i e o Subtotals Total

69 0 13 0 82 228

0 12 0 134 146

OJ w OJ w in Kojiki corresponds to EMC *®w and *w (Table 7.8). EMC *®w and *w are in complementary distribution: EMC *®w precedes OJ a and e and EMC *w precedes OJ i and o. But I will not reconstruct two allophones for OJ w, because *®w and *w were in complementary distribution in EMC. EMC *®w appeared only before non-high (i.e., low and mid) vowels (*a, *e, etc.) and EMC *w appeared only before high vowels (*i, *t, *u). If OJ wa and we had (1) initial *w- and (2) non-high ‘vowels’ (e.g., *a and *e), then the closest EMC equivalents would have initial *®w- coupled with non-high vowels. I do not think one should project a distributional peculiarity of EMC into OJ.2 It is also possible that the variety of Chinese underlying the Kojiki orthography had *w instead of *®w. EMC *®w phonograms may have been borrowed before the shift of *w to *®w before non-high vowels. I will reconstruct OJ w in Kojiki as a labial glide *w. OJ w in Shoki corresponds to CLMC *h®, *h®w, *q, *qw, and *w (Table 7.9). Surprisingly, the most common CLMC initial is the glottal stop *q [}]. CLMC initials without any *w account for 52.3 per cent (114/218) of the total. How can we account for this bizarre state of affairs? Was OJ w really *w? Why would anyone write a labial glide *w with a glottal stop *q [}] or a glottal fricative *h®? The answers to the above questions lie in the finals rather than the initials of the Shoki phonograms for OJ w-initial syllables. All of the phonograms Table 7.9 OJ w in Shoki poetry OJ \ CLMC

*h®

*h®w

*q

*qw

*w

a i e o Subtotals Total

0 0 0 5 5 218

51 0 6 6 63

0 3 0 106 109

24 2 1 1 28

0 8 5 0 13

170

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

with *w-less CLMC initials (i.e., *q and *h®) have EMC, SK, and SV readings with labial glides or vowels (7.2). (7.2) OJ wo wo wo wo wi

Ph.






Freq. 47 34 25 5 3

EMC *qo *qo *qoq *®o *quy

Go CLMC u *qu *qu *qwo, go *h®wi *q-

Kan wo wo wo ko wi

SK wo [o] wo [o] wo [o] hwo [ho] wi [uy]

SV ô [}oo] ô [}oo] 7 [}oo] h4 [hoo] uy [}wii] oai [}waay]

Coblin (1994: 153, 245) notes that Amoghavajra (705–774) of the northwest used the first phonogram
to write Sanskrit o and e and that the fifth phonogram
is transcribed as Tibetan u in Tiandi bayang shenzhoujing. The other three phonograms do not appear in the northwestern LMC transcriptional materials. One might conclude that OJ wi was *wi and that OJ wo was *o. However, there is no strong evidence for a distinction between *qo and *wo in LMC. (Although SV alone distinguishes between [}] and [v] before o-like vowels, SV readings such as vo and vô always derive from LMC *mv- rather than LMC *w-.) CLMC *q- < EMC *qo (cf.
SK wo [o], SV ô [}oo] ) was the closest equivalent of OJ wo. I will reconstruct OJ w as *w.

OJ t OJ t in Kojiki corresponds to EMC *t, *th, *tr, and *d (Table 7.10). 80.5 per cent (566/703) of OJ t were unambiguously written with EMC *t. All 60 attestations of OJ te in Kojiki were written with the phonogram H which is not listed in Guangyun. Tôdô (1978: 431) speculates that H is a variant of
, which is listed in Guangyun with two fanqie, one for EMC *tey and another for EMC *tri. If I treated H as an EMC *t-initial phonogram, then the figure for EMC *t in Kojiki would rise up to 89.0 per cent (626/703). Table 7.10 OJ t in Kojiki poetry OJ \ EMC

*t

*t?

*th

*tr

*d

a i u e wo o Subtotals Total

222 0 135 0 14 195 566 703

0 0 0 60 0 0 60

1 0 0 0 0 0 1

0 74 0 0 0 0 74

0 0 1 0 0 1 2

171

OLD JAPANESE

Table 7.11 OJ t in Shoki poetry OJ \ CLMC

*t

*th

*t®

*n(d)

*tr

*thr

*t®r

*ch/*c®/*t/*thr

a i u e wo o Subtotals Total

101 0 50 66 11 86 314 509 (+78)

2 0 49 0 0 0 51

8 (+45) 0 4 3 12 36 (+33) 63 (+78)

1 0 0 1 0 0 2

0 62 0 0 0 0 62

0 1 0 0 0 0 1

0 3 0 0 0 0 3

13 0 0 0 0 0 13

But what about the remaining 10.7 per cent (74/703) of OJ t in Kojiki which corresponded to EMC *tr? All 74 of these cases involve the OJ syllable ti, though skeptics may wish to include the 60 cases of H ≅
EMC *tey/ *tri. OJ ti was never written with EMC *t. Should I reconstruct a retroflex allophone *tr [â] in OJ before the ‘vowel’ i (and the ‘vowel’ e?) and a dental allophone *t elsewhere? It may be better to ask about the distribution of EMC *t and *tr first. EMC *t could never occur before the high front vowel *i. Thus EMC had no syllables such as *ti or *tiR, though it did have syllables such as *tri and *triR. The complementary distribution of *t and *tr is a trait of EMC that should not be projected onto OJ. OJ ti (*tV) was written with EMC
*triR(h) and
EMC *triRh in Kojiki because there were no phonograms with readings like EMC *tiR. We cannot always assume that the Chinese readings of phonograms were exact phonetic equivalents of OJ syllables. Hence I will reconstruct OJ t in Kojiki as *t. OJ t in Shoki corresponds to nine CLMC initials (Table 7.11). Like EMC *tr in Kojiki, CLMC *tr in Shoki is found only in phonograms for OJ ti. Like EMC, CLMC probably did not have a *ti, but it did have many similar syllables with *tr- [â] (
TKCW and sixteenth-century SK ti > later SK ci, SV tri, Tibetan transcription ci) and one anomalous syllable with *t®- (
SK ti, SV $ha, Tibetan transcription di). The use of CLMC retroflex initials for OJ ti probably tells us more about the phonotactics of CLMC than OJ. Thus I reject *tr as a possible allophone for OJ t. Three other high-frequency CLMC initials corresponding to OJ t are *t, *th, and *t®. *th is restricted to phonograms for the syllables ta and tu. It seems arbitrary to reconstruct *th only before OJ a and u, so I reject *th. *t® appears to be more infrequent than it actually is only because it is predominantly found among the variant phonograms that I have excluded from this study. I have added the figures for variants in parentheses in Table 7.11. These variants do not alter the general picture. Even if I include the 78 probable cases of CLMC *t® for OJ t), CLMC *t still outnumbers CLMC *t® by a ratio of two to one (314 to 141 = 63 + 78). Also, 172

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

CLMC *t® mostly appears in phonograms for OJ ta, two, and to. It would be arbitrary to reconstruct *t® only before these three ‘vowels.’ The statistics for *t may not be overwhelming – 61.7 per cent (314/509) – but *t is less marked as well as more common than *t®. Therefore I will reconstruct OJ t as a voiceless unaspirated dental stop *t in both Kojiki and Shoki.

OJ d OJ d in Kojiki corresponds to EMC *t, *d, *dr, and *9 (Table 7.12). There are only two strong possibilities: *d and *dr. Half the instances of *dr involve phonograms for OJ di. EMC had only one syllable combining a dental initial with a high front vowel (
*dih). In Kojiki poetry, OJ di was written with
EMC *dri and
EMC *drt(h). These graphs were the closest EMC counterparts to OJ di apart from EMC
*dih, which is found as a phonogram for OJ di in the prose of Kojiki and in Man’yôshû. If OJ di really did have a retroflex initial *dr- [¶ ], why write it with a dental initial phonogram such as EMC
*dih? We find EMC *d corresponding to d in OJ da, du, and dwo. Why not reconstruct *d for OJ di as well? I cannot explain why
EMC *drwiRn(h) and
EMC *drtR / *9tR with the EMC initial *dr- are the sole phonograms for OJ de and do in the poetry of Kojiki. I would have expected phonograms with the EMC initial *d-. But before I reconstruct *dr for OJ de and do, I should note that, in the prose of Kojiki, EMC dental-initial phonograms coexist alongside the aforementioned *dr-initial phonograms for these syllables: (7.3) OJ de OJ do


EMC *denh / *tenh,
EMC *tey
EMC *dRY,
EMC *dRY

Hence I think it is possible to reconstruct OJ d as *d for all d-initial syllables in Kojiki. OJ d in Shoki corresponds to six CLMC initials (Table 7.13). Most of the initials are partly or wholly voiced: *t®, n(d), *nd, *ndr. As we would expect Table 7.12 OJ d in Kojiki poetry OJ \ EMC

*t

*d

*dr

*dr/*9

a i u e wo o Subtotals Total

1 0 0 0 0 1 2 143

55 0 46 0 2 0 103

0 9 0 14 0 0 23

0 0 0 0 0 15 15

173

OLD JAPANESE

Table 7.13 OJ d in Shoki poetry OJ \ CLMC

*t

*th

*t®

*n(d)

*nd

*ndr

a i u e wo o Subtotals Total

6 0 1 2 0 1 10 120 (+5)

1 0 1 0 0 0 2

1 (+5) 0 35 6 2 5 49 (+5)

12 1 1 13 4 12 43

12 0 0 0 0 0 12

0 4 0 0 0 0 4

from our previous experience with the phonograms for OJ ti and di, we find an CLMC retroflex initial *ndr [å¶ ] for 80 per cent (4/5) of OJ di in Shoki. There was no CLMC *ndi. The one definite case of *nd for OJ di (NS 31.5.2) involves the phonogram
which probably had a mid vowel in CLMC (cf. Tibetan transcription ’de *[nde]? and SV nê [nee], nq [nee] ). If OJ d were *nd, then CLMC syllables such as *ndri and *nde would have been the closest equivalents to OJ di (= *ndV). I will not project the distributional idiosyncrasies of CLMC into OJ by reconstructing a retroflex allophone of OJ d before i. The only candidates left are *n(d), *nd, and *t®. None is quite like the *d that I reconstructed for OJ d in Kojiki. *n(d) and *nd together (45.8 per cent = 55/120) are somewhat more common than *t® (40.8 per cent = 49/120), but there is no obvious majority candidate. If I add five variants with probable *t® to the 49 cases of definite *t®- in Table 7.13, the total of *t® phonograms (43.2 per cent = 54/125 = 120 + 5) is nearly equal to the total of *n(d) and *nd phonograms (44.0 per cent = 55/125 = 5 + 120). Any reconstruction of OJ d for Shoki would have to account for both the nasal and non-nasal correspondences. I propose a solution parallel to that given for OJ b in section 7.3: 1 2 3

OJ d was generally a prenasalized voiced unaspirated dental stop *nd. ‘Vowels’ preceding *nd may have been nasalized (*~). *nd may have been reduced to *d if a preceding ‘vowel’ was nasalized (*~). OJ *nd may have had the following allophones in free variation:

(7.4) (1) *[nd]

(2) *[~nd]

(3) *[~d]

Since nearly all instances of OJ *nd in Shoki are intervocalic,3 almost any OJ *nd could potentially be reduced to OJ *~d. There was no *~d in CLMC, so scribes generally wrote this non-nasal voiced stop allophone of OJ d as 174

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

CLMC *nd(r) (a prenasalized voiced stop) or *t® (a non-nasal stop with voiceless onset and voiced aspiration). Although I could write my reconstruction of OJ d as *~d or even as *d or *nt without specifying nasality or voicing, I will continue to write it as *nd.

OJ n OJ n in Kojiki corresponds to EMC *n and *ã (Table 7.14). EMC *ã corresponds to OJ n before i and EMC *n corresponds to OJ n elsewhere. At face value, this would imply that OJ n was *ã before i and *n elsewhere. Although the MSJ syllable ni is [nyi], I am not sure whether I should project a palatal or palatalized allophone of n back into OJ. The LMJ syllable ni did not have a palatal initial, for the Portuguese did not romanize it as *nhi *[ ãi] (cf. the Portuguese-based practice of writing Vietnamese [ ãii] as nhi). There was no EMC *ni. The Kojiki phonograms for ni (
EMC *ãiRq and
EMC *ãiRq) may be inexact attempts to write an OJ *ni. Although I cannot rule out a palatalized or palatal initial before OJ i, it is simpler to reconstruct *n before all vowels. OJ n in Shoki corresponds to CLMC *n(d), *nd, *ndr, and *ã9 (Table 7.15). This would seem to contradict my earlier reconstruction of OJ n as *n. Table 7.14 OJ n in Kojiki poetry OJ \ EMC

*n

*ã

a i u e wo o Subtotals Totals

126 0 18 49 12 272 477 669

0 192 0 0 0 0 192

Table 7.15 OJ n in Shoki poetry OJ \CLMC

*n(d)

*nd

*nd(r)

*ã9

a i u e wo o Subtotals Total

0 0 0 0 0 252 252 620

138 1 26 21 3 12 201

0 4 0 0 0 0 4

0 163 0 0 0 0 163

175

OLD JAPANESE

However, I have already explained that syllables with initial nasals and non-nasal finals were impossible in CLMC (p. 169). Let us assume that OJ n was *n. The scribes of Shoki could write OJ *nV syllables either with *nd/*ã9-initial, non-nasal final phonograms or with *n-initial nasal-final syllable phonograms. The former solution predominated for all OJ n-syllables with the possible exception of OJ no. 95.5 per cent (252/264) of no in Shoki are written with the phonogram=
whose CLMC initial may have been either *n- or *nd-.
appears in Tibetan transcriptions with both ’d (*[nd] ) and (’)n. If we assume that the CLMC initial of
was *nd-, then all OJ n-syllables in Shoki would be written with CLMC prenasalized obstruents (*nd, *ndr, and *ã9). Why not reconstruct OJ n as a prenasalized obstruent *nd? Although I have already reconstructed OJ d as *nd, one might argue for a phonemic split of *nd: i.e., some OJ *nd became post-OJ n and others became post-OJ d. This solution is not only messy but also contradictory to what we already know about OJ d = *nd. OJ d = *nd corresponded exclusively to EMC non-nasal obstruents in Kojiki and corresponded frequently to the CLMC non-nasal obstruent *t® in Shoki. Yet we never find OJ n corresponding to a MC non-nasal obstruent in either text. Since the correspondence patterns for OJ d and n are quite distinct, they must be reconstructed separately. I reconstruct OJ n as *n for Shoki because it is always written with prenasalized or nasal CLMC initials. The high frequency of CLMC *ã9 for the initial of OJ ni in Shoki may suggest a palatal allophone of OJ n. However, recall that EMC had no syllable *ni. Such a *ni would have developed into an equally nonexistent CLMC *ndi. CLMC *ndr- and *ã9- were the only choices for OJ ni available to Shoki scribes, who preferred the latter to the former. But the high frequency of CLMC *ã9- may tell us only that the n of OJ ni was non-retroflex, not that it was specifically palatal. Although I accept the possibility of a palatal allophone of n in OJ ni, it would be simplest to reconstruct OJ n as a dental nasal *n before all vowels.

OJ r OJ r in both Kojiki and Shoki corresponds to MC l (Table 7.16). On the basis of the data, I could reconstruct OJ r as *l. Lange (1973: 124) and Mori (1991: 135) have also reconstructed OJ r as *l. However, other data point toward an alternative reconstruction *r. The Portuguese transcribed the LMJ reflex of OJ r as r and MSJ r is still an [r]. I could posit a change of *l to *r between OJ and LMJ, but I would rather not do so. Furthermore, neither EMC nor CLMC had an *r. Eastern Han *r had become *l in early LOC (Starostin 1989: 485). This *l remained unchanged in EMC and CLMC. Although Tôdô (1978) and Pulleyblank (1984, 1991a) have reconstructed 176

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

Table 7.16 OJ r in Kojiki and Shoki poetry OJ

EMC *l (Kojiki)

CLMC *l (Shoki)

a i u e wo o Totals

102 144 87 53 11 28 425

120 104 74 55 9 31 393

the CLMC
‘sun’ initial as *r, their *r corresponds to zh [9] rather than to r in the Tibetan transcriptions of northwestern Chinese (Luo 1933a: 19). The Uighurs transcribed the northwestern LMC
‘sun’ initial as sh, zh, or y, and not as r, which did exist in their alphabet (Barat n.d.: 44 –45). I reconstruct the CLMC
‘sun’ initial as *ã9. Sanskrit r was transcribed with CLMC *l, not CLMC *ã9. By analogy, OJ *r could also have been transcribed with CLMC *l. Given what we know of post-OJ, it would be simplest to reconstruct an *r for OJ r. This *r would then continue more or less unchanged in later stages of Japanese.

OJ s OJ s in Kojiki corresponds to five EMC initials (Table 7.17).
, the sole Kojiki phonogram for OJ so, has two EMC readings, *tsRY and *dzRY. Since none of the other phonograms for OJ s-syllables in Kojiki has voiced EMC initials, I assume that
was chosen for its voiceless initial reading *tsRY rather than its voiced initial reading *dzRY. Thus I will ignore *dz as a candidate for OJ s. The remaining candidates are found in near-complementary distribution. If I take the evidence in Table 7.17 at face value, I should reconstruct OJ s as: Table 7.17 OJ s in Kojiki poetry OJ \ EMC

*s

*ts

*ts/*dz

*\

*c

a i u e wo o Subtotals Total

0 171 80 0 15 0 266 570

117 0 0 0 0 0 117

0 0 0 0 0 48 48

0 0 0 41 0 0 41

0 98 0 0 0 0 98

177

OLD JAPANESE

Table 7.18 Phonograms for OJ sa in Kojiki prose Ph

Frequency

EMC

EMC initial type

OJ








114/156 34/156 4/156 3/156 1/156

*tsah *Zæ *tsanh *stRY *tsaq/ h

dental affricate retroflex fricative dental affricate dental fricative dental affricate

sa sa sanu saga sa

1 2 3 4

*ts before a and o *c and/or *s before i *s before u and wo *\ before e

This is an exceedingly inelegant solution. Does any language exist with four allophones of /s/ distributed in such a manner? LMJ and MSJ, the descendants of OJ, have only two allophones of /s/, [\] before palatal vowel(s) and [s] elsewhere. The motivation for the apparent allophony of /s/ in Kojiki is far less clear. I know of no language with a similar pattern of allophony. I would like to reconstruct *s not only for OJ su and swo but also for OJ sa and so and perhaps even for OJ si and se. In the prose of Kojiki, OJ sa was spelled with both affricates (*ts) and fricatives (*Z, *s) (Table 7.18). Although the majority of OJ sa are still written with *ts, we cannot ignore the 23.7 per cent (37/156) which are written with fricative-initial phonograms. We cannot say that OJ sa was written with *ts throughout Kojiki. One could try to claim that OJ sa was pronounced only as an affricate *ts in poetry but could be pronounced as an affricate *ts or a fricative *s in prose, but this sounds arbitrary and bizarre. Moreover, the Suiko Period phonograms for OJ sa include a mix of affricate and fricative initials (Ôno Tôru 1962: 93; Omodaka 1967: 894) (Table 7.19). I do not have statistics for the frequency of these three phonograms. Ôno Tôru (1962: 93, 141) considers both
EMC *tsah and
EMC *Zæ to be “commonly used” ( jôyô) Suiko phonograms for sa. Thus we cannot disregard
EMC *Zæ as a rare phonogram during the Suiko Period.

Table 7.19 Suiko Period phonograms for OJ sa Ph

EMC

EMC initial type

OJ






*tsah *Zæ *tsak

dental affricate retroflex fricative dental affricate

sa sa sa

178

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

According to Ôno’s (1962: 141) chronological table of phonogram usage
EMC *Zæ did wane in popularity from 700 CE onward. This would seem to imply a fortition of a Suiko fricative to a Nara affricate *ts – and a later lenition of *ts to a post-OJ fricative [s]. Those in favor of such a fortition could also point out that the Suiko phonogram for OJ so was
EMC *st < LOC *sR with a fricative initial, whereas the Nara phonogram was
*tsRY/ *dzRY with an affricate initial (Ôno 1962: 141). However, I would rather not posit a scenario of back-and-forth changes: (7.5) Suiko *s (for OJ sa, so) > Nara *ts (for OJ sa, so) > Post-OJ [s] (for sa, so) I still cannot explain why fricatives would become affricates only before OJ a and o. If I took the EMC initials at face value, I would have to claim that Suiko *ts became Nara *s before OJ u – the mirror image of what we would expect by analogy with (7.5). There is no Suiko phonogram for OJ su (Ôno Tôru 1962: 50), but the Suiko phonogram for OJ suku,
EMC *tsuRwk, has an initial affricate *ts. In the poetry of Kojiki, OJ su is spelled with
EMC *suR, which has an initial fricative *s. Affricate-initial phonograms for OJ su are rare in the text of Kojiki:
EMC *suR appears 114 times, but
EMC *cuw appears three times,
EMC *cuw appears twice, and
EMC *cuw appears only once. Were OJ sa and so developing affricate initials while OJ su was losing its affricate initials? How probable are simultaneous changes in opposite directions? (7.6) Suiko *s (for OJ sa, so) > Nara *ts (for OJ sa, so) > Post-OJ [s] (for sa, so) but Suiko *ts (for OJ su) > Nara *s (for OJ su) > Post-OJ [s] (for su) Were scribes accurately recording such strange shifts? So far, I have implicitly assumed that the Japanese could distinguish between the various *s-like EMC initials and deliberately used them to write the four allophones of their native /s/. What if we assumed the reverse scenario? What if the Japanese only had a simple *s without any allophonic variation? What if all the *s-like initials of EMC (or, more precisely, the System D version of EMC) sounded like native *s to them? Sandness also supports reconstructing only a simple *s for OJ. She points out (1985: 10): It is important to remember that the substitution of [s] for [ts] in foreign words is common among people whose native language lacks an initial /ts/. For example, many English-speakers have trouble learning even to hear, much less reproduce, the distinction between Japanese sugi “cedar” and tsugi “next” or between suite iru “is emptied out” or tsuite iru “is attached.” 179

OLD JAPANESE

Like modern English speakers who substitute their [s] for Japanese [ts], OJ speakers could have substituted *s for Chinese *ts. The Japanese may have regarded such characters as *s-initial phonograms appropriate for writing OJ *s-initial syllables. Moreover, phonograms such as
EMC *tsah and
*tsRY may have been the best available substitutes for phonetically more appropriate characters such as
EMC *sa (which was of very low frequency) and
EMC *sRY (which meant ‘monk’ and may have been taboo in secular transcription). Hence I reject *ts as a possible allophone of OJ /s/. Did OJ /s/ have two allophones *c and *s before OJ i ? Or are these allophones actually phonemes? One might try to reconstruct two kinds of OJ si: one with an initial *s and another with an initial *c. This would resemble Mabuchi’s (1957) proposal of two types of OJ si: si A, written with EMC dentals (*s and *ts), and si B, written with EMC non-dentals (*Z and *c). Such a proposal is unacceptable because the phonograms for the two ‘types’ of si are interchangeable (Bentley 1997: 97). Bentley (1997: 95) thinks that “the ancient writers used both types [of phonograms for OJ si] because they sounded similar to the native ears.” I believe that Bentley is correct. It seems wasteful to reconstruct *c solely for OJ si (and *\ solely for OJ se). It would be more economical to reconstruct *s for all s-initial syllables, though a palatal allophone *\ of OJ /s/ may have existed before the ‘vowels’ (i and e). OJ s in Shoki corresponds to twelve different CLMC initials (Table 7.20). It is extremely unlikely that OJ s had so many allophones. Several or even all of these CLMC initials must correspond to a single OJ allophone. The rather unlikely velar initial *k(®) occurs only in one instance of the phonogram
CLMC *k(®)- for OJ si. Scribes may have assumed that
was read with a dental initial like its phonetic element,
CLMC *ts®-, a phonogram for OJ zi. The remaining CLMC initials are all *s-like. The five most frequent initials are: Table 7.20 OJ s in Shoki poetry OJ \ CLMC

*s

*s®

*ts

*tsh

*ts(®)

*ts(®)/*cr

*\

*c

Other

a i u e wo o Subtotals Total

8 30 52 11 14 0 115 441

0 53 0 0 0 0 53

64 3 0 0 0 0 67

14 0 0 0 0 0 14

0 0 0 0 0 29 29

0 0 0 9 0 0 9

3 36 13 15 0 0 67

0 58 4 8 0 2 72

2 2 4 0 0 7 15

Note ‘Other’ includes all unambiguous initials that occur ten times or fewer in Shoki: *ts®, *Z, *chr, *c®, *k(®).

180

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

1 2 3; tie 3; tie 5

*s / _ a, i, u, e, wo (26.1 per cent = 115/441) *c / _ i, u, e, o (16.3 per cent = 72/441) *\ / _ a, i, u, e (15.2 per cent = 67/441) *ts / _ a, i (15.2 per cent = 67/441) *s® / _ i (12.0 per cent = 53/441)

None of these initials appears before all six OJ ‘vowels.’ Only *s appears before five of the six OJ ‘vowels.’ Dentals (*s, *s®, *ts, *tsh, *ts®) are the most common at 64.4 per cent (284/441). They outnumber palatals (*\, *c, *c®) by a two to one ratio (64.4 per cent to 32.7 per cent = 144/441). Retroflexes (*Z, *cr, *chr) amount to only 2.7 per cent (12/441) and can be ignored entirely. In terms of manners of articulation, fricatives (*s, *s®, *Z, *\; 53.7 per cent = 237/441) are more common than affricates (*ts, *tsh, *ts®, *cr, *chr, *c, *c®; 46.0 per cent = 203/441). Non-aspirates (*s, *ts, *Z, *cr, *\, *c; 73.2 per cent = 323/441) are more common than voiceless (*tsh, *chr; 3.4 per cent = 15/441) or voiced aspirates (*s®, *ts®, *c®; 23.1 per cent = 102/441). On the basis of the above statistics, the ‘average’ CLMC initial for OJ s in Shoki is an unaspirated dental fricative: i.e., *s, which turns out to be the most common initial and the only initial that appears before five of the six OJ ‘vowels.’ However, *s is not the most frequent CLMC initial for OJ sa, si, se, or so in Shoki. Yet I am still reluctant to reconstruct four allophones of OJ s (*ts, *ts, *\, *c) for reasons I have already discussed. Perhaps looking only at Mori’s (1991) ‘alpha section’ of Shoki may give us some insights into this problem. I have already done so using Mori’s own statistics (p. 59), but this time I will try to do so using my statistics. Mori believes that the alpha section of Shoki was written by Chinese scribes. Even if those scribes were not native speakers of CLMC, we would expect them to do a reasonably good job of transcribing OJ s. If they could distinguish between fricatives and affricates in their own language, they should have been able to distinguish between fricatives and affricates in OJ. Table 7.21 shows us how well they did. Table 7.21 OJ s in Shoki alpha section poetry OJ \ CLMC

*s

*ts

*ts/*ts®

*ts®

*Z

*\

*c

a i u e wo o Subtotals Total

8 26 29 10 5 0 78 191

15 0 0 0 0 0 15

0 0 0 0 0 11 11

0 0 0 0 0 1 1

0 0 0 0 0 1 1

0 21 0 4 0 0 25

0 52 0 8 0 0 60

181

OLD JAPANESE

I doubt that the scribes of the alpha section were native Chinese speakers. Although we cannot expect 100 per cent accuracy from any scribes, the OJ syllables sa, si, se, and so are written with both fricatives and affricates. Only su and swo are consistently written with fricatives. Should I take the alpha section at face value and reconstruct: 1 2 3

*ts for OJ sa and so allophonic variation between *s, *\, and *c for OJ si and se *s for OJ su and swo?

I have no objections to (3). I find (1) and part of (2) harder to accept. Why would all ‘vowels’ other than u and wo be preceded by affricate allophones (i.e., *ts and *c)? I would like to think that the variation is an unintentional byproduct of scribes unable to tell apart the many *s-like initials of CLMC. Yet it is difficult to ignore the fact that the same MC initials are correlated with the same OJ ‘vowels’ in both Kojiki and Shoki. We do not find completely random variation. There are definite patterns of phonogram selection at work: *s with u and wo, *ts with a and o, and *s, *\, or *c with i and e. Do these patterns necessarily reflect actual OJ pronunciation, or can they be explained in some other way? I think that the patterns resulted from the continuation of earlier orthographic traditions. The Shoki scribes did not invent an entirely new writing system from scratch. Their orthography was an update of earlier orthographies rather than a wholly separate entity. There is considerable overlap between the orthographies of Kojiki and Shoki. All but one of the phonograms for OJ s-syllables in the poetry of Kojiki reappear in the poetry of Shoki. In almost all cases, these ‘carryover’ phonograms (e.g.,
*tsah) are among the most common phonograms for those syllables in the poetry of Shoki. Although each scribe could have chosen entirely different phonograms for any OJ syllable, most chose those already chosen by scribes before them. Bentley (1997: 157) has proposed that “set spellings” existed for words as well as for syllables. I hypothesize that those who originated these “set spellings” must have been unable to distinguish between the various MC *s-like initials. Later scribes simply repeated their predecessors’ arbitrary choices of *ts-initial phonograms for OJ sa and so, *c- and *s-initial phonograms for OJ si, *\-initial phonograms for OJ se, and *s-initial phonograms. Even if immigrant scribes and native scribes skilled in OSK and Chinese itself thought these spellings poorly matched the actual sounds of OJ, they were not in a position to rock the boat. As Bentley (1997: 158) pointed out, stable societies need stable orthographies. Though hardly entirely revolutionary, the Shoki orthography may have been rejected by later scribes because its modifications to the preexisting, long-established system were too radical. 182

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

Thus the patterns of spelling that we have seen above are at least partly the result of arbitrary conventions rather than conscious attempts to record the supposed chaotic allophony of OJ s. The early seventh-century spellings of s-syllables (EMC *s for OJ sa and so but EMC *ts for OJ su) operated by a different set of arbitrary conventions which were abandoned by the eighth century. Taking these conventions literally can only result in the reconstruction of strangely distributed allophones and nonsensical sound changes such as those in (7.6). Judging from the paucity of Suiko Period transcriptions compared with the wealth of Nara Period transcriptions, standardization inevitably accompanied the increasing use of writing. The above solution is admittedly rather speculative because we have no contemporary records telling us about the standardization of phonogram writing. I think the process was gradual and perhaps even subconscious. Scribes adjusted their spelling habits to reflect the latest SJ readings while still being influenced by the habits of their peers and their own old habits. We may never know the details of the process, but we cannot regard the convergence in spellings seen in the tables of Ôno Tôru (1962: 141–142) and Bentley (1997: 159–161) as a coincidence. There definitely was a tradition at work, and, once it was set, it was relatively impervious to innovations. Non-traditional Shoki-style spellings are almost nonexistent in post-Shoki works. Chaotic writing systems are more probable than chaotic sound systems. It is unlikely that the OJ writing system consistently recorded four or more allophones of OJ s. It is more likely that OJ had an *s which was inconsistently recorded by the OJ writing system. Thus I reconstruct OJ s as a voiceless dental fricative *s (with or without a palatal allophone *\ before the ‘vowels’ i and e) for both Kojiki and Shoki.

OJ z OJ z in Kojiki corresponds to six EMC initials (Table 7.22). Owing to the small number of z-initial syllables in Kojiki, I have added all other variants into the total to avoid skewing the statistics: Table 7.22 OJ z in Kojiki poetry OJ \ EMC

*z

*dz

*jr

*\

*j

*y/*zy

a i u e o Subtotals Total

0 0 0 0 2 (+5) 2 (+5) 45 (+14)

0 0 0 0 0 (+1) 0 (+1)

0 6 (+3) 0 0 0 6 (+3)

1 0 0 0 0 1

0 0 28 4 0 32

4 (+5) 0 0 0 0 4 (+5)

183

OLD JAPANESE

(1) All definite cases of *y/*z involve the phonogram
EMC *zya / *ya. Four of the five variants involve
EMC *ya, a graphic variant of
EMC *zya / *ya which lacks a *z-initial reading. The fifth variant involves
EMC *na/q/h, an unrelated but graphically similar character. I will treat
EMC *ya and
EMC *na/q/h as scribal errors for
EMC *zya. Since the phonograms for the other z-syllables all have EMC obstruent initials, I will disregard the *y-initial reading of
. (2) One instance of OJ zo in Kojiki (78.9.2) is written as
EMC *dzwRn in some texts and as
EMC *dzRyq/h in others. Since either character has initial *dz, I will assume that the original phonogram for (78.9.2) also had initial *dz. (3) OJ zi in Kojiki is written as
EMC *jrtq > later EMC *jriq. I will regard all three variants (
EMC *thoq,
EMC *jtRYq/h, and
EMC *cth > later EMC *cih) as scribal errors for
, though it is remotely possible that
is original. (4) OJ zo in Kojiki definitely appears twice as
EMC *ztRq. Some texts have five more instances of
whereas others have pseudographs such as b which are obvious scribal errors for
. I will emend these five pseudographs as
. If I take the data in Table 7.26 literally and ignore the marginal cases (EMC *\ and *dz), I could reconstruct the following allophones of OJ z: 1 2

3

*zy for OJ za. *j for OJ zi, zu, and ze. (I will not reconstruct *jr [¶| ] for OJ zi, since nothing so far leads me to believe that OJ ever had retroflex consonants. We have already seen that OJ ti (*t-) and di (*nd-) were written with phonograms such as
EMC *triR(h) and
EMC *dri because EMC had no such syllables as *ti or *di.) *z for OJ zo.

The last proposal is easy to accept, but the first and second are odd. Go-on distinguishes between z and zy before a, u, and o. What were the OJ sources of z and zy? If z were *zy before a and *j before u, how would Go-on zya, zyu have been distinguished from Go-on za, zu in the OJ period? It would be simpler to assume that Go-on zy < OJ Go-on *zy whereas Go-on z < OJ Go-on *z before a, u, and o. For maximal simplicity, I could reconstruct OJ z as *z before all vowels, though I cannot exclude the possibility of a palatal allophone (*j or *9) before i and e. However, Shoki casts some doubt on a simple *z. OJ z generally corresponds to CLMC *ã9 in Shoki (Table 7.23). 71.2 per cent (37/52) of z-syllables in Shoki are written with CLMC *ã9. Only OJ za and ze are never written with CLMC *ã9. This does not necessarily mean that the initials of OJ za and ze lacked nasality. EMC did not allow *ã before the vowels *a and *e. Judging from SX, LMC also did not allow *ã (CLMC *ã9) before the vowels *a and *e. SV has no syllables such as *nha [ ãaa], *nhê [ ãee], or 184

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

Table 7.23 OJ z in Shoki poetry OJ \ CLMC

*s®

*ts

*ts®

*cr

*c®r

*c®

*ã9

a i u e o Subtotals Total

0 0 0 0 2 2 52

0 0 0 0 1 1

2 0 0 0 1 3

3 0 0 0 0 3

0 0 2 1 0 3

0 2 0 0 1 3

0 7 28 0 2 37

*nhe [ ãεε] derived from LMC *ã-initial syllables. SK has no syllables such as *za [ ãa] / [za] or *zyey [ãyRy] / [zyRy] derived from LMC *ã(9)-initial syllables. Kan-on has no syllables such as za or ze. Since there were no CLMC syllables such as *ã9a or *ã9e, Shoki scribes were forced to use phonograms with initials other than *ã9 to write OJ za and ze. There is no need to reconstruct non-nasal allophones of OJ z specifically for OJ za and ze. The absence of *ã9-initial phonograms for those syllables reflects the phonotactics of CLMC rather than the phonotactics of OJ. But how do we explain the fact that OJ zi, zu, and zo are sometimes written with initials other than CLMC *ã9? I reconstructed OJ z in Kojiki as a simple *z. If OJ z were also a simple *z in Shoki, it would be impossible to write this *z accurately. CLMC had no fully voiced obstruents such as *z. The closest CLMC equivalents of foreign *z were ‘muddy’ obstruents (*s®, etc.) and *ã9 (< EMC *ã). Skeptics might point out that CLMC *ã9 appears in Tibetan transcriptions as zh [9], not ’zh ( [ ã9]? or [®9]?) and argue that it may not have been prenasalized at all. Nonetheless, we must remember that the Tibetan transcriptions are from the ninth and tenth centuries, one to two centuries after the borrowing of Kan-on. There must have been a prenasalized stage somewhere between the *ã of sixth-century EMC and the *9 of ninth-century CLMC: (7.7) EMC *ã

> *ãj

> *ã9

>

late CLMC *9

The late seventh-century CLMC known to Shoki scribes probably had * ãj or *ã9. It is difficult to believe that EMC *ã changed directly to CLMC *9. All other CLMC nasals are written with an a.chung ( ) which may or may not have represented prenasalization in Tibetan transcriptions. EMC *ã may have initially developed prenasalization like all the other nasals at first, only to lose it later (*ã9 > *9): (7.8) EMC Early CLMC *m, *n, *nr, *Y *mb/*mv, *nd, *ndr, *Yg but *ã *ãj > *ã9 185

Late CLMC Tib. tr. ’b, ’d, ’j, ’g Tib. tr. zh

OLD JAPANESE

It really does not matter whether CLMC *ã9 was prenasalized or not in the late seventh century, since CLMC *( ã)9 would have been an acceptable near-equivalent for an OJ *z in either case. To be certain that OJ z was prenasalized, we must look at later stages of the language. On p. 75, I have already discussed evidence from LMJ and modern dialects for the prenasalization of z. I could claim that an originally non-prenasalized OJ z somehow prenasalized and then lost its prenasalization (*z > *nz > [z] ), but I would rather claim that OJ z was prenasalized and that the prenasalization found in LMJ and some modern dialects is a retention and not a post-OJ innovation. The prenasalization of OJ z need not have been manifested as a segment (i.e., *n or *ã). As Ramsey and Unger (1972) have suggested, vowels preceding OJ voiced obstruents could have been nasalized. Following their suggestion, I propose a solution parallel to those given for OJ b and d (pp. 168 and 174): (1) OJ z was generally a prenasalized voiced unaspirated dental fricative *nz. It could not have been palatal because Kan-on renders CLMC *( ã)9 as zy before non-front vowels. If OJ z were *ã9 or ãj, we would find Kan-on z, not zy, in this environment. Kan-on zya, zyu, zyo tell us that OJ z by itself was not sufficient to imitate CLMC *( ã)9. (2) ‘Vowels’ preceding *nz may have been nasalized (*~). (3) *nz may have been reduced to *z if a preceding ‘vowel’ was nasalized (*~ ). OJ *nz may have had the following allophones in free variation: (7.9) (1) *[nz]

(2) *[~nz]

(3) *[~z]

If OJ z had palatal allophones before front ‘vowels,’ they would be realized as: (7.10) (1) *[ ãj] (4) *[ ã9]

(2) *[~ãj] (5) *[~ã9]

(3) *[~j] (6) *[~9]

and/or

Perhaps some if not all of the cases of OJ zi, zu, and zo written with nonnasal CLMC initials reflect allophones of OJ z such as *[~z], *[~j], or *[~9] without *[n] or *[ ã]. I will hereafter write my reconstruction of OJ z as *nz, though one could write it as *z or *ns without specifying either its prenasalization or its voicing.

OJ y OJ y corresponds only to MC *y in both Kojiki and Shoki (Table 7.24). I will reconstruct OJ y as *y for both texts.

186

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

Table 7.24 OJ y in Kojiki and Shoki poetry OJ

EMC *y (Kojiki)

CLMC *y (Shoki)

a u e wo o Totals

154 52 25 16 29 276

124 39 25 9 49 246

Table 7.25 OJ k in Kojiki poetry OJ \ EMC

*k

*kò

*kh

*g

*Y

*h

*®

a yi iy u ey ye wo o Subtotals Total

127 0 24 133 0 0 38 0 322 730

68 0 0 0 0 0 0 0 68

1 1 0 0 26 0 0 0 28

0 154 0 0 0 36 0 0 190

0 0 1 0 0 0 0 0 1

0 0 0 0 0 0 0 116 116

4 0 0 0 0 0 1 0 5

OJ k OJ k in Kojiki corresponds to seven EMC initials (Table 7.25). *k is clearly the EMC consonant that most frequently corresponds to OJ k. However, there are a number of oddities in Table 7.25 that require explanation. (1) *k and *kò for OJ ka: In Pulleyblank’s (1984: 105, 167–168) MC reconstruction, *k has a back allophone before non-high vowels. This is evident from Uighur transcriptions of MC which distinguish between the two allophones: the back allophone is written as x or q and the front allophone is written as k (Csongor 1949, 1962; Pulleyblank 1965; Barat n.d.). Although Uighur has x and q before non-palatal vowels and k before palatal vowels, the Uighur consistently transcribed Sanskrit k as k before all vowels, palatal or otherwise (Pulleyblank 1965: 200). If MC *k were simply [k] like Sanskrit k in all environments, why did the Uighurs not transcribe MC *k as k in all environments? The Chinese did not perceive Sanskrit k before a to be like their own back allophone of *k before *a (i.e., *[q]?), so they transcribed Sanskrit ka with special phonograms such as
EMC *kòa ( p. 148).

187

OLD JAPANESE

At one point I wondered whether pre-OJ had a back allophone of k before non-palatal vowels and a front allophone of k before palatal vowels which was breaking down in OJ. In Kojiki, we find both
EMC *kæ (*[qæ]?) and
EMC *kòa (*[kòa] ) as phonograms for OJ ka. If OJ ka had initial *q, we would expect only to find the phonogram
EMC *kæ (*[qæ]?). Were *q and *k in free variation before OJ a? I would rather not reconstruct such variation. If the Japanese had no *q : *k distinction in their language, they would have borrowed both allophones of EMC *k as *k. They probably regarded
EMC *kæ (*[qæ]?) and
EMC *kòa (*[kòa] ) as interchangeable phonograms for OJ ka with the same initial. I know of no Japonic evidence for a uvular stop *q, so I will not reconstruct one. (2) OJ kyi and kye are almost always spelled with EMC *g. If one ignores the phonograms, there is no reason to believe that OJ kyi and kye ever had voiced initials. OJ kyi and kye correspond to LMJ and MSJ ki and ke (see Martin 1987: 88–93 for exceptions). There is nothing in Ryukyuan pointing toward earlier voicing in these syllables (unless one wants to claim that Ryukyuan and OJ independently lost voicing in these particular syllables after the two branches of Japonic split). On the basis of Japanese-internal evidence, I will disregard EMC *g and reconstruct *k for OJ kyi and kye, since *k-initial phonograms such as
EMC *khyih (which appears only once for OJ kyi in Kojiki 4.20.4) did exist.4 (3) OJ key is exclusively spelled with EMC *kh: all 26 cases involve the phonogram
EMC *khtyh. I do not see any reason why only OJ key would have an aspirated allophone. OJ key was transcribed during the Suiko Period with both aspirated and unaspirated phonograms:
EMC *ktR < LOC *kaò B,
EMC *hty < LOC *hty B,
EMC *ktR < LOC *kaò B, and
EMC *khtyh < LOC *khtyhB (Ôno Tôru 1962: 50). The fourth phonogram,
EMC *khtyh, was chosen to be the sole phonogram for OJ key in Kojiki, giving the misleading impression that OJ key had an aspirated initial. I will disregard the aspiration of
EMC *khtyh and reconstruct the initial of OJ key as unaspirated *k. (4) OJ ko is written only with the phonogram
EMC *htRq, and OJ ka and kwo are sporadically spelled with *®-. Does this mean that OJ ko (and rarely OJ ka and kwo) had a fricative initial that later hardened into the k attested in all later stages? In the Suiko inscriptions, OJ ko was also transcribed as
EMC *ktq < LOC *kRòqB. If we assume that the *h of EMC *htRq represented an initial *h in OJ ko, we will have to claim that the initial of this syllable was *k during the Suiko Period, lenited to *h during the Nara Period, and then reverted to the k attested in LMJ. Why would this unusual fortition occur only to the syllable ko? It is highly improbable that the use of fricative-initial characters reflects fricative allophones of OJ k which appeared only before certain ‘vowels’ (i.e., o , a, and wo). It is more probable that scribes chose phonograms 188

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

Table 7.26 OJ k in Shoki poetry OJ \ CLMC

*c /*k

*k

*kh

*k®

*k®ò

*kh/q

*h

Other

a yi iy u ye ey wo o Subtotals Total

0 89 0 0 0 0 0 0 89 698

129 1 21 105 8 10 55 30 359

3 23 2 0 1 16 0 4 49

0 41 0 0 2 0 0 4 47

41 0 0 0 0 0 0 0 41

0 0 0 42 0 0 0 0 42

12 0 0 0 0 0 0 50 62

0 2 (*c /*k®) 0 0 4 (*k(h) ) 0 3 (*h®) 0 9

Note ‘Other’ includes all CLMC initials which appear four or fewer times: *c/*k®, *k(h), and *h®.

according to SJ readings which lacked distinctions between *k and *h and between *g and *®. In all strata of SJ, Chinese glottal fricatives are always borrowed as k and g. (Recall that MSJ h ultimately originates from OJ p = *p.) Let us suppose that OJ k was *k in all environments. Readings of characters such as
EMC *htRq would have been borrowed with OJ k = *k. To Japanese without firsthand knowledge of Chinese,
EMC *htRq (cf. Go-on ko) was a *k-initial phonogram just like
EMC *ktq < LOC *kRòqB (cf. Go-on ko). Hence I doubt that fricative-initial phonograms for OJ k-syllables are evidence for OJ *h or *®.5 In spite of the diversity of EMC initials in Table 7.25, I will reconstruct OJ k in Kojiki as a voiceless velar stop *k on the basis of the above arguments. OJ k in Shoki corresponds to eight CLMC initials (Table 7.26). *k vastly outnumbers all other initials and is the only initial that appears before all OJ ‘vowels.’ I will now proceed to argue against using any of the other initials as evidence for non-*k allophones of OJ k. (1) *c/*k and *c/*k®: these cases involve only two phonograms for OJ kyi. Each of these phonograms had two readings in EMC, one with an initial palatal and one with an initial velar:
EMC *ciR / *kyiR and
EMC *ciR / *gyiR. Double readings also existed in CLMC:
has the Kan-on readings si < CLMC *c- and ki < CLMC *k- and
has the Kan-on readings si < CLMC *c- and ki < CLMC *k®-. Since there are no cases of phonograms with only palatal-initial readings for OJ kyi, I will assume that
CLMC *c- / *k- and
CLMC *c- / *k®- were chosen as phonograms for OJ kyi on the basis of their velar-initial readings. I reject *c as a possible allophone of OJ k before the ‘vowel’ yi. (2) *kh and *k®: aspirated initials are far rarer than *k and do not appear before all ‘vowels.’ The Japanese probably could not distinguish between *k, *kh, and *k® (which all corresponded to Kan-on k) and used phonograms 189

OLD JAPANESE

Table 7.27 OJ k in Shoki alpha section poetry OJ \ CLMC

*c/*k

*k

*k(h)

*kh

*k®

a yi iy u ye ey wo o Subtotals Total

0 49 0 0 0 0 0 0 49 322

83 0 10 80 2 7 29 22 233

0 0 0 0 4 0 0 0 4

1 7 0 0 1 7 0 0 16

0 16 0 0 0 0 0 4 20

with all three initials interchangeably. Chinese speakers would of course be able to distinguish between all three initials and would not use *k-, *kh-, and *k®-initial phonograms interchangeably. Mori (1991) credited Chinese scribes with the low frequency of aspirate phonograms in his alpha section of Shoki. My own statistics for OJ k in the alpha section confirm this low frequency (Table 7.27). Notice how nearly two-thirds of the aspirates (23/36) are found among phonograms for OJ kyi and roughly half of the remaining third (7/36) are found among phonograms for OJ key. I doubt that Chinese speakers would tend to perceive an aspirated allophone *kh before the ‘vowels’ yi and ey but not before other vowels. I know of no other languages with this pattern of allophony and would not reconstruct aspirated allophones for any OJ k-initial syllables. (3) CLMC *h and *h®: I do think that the scribes of the alpha section were more knowledgeable about Chinese than the more conservative scribes of the beta section. As Mori (1991) also pointed out, OJ k is never written with CLMC fricatives in the alpha section. This shows that OJ k probably had no fricative allophones such as *h or *®. I suspect that Japanese and Korean peninsular students of Chinese found a distinction between CLMC *k and *h to be easier to learn than the distinctions between *k, *kh, and *k®. So far we have seen nothing suggesting the existence of phonemic aspiration in OJ. Eom (1994) denies that any of the three major branches of OSK (p. 110) had aspiration. The poor degree of correlation between aspiration in LMC and aspiration in SK proper (i.e., the last stratum of Sino-Silla) is probably proof of the partial or total absence of aspiration in the Silla language. Without aspiration in their own languages, non-native speakers of Chinese were more likely to make mistakes involving aspiration than mistakes involving velar stops and glottal fricatives. (4) CLMC *kh/*q [}]: all 42 of these cases involve a single beta section phonogram
which has two readings, CLMC *kh- (cf. SV khu [xuu] ) and CLMC *q- [}] (cf. SV âu [}Rw] ). Since OJ ku is written 105 times in Shoki with 190

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

CLMC *k, I will assume that the phonogram
was chosen for its velarinitial reading *kh- and not its glottal-initial reading *q- [}]. (5) CLMC *k®ò: all 41 of these cases involve a single beta section phonogram
CLMC *k®ò- < EMC *gòa which was originally devised by the Chinese as a transcription character for Indic ga, ga, gha, and gha. Once I would have assumed that
CLMC *k®ò- was evidence for a velar allophone *k of OJ k which would have been in free variation with a uvular allophone *q represented by phonograms like
CLMC *k- (*[q] ), but I have already rejected this hypothesis above.
CLMC *k®ò- never appears in the alpha section. If the scribes of the alpha section were Chinese as Mori (1991) supposes, they would have used
CLMC *k®ò- or, more likely,
CLMC *kò- < EMC *kòa, the phonogram for Indic ka and ka, to transcribe OJ ka (*[kV] ). I suppose one could claim that the alpha section scribes were well versed enough in Chinese not to make the ‘mistake’ of writing an OJ uvular stop *q with CLMC *kò- or *k®ò, but I am still not enthusiastic about reconstructing an OJ *q solely on the basis of phonogram evidence. It is simpler to assume that OJ k was velar like its reflexes in LMJ and MSJ. I will reconstruct OJ k as an voiced unaspirated velar stop *k in all environments for both Kojiki and Shoki.

OJ g OJ g in Kojiki corresponds to EMC *kò, g, *Y, and *® (Table 7.28). Although OJ g most frequently corresponds to EMC *®, this initial appears only in phonograms for OJ ga (
LOC *gahA > EMC *®ah,
LOC *ga(h)A > EMC *®a(h) ) and perhaps OJ gwo (two probable cases of
LOC *goA > EMC *®o).6 Note that all three phonograms have LOC readings with a voiced velar stop *g. Perhaps these characters were chosen to be phonograms for OJ ga and gwo on the basis of their LOC initial *g. (Cf. the choice of LOC *w-initial phonograms for OJ w-syllables on p. 170.) Table 7.28 OJ g in Kojiki poetry OJ \ EMC

*kò

*Y

*g

*®

a yi iy u ye ey wo o Subtotals Total

1 0 0 0 0 0 0 0 1 196 (+2)

0 16 4 0 1 6 0 0 27

0 1 0 16 0 0 0 11 28

140 0 0 0 0 0 (2) 0 140 (+2)

191

OLD JAPANESE

Literalists who take Table 7.28 at face value might want to reconstruct the following pattern of allophony: 1 2 3

*® (or *g) before OJ a and wo *g before OJ u and o *Y elsewhere (before OJ yi, iy, ye, and ey).

If I conflate (1) and (2), I need to explain why OJ g had a nasal allophone only before four ‘vowels’. I know of no correlations between vowels and the prenasalization of reflexes of OJ g in later varieties of Japanese. Is there any evidence for the nasality of OJ g before OJ a, u, wo, and o? In the prose of Kojiki, the phonogram
EMC *htRY represents three OJ two-syllable sequences: kaga, kagu, and kago.
EMC *stRY(h) represents the OJ two-syllable sequence saga in the prose of Kojiki. OJ ga is also written once in the prose of Kojiki as
EMC *Yaq. If the initials of syllables such as OJ ga, gu, and go had no nasality, why spell them with EMC *Y? Why not consistently spell them with EMC *g or *® (< LOC *g)? In the Suiko inscriptions, OJ ga was spelled with
EMC *YiR as well as
EMC *giR and
EMC *kæ (Ôno Tôru 1962: 50), so the use of Chinese *Y to transcribe the initial of g was not an innovation of Kojiki. OJ g is transcribed with a mix of nasals and non-nasals throughout the history of Japanese phonogram writing. In texts other than Shoki (see below), this mixture reflected the absence of a prenasalized voiced velar stop *Yg in LOC and EMC. Foreign *Y could approximate the nasal element of OJ g = *Yg while foreign *g could approximate its non-nasal element. The orthography of Kojiki poetry implies a pattern of allophony (*g~*Y) that is contradicted by all other evidence. No later stage of Japanese exhibits such allophony. Hence I will reconstruct OJ g for Kojiki as a prenasalized voiced velar stop *Yg. OJ g in Shoki corresponds to CLMC *c/*k, *k®, and *Yg (Table 7.29). Unlike the orthographies of other OJ phonogram texts, the orthography of Table 7.29 OJ g in Shoki poetry OJ \ CLMC

*c/*k

*k®

*Yg

a yi iy u ey wo o Subtotals Total

0 1 0 0 0 0 0 1 173

0 0 0 2 0 0 1 3

106 16 13 7 9 12 6 169

192

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

Shoki was based on CLMC, which did have a prenasalized voiced velar stop *Yg. The 97.8 per cent (169/173) correlation between CLMC *Yg and OJ g in Shoki confirms my earlier reconstruction of OJ g as *Yg. The exceptions may reflect an allophone such as *~g. Perhaps OJ g had allophony similar to the allophony proposed for OJ b, d, and z on pp. 168, 174, and 186: 1 2 3

OJ g was generally a prenasalized voiced unaspirated velar stop *Yg. ‘Vowels’ preceding *Yg may have been nasalized (*~). *Yg may have been reduced to *g if a preceding ‘vowel’ was nasalized (*~). OJ *Yg may have been phonetically realized in at least three ways:

(7.11) (1) *[Yg]

(2) *[~Yg] (3) *[~g]

Why do I not reconstruct *Y or *~Y as an allophone of OJ g? It is not as if an OJ *Y would have no parallels in later stages of the language. In MSJ, *Y is an intervocalic allophone of /g/. Kamei (1956) has reconstructed OJ g as *~Y in intervocalic position and Kiyose (1991) has reconstructed it as *Y in all positions. I object to the reconstruction of a velar nasal allophone *(~)Y for two reasons. (1) Scribes consistently transcribed the two other OJ nasals (m and n) only with LOC or EMC nasals in non-Shoki OJ phonogram texts. However, they transcribed OJ g with both nasals (*Y) and non-nasals (*g), perhaps because OJ g (*Yg) was somehow different from ordinary nasals such as OJ m (*m) and n (*n). (2) The SJ reflexes of MC *Y are vowels (i and u), not g-vowel sequences. It is true that MC *-Y codas do correspond to OJ g. However, we must not forget that the OJ sounds represented by phonograms and the OJ period SJ readings of phonograms were not necessarily one and the same.
EMC *htRY was chosen as an approximation of OJ kagV (*kVYgV). Its OJ period Go-on reading was probably not identical to its readings as a phonogram (i.e., OJ kaga, kagu, and kago), since its Go-on reading is kau, not *kagu. I have hypothesized that OJ had a short-lived phoneme *Y found only in SJ loans which was later lost:
EMC *htRY > OJ Go-on *kaYu > post-OJ Go-on kau. According to a more mainstream view, MC *Y was borrowed as OJ -u and -Z (Hamada 1952, Miller 1967, Frellesvig 1995). Erickson (1998) denies the existence of nasalized vowel phonemes in OJ and implies that MC *Y was borrowed as non-nasalized OJ -i and -u. In any case, OJ g must not have been appropriate for rendering MC *Y. If one claims that MC *Y was borrowed as OJ g, then one must explain why the usual reflexes of OJ g are zero in SJ forms and g in native words. I do not understand what would motivate the deletion of OJ g in SJ. I would rather claim that MC *Y was borrowed as something other than OJ g (my *Yg) to avoid positing such divergent developments. 193

OLD JAPANESE

Although I have proposed allophones such as *~g for OJ g, I will write my reconstruction of OJ g as *Yg. One could also write OJ g less redundantly as *g or *nk without specifying either nasality or voicing.

OJ 2 The OJ ‘zero initial’ in Kojiki corresponds to EMC *q and *w (Table 7.30). I have included all attestations of OJ e in Kojiki: five cases of
EMC *qRyh in the text (I: Age of the Gods) and two cases of
EMC *qæh in the poetry (9.18.1, 9.18.2). EMC *q [}] and EMC *w are in complementary distribution in Table 7.30. Should I reconstruct OJ $ as *w before u and as *q [}] elsewhere? All 56 instances of OJ u in the poetry of Kojiki are spelled with the phonogram
EMC *wuRq. This phonogram may have been arbitrarily chosen among several phonograms already in use to be the one and only phonogram for OJ u in Kojiki poetry. In the prose of Kojiki, we find OJ u spelled once with
EMC *qo(h) / *qwæ in a gloss for OJ ukey ‘food’
 EMC *qo(h) khtyh (I: Age of the Gods). This word does not appear elsewhere in Kojiki, but it does reappear in Shoki as CLMC
 *w- k- (cf. SV vu cai [vuu kaay] ). Ôno Tôru (1962: 88, 141) considers
EMC *qo(h) (rarely EMC *qwæ) to be the principal phonogram for OJ u in the Suiko inscriptions. The only other Suiko phonogram for OJ u,
EMC *wuwq, appears only four times (Ôno Tôru 1962: 52). OJ i is mostly written as
EMC *qyi (as in Kojiki) in the Suiko inscriptions, but it is also written three times as
EMC *yi (Ôno Tôru 1962: 52). Should we conclude that there was variation between *w~*q for OJ u and between *y~*q for OJ i in the Suiko Period which was lost by the time Kojiki was written? I would rather not interpret the evidence so literally. I will reconstruct OJ $ as *$ for Kojiki, leaving open the question of its exact phonetic manifestation. I think that OJ $ may have been a true zero initial. OJ speakers may not have been able to distinguish between *wu and *}u or *yi and *}i. The Shoki evidence corroborates my previous claims. OJ $ corresponds to *q and *y before i, *q and *w before u, and only to *q elsewhere (Table 7.31). Table 7.30 OJ $ in Kojiki poetry OJ \ EMC

*q

*w

a i u e o Subtotals Total

120 118 0 2 (+5) 75 315 (+5) 371+5

0 0 56 0 0 56

194

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

Table 7.31 OJ $ in Shoki poetry OJ \ CLMC

*q

*w

*y

a i u e o Subtotals Total

124 80 5 2 73 284 379

0 0 61 0 0 61

0 34 0 0 0 34

Table 7.32 OJ $ in Shoki alpha section poetry OJ \ CLMC

*q

*w

*y

a i u e o Subtotals Total

53 33 4 1 37 128 153

0 0 18 0 0 18

0 7 0 0 0 7

I see no need to reconstruct allophonic variation between initial *q and *y for i or between initial *q and *w for u. We can easily find words spelled both with and without CLMC glottal stops in Shoki. Even the alpha section of Shoki mixes glottal stop and non-glottal-stop spellings of OJ i and u, though a distinction between *q, *y, and *w was phonemic in CLMC (Table 7.32). If the scribes of Shoki could not distinguish between *w and *q, they could not have been native speakers of Chinese. I am finding Mori’s hypothesis of Chinese scribes for the alpha section increasingly difficult to accept. Unger (1993: 6, 22) reconstructed the i of OJ words such as kai ‘oar’ and kui ‘regret’ as */yi/, but the phonogram evidence does not strongly support his position. OJ kai does not appear in Shoki, but OJ kui does appear as
 CLMC *k- q- (cf. SV cu y [kuu }ii] ) with a medial glottal stop (*q) instead of the expected medial *-y- (NS 124.7). OJ kai is always spelled in Man’yôshû with a medial glottal stop as either
 EMC *kæ qyi (153.9, etc.) or
 EMC *®ah (< LOC *gah) qyi (1052.5). These spellings do not necessarily disprove Unger’s view. One could argue that
EMC *qyi was not inappropriate for Unger’s OJ */yi/ since its EMC reading did have a medial *-y-. Nevertheless, EMC and CLMC did have similar syllables with initial *y such as
CLMC *y- (cf. SV di [zii] < *yi) < EMC *yi, so scribes did have alternatives to
EMC *qyi for the transcription of OJ */yi/. Unger (1993: 6, 22) also reconstructed the second u of OJ ku-u ‘kick-FIN’ as */wu /. However, since only the infinitive form of OJ ku-u, OJ kuw-e 195

OLD JAPANESE

Table 7.33 The OJ consonant inventory

Voiceless obstruents Prenasalized voiced obstruents Nasals Liquid and glides

Labial

Dental

*p *mb *m *w

*t *nd

Palatal *s *nz

*n *r

Velar *k *Yg

*y

‘kick-INF,’ appears in phonograms (
 CLMC *k- qw- < EMC *kuR qwiRyh; NS Age of the Gods I), I cannot directly test his reconstruction of */wu/. Furthermore, since OJ u is generally written with MC *w after the Suiko Period,7 the second syllable of OJ ku-u would probably have been spelled with a *w-initial phonogram, regardless of whether it had an initial *w or not. LMJ and MSJ do not have any syllables such as *yi or *wu, so I will not reconstruct such syllables in OJ. I will reconstruct OJ $ as *$ for both Kojiki and Shoki.

Summary I have reconstructed thirteen consonants for OJ (Table 7.33). This system is similar to those of LMJ and MSJ. The main differences are as follows. (1) OJ had a labial stop *p corresponding to the LMJ labial fricative f [φ] and the MSJ glottal fricative h (which has a labial allophone f [φ] before u [T] ). (2) The OJ prenasalized voiced obstruents (*mb, *nd, *nz, *Yg) have lost their prenasalization in MSJ, though MSJ /g/ does have an intervocalic nasal allophone [Y]. (3) Unlike LMJ and MSJ, OJ may have lacked palatal allophones of s and z (i.e., *\ and *ã9 or *ãj) before front ‘vowels.’ The OJ consonant system does not offer us any real surprises. But, as we will see in the next chapter, the same cannot be said of the OJ vowel system.

Notes 1 I cannot cite CLMC finals because I have only reconstructed the CLMC initials. I will cite LMC-based SK and SV readings whenever I need to refer to CLMC finals. 2 One might expect the graphs for OJ wo to have the EMC initial *®w-, assuming that OJ o was a non-high vowel. Yet both Kojiki graphs for OJ wo (
EMC *wuRn,
EMC *wuRnq/h) in fact had the EMC initial *w-. Does this necessarily mean that OJ o was a high vowel? No, for these graphs may have been chosen for their EMC readings’ schwas. I present evidence for *R in this EMC rhyme in (8.59).

196

THE RECONSTRUCTION OF OLD JAPANESE CONSONANTS

3 Ôno (1953a: 211) and the Nihon koten bungaku taikei edition of Shoki interpret phrase-initial
(NS 40.6.1) as da ‘who’ rather than as ta ‘id.’ 4 I do not understand why *g-initial phonograms were used in these cases. 5 I do not know why voiced initial phonograms such as
EMC *®ah (cf. Go-on ga) and
EMC *®o (cf. Go-on go) were used for OJ ka and kwo. 6 This phonogram appears in two phrases (KJK 9.19, 9.22) interpreted as either siyagwo siya (Shintô taikei ed. of Kojiki) or siya kwosiya (Kodai kayô shû ed. of Kojiki). 7 Ôno Tôru (1962: 145) considers
EMC *wuRq > CLMC *w- (cf. SV vZ [vuu] ) to be the “sole commonly used phonogram” for OJ u in the eighth century.

197

OLD JAPANESE

8 THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Overview In this chapter, I will reconstruct the ‘vowels’ of OJ in the following sequence. First, I will reconstruct what I consider to be the primary ‘vowels’ of OJ: a, yi, u, and o. Much of the work on pre-OJ internal reconstruction is based upon the assumption that the pre-OJ predecessors of these four ‘vowels’ (*a, *i, *u, *R) combined into diphthongs which developed into the remaining OJ ‘vowels’ ( ye, ey, iy, wo) (p. 81). If OJ a, yi, u, and o turn out to be quite different from *a, *i, *u, and *R, then I will have to reevaluate preexisting hypotheses of diphthongal origins for the other OJ ‘vowels.’ Then I will reconstruct what I consider to be the secondary ‘vowels’ of OJ: iy, ye, ey, and wo. I will see whether their reconstructed phonetic forms are consistent with theories of their derivations from earlier combinations of the primary ‘vowels.’ Finally, I will determine whether the OJ C-type ‘vowels’ (i, e, o) can be identified as A- or B-type ‘vowels’ in each syllable that they appear in. There is no guarantee that all syllables ending in a C-type ‘vowel’ actually had the same ‘vowel.’ Some syllables may have had A-type ‘vowels’ and others may have had B-type ‘vowels.’ I will consider the possibility that the C-type ‘vowels’ may truly have been distinct from A- and B-type vowels. I will also evaluate Mabuchi’s (1957) and Bentley’s (1997) attempts to find further A/B-type distinctions among syllables with C-type ‘vowels.’ As in the previous chapter, I use the terms ‘Kojiki’ and ‘Shoki’ to refer only to the poetry of those two works unless otherwise specified.

OJ a OJ a in Kojiki corresponds to five EMC rhyme categories (Table 8.1). The most common rhyme category is EMC *-a. The other rhyme categories are found only for one or two syllables: *-æ (OJ ka, ma), *-ya (OJ za, ya), *-tRY (OJ ra), *-òa (OJ ka, ga), and *-aY (OJ ta). 198

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.1 OJ a in Kojiki poetry OJ \ EMC

*-a

*-aY

*-òa

*-æ

*-tRY

*-ya

p b m w t d n r s z y k g $ Subtotals Total

208 66 4 69 222 56 126 1 117 0 0 5 140 120 1134 1785

0 0 0 0 1 0 0 0 0 0 0 0 0 0 1

0 0 0 0 0 0 0 0 0 0 0 68 1 0 69

0 0 194 0 0 0 0 0 0 0 0 127 0 0 321

0 0 0 0 0 0 0 101 0 0 0 0 0 0 101

0 0 0 0 0 0 0 0 0 5 154 0 0 0 159

As explained in Chapter 6, I will not use my modified version of Pulleyblank’s (1984, 1991a) EMC reconstruction as primary evidence for the values of OJ ‘vowels.’ The finals of Pulleyblank’s EMC reconstruction were partly based on his interpretation of OJ phonograms. To avoid circularity, I will rely on attested evidence rather than reconstructions, though I will use reconstructions as easily typable tags for Qieyun/Guangyun rhyme names. For example, ‘*-a’ refers to the Qieyun/Guangyun
‘song’ rhyme. See Pulleyblank (1984: 233–236) for a list of Qieyun/Guangyun rhymes and their EMC reconstructions. I will reconstruct the OJ ‘vowel’ a in Kojiki as a low unrounded vowel *a for the following reasons. (1) OSV a [aa] corresponds to EMC *-a. (8.1) Sg
‘seat’
‘disaster’

EMC *dzwa *®waq < LOC *waqA

OSV SV tòa [twaa] tia [twaa] vf [vaa] hia [hwaa]

I have not been able to find any OSV words corresponding to OJ a-phonograms. (2) OJ a-phonograms appear as nôm phonograms for Vietnamese a-syllables:

199

OLD JAPANESE

(8.2) OJ a ta ta da ra wa

Ph







EMC *qa *ta *tha *da *la *®wa

Nôm a [}aa] $a [ßaa] tha, thà [thaa] $à [ßaa] la, là [laa] hòa [hwaa], òa [}waa]

SV a [}aa] $a [ßaa] tha [thaa] $à [ßaa] la [laa] hòa [hwaa]

(3) OJ a-phonograms serve as neutral vowel bases for vietographs generally representing Vietnamese a-syllables: (8.3) OJ Ph ta
na
ra


EMC *ta *nah *la

Derived vietographs i $á ‘stone’ j da ‘skin’ k nf ‘mask’ l ra ‘go out’

(4) OJ a-phonograms have EMC-based Colloquial Taiwanese (CT) readings ending in -a or -oa [wa]: (8.4) OJ a ka ga na

Ph





EMC *qa *khaq *®a *nah

CT a1 khoa 3 oa 2 na 5

(5) OJ a-phonograms appear among Jñanagupta’s (Jn) EMC-based late fifth-century CE transcriptions of Indic syllables ending in -a [£] or -a [aa]: (8.5) OJ a ta ta da na pa ba ra

Ph









EMC *qa *ta *tha *da *na(q/h) *pa *ba *la

Ind. tr. (Jn) a, a ta, ta tha da, dha, dha na, na pa, va ba, bha, bha, va ra, ra

The data in (8.1– 8.5) included only phonograms ending in EMC *-a. However, five OJ a-syllables (ma, ra, za, ya, ka) are predominantly or entirely written with phonograms of other rhymes. Should I reconstruct special allophones of OJ a for these syllables? (1) The *-æ phonograms for OJ ka and ma:
EMC *kæ,
EMC *mæ. I know of no Japanese-internal evidence which would indicate a special 200

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

vowel allophone for OJ ka and ma. There is no property unique to *k and *m that might motivate the development of such an allophone. Hence I will reconstruct both syllables with *-a as OJ *ka and *ma. Note that these syllables also do appear in Kojiki with EMC *-a phonograms:
EMC *kha (cf. CT khoa 3 [khwa]; 42.27.4) and
EMC *ma (Jñanagupta’s phonogram for Ind ma; 15.5.5, etc.). (2) The *-òa phonogram
EMC *kòa for OJ ka. I have already mentioned that this character was originally created to transcribe Indic ka and ka which were syllables without any equivalents in EMC (pp. 148 and 187). Therefore
presumably represented an OJ *ka. (3) The *-ya phonograms
EMC *zya/*ya and
EMC *yah for OJ za and ya.
EMC *zya /*ya is homophonous with
EMC *ya, which Jñanagupta used to transcribe Ind ya and ya.
EMC *yah appears in Jñanagupta’s transcriptions as a phonogram for Ind ya. Thus I see no problem with reconstructing OJ za and ya as *za and *ya. See p. 184 for the reasons why I do not reconstruct OJ za as *zya. (4) The *-tRY phonogram
EMC *ltRY for OJ ra. In the poetry of Kojiki, OJ ra is always written as
EMC *ltRY except for one indisputable instance of
EMC *la (52.2.5). Both of these phonograms are interchangeable in Korean peninsular transcriptions. They are, however, not interchangeable in Chinese transcriptions. I suspect that the use of both phonograms for *la or *ra originated on the Korean peninsula and spread to Japan. I am not certain why
EMC *ltRY would have been used to transcribe open syllables. However,
EMC *ltRY would be a good choice for foreign *la or *ra if one ignores the coda and looks at its LOC form *laYB. There is no need to interpret
EMC *ltRY as evidence for a special allophone of OJ a. I will reconstruct OJ ra as *ra. (5) The *-aY phonogram
EMC *taY for OJ ta. This phonogram appears only once in Kojiki in the name ‘Tagyimati’: (8.6)


! EMC *taY YyiRyh mæ tri (KJK 77.5)

The final *-Y of
may reflect the prenasalization of the following OJ g = *Yg. None of the exceptional rhymes in Kojiki offers significant counterevidence against my reconstruction of OJ a as *a before all initials. The a-phonograms of Shoki also do not offer any such counterevidence even though they belong to twelve different EMC rhyme categories (Table 8.2). Most OJ a-phonograms in Shoki belong to the EMC *-a rhyme category. The majority of the remaining phonograms belong to EMC rhyme categories containing the vowel *a. These rhyme categories apparently had LMC reflexes with an *a-like vowel, judging from SK and SV (Table 8.3). Most of the OJ a-phonograms (1) are read with a in LMC-based SK and SV, (2) were transcribed with a in Tibetan, and (3) were used by Amoghavajra (Am; 705–774 CE) to transcribe Indic syllables ending in a and a. In 201

OLD JAPANESE

Table 8.2 OJ a in Shoki poetry OJ \ EMC

*-a

*-ak *-an *-aY

*-at

*-ay *-æ *-εy

*-iR *-tRY *-Ry

*-ya

p b m w t d n r s z y k g $ Subtotals Total

201 37 114 75 109 19 137 120 86 0 0 165 106 124 1,293 1,548

0 0 0 0 0 0 0 1 0 0 0 0 0 0 1

3 0 0 0 0 0 0 0 0 0 0 0 0 0 3

0 0 2 0 0 0 0 0 0 0 0 0 0 0 2

0 0 0 0 1 1 0 0 0 0 0 0 0 0 2

0 0 0 0 0 0 0 0 0 0 0 18 0 0 18

0 1 0 0 0 0 0 0 0 0 0 0 0 0 1

0 0 0 0 13 0 0 0 3 0 124 0 0 0 140

OJ \ EMC

1,548

*-ak

*-an *-aY *-at

*-εy

*-iR *-tRY *-Ry *-ya

0 0 28 0 2 13 0 0 0 2 0 0 0 0 45

0 0 35 0 0 0 0 0 2 0 0 2 0 0 39

*-ay *-æ

0 0 0 0 0 0 0 0 0 3 0 0 0 0 3

0 0 0 0 0 0 1 0 0 0 0 0 0 0 1

Table 8.3 SX equivalents of rhyme categories for Shoki a-phonograms EMC *-(y)a *-ak

*-an

*-aY

*-at

*-ay

*-æ

*-εy

*-iR *-tRY -i

Kan

-(y)a

-aku

-an

-au

-atu

-ai

-a

-ai

SK

-(y)a

-ak

-an

-ang

-al

-ay

-a

-ay, -oy -i

SV

-a [aa]

-ac -an -ang -at -ai -a -ai [aak] [aan] [aaY] [aat] [aay] [aa] [aay]

-i [ii]

*-Ry

-au

-ai

-ang

-ay

-edng -ai [tRY] [aay]

Table 8.4, I list the most common phonograms for OJ a-syllables along with their SX readings, their Tibetan transcriptions, and their Indic sound values in Amoghavajra’s transcriptions. I have also listed the most common phonograms which did not belong to the EMC *-a rhyme category. 61.8 per cent (957/1,548) of the total number of a-syllables in Shoki are written with these nineteen phonograms. The only OJ a-phonogram that does not fit into this pattern is: (8.7)
EMC *myiR > CLMC *mb-: Kan bi, SK mi, SV mi for OJ ba. It appears once in all Shoki texts (19.4.7) and appears two more times (7.9.5, 7.10.4) in some texts as a variant of the most common phonogram for OJ ba: (8.8)
EMC *maq > CLMC *mb-: Kan ba, SK ma, SV mk 202

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.4 Common a-phonograms in Shoki poetry OJ

Ph

EMC

CLMC

Kan

SK

SV

Tib. tr.

Ind. tr. (Am)

pa ba





*pa *maq

*p*mb-

ha ba

pha ma

ba mk

pa –

ma











*ma *mæ *maYq *®wa *ta *naq *naYh *na

*mb*mb*m(b)*h®w*t*nd*n(d)*nd-

ba ba bau kwa ta da dau da

ma ma mang hwa ta na nang na

ma ma mãng hòa $a nã náng na

ba, ’ba ’ba, ma – hwa ta, da – – ’da












*la *tsah *crtRY *yah *ka *gòa *kεyh *Yah *qa

*l*ts*cr*y*k*k®*k*Yg*q-

ra sa sau ya ka kya kai ga a

la ca cang ya ka ka kay a a

la tá trang df kha già giái, giDi ngã a

la, la dza – ya – ga – ’ga a

– ba, ma, ma, (m)ba, (m)bha ma, (m)ba ma ma – ta, ta, ∞a da, da – na, (n)da, (n)da, (r)da ¶a, la, ra, ra ca – ya – gha, gha – – a, a

wa ta da na ra sa za ya ka ga a

Since the two characters are graphically very similar, all three cases of
EMC *myiR > CLMC *mb- for OJ ba are probably scribal errors for
EMC *ma > CLMC *mb-. 4.8 per cent (75/1548) of OJ a-syllables are written with MC closed syllable phonograms: e.g.,
EMC *maY > CLMC *m(b)- for OJ ma. Such phonograms are not evidence for OJ *-aY, etc., because they are interchangeable with MC open syllable phonograms (8.9). (8.9) OJ ma ‘true’

Spellings
EMC *maY > CLMC *m(b)- (cf. SK mang) (NS 96.9.1)
EMC *ma > CLMC *mb- (cf. SK ma) (NS 103.6.4)
EMC *mæ > CLMC *mb- (cf. SK ma) (NS 128.3.1)

I will reconstruct OJ a in both Kojiki and Shoki as a low unrounded vowel *a.

OJ yi OJ yi (‘A-type i’) in Kojiki corresponds to six EMC rhyme categories (Table 8.5). The most common of these categories is *-i, though no single category is predominantly used to write all five yi-syllables ( pyi, byi, myi, kyi, gyi). 203

OLD JAPANESE

Table 8.5 OJ yi in Kojiki poetry OJ \ EMC

*-i

*-iR

*-yi

*-yi(t)

*-yiR

*-yiRy

p b m k g Subtotals Total

0 0 179 0 0 179 497

0 0 0 4 0 4

1 15 0 1 0 17

127 1 0 0 0 128

0 0 1 150 1 152

0 0 0 0 17 17

All cases of *-yit involve the phonogram
, which can be read either with or without a final stop *-t: EMC *pyiq, *pyih, *byi, *byih, *byit. (Note that *-q and *-h are orthographic tone markers and not segments: EMC *pyiq may have been something like *[pyi] with the Rising Tone.) I will treat
as a *-yi final phonogram. A literalist might want to reconstruct five different allophones of OJ yi corresponding to the five different EMC rhyme categories in Table 8.5. OJ pyi, byi, myi, kyi, and gyi would then each have their own allophone of OJ yi. This sounds highly improbable. I will reconstruct OJ yi in Kojiki as a high front vowel *i for the following reasons: (1) OSV [ii] and [iR] correspond to EMC *-i, -iR, and *-yi. (8.10) Sg EMC
‘two’ *ãih
‘pond’ *driR
‘compare’ *pyiq

OSV nhì [ ãii] $ìa [ß iR] ví [vii]

SV nhh [ ãii] $ì [ß ii] tm [tii] < LMC *py-

(2) The evidence for the ‘vowel’ of OJ gyi is puzzling at first glance. The most frequent graph for OJ gyi in the poetry of Kojiki is
EMC *YyiRyh, whose OSV cognate has the final -ê [ee]. (8.11) Sg EMC
‘art, skill’ *YyiRyh

OSV SV nghV [Yee] nghq [Yee]

The vowel of OSV -ê [ee] is lower than i in OSV -i [ii] and -ia [iR], but nevertheless it is non-low and palatal. This graph may have once represented a pre-OJ *Yge which raised to OJ gyi (p. 79). OJ gyi is also spelled once with (8.12) OJ Ph EMC gyi
*gyiR

(32.2.2) 204

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

This graph also represents OJ gyi in the text of Kojiki, where the name of the god Ninigyi has one spelling with
*gyiR and another with
EMC *YyiRyh: (8.13)
 EMC *ãin gyiR

 EMC *ãiRq ãiRq YyiRyh

(KJK Preface) (KJK I: Age of the Gods)

If OJ gyi really had a ‘vowel’ distinct from that of OJ kyi, it would never have been spelled with
EMC *gyiR, the usual phonogram for OJ kyi. I have not been able to find any OSV evidence for EMC *-yiR. (3)
EMC *miq, the main phonogram for OJ myi, appears as a palatal base for the following vietographs which have palatal vowels in two out of three cases: (8.14)
mg ‘mother’

m mía ‘sugar cane’

(but n mN ‘open’)

(4) Jñanagupta transcribed Indic syllables containing palatal vowels (i, c, e) using OJ yi-phonograms. (8.15) OJ pyi byi myi kyi

Ph





EMC *pyiq/h *byi *myiR *gyiR

Ind tr (Jn) pu (sic!) pi, bhc, vi, vc, ve mi, me gi ( fanqie:
 EMC *gtR + *khyih = *gih), khye, gra (sic!)

Pulleyblank (1984: 148) notes that Jñanagupta uses EMC *-i and *-iR phonograms interchangeably for Ind i and c. In (8.15), Ind -i corresponds to both EMC *-yi and *-yiR. Pulleyblank interprets this as evidence for the merger of these two rhyme classes in northern EMC. Southern EMC, the source of CT and OSV, kept these categories distinct: (8.16) EMC CT *-(y)iR -ia [ya], -i *-(y)i -ai [ay], -i

OSV -ia [iR] -i [ii]

Go -i (usual), -e (rare) -i

Judging from Jñanagupta’s transcriptions, the two categories had merged into *-i in the north by the late sixth century. This merger is evident not only in Go-on (8.16) but also in Kojiki. For example, the second syllable of the OJ word kyimyi ‘lord’ can be spelled with phonograms of both the EMC *-iR and *-i rhyme categories: (8.17)
 EMC *gyiR mi > late 6th c. n. EMC *gyi mi (KJK 7.4)
 EMC *gyiR myiR > late 6th c. n. EMC *gyi myi (KJK 65.4) 205

OLD JAPANESE

Table 8.6 OJ yi in Shoki poetry OJ \ EMC

*-i

*-iR

*-yi

*-yiR

*-yiRy

yit

p b m k g Subtotals Total

0 0 9 15 0 24 442

0 0 0 0 3 3

94 10 1 3 0 108

9 5 141 137 1 293

0 0 0 0 13 13

0 0 0 1 0 1

Table 8.7 Common yi-phonograms in Shoki poetry OJ

Ph

EMC

CLMC

Kan

SK

SV

Tib. tr.

Ind. tr. (Am)

pyi








*pyiq *pyih *myih *myiRq *myiR

pi pi mi, moy mi mi mi mi ki ki yey

tm tí mh mW ri [sic] mW di – kì, kz nghq

pi – – mi, me –

*myiR *kyiRq *gyiR *YyiRyh

hi hi bi bi bi bi bi ki ki gei

bt, ’byi – ’bt – –







*p(yiq) *p(yih) *mb(yih) *mb(yiq) *mb(yi) *mb(yiq) *mb(yi) *k(yiq) *k®(yi) *Yg-

’byi, myi – gi, gyi ’ge’t

– ki, kc – –

byi myi kyi gyi

Thus there is no need to reconstruct a vowel other than *-i for OJ yisyllables represented by phonograms of the *-iR rhyme category. The *-iR : *-i distinction was no longer relevant for the late variety of EMC that was the basis of most of the orthography of Kojiki. It is simplest to reconstruct *i as the vowel of all OJ yi-syllables in Kojiki. OJ yi in Shoki corresponds to six EMC rhyme categories (Table 8.6). The two most frequent final categories are *-yiR (66.3 per cent = 293/442) and *-yi (24.4 per cent = 108/442). Since I have already demonstrated that these categories had merged in Jñanagupta’s sixth-century EMC, I can treat them as a single final *-yi (90.7 per cent = 401/442). In Table 8.7, I list the LMC-based SX readings and the transcriptional data for the most common Shoki yi-phonograms. I have provisionally supplied CLMC finals in parentheses whenever possible on the basis of the EMC final mergers discussed above. The above nine phonograms are quite representative, for they account for 81.9 per cent (362/442) of the total. They all represent LMC Grade IV syllables (p. 97). SK generally only allows the palatal vowel i or the palatal glide y after velar initials of Grade IV LMC syllables. Hence the SK readings of Grade IV=
, 206

THE RECONSTRUCTION OF OLD JAPANESE VOWELS


, and
are ki, ki, and yey [yRy]. Cenwun okphyen lists SK moy [m£y] for Grade IV
as a ‘vulgar’ (swok) reading; the proper reading is SK mi. SV generally palatalizes labials of Grade IV LMC syllables. The SV readings of OJ yi phonograms have dental initials derived from earlier labial-yod clusters (5.16, 5.20, 5.22): (8.18) OJ pyi pyi pyi pyi pyi myi myi

Ph








CLMC *p(yiq) *p(yih) *p(yi) *ph(yih) *p®(yi) *mb(yi) *mb(yi)

Earlier *pyii *pyii *pyii *phyii *pyii *myii *myii

SV > > > > > > >

SV tm tí ti thí tì di ri [sic]

[tii] [tii] [tii] [thii] [tii] [zii] [zii]

The phonograms
,
, and
are of low frequency and hence absent from Table 8.7. SV ri [zii] for
is probably an error for di. Although SV does not consistently have dentals corresponding to LMC Grade IV labial initials, we will later see that such dental reflexes are completely absent in the SV readings of Shoki phonograms for OJ piy, biy, and miy (p. 217) and OJ pey, bey, and mey (p. 227). These dental reflexes will be crucial in distinguishing the OJ palatal A-type ‘vowels’ ( yi, ye) from the OJ non-palatal B-type ‘vowels’ (iy, ey). On the basis of the SV evidence, one could reconstruct OJ yi as (a) *yi with a palatal glide *y or (b) an *i preceded by palatalized consonants (e.g. *pyi, *mb yi, etc.). If such features existed, I will consider them to be subphonemic. I will reconstruct OJ A-type yi as a high front unrounded vowel *i for both Kojiki and Shoki.

OJ u OJ u in Kojiki corresponds to four EMC rhyme categories (Table 8.8). These rhyme categories are in near-complementary distribution: 1 2 3 4

EMC EMC EMC EMC

*-o: *-Rw: *-uw: *-uR:

OJ OJ OJ OJ

pu, tu, nu du mu, ru, zu, yu, ku bu, su, gu, u

The distribution of OJ initial classes among these EMC rhyme categories appears to be random and is not strong evidence for four different allophones of OJ u. The evidence below points to a labial pronunciation of OJ u. (1) Words in all four EMC rhyme categories appear in OSV with labial vowels and/or labial glides. 207

OLD JAPANESE

Table 8.8 OJ u in Kojiki poetry OJ \ EMC

*-o

*-Rw

*-uw

*-uR

p b m t d n r s z y k g $ Subtotals Total

54 0 0 135 0 18 0 0 0 0 0 0 0 207 797

0 0 0 1 46 0 0 0 0 0 0 0 0 47

0 0 72 0 0 0 87 0 28 52 133 0 0 372

0 19 0 0 0 0 0 80 0 0 0 16 56 171

(8.19) Sg EMC OSV OJ cf. Ph EMC OJ


*noq nn [nQQ] –
*no nu



*dRwh *guwq $5 [ßoo] cpu [kRw] du –
*kuwq ku


*buR bùa [(uR] –
*puR, *buR bu

(2) The CT reflexes of all four EMC rhyme categories contain labial vowels and/or labial glides. (8.20) Sg EMC CT OJ


*no lo2 nu


*dRwh tau6 du


*kuwq ku 3 ku


*puR, *buR po1 bu

(3) Jñanagupta used phonograms for OJ u-syllables to transcribe Indic syllables containing u or e: (8.21) OJ Ph tu
nu
mu
yu
ru
(with the

EMC *to *no *muw *yuw *luw radical

Ind tr. (Jn) tu nu mu yu re
‘mouth’ added to indicate r-) 208

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Clearly, OJ u was not an unrounded vowel like MSJ u [T]. There are only a few obstacles left before I can confidently reconstruct OJ u as a rounded *u. First, why did OJ scribes write OJ pu, tu, nu with EMC *-o phonograms (cf. the OSV and CT evidence)? EMC had no *tuw or *nuw, so *to and *no were the best possible matches for OJ tu and nu. However, this does not explain the choice of
EMC *poh (cf. CT po5) for OJ pu, since EMC did have a pu. The answer may lie in pre-OJ vowel raising (4.14). If
EMC *poh were borrowed into Japanese before vowel raising as *po, it would become *pu after raising and be a suitable graph for OJ pu. Even if vowel raising were not relevant, semantics and frequency may have prevented scribes from using EMC *puw graphs: (8.22)
EMC *puw
EMC *puw
EMC *puwq
EMC *puwq
EMC *puwq
EMC *puwh

‘not’ ‘bright (clothing)’ ‘not’ ‘kind of plant’ ‘earthenware vessel’ ‘wealthy’


‘not’ and
‘not’ could be mistaken for Classical Chinese function words.
‘bright (clothing)’ and
‘earthenware vessel’ were relatively obscure characters. I have not listed other even more obscure *puw characters. Unlike Shoki orthography, Kojiki orthography mostly uses easily readable high-frequency characters as phonograms. We will see on p. 253 that
‘wealthy’ is the most common Kojiki phonogram for OJ po owing to its LOC reading *pRòhB. To use
‘wealthy’ for OJ pu as well as OJ po would have been confusing. Second, why did OJ scribes write OJ du with
EMC *dRwh? EMC had no *du, so this was the best possible match for OJ du. Third, why did OJ scribes write OJ bu, su, gu, u with EMC *-uR rhyme words? EMC *-uR originated from LOC *-wo. Perhaps the LOC readings of these phonograms were borrowed with pre-OJ *-o which then raised to OJ u (cf. 4.14). Alternately, scribes simply ignored the final EMC *R (still preserved in OSV) and used these graphs as if they simply ended in *-u. I will reconstruct OJ u in Kojiki as a high back rounded vowel *u. OJ u in Shoki corresponds to eight EMC rhyme categories (Table 8.9). All of the high-frequency rhyme categories contained labial vowels or labial glides, judging from their SK and SV reflexes (Table 8.10). The most frequent category is *-uR. Most of the phonograms for all OJ u-syllables except tu, du, and nu belong to this category. As in Kojiki, OJ tu, du, and nu are generally written with phonograms of the *-o and *-Rw rhyme categories. Like EMC, LMC probably did not have any syllables such as *tu, *du, and *nu. Kan-on has no tu, du, or nu. SV does not have 209

OLD JAPANESE

Table 8.9 OJ u in Shoki poetry OJ \ LMC

*-awY

*-o

*-Rw

*-uR

*-uR/*-Rw

*-uw

Other

p b m t d n r s z y k g $ Subtotals Total

0 0 0 0 0 10 0 0 0 0 0 0 0 10 740

6 2 0 101 3 14 9 4 0 0 0 0 0 139

0 0 10 3 35 0 6 0 0 0 7 0 0 61

51 4 56 0 0 0 38 65 28 27 124 9 65 467

0 0 0 0 0 0 0 0 0 0 9 0 0 9

0 0 0 0 0 0 21 4 2 13 8 0 0 48

0 0 3 (*-uwY) 0 0 2 (*-uRwY) 0 0 0 0 0 0 1 (*-o/*-wæ) 6

Note ‘Other’ includes all categories that appear three times or fewer: *-o/-wæ (once), *-uRwY (twice), *-uY (thrice).

Table 8.10 SX equivalents of major rhyme categories of Shoki u-phonograms EMC Kan SK SV

*-awY -ou -wong [oY] -ông [Rw§]

*-o -o -wo [o] -ô [o]

*-Rw -ou -wu [u] -âu [Rw]

*-uR -u -wu [u] -u [u]

*-uw -iu -(y)wu [(y)u] -u, -eu [u], [tw]

any syllables such as *$u [ßuu]. SV nu [nuu] for
‘wife and children’ is irregular. Its SK reading nwo [no] has the expected mid vowel. SK does have the syllables twu and nwu (
SK twu ‘bean,’
SK nwu ‘mongrel’) but this -wu corresponds to SV -âu [Rw] (cf.
SV $pu ‘bean,’
SV npu ‘mongrel’). Since Korean did not permit the finals -ew [Rw] and -ow [£w], -wu [u] would be the closest possible approximation of an LMC final such as *-Rw. Phonograms such as
EMC *to ‘capital’ (SK two, SV $ô),
EMC *dRwh ‘bean’ (SK twu, SV $pu), and
EMC *nawY ‘agriculture’ (SK nwong, SV nông) were the closest possible approximations of OJ tu, du, nu known to Shoki scribes. Such phonograms are not evidence for the reconstruction of allophones of OJ u after dental initials.

210

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.11 Common u-phonograms in Shoki poetry OJ

Ph

EMC

CLMC

Kan

SK

SV

Tib. tr.

Ind. tr. (Am)

pu






*phuRh *puRh *buR

*fh*f*f®-

hu hu hu

pwu pwu pwu

phó, thú phú phù

– – –


























*boh *muR *muRh *to *thoh *dRwh *dRwh *nawY *no *luRh *luw *suR *\uRq/h *ãuR *ãuRh *yuRh *yuw *kuR *khuR *YuR *YuR *wuR *wuRq

*p®*mbv*mbv*t*th*t®*t®*n(d)*nd*l*l*s*\*ã9*ã9*y*y*k*kh*Yg*Yg*w*w-

ho bu bu to to tou tou nou no ru riu su syu ju ju yu iu ku ku gu gu u u

pwo mwu mwu two two, thwo twu twu nwong nwo lwu lywu sywu sywu ywu ywu ywu ywu kwu kwu wu wu wu wu

bò, bs vZ vj $ô th3 $fu $fu nông nô lZ leu tu du, thâu, thú nho, nhu nhj dj, dZ do, du cu, câu khu ngu ngu vu vZ

– – pu, pu’, phu, phu’, ’bu bu bu, b’u [sic] ’bu – – – – ’dong, ’nong – – li’u su – zhu – yu yi’u ku, khu – – ’gu yu –

bu mu tu du nu ru su zu yu ku gu u

bhu, bhe vo vu tu – – – – nu – – – 5u, 5o – ju, jyo yo, yoh· – ku – – – – –

In Table 8.11, I list the LMC-based SX readings and the transcriptional data for the most common Shoki u-phonograms which comprise 77.0 per cent (569/739) of the total. Amoghavajra’s Indic transcriptions and the Tibetan transcriptions confirm that the phonograms for OJ u represented a non-low back rounded vowel (*u or *o). Most of the data point toward a high vowel (*u) rather than a mid vowel (*o). I will reconstruct OJ u as a high back rounded vowel *u. This vowel and the two previous vowels (*a and *i) all remained intact in LMJ ( [a i u] ). MSJ a and i are still [a] and [i], but MSJ u [T] has lost its rounding.

OJ o OJ o (‘B-type o’) corresponds to a mid back rounded vowel [o] in LMJ and MSJ. It corresponds to five EMC rhyme categories with central unrounded vowels (Table 8.12). All of these rhyme categories derived from LOC rhyme categories with central unrounded vowels (*R, *a):

211

OLD JAPANESE

Table 8.12 OJ o in Kojiki poetry OJ \ EMC

*-t

*-tR

*-RY

*-RY/*-Ry

*-Rw

*-Ry

m t d n r s z y k g Subtotals Total

0 0 0 0 0 0 0 0 0 11 11 853

0 1 15 0 28 0 2 29 116 0 191

0 191 1 270 0 48 0 0 0 0 510

0 4 0 0 0 0 0 0 0 0 4

135 0 0 0 0 0 0 0 0 0 135

0 0 0 2 0 0 0 0 0 0 2

(8.23) LOC *-RòB *-aòB *-RYA *-Rò A *-Rò A

> > > > >

EMC *-t *-tR *-RY *-Rw (irreg.) *-Ry (reg.)

Ph w





LOC *gRòB *haòqB *nRY A *mRò A *nRò A

EMC *gt *htRq *nRY *mRwq *nRyq

OJ go ko no mo no

The o-phonograms
‘mother’ and
‘then’ were probably selected for their LOC readings *mRòqA and *nRòqA rather than for their EMC readings with inappropriate final glides (*w and *y). Therefore
‘mother’ and
‘then’ are not evidence for *Rw and *Ry allophones of OJ o. The OSV and nôm evidence confirms the absence of rounding in the EMC rhyme categories of OJ o-phonograms (and in OJ o itself ). (1) The EMC rhymes of Kojiki o-phonograms correspond to OSV unrounded central vowels:

‘game of go’ ‘allow’ LOC *gRòB *haòqB EMC *gt *htRq OSV cI hEa [kRR] [htR] OJ – ko Ph w
‘game of go’ ‘I’ LOC *gRòB *yaòB EMC *gt *ytR

(8.24) Sg

cf.

OJ

go

yo


‘donkey’ *laòB *ltR lJa [ltR] –
‘backbone’ *laòqB *ltRq ro

212


‘plum’ *mRò A *mRy md [mRR] (NS: mey)
‘mother’ *mRòqA *mRwq (irreg.) mo


‘generation’ *dRòh A *dRyh $Ii [ßRRy] (MYS: de)
‘then’ *nRòqA *nRyq no

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

(2) Kojiki o-phonograms serve as phonograms for Vietnamese words with unrounded central vowels and generally serve as neutral vowel bases for vietographs: (8.25) OJ so to ro

Ph




EMC *tsRY ‘then’ *tRY ‘climb’ *ltR ‘backbone’

Viet ph reading tJng ‘experience’ – –

Derived vietograph – o dâng ‘offer’ p lOa ‘fire,’ rv ‘radiant’

(3) Taiwanese is not very helpful because it lacks an t. The R of its literary and colloquial layers (LT and CT) corresponds to finals with labial vowels and/or glides in other Chinese languages and SX: (8.26)
‘wormwood’

EMC LT CT *haw hR1 R1

Cnt Md hòu hao

SV hao

SK hwo

Taiwanese R probably originated from an earlier *-Q: Amoy, the closest mainland relative of Taiwanese, has an -o corresponding to LT R. The CT reflexes of the EMC rhyme categories of Kojiki o-phonograms are: (8.27) EMC CT

*-t -ai

*-tR -i, -u (usual); -o (rare)

*-RY *-Rw -in -au

*-Ry -e (usual)

Presumably CT -i, -u, -e, -o originated from finals containing earlier nonpalatal, non-labial vowels such as *t or *R. The CT -u and -o reflexes of the EMC *-tR rhyme correspond to an unrounded high back -T zin Chaozhou (Cz) Min (cf. OSV -ea [tR] ). Chaozhou seems to preserve the earlier vocalism better than CT in this instance (Pulleyblank 1984: 216). I have found CT and Chaozhou readings for five Kojiki o-phonograms (8.28). Notice that the Cz readings of four of the phonograms have the final -T: (8.28) Ph EMC CT Cz OSV OJ


‘allow’ *htRq kho3 hT 3 hEa ko


‘I’ *ytR – T2 – yo


‘give’ *ytRq tho6 T3 – yo


‘backbone’ *ltRq lu6 lT 4 lã < OC *rhaqB? ro


‘mother’ *mRwq bu 3 bo3 – mo

I will consider this Chaozhou -T to be weak evidence for an unrounded reconstruction of OJ o. Unlike Pulleyblank (1984: 216), I am uncertain whether this -T is a true retention from EMC, since it looks more like LMC-based SV -e [tt] than EMC-based OSV -ea [tR]. 213

OLD JAPANESE

(4) Jñanagupta almost never used OJ o-phonograms or their nearhomophones in his transcriptions. This tells us that their EMC rhymes (and by extension, OJ o) must not have been very much like anything in Indic languages. The only Kojiki o-phonogram that appears in Jñanagupta’s Indic transcriptions is
EMC *tRY ‘climb’ for Indic ta@g [t£Yg]. This may imply that the OJ to represented by
EMC *tRY ‘climb’ was something like Indic ta [t£]. Jñanagupta also used
EMC *sRY ‘monk’ (borrowed from Sanskrit saúgha [s£Yg®£] ‘monastic order’), to transcribe Indic saú(gh) [s£Y(g®) ] and saúk [s£Yk]. Since
EMC *sRY ‘monk’ appears as a phonogram for OJ so in Man’yôshû (817.2.5) and is a near-homophone of the sole Kojiki phonogram for OJ so,
EMC *tsRY ‘then,’ OJ so may have been similar to Indic sa [s£]. This limited evidence indicates that OJ o was not a rounded vowel like LMJ or MSJ [o]. On the basis of the general absence of rounded vowels in the OSV, nôm, Chaozhou, and Indic transcriptional data, I will reconstruct OJ o in Kojiki as a mid central unrounded vowel *R. Although I could reconstruct OJ o as a high central *t or a high back *T on the basis of the high vowels in OSV, nôm, and Chaozhou, I prefer a mid central *R because (1) it differs only in frontness and rounding from the mid back [o] of LMJ and MSJ and (2) it is closer to Indic a [£]. In Shoki, OJ o corresponds to four EMC rhyme categories which all contain *R (Table 8.13). SK and SV have unrounded vowels in their equivalents of three of the four EMC rhyme categories of Shoki o-phonograms (Table 8.14). The SK and SV vowels ( [t R £ a] ) generally suggest that OJ o was an central or back unrounded vowel. Obviously, OJ o cannot be *a because I have already reconstructed OJ a as *a. OJ o also could not be *ay because the majority of o-phonograms do not have SK or SV readings ending in [y].

Table 8.13 OJ o in Shoki poetry OJ \ EMC

*-tR

*-Rng

*-Rng/*-Ry

*-Ry

*-wRn

t d n r s z y k g Subtotals Total

0 0 0 30 3 5 49 88 7 182 623 (+30)

46 4 252 1 35 1 0 0 0 339

71 1 0 0 0 0 0 0 0 72

5 (+30) 13 11 0 0 0 0 0 0 29 (+30)

0 0 0 0 0 1 0 0 0 1

214

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.14 SX equivalents of rhyme categories for Shoki o-phonograms EMC Kan SK SV

*-tR -yo -(y)e [(y)R] -u’ [tt]

*-Rng -ou -ung [tY] -#ng [aY], -u’ng [tY]

*-Ry -ai -oy [£y], -ay [ay] -ai [aay]

*-wRn -on -won [on] -ôn [on]

But why was OJ o written with EMC *-Ry > CLMC *(-ay) in Shoki? All three EMC *-Ry > CLMC *(-ay) phonograms for Shoki o-syllables are confined to the beta section: (8.29) OJ to, do do do, no

Ph
‘moss’
‘bear’
‘then’

LOC *dRòA *nRòhA *nRòqA

EMC *dRy *dRy *dRy

Go dai dai dai

CLMC *t®(ay) *nd(ay) *nd(ay)

SK thoy nay nay

SV $ài nfi nãi

Some texts have 30 instances of
‘thatch’ in place of the phonogram
‘moss’ for OJ to. In Table 8.13 I have optionally added all thirty instances of
‘thatch’ to the figures for
‘moss.’ Although I do not completely accept Mori’s hypothesis of a dichotomy between a ‘Chinese’ alpha section and a ‘native’ beta section, I do believe that the beta section has more archaic usages than the alpha section. Beta section scribes may have chosen the phonograms in (8.29) for their LOC readings ending in *-Rò A. Even if they were unaware of those LOC readings, scribes may have chosen CLMC *(-ay) phonograms because CLMC apparently had no syllables such as *tR or *ndR. SK has no te, the, ne [tR thR nR] and SV has no $d, thd, nd [ßRR thRR nRR] corresponding to the theoretical LMC syllables *tR, *thR, *nR. Closed syllable phonograms are orthographic compromises rather than evidence for lost final consonants. I will reconstruct OJ to, do, no as *tR, *ndR, *nR. In Table 8.15, I list the LMC-based SX readings and the transcriptional data for the most common Shoki o-phonograms which comprise 89.1 per cent (555/623) of the total. Only one SK reading (cwo for
EMC *jrtR > LMC *c®r- ‘hoe’) in Table 8.15 has a rounded vowel. All other SX readings, including the ‘vulgar’ reading of
‘hoe,’ SK se [sR], have unrounded vowels: SK u, e, o, a [t R £ a] and SV e, ea, #, a [t tR a aa]. The Tibetan transcriptions have a wide range of non-low vowels: i, e, t, u. The variation in transcriptions probably reflect not only northwestern LMC dialect variation but also the lack of exact Tibetan equivalents for the LMC vowels of Shoki o-phonograms. LMC and OJ may have had an *R, but Tibetan did not.

215

OLD JAPANESE

Table 8.15 Common o-phonograms in Shoki poetry OJ

Ph

EMC

CLMC

Kan SK

to to, do





*tRY *tRYq

*t*t-

tou tou




do do, no
no


*dRy *dRY *nRyh *nRyq *nRY

*t®(ay) *t®*nd(ayh) *nd(ayq) *nd-

tai tou dai dai dou












*ltRq *ltRh *tsRY *dzRYh *ãtRh *jrtR *ytRh *ytR *htR

*l*l*ts*ts®*ã9*c®r*y*y*h-

ryo ryo sou sou jo syo yo yo kyo







*htRq *ktRq *YtRq *YtRh

*h*k*Y*Y-

kyo kyo go go

ro so zo yo k

g

SV

Tib. tr.

tung tung

$#ng $=ng ting, teng, ding, ’ding thoy $ài – tung $9ng – nay nfi – nay nãi – nung n#ng ’ding, ’dtng, ning, ’ning, ’neng lye lw – lye l@ lu cung t#ng tsing, tshing cung t'ng – ye nh@ – cwo, se sJ – ye d@ yu ye de – he he hi, hyi, he’i, hye, hu, ’hyu he hEa he ke cO ku’u, gu e ngw ’gi, ’gt, ’ge e ng@ ’gu

Ind. tr. (Am) – – – – – – ù, ê – – – – – – – – – – – –

Amoghavajra almost never used OJ o-phonograms or their nearhomophones in his Indic transcriptions. Such characters probably had CLMC vowels absent in Indic (i.e., *t or *R). The evidence from both the pre-LMC and LMC periods supports my reconstruction of OJ o as a mid central unrounded vowel *R. This vowel later backed and rounded to become the [o] found in LMJ and MSJ. If one diagrams the four primary ‘vowels’ of OJ, *a, *i, and *u form the corners of a triangle and *R is at the center (Table 8.16). Table 8.16 is practically identical to Whitman’s (1985) scheme of ProtoJapanese vowels in Table 4.8. These four vowels supposedly combined to form the four secondary ‘vowels’ of OJ (iy, ye, ey, wo; p. 80). I will test this monophthongization hypothesis in the next four sections (pp. 217–38). Table 8.16 The ‘primary’ vowels of OJ Front High Mid Low

Central

*i

Back *u

*R *a

216

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.17 OJ iy in Kojiki poetry OJ \ EMC

*-i

*-t

*-uy

p b m k g Subtotals Total

0 4 0 0 0 4 61

0 0 0 24 4 28

13 0 15 1 0 29

OJ iy OJ iy (‘B-type i’) is the least frequent ‘vowel’ in Kojiki. It corresponds to three EMC rhyme categories (Table 8.17). The only Kojiki iy-phonogram of the *-i rhyme category is
EMC *bih ‘prepare’ for OJ biy. The vast majority of iy-phonograms are evenly split between the *-t and *-uy rhyme categories. The OSV and nôm evidence indicates that B-type iy was less palatal than A-type yi, as follows. (1) The OSV equivalents of the EMC rhyme categories corresponding to OJ iy (*-i, *-t, *-uy) generally lack palatal vowels: (8.30) Sg

cf.

LOC EMC OSV OJ Ph LOC EMC OJ


‘hunger’ *kiB *ki cd [kRR] –
‘prepare’ *bihB *bih biy




‘rise’ ‘suspect’ ‘taste’ *khRòB *YRòB *mwtyhB *khtq *Yt *muyh khNi [xRRy] ngI [YRR] mùi [muy] – giy miy



‘age’ ‘valuable’ ‘ornate’ ‘fat’ *kRòqB *kwtyhB *phwtqB *bwtyB *ktq *kuyh *phuyq *buy kiy kiy piy piy


‘small’ *mwtyB *muy miy

(2) OJ iy-phonograms serve as phonograms for Vietnamese words with non-palatal vowels (d [RR], u [u] ) and as neutral vowel bases for vietographs: (8.31) OJ giy

Ph LOC
*YRò B ‘suspect’

(MYS: miy)
Cf. miy


Viet ph reading ngI ‘suspect’ (OSV) *mwtyqB ‘tail’ – *mwtyhB ‘taste’ mùi ‘taste’ (OSV) 217

Derived vietograph q ngI ‘suspect’ (OSV) r vv ‘break’ –

OLD JAPANESE

Since I have already reconstructed OJ o as *R, I cannot reconstruct OJ iy as *R on the basis of OSV d [RR]. Neither can I reconstruct OJ iy as *uy (cf. Whitman 1985) or *wi (cf. Unger 1993) on the basis of OSV ui [uy]. I will not reconstruct OJ iy with labial elements such as *u or *w for the following reasons. (a) 92 per cent (23/25) of OJ kiy in Kojiki are written with
EMC *ktq ‘age.’ OJ kiy is written only once with
EMC *kuyh ‘valuable.’ If OJ kiy were *kuy or *kwi, we would expect
EMC *kuyh ‘valuable’ to be the most frequent phonogram for OJ kiy, since
EMC *ktq ‘age’ (cf.
EMC *khtq, OSV khNi [xRRy] ‘rise’) would be a very poor choice for an OJ *kuy or *kwi. (b) We would also expect the Go-on reading of
EMC *kuyh ‘valuable’ to be ki without a -w- since OJ kiy (*kuy / *kwi) became post-OJ ki. But EMC *-uy regularly corresponds to Go-on -wi after velars. The actual Go-on reading of
EMC *kuyh ‘valuable’ is kwi, not ki. EMC *kuy was probably closer to OJ kwi than to OJ kiy. (c) OJ biy and giy are never written with EMC *-uy phonograms, whereas OJ piy and miy are always written with such phonograms (
EMC *phuyq ‘ornate,’
EMC *buy ‘fat,’
EMC *muy ‘small,’
EMC *muyh ‘taste’). It would be strange to posit an *-uy or *-wi allophone of OJ iy only before *p and *m but not before their fellow labial *b. Unlike LOC, EMC did not permit the final *-ty after labials. Neither LOC nor EMC permitted the final *-t after labials. Scribes had to use EMC *-i and EMC *-uy phonograms for OJ piy, biy, miy (*pt, *mbt, *mt?) because LOC and EMC had no syllables such as *pt, *bt, or *mt. There are no OSV forms such as be [ptt], phe [ftt], or me [mtt]. Starostin reconstructs the LOC source of EMC *-uy as *-wty. Since Vietnamese phonotactics forbid finals combining [w] with [t] such as *-uu’i [wty], I cannot test Starostin’s reconstruction using OSV evidence. If his reconstruction is correct, then perhaps phonograms which end in *-uy in EMC (e.g.,
EMC *kuyh ‘valuable’) were chosen to write OJ iy-syllables on the basis of their earlier LOC *-wty readings. I also cannot use CT to test Starostin’s reconstruction of LOC *-wty. As I have already mentioned, CT lacks a t and its R is probably from an earlier *-Q. The only Kojiki iy-phonogram with a CT reading is: (8.32)

EMC CT Go
‘fat’ *buy pui 2 bi

OJ piy

Compare
CT pui 2 [pwi] ‘fat’ for OJ piy in (8.32) to
OSV mùi ‘taste’ for OJ miy in (8.30) and (8.31). OJ piy (*pt) and miy (*mt) were definitely not simply *pi and *mi (cf. Lange 1973), though I doubt that they were *puy and *muy or *pwi and *mwi. Jñanagupta did not use the Kojiki iy-phonograms or their nearhomophones in his Indic transcriptions. There was no need to use characters 218

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

of the EMC *-t and *-uy rhymes, since Indic languages lacked the vowel t and diphthongs such as ui. This negative evidence underscores the relatively non-palatal nature of OJ iy. If OJ iy were a palatal vowel like *i, Jñanagupta would have used the phonograms for OJ iy-syllables or their near-homophones to transcribe Indic palatal vowels such as i and c. I will reconstruct OJ iy in Kojiki as a non-palatal high central vowel *t. In Shoki, OJ iy corresponds to five EMC rhyme categories (Table 8.18). All of the Shoki iy-phonograms belong to Grade III, whereas the majority of Shoki yi-phonograms belong to Grade IV. (The same is true of the Kojiki iy- and yi-phonograms, though it may be anachronistic to apply the Grades to EMC instead of LMC.) By comparing the SK and SV readings of all of the Shoki iy-phonograms in Table 8.19 with the SK and SV (sans tones) readings of the most common Shoki yi-phonograms in Table 8.7, one can see in (8.33) that Grade III (correlated with OJ iy) is less palatal than Grade IV (correlated with OJ yi). Table 8.18 OJ iy in Shoki poetry OJ \ EMC

*-i

*-iR

*-t

*-ty

*-uy

p b m k g Subtotals Total

0 3 0 0 0 3 51

4 0 0 0 0 4

0 0 0 19 13 32

0 0 0 4 0 4

0 0 8 0 0 8

Table 8.19 Iy-phonograms in Shoki poetry OJ

Ph

EMC

CLMC

Kan

SK

SV

Tib. tr.

Ind. tr. (Am)

piy














*piRq *biRh *bih *mih *muyh *muy *ktq *kt *khtyh *ktq *kty(q)

*p(iq) *p®(ih) *p®(ih) *mb(ih) *mbv*mbv*k*k*kh*k*k-

hi hi hi bi bi bi ki ki ki ki ki

phi phi pi mi mi mi kuy kuy kuy kuy kuy

be byi – – – ’byi – – ki, khe – ki, gi

– – – – – vi, ve – – – – –





*kty *Yt

*k*Yg-

ki gi

kuy uy

bm bh bh mh vh vi km, k] cd, ki khí km, k] cd, ki, km, k] cd nghi

kt ’gi’, ’gu

– -g- [sic]

biy miy kiy

giy

219

OLD JAPANESE

(8.33) OJ OJ Type pyi, byi, myi A piy, biy, miy B kyi, gyi A kiy, giy B

LMC Grade IV III IV III

SK pi, mi, moy [pi mi m£y] pi, phi, mi ki, yey kuy, uy [kty ty]

SV ti, mi, mê, di, ri [sic] bi, mi, vi ki, ky, nghê cd, ki, ky, khi, nghi

Clearly, OJ iy was less palatal than OJ yi (*i). Amoghavajra’s transcriptions also confirm this. In Table 8.19, we can see that he almost never used any OJ iy-phonograms to transcribe Indic sounds. He had little reason to do so, since Indic languages lacked the non-palatal vowel *t. Thus OJ iy was unlike anything in Indic. The ninth- and tenth-century Tibetan transcriptions of northwestern LMC do not consistently distinguish Grades III and IV. Perhaps those two Grades had merged or were in the process of merger in late northwestern LMC. In any case, the Tibetan evidence cannot help us to determine the phonetic value of OJ iy. On the basis of the evidence above, I will reconstruct OJ B-type iy as a high central unrounded vowel *t for both Kojiki and Shoki. It is also possible that OJ iy was something like SK uy [ty]. If OJ iy did originate from pre-OJ *ui and *Ri (p. 80), *u and *R would have assimilated to the following *i, which may have become a glide *y: (8.34) *u *R

> >

*t *t

/ /

_i (y) _i (y)

(centralization, loss of rounding) (raising)

*ui and *Ri would then merge into *ty. In the OJ period, this *ty may have been reduced to *t. After the OJ period, A-type yi (*i) and B-type iy (*t or *ty) merged into [i]: (8.35) Pre-OJ *ui > *uy > iy = *ty *Ri > *Ry > iy = *ty *i > *i > yi = *i

OJ Post-OJ > *t > [i] > *t > [i] > *i > [i]

Perhaps an OJ *ty became post-OJ [i] without any intermediate *t stage: (8.36) *ty

> *iy

> [i]

Although I favor *ty over *t, I will write my reconstruction of OJ iy as *t for simplicity.

220

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.20 OJ ye in Kojiki poetry OJ \ EMC

*-æ

*-æy

*-iRn

*-yi

*-yiRy

p b m k g Subtotals Total

0 0 0 0 1 1 65 (+29)

0 0 24 0 0 24

0 1 (+1) 0 0 0 1 (+1)

0 0 0 36 0 36

3 (+28) 0 0 0 0 3 (+28)

OJ ye OJ ye (‘A-type ye’) is the third least frequent ‘vowel’ in Kojiki. It corresponds to five different EMC rhyme categories (Table 8.20). Each of these categories is associated with one and only one OJ syllable. This might imply five different allophones, but I would like to reconstruct as few allophones as possible. The most frequent category is *-yi, which was associated with OJ yi (p. 204). Having already reconstructed OJ yi as *i, I obviously cannot reconstruct OJ ye as *i. Nevertheless the fact that
EMC *gyi ‘grand’ was chosen to be the sole phonogram for OJ kye in Kojiki tells us that the pronunciation of OJ kye must have resembled the pronunciation of OJ kyi (*ki). Since OJ ye corresponds to [e] in LMJ, I will tentatively hypothesize that OJ ye was an upper mid front unrounded vowel like *e. The use of EMC *-yiRy characters as phonograms for OJ pye supports this hypothesis. OJ pye is written three times in all Kojiki texts with the phonogram
EMC *byiRyh ‘money’ (4.19.1, 22.7.2, 52.5.3). It is written twenty-eight times with
EMC *byiRyh ‘money’ in some texts and as
EMC *byiRyh ‘ruin’ in others. Owing to the homophony and graphic similarity of these two characters, it is difficult to determine which characters were original and which characters were scribal errors. Nevertheless, it is certain that all of these instances of OJ pye were being spelled with characters pronounced as *byiRyh in EMC. If I add the twenty-eight ambiguous instances of
/
EMC *byiRyh ‘money’ / ‘ruin’ to the three definite instances of
EMC *byiRyh ‘money’ for OJ pye, *-yiRy is the second most common EMC rhyme category after *-yi among the ye-phonograms of Kojiki. The Kojiki and Shoki phonogram for OJ gyi,
EMC *YyiRy ‘art,’ belongs to this category. The OSV word corresponding to
EMC *YyiRy ‘art’ is nghV [Yee] ‘craft’ with an upper mid front unrounded vowel ê [ee] not unlike the *e that I hypothesized for OJ ye. The OSV ‘readings’ for
EMC *byiRyh ‘money’ and
EMC *byiRyh ‘ruin’ have not survived, but they probably would have ended in -ê [ee] like
OSV nghV ‘craft.’ The ‘vowels’ of OJ gyi (*Ygi) and OJ pye (*pe?) must have been close enough to justify using phonograms of the same EMC rhyme category for both syllables. 221

OLD JAPANESE

The third most frequent rhyme category is represented by the sole phonogram for OJ mye in Kojiki,
EMC *mæyh ‘sell.’ Ôno Tôru (1962: 186) considers
to be the most commonly used phonogram for OJ mye in the Suiko and Nara Periods. The OSV and CT evidence for the EMC *-æy rhyme category after labials points toward an *-e-like pronunciation: (8.37) Sg LOC EMC CT OSV OJ


‘picture’ *wier hA *®wæyh oe6, ui 6 vv [vεε] –


‘sell’ *mier hA *mæyh b(o)e6 – mye


EMC *mæyh ‘sell’ is not a nôm phonogram or a vietograph base, but its phonetic element
EMC *mæyq ‘buy’ is a phonogram for Viet mDi ‘new.’
EMC *mæyq ‘buy’ is also a neutral vowel base: (8.38) s mái [maay] ‘roof ’

t, u máy [may] ‘machine’

This usage is not necessarily evidence for the pre-LMC pronunciation of
EMC *mæyq ‘buy’ or its near-homophone
EMC *mæyh ‘sell.’
EMC *mæyq ‘buy’ may have been chosen as a phonogram for Viet mDi ‘new’ and as a neutral vowel base because of its LMC-based SV reading mãi [maay] (cf. Kan-on bai and SK may). Jñanagupta did not use
EMC *mæyh ‘sell’ or any of its nearhomophones as phonograms for Indic syllables. He transcribed Indic mi and me with
EMC *myiR ‘increasingly’ (a phonogram for OJ myi; see 8.19), not
EMC *mæyh ‘sell.’ Apparently Indic -i and -e [ee] were not much like EMC *-æy. Sanskrit has no vowel such as [æ]. The above data show that
EMC *mæyh ‘sell’ may not have been an ideal phonogram for OJ mye (*me?). OSV -e [εε] is lower than the *e that I had originally hypothesized for OJ ye. CT (o)e looks like a relatively good match for my *e, but this (o)e may be the product of post-EMC sound changes. Nevertheless, I would rather not reconstruct a unique *æ or *ε allophone of OJ ye before OJ m. I will tenatively reconstruct OJ mye as *me. The phonograms for OJ bye and gye are extremely rare in Kojiki but do offer some support for the reconstruction of OJ ye as *e.
EMC *biRnh ‘divide’ for OJ bye (KJK 33.2.4, 91.1.4a), belongs to the *-iRn rhyme category which corresponds to OSV -iên [iRn] (cf. OSV liVn ‘connect’ <
EMC *liRn ‘id.’) and to CT -iN [u ] (cf.
EMC *piRnh / CT piN 5 ‘change’). These OSV and CT finals do not match my reconstruction of a mid front vowel *e for OJ ye, but they do contain a high front vowel [i]. 222

THE RECONSTRUCTION OF OLD JAPANESE VOWELS


EMC *Yæ ‘tusk, ivory’ for OJ gye (KJK 21.5.6) belongs to the *-æ rhyme category which corresponds to OSV -e [εε] and CT -e (cf. OSV hè and CT he6 ‘summer’ <
EMC *®æq ‘id.’). The CT reading of
EMC *Yæ ‘tusk, ivory’ is ge2, which closely matches my reconstruction of OJ gye as *Yge. I will reconstruct OJ ye in Kojiki as a mid front unrounded vowel *e. I will not specify its precise height since it is written with phonograms whose EMC rhyme categories correspond to high, upper mid, and lower mid OSV vowels (i [ii], -iê- [iR], ê [ee], e [εε] ). Perhaps OJ ye had a wide range of allophonic variation. OJ ye in Shoki corresponds to seven EMC rhyme categories (Table 8.21). All but two of these categories (*-æ and *-æy) correspond to SK and SV rhymes containing [i], [e], or [yR] (Table 8.22). *-ey is the most frequent category for all four ye-syllables in Shoki. The EMC *-ey rhyme category would seem to be the most obvious choice for the transcription of OJ ye, which I have just reconstructed as *e. In fact, Jñanagupta used EMC *-ey phonograms to transcribe many Indic syllables ending in -e (Pulleyblank 1984: 198), confirming that the *-ey rhyme category Table 8.21 OJ ye in Shoki poetry OJ \ EMC

*-æ

*-æy

*-ey a

*-yi

*-yiR

*-yiRn

*-yiRyb

p b m k Subtotals Total

3 0 0 1 4 62 (+23)

0 0 3 0 3

15 2 10 12 (+13) 39 (+13)

0 0 0 2 2

4 0 2 0 6

0 0 3 0 3

5 (+10) 0 0 0 5 (+10)

Notes a The most frequent phonogram for OJ kye in Shoki appears as
EMC *key / CLMC *k- ‘chicken’ in some texts and as y EMC *key / CLMC *k- ‘chicken’ in others. These characters are simply graphic variants. I have optionally added all thirteen cases of variation between the two characters to my figures for EMC *-ey. b The second most frequent phonogram for OJ pye in Shoki appears in some texts as
EMC *byiRyh / CLMC *p®- ‘money’ and in others as
EMC *byiRyh / CLMC *p®- ‘ruin.’ I have optionally added all ten cases of variation between the two characters to my figures for EMC *-yiRy.

Table 8.22 SX equivalents of rhyme categories for Shoki ye-phonograms EMC

*-æ

*-æy

*-ey

*-yi

Kan

-a

SK

-a

SV

-a [aa]

-a(i) [aa(y) ]

-a(i)

-ei

-i

-i

-en

-ei

-a(y)

-yey [yRy]

-i

-i

-yen [yRn]

-yey [yRy]

-ê [ee]

-i [ii]

-i [ii]

-iên [iRn]

-ê [ee]

223

*-yiR

*-yiRn

*-yiRy

OLD JAPANESE

was [e]-like in northwestern EMC. Copious examples of EMC *-ey representing Indic e in Jñanagupta’s transcriptions can be found in Coblin (1994: 194–203). Yet not one ye-syllable in Kojiki is ever spelled with an EMC *-ey phonogram. In fact, as we will see in the next section, many OJ ey-syllables in Kojiki are spelled with EMC *-ey. Was OJ ye in Kojiki really an *e? A better question to ask might be: Was EMC *-ey really an *-ey? It obviously was [e]-like in Jñanagupta’s northwestern EMC dialect, but it may have been an *-ay or *-Ry in the non-northwestern EMC dialect that was the basis of Go-on and the Kojiki orthography. Southern Chinese languages, Chinese loans in Vietnamese and Tai, and the early sixth-century transcriptions of Saúghabhara (or Saúghavarman) of the Liang (502–557) provide evidence for an *ay- or *Ry-like pronunciation of the *-ey rhyme category in southern EMC. (1) EMC *-ey regularly corresponds to Cantonese -ai and sometimes corresponds to -ai in CT and Meixian Hakka:
‘chicken’ OC *keA EMC *key(h) CT koe1 [kwe] Hakka kai1 Cnt gài [kay] Go ke OJ (KJK ) – LT ke1 Kan kei OJ (NS) kye

(8.39) Sg


‘steps’ *biyqA *beyq – – baih [pay] bai – pe 2, pe6 hei pye


‘rice’ *mhiyqA *meyq – mi 3 máih [may] mai mey bi 3 bei –


‘enigma’ *miy(h)A *mey(h) – mi 2 màih [may] mai – be 2 bei bye, mye


‘west’ *sRyA *sey sai 1 si 1 sài [say] sai – se1 sei se

(2) One possible OSV word and a few Tai words reflect an *Ry-like pronunciation of the EMC *-ey rhyme category.
EMC *sey ‘west’ has two readings in SV, tây [tRy] < *sRy and a rarer reading tê [tee] < *see. Since EMC *-ey regularly corresponds only to ê [ee], tây may be an OSV borrowing from a southern EMC form such as *sRy (cf. CT sai 1 and Cnt sài). A southern form of
EMC *key ‘chicken’ such as *kRy (cf. Hakka kai 1, Cnt gài) may be the source of Proto-Tai *kRiB (Li 1977: 31, 186, 285) despite the irregular tone correspondence. Manomaivibool (1976: 13, 23, 24) has found three more possible Tai borrowings of Chinese words belonging to the EMC *-ey rhyme category: (8.40) Sg OC
‘ladder’ *c(h)lRyA
‘bottom’ *tRyqA
‘plow’ *riA

EMC *they *teyq *ley 224

Cnt tài [thay] dái [tay] làih [lay]

Proto-Tai (Li 1977) *}dl/rRiA ‘id.’ *tRï C ‘under’ *thlRiA ‘id.’

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Note that the superscript ‘A’ in OC reconstructions refers to syllable types (A and B, corresponding to Starostin 1989’s long and short vowels) whereas the superscript ‘A’ and ‘C’ in Proto-Tai (PT) reconstructions refers to tones. Since all three PT forms in (8.40) have problematic segmental correspondences, I consider PT *kRiB ‘chicken’ to be the best evidence for a final *-Ry in southern Chinese. (3) Unlike Jñanagupta in the northwest, Saúghabhara in the south used characters of the EMC *-ey rhyme category to transcribe Sanskrit ai instead of Sanskrit e (Pulleyblank 1984: 199). This implies that *-ey was phonetically *[ay] or *[Ry] in southern EMC. Pulleyblank reconstructed the phonetic value of his EMC final *-εy (our *-ey) as *[æy]. He believed that Go-on ai for EMC *-ey “followed southern usage” (1984: 199). In (8.39), Go-on ai corresponds to Cantonese, Hakka, and CT -ai. (But note that
EMC *key ‘chicken’ is Go-on ke, not kai. This Go-on -e may be from a northern EMC *-ey or from a southern EMC *-Ry. See p. 224.) Characters pronounced with *-ay or *-Ry in southern EMC would have been poor choices for the transcription of OJ ye-syllables such as pye (*pe), bye (*mbe), etc. Thus scribes using EMC-based orthography to write OJ would favor characters from other rhyme categories for that purpose. However, the Shoki scribes were using an orthography based on the Chang’an dialect of LMC (CLMC), not southern EMC. The general LMC reflex of the EMC *-ey rhyme category had a more [e]-like pronunciation: cf. LMC-based LT -e, SK -yey [yRy] (premodern SK had no [e]), and SV -ê [ee]. Hence characters such as
southern EMC *bayq ‘steps’ (cf. Cnt baih, Go-on bai) which were once inappropriate for the transcription of OJ ye-syllables were now quite appropriate if one chose them for their CLMC pronunciations. In fact,
CLMC *p®- ‘steps’ (cf. LT pe5 and pe6, SK phyey, SV bq) is the most common Shoki phonogram for OJ pye (*pe). The different spellings of OJ ye-syllables in Kojiki and Shoki ultimately result from differences between EMC and CLMC and are not evidence for different pronunciations of OJ ye in the two texts. Not all of the Shoki phonograms for OJ ye-syllables seem to be suitable from an LMC point of view. EMC *-æ and *-æy correspond to finals without front vowels in LMC-based CT, SK, and SV. Three Shoki ye-phonograms belong to these two rhyme categories: (8.41) OJ kye pye mye

Ph
‘family’
‘hegemon’
‘sell’

EMC *kæ *pæh *mæyh

CLMC *k*p*mb-

LT ka1 pa 5 mai 6

Kan ka ha bai

SK ka pha may

SV ca [kaa] bá [(aa] mfi [maay]

These characters are found only in the beta section and are very rare even there:
‘hegemon’ for OJ pye and
‘sell’ for OJ mye appear only three times each and
‘family’ for OJ kye appears only once. 225

OLD JAPANESE

Table 8.23 Ye-phonograms in Shoki poetry OJ

Ph

EMC

CLMC

Kan

SK

SV

Tib. tr.

Ind. tr. (Am)

pye












úy


x



*beyh *byiRy *byiRy *pyiRq *pæh *mey *mey *myiRn *mæyh *myiR *key *key *kheyq *gyi *kheyq *keyh *kæ

*p®*p®*p®*p*p*mb*mb*m(b)*mb*mb*k*k*kh*k®*kh*k*k-

hei hei hei hi ha bei bei ben bai bi kei kei kei ki kei kei ka

phyey phyey phyey pi pha mi mi myen may – kyey kyey kyey kuy kyey kyey ka

bq tq tq tz bá mê mê miên mfi – kê kê khái kì, kz khki kU ca

– – – pyi – – – myen ’be’i – kye – khye – khye kye ka

– – – – – me, (m)be me, (m)be – – – ke – – – – ke –

bye mye

kye

I believe that these characters were used for their EMC sound values. I have already shown that EMC *-æ and *-æy correspond to OSV -e [εε] and CT -e.
EMC *mæyh ‘sell’ for OJ mye appears to be a carryover from System-D-based transcription.
EMC *kæ ‘family’ (cf. CT ke1) for OJ kye and
EMC *pæh ‘hegemon’ for OJ pye are not carryovers, but they do belong to the same rhyme category as
EMC *Yæ ‘tusk, ivory’ (cf. CT ge 2), the Kojiki phonogram for OJ gye (*Yge). Although the beta section is not primarily written in an EMC-based orthography, it does contain conservative elements lacking in the alpha section. The Tibetan and Indic transcriptive data (Table 8.23), though sparse, confirms that at least some of the Shoki ye-phonograms were read with mid or high front vowels in northwestern LMC. I will reconstruct OJ ye as a mid front unrounded vowel *e for both Kojiki and Shoki. Whitman (1985) and Unger (1993) reconstructed OJ ye as a glidevowel combination *ye. Lange (1973) reconstructed OJ ye as *e before velars (k, g) but as *ye before labials ( p, b, m). None of the data shown here supports Lange’s reconstruction of two different sound values of OJ ye. As for Unger’s view, it may have some merit because most of the phonograms in Table 8.23 belong to Grade IV, the most palatal of the Four Grades (p. 97) and are read with a medial -y- in SK. Three of the phonograms for OJ pye are read with the dental initial t in SV derived from an earlier *py. We will see in the next section that none of the phonograms for OJ pey, bey, or mey is ever read with a dental initial (t, th, d ) in SV. If OJ ye originated from pre-OJ *ia and *iR (p. 81), it must have had a glide *y at some point in its history: 226

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

(8.42) Pre-OJ *ia > *ya *iR > *yR

OJ? > *ye > *ye

> >

OJ? *e *e

Post-OJ > [e] > [e]

Whether OJ ye had this glide during the OJ period or not is open to question. There is also a third possibility. OJ ye could have been an *e preceded by palatalized initials: *pye, *mbye, etc. Although I prefer this last solution, I find it hard to rule out the other two. I will continue to write my reconstruction of OJ ye simply as *e without specifying a preceding glide or the palatalization of a preceding consonant.

OJ ey OJ ey (‘B-type ey’) in Kojiki corresponds to four EMC rhyme categories (Table 8.24). The most common of these categories is *-ey (55.3 per cent = 47/85), represented by two phonograms,
EMC *peyh ‘close’ for OJ pey and
EMC *meyq ‘rice.’ In the previous section, we have seen that this rhyme category was probably an *-ay or *-Ry in southern EMC.
EMC *peyh ‘close’ and
EMC *meyq ‘rice’ are pai [pay] and máih [may] in Cantonese. The second most common category is *-ty, represented by the sole Kojiki phonogram for OJ key,
=EMC *khtyh ‘breath’. The OSV form of this word, hdi [hRRy] ‘breath’ leads me to suspect that OJ key was something like *kRy. EMC *-Ry and *-iR, represented by one phonogram each in Kojiki, tie for the least common category. OSV evidence (8.43) confirms Pulleyblank’s reconstruction of EMC *-Ry and enables me to reconstruct OJ bey as *mbRy. (8.43) Sg

cf.


Gloss ‘wait’ ‘generation’ ‘times’

EMC *dRyq *dRyh *bRyh

OSV $%i [ßRRy] $Ii [ßRRy] –

OJ – te (in Suiko texts) bey (Kojiki)

Table 8.24 OJ ey in Kojiki poetry OJ \ EMC

*-ey

*-iR

*-ty

*-Ry

p b m k g Subtotals Total

18 0 29 0 0 47 85

0 0 0 0 6 6

0 0 0 26 0 26

0 6 0 0 0 6

227

OLD JAPANESE

EMC *-iR looks like an unusual choice for OJ ey until one realizes that the sole Kojiki phonogram for OJ gey,
EMC *YiR ‘proper,’ had the OC reading *YayB. Although this word has no known OSV equivalent, its nearhomophone
EMC *YiRh < OC *Y(r)ayhB ‘righteous’ does appear in OSV as ngãi [Yaay]. This would seem to indicate that OJ ey had an *ay allophone after g (*Yg), but there is no reason why ey would have different allophones after k (*k) and g (*Yg). Most of the above evidence points toward *Ry for OJ ey. The OJ ‘vowels’ iy (*t) and ey (*Ry?) must have been similar because EMC *-ty characters transcribed both OJ iy and ey-syllables in the text of Kojiki. The *-ty character
EMC *kty ‘few’ / *ktyq ‘how many’ is the second phonogram in one of the Kojiki spellings of the toponym Sikiy:
 EMC *cth kty (II: Keikô, III: Yûryaku). Yet
EMC *khtyh ‘breath,’ a near-homophone of this character, is the second phonogram of a gloss for the OJ name element wakey:
 EMC *®wa khtyh (I: Age of the Gods). None of the five Kojiki phonograms for OJ ey-syllables appears in Jñanagupta’s transcriptions, though a near-homophone of the Kojiki phonogram for OJ mey (
EMC *meyq ‘rice’; cf. Cnt máih),
EMC *mey ‘confused’ (cf. Cnt màih), does appear as a phonogram for Indic me. However, this does not mean that OJ mey was *me since
EMC *mey ‘confused’ was probably pronounced as *may in southern EMC (cf. Cnt màih, Go-on mai). The Shoki data generally confirm my reconstruction of OJ ey as *Ry. This ‘vowel’ corresponds to seven EMC rhyme categories (Table 8.25). To err on the side of overdifferentiation, I have followed Mineya (1972) in distinguishing between the *-Ry and *-wRy rhymes after labials, though I know of no evidence outside the rhyme dictionary tradition for such a distinction. Pulleyblank (1991a) reconstructs *-Ry only after labials. All 52 cases of *-wRy in Shoki involve labial-initial phonograms. If I follow Pulleyblank and rewrite *-wRy after labials as *-Ry, then the *-Ry phonograms comprise 89.1 per cent (98/110) of the ey-phonograms in Shoki. Judging from SX evidence, EMC *-Ry lowered to *-ay in LMC (Table 8.26). (SK -oy [£y] appears to be an EMC retention in the source dialect of SK.) Without any characters

Table 8.25 OJ ey in Shoki poetry OJ \ EMC

*-ay

*-ey

*-εy

*-i

*-ty

*-Ry

*-wRy

p b m k g Subtotals Total

1 0 0 0 0 1 110

3 0 1 0 0 4

0 0 0 2 0 2

0 0 0 1 0 1

0 0 0 4 0 4

9 3 0 19 9 40

15 6 37 0 0 58

228

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.26 SX equivalents of rhyme categories for Shoki ey-phonograms EMC Kan SK

*-ay -ai -ay

SV

-ai [aay]

*-ey -ei -yey [yRy] -ê [ee]

*-εy -ai -oy, -yey [£y yRy] -ai, -o’i [aay RRy]

*-i -i -uy [ty] -i [ii]

*-ty -i -uy [ty] -i [ii]

*-Ry -ai -ay, -oy [ay £y] -ai, -ôi (/ b_) [aay oy]

*-wRy -ai -ay, -oy [ay £y] -ôi [oy]

read with *-Ry in CLMC, Shoki scribes had no choice but to use CLMC *-ay < EMC *-Ry characters as phonograms for OJ ey-syllables. CLMC *-ay was the closest available equivalent of OJ ey (*Ry). SV has -ôi corresponding to EMC *-(w)Ry after labials:
EMC *bRyq / LMC *p®- / SV bsi [(oy] ‘increase.’ The rounded vowel of SV -ôi probably reflects the southern dialect that was the source of SV. The Cantonese reading of
‘increase’ is púih [phuy]. Such a rounded vowel may not have existed in northern dialects: compare SV bsi and Cnt púih with the Tibetan transcription be’i, SK poy [p£y] and phay [phay], and Kan-on hai. I will not reconstruct OJ ey as *oy after labials. The other rhyme categories are sparsely represented only in the beta section. (1) Two beta section ey-phonograms belong to *-ey, the most common category for Shoki ye-phonograms (Table 8.21). These exceptional instances must reflect the southern EMC pronunciation of the *-ey rhyme as *-ay. The phonogram
EMC *peyh / CLMC *p- ‘close’ (cf. Cnt pai) for OJ pey (43.7.6, 43.9.6, 68.2.5) is a carryover from earlier usage. It is the sole Kojiki phonogram for OJ pey. The phonogram
EMC *mey / CLMC *mb- (cf. Cnt màih) for OJ mey (NS 104.7.6) is a near-homophone of the Kojiki phonogram for OJ mey,
EMC *meyq ‘rice’ (cf. Cnt máih). (2) *-ty is represented four times by the beta section phonogram
EMC *khtyh / CLMC *kh- for OJ key (59.5.6, 66.3.4, 104.8.4, 105.4.1). This is also a carryover from earlier usage, since the LMC reading of this character was closer to *i or *ty than to *Ry: compare SK kuy [kty], SV khí [xii], and LT khi 5 to OSV hdi [hRRy]. (3) *-εy is represented twice by the beta section phonograms
EMC *kεy / CLMC *k- ‘step’ (17.1.4) and
EMC *kεyh ‘abstain’ / CLMC *k(36.2.6) for OJ key (*kRy?). The CLMC reflex of the EMC *-εy rhyme, rendered in Kan-on as -ai, must not have been entirely appropriate for OJ ey (*Ry). The graphs
and
were respectively transcribed in Tibetan as ke’t and ke. The beta section scribes may have chosen
‘step’ and
‘abstain’ for their EMC readings *kεy and *kεyh. EMC *-εy may have phonetically resembled earlier stratum SK -oy [£y] (cf.
SK koy [k£y] ‘step’) and /or earlier stratum SV -di [RRy] (cf.
SV gió’i [zRRy] ‘abstain’). 229

OLD JAPANESE

(4) *-ay is represented only once by the beta section phonogram
EMC *payh / *phayh ‘river name’ (64.3.4) which probably had a low vowel *a in LMC: cf. SK phay [phay], SV bái [(aay] / phái [faay], and LT phai 5 [phay]. This low frequency tells us that OJ ey did not sound much like LMC *-ay, which corresponds to Kan-on -ai (< presumably *ai), not Kan-on -e (< *ye, *ey). Pulleyblank (1984, 1991a) believed that the EMC *-Ry and *-ay categories had merged into a single LMC category *-ay. If such a merger had occurred, both EMC *-Ry > CLMC *-ay and EMC *-ay > CLMC *-ay characters should appear in large numbers as phonograms for OJ ey-syllables in Shoki. Yet this is not the case: EMC *-ay appears only once and EMC *-Ry graphs appear in nearly all other cases. Why would Shoki scribes overwhelmingly prefer EMC *-Ry characters over EMC *-ay characters if their finals were indistinguishable in CLMC? Perhaps CLMC still distinguished between the two rhyme categories. The CLMC reflex of EMC *-ay must have been less like OJ ey (*Ry?) than the CLMC reflex of EMC *-Ry. (5) *-i is represented only once by the beta section phonogram
EMC *kih ‘crammed’ (24.2.2); cf. SK kuy [kty]. A scribe may have assumed that this character was read like a more common character sharing the same phonetic element such as
EMC *khRyh / CLMC *kh- ‘indignant’ (a beta section Shoki phonogram for key; cf. SK kay, SV khái [xaay], and LT khai 5). In Table 8.27, I list the ten ey-phonograms which appear three or more times in Shoki poetry along with their SX readings and the Tibetan and Indic transcriptive data. 81.8 per cent (90/110) of ey-syllables in Shoki are spelled with these phonograms. Amoghavajra’s use of
EMC *pwRy / CLMC *p- ‘cup’ and
EMC *mwRyq / CLMC *mb- ‘every’ for Indic pay- [p£y] and mai [m£y] would be parallel to the Japanese use of those characters as phonograms for OJ pey (*pRy) and mey (*mRy). Table 8.27 Common ey-phonograms in Shoki poetry OJ

Ph

EMC

CLMC

Kan

SK

SV

Tib. tr.

Ind. tr. (Am)

pey pey, bey






*pwRy *bwRy *bRyq

*p*p®*p®-

hai hai hai

bôi b4i bsi

pa’i ba’i be’i

pay– –

mey




*mwRy

*mb-

bai

poy poy poy phay moy

–

–

key






*mwRyq *khRy *kRy

*mb*kh*k-

bai kai kai

mai môi m5i khai cai

’be kha’e, khe –

mai – –

gey






*khty *YRy *YRy

*kh*Yg*Yg-

ki gai gai

khí ngai ngfi

ki, khe – ’ge, ’ge’i

– – –

moy khoy koy hoy kuy ay ay

230

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

The Tibetan transcriptions transcribe the CLMC final corresponding to EMC *-Ry in a number of ways: -a’i, -a’e, -e, -e’i. The a ~ e variation may reflect a late CLMC vowel such as *æ or *ε which was absent in Tibetan. Coblin (1994: 175–183) reconstructed synchronic variation between *-qi and *-εi for this final in ninth- and tenth-century northwestern LMC. He reconstructs an *-qi only for seventh- and eighth-century CLMC. Amoghavajra almost never used this final (Coblin’s mid-Tang CLMC *-qi) to transcribe Indic e, preferring the CLMC reflex of EMC *-ey (Coblin’s mid-Tang CLMC *-iei) for that purpose. This suggests that an *e-like pronunciation of the CLMC reflex of EMC *-Ry developed after Amoghavajra’s time (i.e., the mid to late eighth century). Such a pronunciation would not have existed when Shoki was being written in the early eighth century. Thus the Tibetan transcriptions do not necessarily support reconstructions of OJ ey such as Whitman’s (1985) *ey or Miyake’s (1995) *ε. I will reconstruct OJ ey for both Kojiki and Shoki as a mid central unrounded vowel followed by a palatal glide: *Ry. One may consider this ‘vowel’ to be a combination of o (‘B-type o’ = *R) with yi (‘A-type i’ = *i): *Ri. OJ ey (*Ry) later merged with OJ ye (*e). I do not think that OJ ey = *Ry first became *ey (cf. Whitman 1985) before merging with OJ ye = *e. If *R had fronted to *e before the loss of the following glide, it would have merged with OJ ei = *ey, the probable source of Go-on and Kan-on -ei. But OJ ey became post-OJ e, not post-OJ *ei. My reconstruction of OJ ey as OJ o + OJ yi explains an odd feature of Go-on and Kan-on. The other two branches of SX have finals such as [Ry] (OSV -ây, SK -ey), [RRy] (OSV -o’i), and [£y] (SK -oy) which might be expected to correspond to Go-on and Kan-on -oi < *Ry. However, the index of Tôdô (1978) lists no SJ readings ending in -oi and the index of Morohashi et al. (1981–82) lists only one SJ reading ending in -oi, Tô-sô-on hoi for
EMC *bwRyh ‘dry over heat.’ If OJ ey were *Ry, there would be no Go-on or Kan-on readings ending in -oi < *Ry because all OJ ey = *Ry had shifted to post-OJ e. There are no Kan-on readings ending in -e (implying that CLMC had no finals such as OJ ye = *-e or OJ ey = *-Ry), but some Go-on readings ending in -e may have once ended in OJ ey = *-Ry: (8.44)
EMC *khtyh ‘breath’ > OJ period Go-on *kRy = key > post-OJ Go-on ke (cf. OSV hdi [hRRy] ). Other Go-on readings ending in -e may have once ended in OJ ye = *-e: (8.45)
EMC *kæ ‘family’ > OJ period Go-on *ke = kye > post-OJ Go-on ke (cf. CT ke1) The implications of OJ phonetic reconstruction for OJ period SJ phonology have barely begun to be explored. 231

OLD JAPANESE

Table 8.28 OJ wo in Kojiki poetry OJ \ EMC

*-aw

*-o

*-Rw

*-uRwY

m t d n r s y k g Subtotals Total

37 1 0 0 0 0 0 0 0 38 148 (+13)

0 0 2 12 4 15 0 39 2 74

0 13 (+13) 0 0 7 0 0 0 0 20 (+13)

0 0 0 0 0 0 16 0 0 16

OJ wo OJ wo (‘A-type o’) in Kojiki corresponds to four EMC rhyme categories (Table 8.28). (1) The most frequent category is EMC *-o which corresponds to OSV -o [QQ] and CT -o (p. 208). This suggests that OJ wo may have been a mid back rounded vowel *o not unlike its post-OJ reflex [o]. Curiously, two near-homophones differing only in tone are phonograms for OJ syllables with different vowels: (8.46) OJ Ph EMC CT nu
‘slave’ *no lo2 nwo
‘rage’ *noh – (cf. LT no6 ) There is no evidence indicating that these two words ever had different vowels in any variety of Chinese. The choice of
EMC *no ‘slave’ for OJ nu (*nu) and
EMC *noh ‘rage’ for OJ nwo (*no?) must have been purely arbitrary. (2) The second most common final is *-aw, represented only by two phonograms: (8.47) OJ mwo two

Ph EMC
‘hair’ *maw
‘knife’ *taw

OSV CT – m2, ng 2 dao [zaaw] –


EMC *maw ‘hair’ appears 37 times while
EMC *taw ‘knife’ only appears once (4.33.2). EMC *-aw generally corresponds to CT -au [aw] (with the exception of
CT m2, ng 2 ‘hair’) and to OSV -ao [aaw]. Starostin (1989: 681) reconstructs these two words as OC *mhaaw (my *mhaw A) and OC *taaw (my *tawA). He does not reconstruct rounded vowels for this 232

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

rhyme at any stage of OC or in EMC. Why would the Japanese use words of this rhyme as phonograms for OJ mwo (*mo?) and OJ two (*to?)? I think the answer to this question may lie in the phonology of the languages of the Korean peninsula. [aw] was not a permissible final in Middle Korean. Not even the prescriptive SK of the Tongkwuk cengwun (TKCW ) dictionary listed such a final. The TKCW SK reflex of EMC *-aw was -wow [o(w) ] which corresponded to -wo [o] in non-prescriptive SK. Compare the TKCW and non-TKCW SK readings of ‘hair’ and ‘knife’: (8.48) OJ mwo two

Ph EMC
‘hair’ *maw
‘knife’ *taw

TKCW SK SK mwow [mo(w) ] mwo [mo] twow [to(w) ] two [to]

The Middle Korean hankul letter E w was created solely to write the coda -w of prescriptive SK. There were no native Middle Korean syllables ending in -w. If I project this aversion of a final -w into the Three Kingdoms period, it is possible that the Korean peninsular peoples borrowed EMC *-aw as *-o (cf. SK -wo [o] ). It is also possible that *-aw monophthongized or shifted to *-o or *-ow in one or more Korean languages prior to the transmission of OSK readings to Japan. In either case, perhaps the Japanese initially learned to pronounce
and
as *mo and *to, which closely matched their native syllables mwo (*mo?) and two (*to?). Unfortunately, we know too little about Koguryo, Paekche, and Silla to affirm any of these scenarios. (3) The third most common final, *-Rw, might be interpreted as evidence for the reconstruction of OJ wo as *Rw. This final appears only in two phonograms: (8.49) OJ Ph EMC two
‘bushel’ *tRwq rwo
‘leak’ *lRwh

OSV CT $Pu [ßRw] tau 3 – lau 5, lau 6

Some texts have
EMC *keyh ‘calculate’ instead of
EMC *tRwq ‘bushel’ as a phonogram for two. The former is obviously a scribal error for the latter. I have optionally added all thirteen cases of variation between
‘calculate’ and
‘bushel’ to the total for *-Rw in Table 8.28. Starostin (1989: 685) reconstructs the early LOC readings of
‘bushel’ and
‘leak’ as *toowq (my *towqA) and *lhoowh (my *lhowhA). Perhaps these characters were chosen for their LOC readings with *o. In any case, I will reject the reconstruction of OJ wo as *Rw because EMC *-Rw is not the predominant rhyme category for OJ wo-phonograms. Also, there is no reason why OJ wo would have a *-Rw allophone after t and r but not after other coronal initials: i.e., after d, n, or s. I would expect one allophone after all members of a consonant class (t, d, n, r, s), not after only two random members (t and r). 233

OLD JAPANESE

(4) The least frequent EMC rhyme category is *-uRwY, represented only by the sole Kojiki phonogram for OJ ywo,
EMC *yuRwYh ‘use’ (cf. Go-on yuu). EMC had no syllables such as *yo, *yaw, or *yRw which combined the initial *y with any of the previous three rhyme categories, so
EMC *yuRwYh ‘use’ was chosen despite its nasal final. The OSV equivalent of
EMC *yuRwYh ‘use’ is dùng [zuw§] < *y- ‘id.,’ though the general OSV equivalent of EMC *-uRwY is -uông [uRY]. Neither [u] nor [uR] precisely matches the [o] I would like to see, but they are both labial. I will reconstruct OJ wo in Kojiki as a mid back rounded vowel *o. Over 80 per cent of OJ wo in Shoki correspond to EMC *-o (Table 8.29). EMC *-o in turn corresponds to SK wo [o] and SV ô [oo], which is what we would expect if OJ wo were *o. All but one of the remaining rhyme categories correspond to SK and SV rhymes with labial vowels or glides (Table 8.30). We can ignore the exceptional *-ak category because it is represented only by a single phonogram for OJ two (*to?),
EMC *dak, which has another reading ending in *-o, EMC *doh. Obviously Shoki scribes had the *-o reading in mind (cf. SK two, SV ds, Tibetan transcription do, du) when they chose
‘measure’ as a phonogram for OJ two (*to?). Table 8.29 OJ wo in Shoki poetry OJ \ EMC

*-o

*-o/*-ak

*-Rw

*-uRwY

Other

t d n r s y k g Subtotals Total

12 6 3 6 14 0 58 12 111 135

5 0 0 0 0 0 0 0 5

4 0 0 3 0 0 0 0 7

0 0 0 0 0 7 0 0 7

2 (*-aw) 0 0 0 0 1 (*-iRw) 0 2 (*-o/*-uR) 5

Note ‘Other’ includes all rhyme categories that appear only once or twice: *-aw, *-iRw, *-o/-uR.

Table 8.30 SX equivalents of rhyme categories for Shoki wo-phonograms EMC Kan SK SV

*-ak -aku -ak [ak] -ac [aak]

*-aw -au -wo [o] -ao [aaw]

*-iRw -eu -ywo [yo] -iêu, -ao [iRw aaw]

*-o -o -wo [o] -ô [oo]

234

*-Rw -ou -wu [u] -âu [Rw]

*-uR -u -wu [u] -u [uu]

*-uRwY -you -ywong [yoY] -ung, -ong [uw§ aw§]

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

The rhyme category *-aw is unique to the beta section, where it only appears twice (42.3.3, 71.6.2), represented by
EMC *taw ‘knife’ (cf. Go-on to, tou, SK two [to] ), a phonogram for OJ two (*to?). This is a carryover from earlier usage. OJ two is first spelled as
EMC *taw ‘knife’ in the Suiko inscriptions. Two of the rhyme categories are unique to the alpha section. (1) *-uR is represented twice (75.9.2., 75.11.2) by the phonogram
EMC *YuR ‘pleasure’ (cf. Kan-on gu, SK wu, SV ngu, LT gu 2) which has an alternative reading ending in *-o, EMC *Yo (cf. Kan-on go, SK wo). (2) *-iRw is represented once (74.5.2) by the phonogram
EMC *yiRw ‘distant’ (cf. Kan-on eu, SK ywo [yo], SV dao [zaaw] < *y-, LT iau 2 [yaw] ). Like
EMC *yuRwYh ‘use’ in Kojiki and Shoki (see below), this phonogram was chosen because of the lack of syllables such as *yo in Chinese. LMC probably had no such syllable, because: (1) Its ancestor EMC had no such syllable. There are no CT syllables such as*io [yo] or OSV syllables such as dô [zoo] which would have derived from an EMC *yo. (CT io derives from EMC *yiRw.) (2) SK ywo [yo] corresponds to Kan-on yau and eu, SV dao [zaaw] and diêu [ziRw], and LT iau [yaw], not Kan-on yo, SV *dô [zoo] or LT *io [yo] (see below). (3) SV has no syllable such as *dô [zoo] < LMC *yo. (4) LT has no syllable such as *io [yo] < LMC *yo. The last two rare rhyme categories are shared by both sections of Shoki: (1) *-uRwY:
EMC *yuRwYh ‘use’ and its near-homophone
EMC *yuRwY ‘usual’ appear as phonograms for OJ ywo (*yo?). If one ignores their nasal finals, these characters would have been the best matches available for OJ ywo (*yo?) in CLMC (cf. SK ywong [yoY] for both, SV djng [zuw§] < *y- for ‘use’ and SV dong [zaw§] ~ dung [zuw§] < *y- for ‘usual’). (2) *-Rw:
EMC *tRwq ‘bushel’ and
EMC *lRwh ‘leak’ are phonograms for OJ two (*to?) and rwo (*ro?) in the beta section of Shoki as well as in Kojiki. The alpha section has one instance (119.3.4) of
EMC *lRw ‘building’ (cf. SV lâu [lRw] ) for OJ rwo (*ro?). With the exception of OJ ywo (*yo?) and this one case of OJ rwo (*ro?), OJ wo (*o?) is consistently spelled with phonograms of the EMC *-o category in the alpha section. I list all wo-phonograms which appear more than once in Shoki poetry in Table 8.31. 90.4 per cent (122/135) of the wo-syllables in Shoki are written with these phonograms. The Indic and Tibetan transcriptive data for the characters in Table 8.31 are more or less consistent with my reconstruction of OJ wo as a rounded vowel. The unrounded vowels in the ninth- and tenth-century Tibetan transcriptions (i, t, e) which correspond to Kan-on o (< OJ Kan-on o = *R?) and SV â [R] may be attempts to write late northwestern LMC vowels such as *t and/or *R. If the Tibetan letters i and e do truly represent palatal vowels, such vowels must have developed after the borrowing of Kan-on in the late seventh century. The Kan-on readings lack palatal vowels and glides. 235

OLD JAPANESE

Table 8.31 Common wo-phonograms in Shoki poetry OJ

Ph EMC

CLMC Kan

two




*doh *dak *toh *tRwq *doh *do *taw *no *noh *do *noq *lo *loq *lRwh

*t®*t®*t*t*t®*t®*t*nd*nd*t®*nd*l*l*l-

to taku to tou to to tau do do to do ro ro rou

*so *soh *yuRwYh *yuRwY *koq *koh *koh *ko *®o *Yoh *Yoh *Yoh *YuR

*s*s*y*y*k*k*k*k*h®*Yg*Yg*Yg*Yg-

so so you you ko ko ko ko ko go go go gu






dwo


nwo
rwo




ywo

kwo




gwo


swo

SK

SV

two thak tyek twu two two two nwo nwo two nwo lwo lwo lwu

$s $fc $3 $Tu $s $4 $ao nô ns $4 n5 lô l5 lpu

Tib. tr.

do, du – – te’[u] – – de’u – – duo [sic] – – – li’u, lt’u, le’u swo tô – swo t3 so ywong djng yung ywong dong, dung – kwo c7 go kwo c3 – kwo c3 ko, go, gu kwo cô ko hwo h4 – wo ngs gwa [sic] wo ngô go wo ngj – wu ngu –

Ind. tr. (Am) du, dhu, dhe – to – – – – – de, do – – ro, ì lu – su se, so – – – – – – – – – – –

The overwhelming majority of the evidence supports my reconstruction of OJ wo as a mid back rounded vowel *o. But it cannot tell us whether OJ wo was actually pronounced with a labial glide *w. We do not even know for sure whether MC had a final such as *-wo. No comparative evidence suggests a distinction between *Co and *Cwo in EMC or LMC. Korean and Vietnamese would not be able to reflect such a distinction or perfectly replicate a Chinese *-wo because neither language has Cwo syllables. (The wo of Yale romanization is Korean [o], not [wo].) Indic syllables of the shape Cvo (Indic has no [w] ) are extremely rare, so we are not likely to find any evidence for MC *Cwo syllables in the transcriptions of Jñanagupta and Amoghavajra. wo does not appear in the Tibetan transcriptions, though
‘diagram’ is transcribed as duo (cf. SK two [to], SV $4 [ßoo], LT to2). Without knowing whether *wo was possible in OJ or MC, we can imagine the following four scenarios: 1 2

OJ wo = *o was written with MC *o. OJ wo = *wo was written with MC *wo. 236

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.32 The reconstructed OJ ‘vowels’ thus far Front

Central

Back

High KJK/NS Total Rank

yi = *i 497/442 939/8,001 4th = 11.7%

iy = *t 61/51 112/8,001 8th = 1.4%

u = *u 797/740 1,537/8,001 2nd = 19.2%

Mid KJK/NS Total Rank

ye = *e 65/62 127/8,002 7th = 1.6%

o = *R 853/623 1,476/8,001 3rd = 18.4%

wo = *o 148/135 283/8,001 5th = 3.5%

Low KJK/NS Total Rank

3 4

Diphthong

ey = *Ry = o + yi 85/110 195/8,001 6th = 2.4%

a = *a 1,785/1,548 3,333/8,001 1st = 41.7%

OJ wo = *o was written with MC *wo because MC had no *o. OJ wo = *wo was written with MC *o because MC had no *wo.

I find the first scenario to be the simplest and therefore the most attractive. Regardless of whether MC had a final like *-wo or not, we have not seen any evidence for a contrast between *wo and *o in OJ. (Unger 1993 does reconstruct such a contrast for OJ, but his *wo and *o are equivalent to our *o and *R.) If OJ wo were *wo, its *w would have been predictable and nondistinctive. Thus I will write my reconstruction of OJ wo as *o without indicating this possible glide. I have now reconstructed the eight ‘vowels’ of OJ (seven vowels and one diphthong). In Table 8.32, I list (1) their reconstructions, (2) the total number of their attestations in Kojiki and Shoki (excluding textual variants), and (3) their rank in frequency relative to each other (excluding the C-type ‘vowels’ i, e, o). Not surprisingly, the ‘primary’ vowels (a, yi, u, o) are the four most frequent. The most frequent ‘secondary’ vowel (wo) is only a third as frequent as the fourth place ‘primary’ vowel (yi). These may not be all of the ‘vowels’ of OJ, for I have not yet examined the three C-type ‘vowels’ (i, e, o). In the next three sections (pp. 238–62), I will determine whether I should reconstruct any further ‘vowels.’ Out of the eight ‘vowels’ in Table 8.32, five remained more or less intact in post-OJ: (1) a = *a, (2) yi = *i, (3) u = *u, (4) ye = *e, and (5) wo = *o (Table 8.33). I assume that OJ ‘A-type’ wo (*o) and ‘B-type’ o (*R) merged into *o. If these two vowels had merged into an *R which would later shift to [o], Early Middle Japanese (EMJ) of the ninth century would have a somewhat unbalanced vowel system (Table 8.34). 237

OLD JAPANESE

Table 8.33 The post-OJ vowel system Front

Central

High

yi = [i] < OJ yi, iy, i

Mid Low

e = [e] < OJ ye, ey, e

Back

a = [a] < OJ a

u = [u] (MSJ [T] ) < OJ u o = [o] < OJ wo, o, o

Table 8.34 A rejected EMJ vowel system Front High Mid Low

Central

i = *i e = *e

o = *R a = *a

Back u = *u

Table 8.35 A preferred EMJ vowel system Front High Mid Low

Central

i = *i e = *e

a = *a

Back u = *u o = *o

I would rather reconstruct a more balanced system with *o instead of *R (Table 8.35). The OJ vowel system so far (Table 8.32) is balanced with the exception of the aberrant diphthong ey (*Ry), which I will discuss in my conclusion. We will now see whether I will have to clutter that neat system with more vowels.

OJ i OJ i (‘C-type i’) corresponds to five EMC rhymes in Kojiki (Table 8.36). Two (*-iR, *-yi) earlier corresponded to OJ A-type yi, two (*-t, *-uy) earlier corresponded to OJ B-type iy, and one (*-i) earlier corresponded to both yi and iy. Judging from the correlations between EMC finals, types of OJ i, and syllables with OJ C-type i, one might superficially conclude that: 1 2 3

OJ ti, ni, and i had A-type yi (*i). OJ wi, ri, and zi had B-type iy (*t). OJ di and si could be pronounced with either A-type yi (*i) or B-type iy (*t). 238

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.36 OJ i in Kojiki poetry OJ \ EMC OJ type

*-i A/B

*-iR A

*-t B

*-uy B

*-yi A

w t d n r s z $ Subtotals Total

0 0 8 0 0 0 0 0 9 825

0 75 0 192 0 171 0 0 438

0 0 1 0 144 98 6 0 248

12 0 0 0 0 0 0 0 12

0 0 0 0 0 0 0 118 118

The distribution of EMC finals among the phonograms for OJ i-syllables is almost random. One would have expected either (a) free variation of yi (*i) and iy (*t) or (b) better yet, the predominance of the less marked vowel (i.e., yi = *i). There is no good reason for ti to have A-type yi (*i) while ri has B-type iy (*t). A literal interpretation of the EMC data makes little sense. si is the only OJ i-syllable which is written with large numbers of phonograms implying both A-type yi (*i) and B-type iy (*t). Mabuchi (1957) proposed that Kojiki distinguished between two kinds of si. He considered
‘this’ (implying A-type syi = *si) to be a phonogram for si A and
‘intention’ and
‘kind of fungus’ (both implying B-type siy = *st) to be phonograms for si B. However, these phonograms were interchangeable. In one and the same Kojiki poem, OJ sisi ‘game’ is spelled as
 (implying syi-siy; 98.4) and as
 (implying siy-syi; 98.5) (Bentley 1997: 97). If both *si and *st existed in OJ, they were in free variation. It would be simplest to reconstruct *i after all coronals instead of reconstructing *t as an optional or obligatory allophone after only certain random coronals (e.g., *nd, *r, *s, *z). Historically, some of these i (*i) may have been iy (*t) from an even earlier PJ *Ri or *ui, but the evidence for reconstructing iy (*t) after coronals in the OJ period is not persuasive. The evidence so far for the vowels of OJ i and wi is unambiguous. Each of these two syllables is written with only one phonogram: (8.50) OJ Ph EMC Corresponding OSV rhyme class i
‘he/she/it’ *qyi -i [ii] wi
‘leather’ *wuy -ui [uy] OJ i seems to be A-type yi (*i) and OJ wi seems to be B-type wiy (*wt) (cf.
EMC *kuyh ‘valuable’ for OJ kiy = *kt). The latter would be the sole 239

OLD JAPANESE

Table 8.37 OJ i in Shoki poetry OJ \ EMC OJ type

*-ey ?

*-ey/*-i ?/A/B

*-i A/B

*-iR A/B

*-iR/*-iRyk A/B/?

*-t B

*-uy B

*-yi A

w t d n r s z $ Subtotals Total

0 0 1 1 0 0 0 0 2 748

0 0 0 0 0 0 1 0 1

1 12 1 6 86 16 1 0 123

3 54 3 33 7 32 0 0 132

0 0 0 0 0 0 0 2 2

0 0 0 128 11 221 7 32 399

9 0 0 0 0 0 0 0 9

0 0 0 0 0 0 0 80 80

OJ i-syllable with the vowel *t. I am not comfortable with reconstructing OJ wi as *wt because *t is otherwise so infrequent in OJ. We will soon see that Shoki provides evidence for an alternate reconstruction *wi. OJ i in Shoki corresponds to seven EMC rhyme categories (Table 8.37). *-iRyk is represented only by the alternative reading EMC *yiRyk ‘change’ for
, a phonogram for OJ i which can also be read as EMC *yiRh ‘easy.’ Clearly only the *k-less reading (cf. SK i) is relevant for our purposes. The most frequent category is *-t (53.3 per cent = 399/748). Yet I doubt that at least half of the i-syllables in Shoki had the OJ B-type vowel iy (*t). *-t and the other EMC rhyme categories corresponding to OJ i mostly have equivalents with palatal vowels or palatal glides in the LMC-based strata of SX (Table 8.38). The Chang’an dialect of LMC (CLMC) probably did not preserve EMC *-t after non-grave initials (i.e., coronals). Amoghavajra occasionally used coronal-initial *-t phonograms to write Indic i and c. For example, he wrote

Table 8.38 SX equivalents of rhyme categories for Shoki i-phonograms EMC Kan-on SK

*-ey -ei -yey [yRy]

SV

-ê [ee]

*-i, *-iR, *-t -i -uy [ty] (after k, $ < *Y, h) -o [£] after LMC dental and retroflex sibilants; -i elsewhere -i, -y [ii] -e [tt] after LMC dental and retroflex sibilants

240

*-iRyk -eki -yek [yRk]

*-uy -wi -wuy [uy] > modern -wi [wi]

*-yi -i -i

-ich [ic]

-uy [wi]

-i, -y [ii]

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

the Indic syllable li with
EMC *lt ‘village,’ whose CLMC pronunciation may have been *li (cf. SK li [ri], SV lí, ly [lii], Tibetan transcription l i). As in Kojiki, both
EMC *lt ‘village’ and its homophone
‘reason’ are phonograms for OJ ri in Shoki. The most common phonogram for OJ ri in Shoki is
EMC *lih ‘profit’ > CLMC *l(ih) (cf. SK li, SV lh [lii], Tibetan transcription li; SV l%i [lRRy] is irregular and probably OSV in origin). One could argue that Shoki scribes were forced to use CLMC coronal-initial *-i characters to transcribe OJ coronal-initial *-t syllables (*tt, *ndt, nt, *rt, *st, *nzt) because CLMC had no syllables such as *nt, *rt, *st, or *nzt. However, it would be simpler to assume that the vowel of OJ ti, di, ni, ri, si, zi was an *i which was written with a CLMC *-i. The vowel of OJ wi may have also been an *-i. In Shoki, this syllable is spelled with seven different phonograms: (8.51) OJ Ph wi
‘power’ wi wi wi wi wi wi

EMC *quy

Kan SK wi wuy [uy]
‘say’ *wuyh wi wuy
‘winding’ *qwiR wi wuy ‘bend’ *qwiRq wi wuy
‘great’ *wuyq wi wuy
‘leather’ *wuy wi wuy
‘do’ *wiR wi wuy ‘for’ *wiRh wi wuy
‘rank’ *wih wi wuy

SV uy [}wi] oai [}waay] vh [vii] uy [}wii] u] [}wii] vu [vii] vi [vii] vi [vii] vh [vii] vh [vii]

Tib. tr. u’i, u

Freq 3

u, yu, ’yu – – – wu’t ’u, yu, ’yu ’u ’u

3 2 2 1 1 1

69.2 per cent (9/13) of OJ wi in Shoki are written with EMC *-uy phonograms (implying B-type wiy = *wt) and the remaining 30.8 per cent (4/13) are written with EMC *-wi(R) phonograms (implying A-type wyi = *wi). Vietnamese, which distinguishes between -ui [uy] and -uy [wi] in native words, has only /wi/ (vi [vii] by itself and uy [wii] after consonants) in SV. If CLMC were like the southern LMC dialect that was the source of SV, it would have merged EMC *-uy, *-wi, and *-wiR into a single rhyme *-wi. The SK evidence is ambiguous because Korean does not distinguish between [uy] and [wi]. SK wuy [uy] might reflect a northern LMC dialect with either *uy or *wi. The u and yu of the Tibetan transcriptions might reflect a late stage of northwestern LMC in which *uy or *wi had monophthongized to *ü (cf. some Korean speakers’ pronunciation of /wi/ as [ü] ). I am uncertain whether EMC *-uy, *-wi, and *-wiR merged into *wi or *uy. This CLMC *wi or *uy would have been the closest available equivalent to OJ wi (*wi or *wt?). I cannot completely rule out the possibility of OJ wi being *wt, but it is far more likely that this syllable was *wi with the more common vowel *i. 241

OLD JAPANESE

OJ i (*i?) is written in both sections of Shoki with EMC *-yi and *-t phonograms: (8.52) OJ i i i i

Ph
‘he/she/it’
‘strange’
‘using’
‘easy’ (
‘change’

EMC *qyi *yth *ytq *yiRh *yiRyk

Kan i i i i eki

SK i i i i yek

SV y [}ii] dh [zii] du [zii] dh [zii] dhch [zic]

Tib. tr. }yi ( ) yi, yt ’i, yt, ye – ytg)

Freq 80 19 13 2

The LMC source dialects of SK and SV seem to have merged the two rhymes into a single rhyme like *-i. Whether CLMC still maintained such a distinction is open to question. Amoghavajra transcribed Indic i and c with the *-yi phonogram
EMC *qyi ‘he/she/it’ but not with *-t phonograms such as
EMC *yth ‘strange.’ This may or may not be evidence for a distinction between *qyi and *yt in CLMC. Whatever the situation may have been in CLMC, OJ i is predominantly spelled with the *-yi phonogram
‘he/she/it’ in Shoki as well as in Kojiki, so I will reconstruct OJ i as *i. I will treat all cases of OJ C-type i as if they were A-type yi (*i), though some of those vowels originated from other pre-OJ sources (i.e., *Ri, *ui, *e). There is no need to reconstruct a special OJ C-type vowel distinct from A-type yi (*i) or B-type iy (*t). Nearly all of the EMC rhyme categories associated with OJ C-type i-phonograms are also associated with OJ A-type yi and OJ B-type iy-phonograms. The SX and transcriptive data for the most common i-phonograms accounting for 92.4 per cent (691/748) of the total generally support the reconstruction of OJ i as *i (Table 8.39). Some of the data, however, imply a non-palatal value for OJ i at first sight. (1) The SK and SV readings for characters with CLMC dental and retroflex sibilant initials (*s, *s®, *Z®) have non-palatal vowels: SK o [£] and SV e [tt]. But characters read with these vowels in SK and SV did not necessarily represent OJ *st and *nzt. Such readings reflect dialects of late LMC which depalatalized *i to *t after dental and retroflex sibilants (p. 108). This late LMC *t (or Pulleyblank’s LMC *" and *ù) did not yet exist when Shoki was being written. Amoghavajra transcribed Indic si with
EMC *siR > CLMC *si (?) ‘this’ in the mid to late eighth century, decades after Shoki was completed. If
EMC *siR ‘this’ had already become CLMC *st or *s" (cf. the ninth- to tenth-century Tibetan transcription st) in the eighth century, Amoghavajra would have indicated that the character was to be pronounced as *si instead of *st by adding a fanqie spelling such as
 CLMC *st + *\i (cf. SK so + si, SV te + thi, LT su1 + si 1) = *si. The OJ syllable si represented by
EMC *siR ‘this’ was probably more like Indic si than SK so [s£] or SV te [ttt]. 242

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.39 Common i-phonograms in Shoki poetry OJ

Ph

EMC

CLMC

Kan

SK

SV

Tib. tr.

Ind. tr. (Am)

wi











z



*quy *wuyh *qwiR(q) *triRh *triR(h) *trih *nriRq *ney(h) *nrih *ãth *ntq *ãiRq *lih

*q*w*q*tr*tr*tr*ndr*nd*ndr*ã9*nd*ã9*l-

wi wi wi ti ti ti di dei di zi di zi ri

wuy wuy wuy ci ci chi ni ni ni i ni i li

u’i, u u, yu, ’yu – ci, ct, ji ci, ct, ji ci – ’de – – ni zhi, zht, zhe li

– – – ∞c ∞i – – de îi, (ñ)îi – -dy-, di, dc – –










*ltq *lih *ct *zt *cth *siR *\tq

*l*l*c*s®*c*s*\-

ri – si si si si si

li – ci so ci so si

*\iq

*\-

si












*\iR *zth *st(h) *ãth *jrth *ãth *qyi *yth *yth

*\*s®*s*ã9*Z®*ã9*q*y*y-

si si si zi si zi i i i

ci (irreg.) si so so i so i i i i

li, lt – ci, chi, ji, jt – – st she, shi, ’shi, sht –

– – – – – si –




uy vh uy, u] trí tri, trí trí nx nê, nq nq, nh, nhh nhh nhu nhu lh l%i (OSV?) lí, ly – chi tJ chí te thm thu] (irreg.) thm – t@ te, tE nhh su nhh y dh du

– st si, st – shi – }yi yi, yt ’i, yt, ye

– – – – – – i, c – –

ti di ni ri

si

zi i

5i

Note The modern SK reading of
is si. ci may be an error in the Cenwun okphyen.

(2) The Tibetan transcriptions often have e and t as well as i. These may be attempts to write late northwestern LMC *R and *t (or Pulleyblank’s LMC *" and *ù). They are not strong evidence for reconstructing OJ i as *t because they may reflect vocalic innovations that took place after the writing of Shoki. Amoghavajra used characters transcribed in Tibetan with t to write Indic syllables ending in i and c. As I just explained, if these characters had been pronounced with a final *-t in CLMC, he would have supplied fanqie indicating that they should be pronounced with a final *-i: e.g.,
 CLMC *st + *\i = *si, etc. I will reconstruct OJ C-type i as a high front unrounded vowel *i, though a high central allophone *t is also possible. 243

OLD JAPANESE

Table 8.40 OJ e in Kojiki poetry OJ \ EMC OJ type

*-æ A

*-ey B

w t d n r s z y $ Subtotals Total

0 0 0 0 0 0 0 49 0 53 0 0 0 0 0 0 (2) 0 (2) 102 261 (+5)

*-ey? B

*-iR B

*-iRn A

*-iRy ?

*-Ry B

*-wiRn A?

*-wey B?

0 60 0 0 0 0 0 0 0 60

0 0 0 0 0 0 4 0 0 4

0 0 0 0 0 0 0 25 0 25

0 0 0 0 0 41 0 0 0 41

0 0 0 0 0 0 0 0 (5) (5)

0 0 14 0 0 0 0 0 0 14

13 0 0 0 0 0 0 0 0 13

OJ e OJ e (‘C-type e’) in Kojiki corresponds to eight EMC rhyme categories (Table 8.40). We have encountered five of these categories (*-æ, *-ey, *-iR, *-iRn, and *-Ry) before in Tables 8.20 and 8.24. Two previously unencountered categories (*-wiRn) and (*-wey) are variants of *-iRn and *-ey, previously associated with phonograms for OJ ye and ey-syllables in Kojiki. The remaining previously unencountered category (*-iRy) is a variant of the *-yiRy category represented in Kojiki by the phonograms
EMC *YyiRy ‘art, skill’ for OJ gyi (*Ygi) and
EMC *byiRyh ‘money’ (and/or
EMC *byiRyh ‘ruin’) for OJ bye (*mbe). The absence of categories entirely unlike others we have seen before suggests that OJ C-type e was not a ‘vowel’ distinct from OJ A-type ye (*e) and OJ B-type ey (*Ry). If I literally interpret the correlations between EMC finals, types of OJ e, and syllables with OJ C-type e, I could reconstruct: 1 2 3

A-type ye (*e) for OJ de, se, and ye B-type ey (*Ry) for OJ we, te, ne, re, and ze Allophonic variation between the two ‘vowels’ (*e ~ *Ry) for OJ e.

I could even claim that OJ de had a medial glide *-w- on the basis of
EMC *drwiRn(h), the sole phonogram for OJ de in Kojiki poetry. I do not know why a character with medial *-w- was selected. I doubt that the *-wreflected anything in Japanese. The *-w- may not have even existed in the OSK pronunciation of the word, though its prescriptive Tongkwuk cengwun readings are ttwyen and ttwyen’. The SK reading given in the Cenwun okphyen dictionary is cyen [tsyRn] without a medial *-w-. I will ignore the *-w- and treat OJ de like a Type-A syllable. But it seems arbitrary to reconstruct 244

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

OJ de as Type A *nde while reconstructing other OJ dental-initial syllables (te, ne, re) as Type B (*tRy, *nRy, *rRy). If we look beyond Kojiki, we will find that OJ de is spelled in the Taihô census of 702 CE and Man’yôshû (17.6.4, 17.8.5) as
EMC *dRyh ‘generation’ (cf. OSV $Ii [ßRRy] ‘id.’), implying B-type dey (*ndRy). Were A-type dye (*nde) and B-type dey (*ndRy) in free variation? Could I reconstruct *-Ry after all dentals? We will later see that Shoki points toward *e after all coronals in OJ. Like the majority of OJ coronal-initial e-syllables, OJ ze at first appears to have had B-type ey (*Ry) as its ‘vowel.’ Its sole phonogram is
EMC *jiRq ‘this,’ which belongs to the same EMC rhyme category as
EMC *YiR ‘proper’ < OC *YayB (cf. OSV ngay [Yay] ‘righteous’), the Kojiki phonogram for OJ gey (*YgRy). One might think that
EMC *jiRq ‘this’ was chosen for an OC reading like *j(h)ayqB by analogy with
OC *YayB ‘proper.’ However, it turns out that
EMC *YiR ‘proper’ belongs to the OC
*-ay ‘song’ class whereas
EMC *jiRq belongs to the OC
*-e ‘branch’ class: its OC reading is *d(h)eqB. Its LOC reading *j(h)eqB is probably the basis of Go-on ze. DharmakÖema (Dk), an Indian monk living in northwestern China in the fifth century, used near-homophones of this character to transcribe Indic syllables ending in -i, -e, and -ya [y£]: (8.53) OJ Ph si
‘this’ (Taihô census: se; NS: si)
‘bestow’ (Inariyama: kye)
‘branch’

LOC *seB *\eB *ceB

EMC *siR *\iR *ciR

Ind (Dk) si, se 5ya [\y£] ci, ce

Of course, Ind -e resembles LOC *-eB and OJ A-type ye (*e). Ind -ya [y£] resembles not only the regular OSV and CT equivalents of EMC *-iR (OSV -ia [iR] and CT -ia [ya] ) but also pre-OJ *ia and *iR, the diphthongal sources of OJ ye (*e). The earliest Go-on readings for
EMC *\iR ‘bestow’ and
EMC *jiRq ‘this’ may have been *siR (or *se) and *nziR (or *nze), which later developed into post-OJ Go-on se and ze. Presumably native OJ ze was similar to *nze, the OJ period Go-on reading of
EMC *jiRq ‘this.’ I will reconstruct OJ ze as *nze. I would like to reconstruct OJ se as *se to match OJ ze (*nze). Both of the Kojiki phonograms for OJ se (
EMC *\iRyh ‘power’ and
EMC *\iRyh ‘world’) have the EMC initial *\. If the initial of OJ se were also *\, a front mid vowel could have caused it to develop from an earlier *s: (8.54) *se > *\e *s would probably not have palatalized before a schwa: i.e., *sRy > *\Ry > *\e. The OSV equivalent of the EMC *-iRyh rhyme may be -ê [ee]: cf. OSV nghV [Yee] ‘craft’ from
EMC *YyiRyh ‘art, skill’ with the *-yiRyh rhyme. 245

OLD JAPANESE

DharmakÖema transcribed Indic 5i [\i] with
EMC *\iRyh ‘world’. Jñanagupta does not use the EMC *-iRyh rhyme in his transcriptions. However, I would not consider DharmakÖema’s transcription to be evidence for reconstructing OJ se as *\i instead of *se or *\e. OJ we is written with
EMC *®weyh < LOC *w(h)ieyhA ‘grace.’ It belongs to a sparsely populated EMC rhyme (*-wey) with no known OSV equivalent. The Kojiki phonograms for OJ pey (*pRy) and mey (*mRy),
EMC *peyh ‘close’ and
EMC *meyq ‘rice,’ belong to a similar rhyme (EMC *-ey) corresponding to (1) [ay]-like rhymes in southern Chinese languages, (2) OSV -ây [Ry] and Proto-Tai *Ri, and (3) Indic -ai in Sa¢ghabhara’s transcriptions (p. 225). If the southern pronunciation of EMC *-ey was *-ay or *-Ry, then the southern pronunciation of EMC *-wey might have been *-way or *-wRy. Hence
southern EMC *way ~ *wRy ‘grace’ (cf. Cnt waih [way] ) may have represented OJ B-type wey (*wRy). Go-on we for
‘grace’ may derive from OJ period Go-on *wRy. I am not entirely certain that OJ we was *wRy because the CT reflexes of EMC *-wey are -e and -i: the CT readings of
EMC *kwey ‘sceptre’ and
EMC *®wey ‘garden plot’ are ke1 and ki 2, not *koai 1 [kway] and *koai 2 [kway]. I will later cite Shoki evidence for the reconstruction of OJ we as A-type wye (*we). The sole Kojiki phonogram for OJ ye is
EMC *yiRn ‘prolong.’ I will reconstruct OJ ye as *ye for the following reasons. (1)
EMC *yiRn ‘prolong’ belongs to the same EMC rhyme category as
EMC *biRnh ‘divide,’ the Kojiki phonogram for OJ bye (*mbe). This rhyme category (EMC *-iRn) corresponds to OSV -iên [iRn] and CT -iN [u]. (2) The CT reading of
EMC *yiRn ‘prolong’ is an exception to this generalization: CT chhian2 [tshyyεn] ends in -ian [yεn], not -iN [u]. Nonetheless, it does contain a front vowel [ε]. (3) DharmakÖema used
EMC *yiRn ‘prolong’ to transcribe the Indic syllables ya [y£], yaá [y£å ], and yaú(t) [y£n(t) ]. Compare Indic ya [y£] to pre-OJ *iR and *ia, the diphthongal sources of OJ *e. OJ e appears only seven times in Kojiki: twice in poetry (9.18.1, 9.18.2) and five times in prose (I: Age of the Gods). The poetic attestations of e are limited to the single exclamation ee whose spelling varies from text to text: (8.55a)



EMC *qæh qæh

(NKBT Kodai kayô shû, Nobuyoshi ms.)

(8.55b)
I =



EMC *qæh qæh

(Shintô taikei, Gohon ms.)

(8.55c)
I =



EMC *dep dep

(Shinpuku-ji, Kanenaga, Kan’ei mss.; Kojiki-den)

I is merely a symbol representing the repetition of the preceding character, so there is no real difference between (8.55a) and (8.55b). I will reject (8.55c) 246

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

because
EMC *dep ‘pile up’ is not attested elsewhere as a phonogram for any OJ syllable. It may be a scribal error for some other character. Furthermore, this ee is parallel to the exclamation aa (spelled as

EMC *qa qa) later in the poem (9.21). The two remaining spellings (8.55a and 8.55b) contain
EMC *qæh ‘second,’ which belongs to the same rhyme category as
EMC *Yæ ‘tusk, ivory’ (cf. CT ge 2), the Kojiki phonogram for OJ gye (*Yge). EMC *-æ corresponds to OSV -e [εε] and CT -e, so I could reconstruct OJ e as *e. Yet OJ e is consistently spelled in the text of Kojiki with the phonogram
EMC *qRyh ‘love’ (cf. SK oy [£y] ) instead of
EMC *qæh ‘second.’ This phonogram belongs to the same rhyme category as
EMC *bRyq ‘multiply,’ the Kojiki phonogram for OJ bey (*mbRy). It rhymes precisely with
EMC *dRy ‘generation’ (cf. OSV $Ii [ßRRy] ), a phonogram for OJ de (*ndRy). All of this suggests that OJ e could also be pronounced as *Ry. It is difficult to draw conclusions about such an extremely rare syllable. Perhaps OJ e was pronounced as *Ry except in (a) marginal native words such as onomatopoeia and exclamations (i.e., OJ ee = *ee) and (b) Chinese loans such as
EMC *qæh > OJ Go-on *e? > Go-on e ‘second.’ A contrast between *Ry and *e contrast could have existed in Chinese loans (e.g., between
EMC *qæh > OJ Go-on *e? > Go-on e ‘second’ and
EMC *qty > OJ Go-on *Ry? > Go-on e ‘clothing’), but there is no evidence for such a contrast in native words. Thus far I have reconstructed OJ e as: 1 2 3

A-type ye (*e) in de, se, ze, ye B-type ey (*Ry) in we, te, ne, re either *e or *Ry in e.

This is still litle better than my original face-value interpretation because of the random split of dental-initial syllables between Type A and Type B. The Shoki data are needed to resolve this problem. In Shoki, OJ e corresponds to eight EMC rhymes (Table 8.41). One (*-ey) is primarily associated with Shoki ye-phonograms, four (*-εy, *-i, *-Ry, *-wRy) are associated with Shoki ey-phonograms, and the remaining three (*-et, *-iRn, *-iRy) are unique to Shoki e-phonograms. Two of the latter rhymes (*-iRn and *-iRy) are similar to rhymes associated with Shoki ye-phonograms (*-yiRn and *-yiRy). 88.2 per cent (231/262) of the e-syllables in Shoki are spelled with either *-ey or *-iRy. This figure rises to 92.4 per cent (242/262) if we treat all 11 instances of the phonogram H ‘phonogram for Jpn te’ as a variant of
EMC *tey / Kan-on tei ‘base.’ The rhymes *-ey and *-iRy had probably merged into a single [e]-like LMC rhyme because: 1 2

Amoghavajra used both rhymes to render Indic e. Their SX equivalents are indistinguishable (Table 8.42). 247

OLD JAPANESE

Table 8.41 OJ e in Shoki poetry OJ \ EMC OJ type

*-et ?

*-ey A

*-ey? A?

*-ey/*-εy A/B

*-i B

*-iRn A?

*-iRy A?

*-Ry B

*-wRy B

w t d n r s z y $ Subtotals Total

0 0 6 0 0 0 0 0 0 6 262

6 59 15 30 14 11 0 0 0 135

0 11 0 0 0 0 0 0 0 11

0 0 0 0 0 9 0 0 0 9

0 0 0 1 0 0 0 0 0 1

0 0 0 0 0 0 0 1 0 1

5 0 0 0 43 23 1 24 0 96

0 0 0 0 0 0 0 0 2 2

1 0 0 0 0 0 0 0 0 1

Table 8.42 SX equivalents of rhyme categories for Shoki e-phonograms EMC Kan SK SV

*-et -etu -yel [yRl] -êt [et]

*-ey -ei -yey [yRy] -ê [ee]

*-εy -ai -ay, -oy [ay £y] -ai, -o’i [aay RRy]

*-i -i -uy, -o, -i [ty £ i] -i, -y, -u’ [ii ii tt]

*-iRn -en -yen [yRn] -iên [iRn]

*-iRy -ei -yey [yRy] -ê [ee]

*-Ry -ai -ay, -oy [ay £y] -ai [aay]

*-wRy -(w)ai -ay, -woy [ay oy] -ôi [oy]

The EMC rhymes *-ey and *-iRy are the most common rhymes for phonograms of all Shoki e-syllables except e itself. Therefore I could reconstruct the vowel *e for all OJ e-syllables in Shoki other than e. The SX and transcriptive evidence for all e-phonograms appearing more than once generally agree with this reconstruction (Table 8.43). The above phonograms account for 95.4 per cent (250/262) of the total. OJ e is the only e-syllable in Shoki which is never written with phonograms of the *-ey and *-iRy rhymes. It is always written in Shoki poetry with the phonogram
EMC *qRyh ‘love.’ In the prose of Shoki, it is also written with two other phonograms: (8.56) OJ Ph e
‘dust’ e
‘mourn’

EMC *qRy *qRy

CLMC *q(ay) *q(ay)

Kan ai ai

SK SV ay ai [aay] oy [£y] ai [aay]

All three phonograms belong to the EMC *-Ry rhyme (cf. OSV -o’i [RRy] ) which probably lowered to *-ay in CLMC (p. 228). There is no evidence in Shoki for a *e-like pronunciation of the OJ syllable e. The exclamation ee 248

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.43 Common e-phonograms in Shoki poetry OJ

Ph

EMC

CLMC

Kan

SK

SV

Tib. tr.

Ind. tr. (Am)

we




H



*®weyh *wiRyh *tey (probably *teyq *net

*h®*w*ta variant *t*nd-

kei wei tei of
) tei detu

hyey wuy tye

huq vq $ê

hywe, hwe ’we’i –

– – – ti, tc –

*dey

*t®-

tei

$Y nát niUt $V

– ’der




ci, tyey nal nyel si, tyey

–













*tey *ney(h) *neyq *liRyh *leyq *ley *\iRyh *dzey(h) *crεy *tsi *ciRyh *\iRyh

*t*nd*nd*l*l*l*\*ts®*cr*ts*c*\-

tei dei dei rei rei rei sei sei sai si sei sei

tyey ni ni, nyey lyey lyey lye syey cyey coy co cyey syey

$ê nê, nq nW lq lW lê thU tV trai, chay – chU thU




*sey

*s-

sei

sye

tây, tê








*sey *seyh *jiRyh *yiRyh *qRyh

*s*s*c®*y*q-

sei sei sei ei ai

sye syey sye yey oy

tê tU phq [sic] dq, duq ái

ti, te, de, de’i, de’t – ’de – – le, le’i – zhe – – – – she, she’i, she’t se, sye, ’she – se – – a, e, a’i

te te, de

de, ne ne re se

ze ye e

– de – – – le 5e, 5ai – – – ce – – – se – yi, ye, yai ai, ai [sic], e

Note The syllable ze appears only twice in Shoki poetry. All texts spell one attestation (8.1.4) with
EMC *jiRyh ‘divine by stalks.’ Some texts spell the second attestation (78.1.4) with this graph while others spell it with the similar but homophonous phonogram
EMC *jiRyh ‘bite’ (Kan-on sei, SK sye, SV thq).

(*ee?, spelled

EMC *qæh qæh) in Kojiki poem 9 does not appear in the Shoki version of the poem (NS 7). I will reconstruct OJ e in Shoki as *Ry. Amoghavajra does spell Indic e as well as ai and ai [sic] with
EMC *qRyh / CLMC *q(ayh) ‘love,’ but this does not necessarily mean that I should reconstruct OJ e in Shoki as *e. If
‘love’ were pronounced as *e in CLMC, its Kan-on reading should be e < OJ period Kan-on *e, not ai < OJ period Kan-on *ay. Kan-on has no syllable such as e which would have derived from OJ period Kan-on *e or *Ry. This suggests that CLMC had no syllables such as *e or *Ry.
CLMC *q(ayh) ‘love’ was only an approximation of Indic e. One might want to argue that I could reconstruct the vowel in other OJ e-syllables as *Ry because Amoghavajra spells Indic 5ai [\ay] and yai [yay] 249

OLD JAPANESE

with Shoki phonograms for OJ se and ye (
‘power’ and
‘drag’). However, Amoghavajra may have been forced to use these phonograms because LMC in general lacked syllables such as *\ay and *yay. Kan-on lacks the syllables syai and yai. There are no SK syllables such as syay or yay. SV has no syllables such as dai, day, dây [zaay zay zRy] < *yaay, *yay, *yRy. SV thay, thây [thay thRy] < *\ay, *\Ry do not exist. SV thai [thaay] could not derive from an LMC syllable such as *\aay because it corresponds to Kan-on tai, sai and SK thay, chay < LMC *th-, *tsh-. The -yey [yRy] found among the SK readings of Shoki e-phonograms is not necessarily proof for the reconstruction of OJ e as *Ry because premodern SK had no [e]. I will continue to reconstruct the vowel *e for the majority of OJ e-syllables in Shoki. Should I continue to reconstruct *-Ry for some of the OJ e-syllables in Kojiki? I do not want to claim that the writers of Kojiki and Shoki had different pronunciations of this vowel. The two works were compiled during the same period and shared many poems. Surely they more or less reflect the same phonology. I think it would be simplest to assume that OJ e was *e except in the syllable e (*Ry). The use of EMC *-ey (southern EMC *-ay ~ *-Ry) phonograms in Kojiki may simply be a carryover from pre-OJ usage rather than evidence for OJ *Ry. The only e-syllables in the Suiko inscriptions are te, ne, and re, spelled as H ‘phonogram for Jpn te’ (a variant of
EMC *tey; cf.
EMC *tey, Cnt dài [tay] ‘low’),
EMC *neyh ‘mire’ (cf.
EMC *neyh, Cnt nàih [nay] ‘mire’), and
EMC *leyq ‘rites’ (cf. Cnt láih [lay] ). Perhaps OJ te, ne, re were *tRy, *nRy, *rRy in the early seventh century, but they may have become *te, *ne, *re by the early eighth century. The innovative Shoki orthography reflected this shift whereas the more conservative Kojiki orthography did not. OJ e = *Ry was the last e-syllable which had not yet shifted to *e. Eventually it too would monophthongize to *e (cf. Middle Korean ey [Ry] > Modern Korean [e] ). After the OJ period, it would merge with OJ ye = *ye into a single syllable *ye.

OJ o OJ o (‘C-type o’) in Kojiki corresponds to six EMC rhyme categories (Table 8.44). The rhymes *-t and *-tR were previously found among the phonograms for Kojiki o-syllables. OJ o is exclusively written with three phonograms belonging to these rhymes: (8.57) OJ o o o

Ph
‘silt’
‘idea’
‘in’

EMC OSV equiv. of rhyme category Freq *qtR(h) -ea [tR] 44 *qth -d [RR] 33 *qtR -ea [tR] 1 250

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.44 OJ o in Kojiki poetry OJ \ EMC OJ type

*-t B

*-tR B

*-o A

*-uRn

*-uw

*-wRn

p b w $ Subtotals Total

0 0 0 33 33 280

0 0 0 42 42

1 0 0 0 1

1 6 134 0 141

42 0 0 0 42

21 0 0 0 21

These characters are all read with unrounded vowels (R and t) as nôm phonograms: (8.58) OJ o o o

Ph
‘silt’
‘idea’
‘in’

EMC *qtR(h) *qth *qtR

Nôm readings dd [zRR] ‘dirty’ Py [}Ry] ‘that’ (cf. LOC *qRòhB ) N [}RR] ‘in,’ J [}tt] ‘yes,’ ea [}tR] ‘to like’

I am ignoring a rare alternative reading of
, EMC *qo < OC *qaA ‘ah!,’ which belongs to the EMC *-o rhyme category corresponding to OSV -o [QQ] and CT -o. If OJ o were *o, it would not have been written only once with an ambiguous *-tR/*-o character. It would have been predominantly written with unambiguous *-o characters such as
EMC *qo ‘crow.’ I will reconstruct the OJ syllable o as B-type o (*R). Mabuchi has proposed that there were two types of OJ o in Kojiki: 1 2

oA, written with
EMC *qth ‘idea’ oB, written with
EMC *qtR(h) ‘silt’.

Bentley (1997: 92) has criticized this proposal: It should be stated from the outset that the case with o is less than convincing, simply because of the inconsistency with the orthography . . . It is thus interesting that even Ishizuka [Tatsumaro] (1929, 1:20) had noticed this orthographical overlap, “There are cases where
[EMC *qth ‘idea’] and
[EMC *qtR(h) ‘silt’] are used interchangeably.” As a simple example of this confusion, consider that the [OJ] word opo- [‘great’] appears in poetry and proper names a total of 61 times in the text of Kojiki. The initial syllable is written with
[EMC *qth ‘idea’] 54 times, while it appears with
[EMC *qtR(h) ‘silt’] only seven times . . . If we are to believe that these two graphs represent phonetically different phonemes, why the apparent confusion? 251

OLD JAPANESE

Even Mabuchi (1957: 83) himself admits that the two graphs were sometimes used interchangeably. The phonological evidence for this distinction is scant. The two graphs involved do belong to different EMC rhyme categories (*-t and *-tR; cf. OSV -o’ [RR] and -u’a [tR] ). One could reconstruct OJ oA (written with *-t) as *t and OJ oB (written with *-tR) as *R, but then one would have to propose an allophonic lowering of *t to *R when not preceded by an initial consonant. Such an awkward rule would be necessary to prevent the vowel *t in OJ kiy, giy, piy, biy, miy (*kt, *Ygt, *pt, *mbt, *mt) from lowering to *R. (The vowel *t of these syllables fronted to *i in post-OJ.) I would rather not posit such a rule. Neither would I want to try to explain the overlapping usage of
EMC *qth ‘idea’ and
EMC *qtR(h) ‘silt’ for OJ o by claiming that the zero-initial syllables *t and *R were in free variation, even though OJ *Ct and *CR syllables were not in free variation. I will continue to reconstruct only a single variety of OJ o (*R). OJ wo is spelled with two phonograms of the *-uRn rhyme category: (8.59) OJ Ph EMC OSV equivs. of rhymes Freq wo
‘long gown’ *wuRn -uôn [uRn], -edn [tRn] 110 wo
‘distant’ *wuRnq -uôn [uRn], -edn [tRn] 24 ‘avoid’ *wuRnh -uôn [uRn], -edn [tRn] Unlike EMC, OSV distinguishes between -uôn [uRn] and -edn [tRn] after labial initials. This may reflect an LOC or southern EMC distinction absent from the Qieyun ‘language.’
EMC *wuRn ‘long gown’ is homophonous with
EMC *wuRn ‘garden,’ corresponding to OSV veIn [vtRn] ‘id.’ One vietograph for veIn ‘garden’ contains
EMC *wuRn ‘long gown’ as its base (phonetic element): v. Another homophone of
EMC *wuRn ‘long gown,’
EMC *wuRn ‘gibbon,’ corresponds to OSV ve%n [vtRn] ‘id.’ with an irregular n'ng tone. All this suggests that
EMC *wuRn ‘long gown’ and its near-homophone
EMC *wuRnq/h ‘distant’/‘avoid’ were pronounced as *wtRn and *wtRnq/h. I will tentatively reconstruct the OJ syllable represented by these two characters as *wR. OJ bo appears only in forms of the verb nobor- ‘climb.’ It is spelled with a single phonogram of the *-uRn rhyme category: (8.60) OJ Ph bo
‘trouble’

EMC *buRn

OSV Freq bu4n [(uRn] ‘sad’ 6

This phonogram could conceivably represent either *mbo or *mbR. On the one hand, Ôno Tôru (1962: 51) believes that
EMC *buRm ‘all’ (cf. its homophone
EMC *buRm / OSV bu4m [(uRm] ‘sail’) was a Suiko Period phonogram for A-type pwo (*po). Since
EMC *buRn ‘trouble’ is nearly homophonous with
EMC *buRm ‘all,’ I could reconstruct OJ bo as *mbo. 252

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

On the other hand, the vowel *R is far more common than the vowel *o (Table 8.32). B-type mo (*mR) outnumbers A-type mwo (*mo) by a 3.65 : 1 (135 : 37) ratio in Kojiki. Any random OJ o is most likely to be a B-type o (*R). For the time being, I will reconstruct OJ bo as *mbR, though *mbo is also possible. OJ po is written with phonograms from four different EMC rhyme categories: (8.61) OJ Ph po
‘wealth’

po po po


‘book, origin’
‘phonogram for Indic bo’
‘trouble’

EMC *puwh < LOC *pRòhB *pwRn *bo *buRn

OSV cf. v% [vRR] ‘wife’ <
LOC *bRòqB

Freq 42

v3n [von] ‘origin’ (The OSV equiv. of this EMC rhyme is -o [QQ].) bu4n [(uRn] ‘sad’

21 1 1

On the basis of the EMC forms alone one might want to reconstruct OJ po as *po, though one could argue that
EMC *pwRn ‘origin’ < OC *pRnB and
EMC *buRn ‘trouble’ could have represented OJ *pR. Nagata (1934, 1935a, 1935b), Mabuchi (1957), and Bentley (1997) have claimed that there were two types of OJ po in Kojiki. I once rejected this claim because the EMC readings for all of the above characters contained labial vowels (*u, *o) and/or labial glides (*w). I saw no clear-cut evidence for reconstructing OJ po as *pR with a non-labial vowel *R. Yet
EMC *buwq < *bRòqB ‘wife,’ a near-homophone of the pophonogram
EMC *puwh ‘wealth,’ corresponds to LOC v% [vRR] with an unrounded vowel. Perhaps two pronunciations of OJ po existed: *pR, spelled with
LOC *pRòhB ‘wealth,’ and *po, spelled with
EMC *pwRn ‘origin.’ These two pronunciations probably did not represent two phonemically distinct syllables in eighth-century OJ. Bentley (1997: 170) thinks that the spellings of OJ po in Kojiki and Shoki reflect a distinction that was already lost “by the beginning of the seventh century (ca. 600).” Kojiki does not always preserve this distinction. For example, OJ moto posi ‘turn around’ was spelled twice with
EMC *pwRn ‘origin’ (implying A-type po = *po) as: (8.62)


! EMC *mRwq tRY pwRnq siR

(KJK 39.10, 109.4)

whereas the related verbs OJ moto pori ‘go around’ and OJ moto poro pu ‘go around’ were spelled with
LOC *pRòhB ‘wealth’ (implying B-type po = *pR) as: (8.63)


! EMC *mRw tRY puwh (LOC *pRòhB) ltq 253

(KJK 13.3)

OLD JAPANESE

(8.64)



!

EMC *mRw tRY puwh (LOC *pRòhB) ltRq poh (KJK 13.4, 34.4)

Presumably all three verbs share the same root (reconstructed by Martin 1987: 727 as *mo-topo- = my *mR-tRpR-). Bentley (1997: 133) considers the spelling in (8.62) for OJ moto posi ‘turn around’ to be a “misspelling.” The ‘correct’ spelling might have been (8.65)


! EMC *mRw tRY puwh (LOC *pRòhB ) siR

with
LOC *pRòhB ‘wealth’ instead of
EMC *pwRn ‘origin.’ Nevertheless, most OJ words containing the syllable po in Kojiki are consistently spelled either with phonograms implying *po (‘A-type’ pwophonograms) or with phonograms implying *pR (‘B-type’ po-phonograms) (Bentley 1997: 82–83). Below I list all of the po-phonograms in the poetry and prose of Kojiki along with their A/B-type classifications according to Bentley (1997: 77). Phonograms found only in the prose of Kojiki are marked with an asterisk (*). (8.66) OJ / Type pwo A pwo A po B po B po B

po B po B

Ph

EMC


‘book, origin’
‘phonogram for Indic bo’
‘wealth’

*pwRn < LOC *pRnqA (!) *bo -o [QQ]


‘trouble’
* ‘bold’ ‘place name’ ‘turn’
* ‘hedge’ ‘growth’
* ‘goods’

*puwh < LOC *pRòhB *buRn *pa *phan *phuRn *puRn *buRn *phimq < OC *phrRmqB

OSV equivs. of EMC rhymes -ôn [on]

-d [RR] -uôn, -edn [uRn tRn] -a [aa] -an [aan] -uôn, [uRn tRn] -uôn, -edn [uRn tRn] -uôn, -edn [uRn tRn] -im [im]

It is difficult to attribute the distinction between the two types of phonograms to chance. If A-type
EMC *pwRn ‘origin’ and B-type
LOC *pRòhB ‘wealth’ were truly interchangeable graphs, we would expect to see the second syllable of the high frequency OJ prefix opo- ‘large’ spelled with both graphs. But the po of OJ opo- is always spelled with B-type
LOC *pRòhB ‘wealth,’ implying that it had once been *pR. Skeptics may wish to point out that the A-type phonogram
‘origin’ belonged to the OC
*-Rn/*-Rr ‘literary’ rhyme. The medial *-w- in its 254

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

EMC reading is an innovation absent in its OC and LOC readings (*pRrqA and *pRnqA). We cannot be absolutely certain that
‘origin’ was chosen for its EMC reading *pwRnq (cf. OSV v3n [von] ) rather than its LOC reading *pRnqA. If
‘origin’ is actually a phonogram for OJ *pR, then we would have to explain the spelling distinctions between
‘origin’ and
‘wealth’ as the byproduct of arbitrary traditions which had nothing to do with phonology. The remaining A-type phonogram for OJ po,
EMC *bo ‘phonogram for Indic bo’ is so rare that it cannot be considered to be sufficient evidence for a *po distinct from *pR. It appears only once in poetry and six times in the text (I: Age of the Gods): (8.67a)
 OJ pokyi ‘pray’ (once; KJK 39.7) (8.67b)
 OJ Ame no popyi ‘name of a god’ (five times; I: Age of the Gods) (8.67c)
OJ po (once; partial gloss for the above name; I: Age of the Gods) OJ pokyi ‘pray’ is also spelled as
 with
‘origin’ (for OJ *po or *pR?) in poetry (KJK 39.8) and in the text (KJK III: Nintoku). We could then simply regard the one spelling of this word with
EMC *bo ‘phonogram for Indic bo’ as an isolated anomaly or as further evidence for reconstructing OJ pokyi ‘pray’ as *poki. The issues involved in determining whether two types of po existed in Kojiki are highly complex. I advise the interested reader to consult Bentley’s (1997) MA thesis on OJ po and mo for a more detailed discussion of the subject. I will not reconstruct a phonetic value for OJ po until I examine the Shoki data. As for the other OJ o-syllables (o, bo, wo), I have reconstructed *R for all of them, though the evidence for OJ bo (*mbR?) is open to interpretation. OJ o in Shoki corresponds to a wide variety of EMC rhyme categories. Five of these rhyme categories (*-ik, *-it, *-uRn, *-uy, *-wan) appear only once: (8.68) OJ Ph o
‘100,000’ o
‘second Heavenly Stem’ bo
‘trouble’ po
‘spend’ wo
‘alarmed’

EMC *qik *qit

CLMC Kan SK *qyoku ek [Rk] *qitu ul [tl]

*buRn *f®*buy *f®*qwanh *q-

SV ú’c [}tk] Pt [}Rt]

han hi wan

pen [pRn] phiVn [fiRn] pi phí, phh [fii] wan oán [}waan]

hai kwai

poy ?

One category appears only twice: (8.69) po
‘support’ wo
‘return’

*bwRy *®wRy

*p®*h®255

b4i [(oy] h4i [hoy]

OLD JAPANESE

Table 8.45 OJ o in Shoki poetry OJ \ LMC OJ type

*-aw A

*-tR B

*-o A

*-RY B

*-Rw A

p b m w $ Subtotals Total

13 0 2 20 17 0 0 0 8 1 28 0 30 0 84 0 0 111 0 0 0 71 0 0 0 41 71 143 28 102 423 (430 including the data in 8.69)

*-Ry B

*-uRY ?

*-uw A

*-wRY ?

4 0 0 0 0 4

0 0 15 0 0 15

0 0 14 0 0 14

0 0 0 5 0 5

Table 8.46 SX equivalents of rhyme categories for Shoki o-phonograms EMC Kan SK SV

*-aw -au -wo [o] -ao [aaw]

*-tR -yo -e [R] -e [tt]

*-o -o -wo [o] -ô [oo]

*-RY -ou -ung [tY] -#ng, -eng [aY] [tY]

*-Rw -ou -wu [u] -âu [Rw]

*-Ry -ai -ay, -oy [ay] [£y] -ai [aay]

*-uRY -(w)au -ang [aY] -uông [uRY]

*-uw -iu -wu [u] -eu [tw]

*-wRY -ou -wong [oY] -o#ng [waY]

I list all categories which appear more than once in Table 8.45. The categories in Table 8.45 correspond to the SX rhymes in Table 8.46. OJ o in Shoki is almost exclusively spelled with phonograms of the EMC *-tR category: (8.70) OJ o o o

Ph
‘in’
‘feast’
‘silt’

EMC *qtR *qtRh *qtR(h)

CLMC *q*q*q-

Kan yo yo yo

SK SV e [R] e [tt] e [R] E [tt] e [R] E [tt]

Freq 52 17 2

(Once again, I am ignoring a rare alternative reading of
, EMC *qo ‘ah!’) The SK and SV readings for these characters have unrounded vowels ( [R] and [tt] ) while the Kan-on readings have an initial y- that corresponds to zero in the other two SX branches. The SX readings may reflect front rounded vowels in different LMC dialects: (8.71) EMC LMC Borrowed as *-tR > *-ö? (north?) > SK -e [R] *-tR > *-yö? (CLMC) > Kan -yo (*yR?) *-tR > *-ü? (south?) > SV -e [tt] 256

Modern Chn reflexes Beijing Mdn yu [ü] Xi’an Mdn [ü] Cnt yu [ü]

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

I would prefer to assume that they may be near-exact copies of the original LMC finals: (8.72) EMC LMC *-tR > *-R? (north?) *-tR > *-yR? (CLMC) *-tR > *-t? (south?)

Borrowed as > SK -e [R] > Kan -yo (*yR?) > SV -e [tt]

Modern Chn reflexes Beijing Mdn yu [ü] Xi’an Mdn [ü] Cnt yu [ü]

Pulleyblank (1991a: 11) reconstructs the LMC reflex of EMC *-tR as *-iR ~ *-üR. Coblin (1994: 54, 79) reconstructs the seventh- and eighth-century CLMC reflex of EMC *-tR as a mid front rounded vowel *-ø (cf. 8.71). Amoghavajra does not use any of the Shoki phonograms for OJ o or their near-homophones in his transcriptions. The CLMC readings of these characters must not have resembled anything in Indic, which lacked front rounded vowels (ü, ö) and the diphthong [tR]. Moreover, they were borrowed as Kan-on yo, so they must not have been exact matches for OJ o. There is no syllable o in premodern Kan-on orthography. (Modern Kan-on o was once spelled as wo.) I will reconstruct the OJ syllable o as *R in Shoki as well as Kojiki. This *R is unrounded and non-low like SK e [R] and SV u’ [tt]. It is also acoustically similar to [ö]. (Cf. Vietnamese bd [(RR] < French beurre [bœ*] ‘butter.’) The two rare phonograms for OJ o in (8.68),
‘100,000’ and
‘second Heavenly Stem,’ have SK and SV readings with unrounded non-low vowels (SK e [R], SV â [R], and SV u’ [t] ) which are consistent with my reconstruction of OJ o as *R. Both appear in the beta section (NS 2.2.1, 4.1.1) and may have been chosen for their Go-on readings:
Go-on oku < OJ Go-on *Rku? (cf. SK ek [Rk], LOC *qtkB < OC *qRkB ) and
Go-on otu < OJ Go-on *Rtu? (cf. SK ul [tl], LOC *qitB < OC *qrRtB ). 95.7 per cent (111/116) of OJ wo in Shoki are written with phonograms of the EMC *-o category which corresponds to SK wo [o] and SV ô [oo]: (8.73) OJ wo wo wo wo

Ph
‘crow’
‘alas’
‘wall’
‘particle’

EMC *qo *qo *qoq *®o

CLMC *q*q*q*h®-

Kan wo wo wo ko

SK wo [o] wo [o] wo [o] hwo [ho]

SV ô [oo] ô [oo] 7 [oo] h4 [hoo]

Freq 47 34 25 6

The other phonograms for OJ wo have SV and SK readings with labial glides (SV -o- [w] ) or vowels (SK wo [o], SV ô [o] ): (8.74) OJ wo wo wo

Ph
‘great’
‘return’
‘alarmed’

EMC *®wRY *®wRy *qwanh

CLMC *h®*h®*q257

Kan kou kwai wan

SK hwoyng hwoy wan

SV ho9ng [hwaY] h4i [hoy] oán [}waan]

Freq 4 1 1

OLD JAPANESE

The Shoki evidence strongly contradicts my previous reconstruction of OJ wo as *wR. OJ wo appears to have been *wo with a rounded vowel. How can I explain the vast discrepancy in spelling between the two texts? I once hypothesized that the Kojiki orthography reflected an older *wR that no longer existed, whereas the Shoki orthography reflected the then-current pronunciation *wo. If so, one would expect OJ wo to be spelled with characters such as
EMC *wuRn ‘long gown’ (cf. OSV vu’o’n [vtRn] for its homophone
EMC *wuRn ‘garden’) in pre-Kojiki texts. However, in the Suiko Period inscriptions, OJ wo was spelled with
EMC *®o < LOC *woA ‘question particle,’ which is also a Shoki phonogram for wo (8.73). The pronunciation *wo must have existed since the early seventh century. Perhaps *wo and *wR were once in free variation but had merged into *wo by the eighth century. I will reconstruct OJ wo as *wo. OJ po, bo, mo are spelled in Shoki with phonograms implying both rounded and unrounded vowels (Table 8.47). Should I reconstruct allophonic variation between *o and *R for these syllables: i.e., *po ~ *pR, *mbo ~ *mbR, *mo ~ *mR? Or should I reconstruct a phonemic distinction between *o and *R after *p, *b, *m? In his MA thesis, Bentley (1997) has proposed that Shoki partly preserved distinctions between (1) OJ *po and *pR and (2) OJ *mo and *mR. One cannot see these distinctions if one looks at Shoki as a whole. Bentley claims that they are present only in the beta section, which is written in a relatively more conservative orthography. He thinks that these orthographic distinctions represented phonological distinctions lost in eighth-century OJ. I cannot present his entire argument here, but I will summarize the highlights. (1) Bentley (1997) assumes that there were orthographic distinctions between (a) two kinds of OJ po (our *po and *pR) and (b) two kinds of OJ mo (our *mo and *mR) in Kojiki. I have already outlined the evidence for the former distinction above. Ishizuka Tatsumaro first discovered the latter distinction in his Kanazukai oku no yamamichi (The Mountain Road into the Secrets of Kana Usage; c. 1795). It was later confirmed by Ikegami (1932) and Arisaka (1932) and has been accepted by nearly all scholars ever since. (2) Bentley (1997) examined the poems shared by Kojiki and the beta section of Shoki and found that: (a) The Kojiki phonogram for A-type pwo (*po),
EMC *pwRn ‘origin’ (cf. OSV v3n [von] ) corresponds to the Shoki beta section phonogram
EMC *pawq / CLMC *p- ‘protect’ (cf. SK pwo [po], SV bko [(aaw] ). (b) The Kojiki phonogram for B-type po (*pR),
LOC *pRòhB ‘wealth’ (cf. OSV v% [vRR] ‘wife’ <
LOC *bRòqB ‘id.’) corresponds to the Shoki beta section phonogram
EMC *bRY / CLMC *p®- ‘friend’ (cf. SK pung [ptY], SV b9ng [(aY] ).

258

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

(c) The Kojiki phonogram for A-type mwo (*mo),
EMC *maw ‘hair’ (cf. SK mwo [mo] ), corresponds to the Shoki beta section phonogram
EMC *maw / CLMC *mb- ‘hair’ (cf. SK mwo [mo], SV mao [maaw] ). (d) The Kojiki phonogram for B-type mo,
EMC *mRwq < LOC *mRòqA ‘mother,’ corresponds to three Shoki beta section phonograms (8.75) which have SV readings with unrounded vowels (â [R] and o [aw] of the final -ong [aw§] ): (8.75) OJ Ph mo B
‘lush’ mo B
‘hope’ mo B
‘mother’

EMC *mRwh *muRYh *mRwq

CLMC *mb*m(b)v*mb-

Kan bou bau bou

SK mwu mang mwu

SV Freq (β) mpu 46 vong 15 mRu 5

Only one of the SK readings has an unrounded vowel (
SK mang ‘hope’). The other two SK readings have rounded vowels (
SK mwu [mu] ‘lush’ and
SK mwu [mu] ‘mother’; the latter is read as SK mwo [mo] in modern Korean). This does not necessarily mean that the LMC readings also had rounded vowels because Korean does not permit [Rw] as a final. SK -wu [u] (-wuw in the prescriptive SK of the Tongkwuk cengwun dictionary) regularly corresponds to SV -âu [Rw] (Table 8.46). The unrounded vowels of SV and the -a- of
SK mang ‘hope’ are similar to the unrounded *R that I reconstructed for OJ B-type mo (p. 216). Bentley (1997: 30) also reconstructs “a central, unrounded vowel [R]” for OJ B-type mo. (3) Bentley regarded these correlations between Kojiki and beta section Shoki phonograms as vestiges for an earlier distinction between A-type wo (our *o) and B-type o (our *R) before OJ *p and *m. He does not think these correlations are due to chance: “To assign some amount of luck (or twisted fortuity) to the various poets and later compilers is absurd” (1997: 40). (4) The distinction between *o and *R before OJ *p and *m could not have been phonemic in the eighth century. Out of a total of 233 instances of OJ po and mo in the beta section of Shoki, Bentley (1997: 37–46, 134–136) has found that only 23 (9.9 per cent = 23/233) instances are spelled with the ‘wrong’ phonograms if one considers the Kojiki spellings to be ‘correct.’ There is no evidence for a distinction between *o and *R before OJ *p and *m in the alpha section of Shoki. Alpha section scribes used A- and B-type phonograms indiscriminately when writing OJ po and mo: (8.76) OJ sipwo ‘tide’ *sipo

Kojiki
 (108.1) A-type
EMC *pwRn ‘origin’ (cf. OSV v3n [von])

259

Shoki alpha section
=(87.1) B-type?
EMC *bRw / CLMC *p®- ‘collect’ (cf. SV biVu [(iRw])

OLD JAPANESE

opo- ‘great’ *RpR

imwo ‘beloved girl’ *imo

moto ‘base’ *mRtR


 (36.3) B-type
EMC *puwh < LOC *pRòhB ‘wealth’ (cf. OSV v% [vRR] ‘wife’
LOC *bRòqB ‘id.’)
 (8.4)


 (100.3, 101.3) A-type?
EMC *poq / CLMC *p- ‘record book’ (cf. SK pwo [po], SV ph7 [foo])
 (114.3)

A-type
EMC *maw ‘hair’ (cf. SK mwo [mo] )

B-type
EMC *mRw / CLMC *mb- ‘mother’ (cf. SV mRu [mRw] )
 (11.4)
 (114.1) B-type
EMC *mRw A-type?
EMC *mo < LOC *mRòqA ‘mother’ / CLMC *mb- ‘model’ (cf. SK mwo [mo], SV mô [moo] )

Bentley (1997: 65) regards instances in which Kojiki and Shoki alpha section spellings of mwo coincide as “chance (or ‘lucky’) etymological spelling.” The alpha section scribes were presumably using their innovative orthography to record their own phonology rather than the phonology implied by previous orthographic traditions. Their inconsistent spelling may indicate that *o and *R were in free variation after *p and *m. I will reconstruct OJ po and mo as *po and *mo, which may have phonetically varied between *[po mo] and *[pR mR]. I choose the symbol *o to represent both types of pronunciations because *R shifted to *o in post-OJ. OJ *o and *R might have been phonemically distinct after *p and *m in SJ readings: (8.77)
Kan ho < OJ Kan *po? ‘record book’ (cf. SK pwo [po], SV ph7 [foo] ) vs.
Kan hou < OJ Kan *pRw? ‘friend’ (cf. SK pung [ptY], SV b9ng [(aY] )
Kan mo < OJ Kan *mo? ‘model’ (cf. SK mwo [mo], SV mô [moo] ) vs.
Kan mo < OJ Kan *mR? ‘mother’ (cf. SV mRu [mRw] ) OJ bo is written with only three phonograms in Shoki poetry: (8.78) OJ bo bo bo

Ph
‘friend’
‘trouble’
‘collect’

EMC *bRY *buRn *bRw

CLMC *p®*p®*p®-

Kan hou han hou

SK pung [ptY] pen [pRn] pwu [pu]

SV b9ng [(aY] phiVn [fiRn] biVu [(iRw]

Freq 8 1 1

Five of the six SK and SV readings have unrounded vowels (SK u [t], e [R] and SV # [a], iê [iR] ). The rounded vowel of SK pwu [pu] for
‘collect’ probably reflects an LMC *-Rw. (Korean phonotactics forbid the final [Rw].) 260

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

Table 8.47 Common o-phonograms in Shoki poetry OJ

Ph

EMC

CLMC

Kan

SK

SV

Tib. tr.

Ind. tr. (Am)

po












*bRY *bRw *pawq *bRY *mRwh *mRwq *maw *muRYh *muw

*p®*p®*p*p®*mb*mb*mb*m(b)v*mb-

hou hou hau hou bou bou bau bau bou

b9ng biVu bko b9ng mpu mRu mao vong meu

– – – – ’bu bu, ’bu ’be’u ’phwang –

– – – – – mu, vu – – (ú)bu [(m)bu]












*moh *mo *moh *qo *qo *qoq *®o *qtR *qtRh

*mb*mb*mb*q*q*q*h®*q*q-

bo bo bo wo wo wo ko yo yo

pung pwu pwo pung mwu mwu mwo mang mwu mwo mwo mwo mwo wo wo wo hwo e e

ms mô ms ô ô 7 h4 e E

– ma [sic] – – – – – i, u u’o

bu, mo mo – – – u – – –

bo mo

wo

o

Having reconstructed OJ po, mo, wo as *po, *mo, *wo, it would be odd to reconstruct OJ bo as *mbR, the only labial-initial syllable followed by an unrounded vowel. Purely for the sake of parallelism, I will write my reconstruction of OJ bo as *mbo, though this *mbo could have been pronounced as *[mbR]. If OJ bo were *[mbo], Shoki scribes could have written it with phonograms such as
EMC *mo / CLMC *mb- ‘model’ (cf. SK mwo [mo], SV mô [moo] ). *[mbo] may have existed only in OJ period SJ readings. The OJ Kan-on reading of
‘model’ may have been *mbo (> post-OJ Kan bo), which could have contrasted with the OJ Kan-on reading of
‘mother,’ *mbRw (> post-OJ Kan bou). The Indic and Tibetan transcriptive data for o-phonograms are too scanty to be of much use. In Table 8.47, I list all o-phonograms which appear more than five times in Shoki poetry. Out of these eighteen phonograms constituting 93.7 per cent (403/430) of the total, only seven have Tibetan transcriptions and only five appear in Amoghavajra’s Indic transcriptions. Two of Bentley’s (1997) phonograms for B-type mo have interesting Tibetan transcriptions and/or Indic transcription values.
EMC *mRwq / CLMC *mb- ‘mother’ and
EMC *mRwh / CLMC *mb- ‘lush,’ are transcribed with -u in Tibetan script. Amoghavajra used the former character to transcribe Indic mu and vu. Coblin (1994: 262) has reconstructed their seventh-century CLMC readings as *mou and *mou (cf. Kan-on bou and bou). If these two characters did have rounded vowels in CLMC, then their status as phonograms for B-type mo (*mR) in the beta section of Shoki might be questionable. 261

OLD JAPANESE

Then again, how would we explain the correlations between these two phonograms and the Kojiki phonogram for mo (*mR),
‘mother’ EMC *mRwq < LOC *mRòqA? Surely the beta section scribes were aware of the Go-on readings of
‘mother’ and
‘lush’ (Go mo and mou < OJ Go *mR and *mRw? < LOC *mRòqA and EMC *mRwh). Perhaps they selected these two phonograms on the basis of these older readings without rounded vowels and disregarded the newer Kan-on readings with rounded vowels (Kan bou and bou < OJ Kan *mbow and *mbow? < CLMC *mbow and *mbowh?). The issues concerning the reconstruction of OJ po and mo are extremely complicated. I recommend those interested in the subject to consult Bentley’s (1997) MA thesis on these two syllables. For now, I will reconstruct OJ o as: 1 2 3

*R in o (which could be rewritten as B-type o ) *o in wo *R ~ *o in free variation in po, bo, and mo. If I must choose one symbol for the vowel of these three syllables, it would be *o, though there is no phonogram evidence for an *o-pronunciation of bo.

The Kojiki evidence indicates that in pre-eighth-century Japanese: (1) Pre-OJ wo may have been pronounced as *[wR] (though the Suiko Period orthography indicates that wo was already being pronounced as *[wo] as in the eighth century). (2) Pre-OJ *o and *R were phonemically distinct after *p and *m in native words. Both of these vowels merged into [o] in post-OJ.

Summary I have reconstructed the eight ‘vowel’ system for OJ in Table 8.48. The reconstructed values of four of these ‘vowels’ (iy, ye, ey, wo) do not conflict with theories of their diphthongal origins: Table 8.48 The ‘vowels’ of OJ

High Mid Low

Front

Central

Back

Diphthong

‘A-type i’ yi = *i ‘A-type e’ ye = *e

‘B-type i’ iy = *t ‘B-type o’ o = *R a = *a

u = *u ‘B-type o’ wo = *o

‘B-type e’ ey = *Ry = o + yi

262

THE RECONSTRUCTION OF OLD JAPANESE VOWELS

(8.79) Pre-OJ *ui, *Ri *ai *ia, *iR *ua, *uR

> > > >

OJ *t ( [ty]?) = iy (‘B-type i’) *Ry = ey (‘B-type e’) *e = ye (‘A-type e’) *o = wo (‘A-type o’)

I have not reconstructed any OJ C-type ‘vowels’ (i, e, o) distinct from the OJ A-type and B-type ‘vowels.’ If the OJ C-type ‘vowels’ were truly separate entities, they would have had their own unique set of correspondences with EMC rhyme categories: (8.80) OJ EMC a, u : *-a, *-uw . . . ‘A-type’ vowels: yi, ye, wo : *-i, *-ey, *-o . . . ‘B-type’ vowels: iy, ey, o : *-ty, *-Ry, *-t . . . ‘C-type’ vowels: i, e, o : *-X, *-Y, *-Z . . . But in reality their correspondences overlap with those of the OJ A and B-type ‘vowels’: (8.81) OJ a, u ‘A-type’ vowels: yi, ye, wo ‘B-type’ vowels: iy, ey, o ‘C-type’ vowels: i, e, o

: : : :

EMC *-a, *-uw . . . *-i, *-ey, *-o . . . *-ty, *-Ry, *-t . . . *-i, *-ey, *-o, *-ty, *-Ry, *-t . . .

Hence I reanalyzed the C-type ‘vowels’ as A- and B-type ‘vowels.’ Some syllables containing C-type ‘vowels’ were consistently pronounced with A- or B-type ‘vowels’ while others could be pronounced with either type of ‘vowel’ (Table 8.49). In EMJ, the eight ‘vowels’ of OJ had been reduced to five: a, i, u, e, o (*a, *i, *u, *e, *o). Jerome Packard and an unidentified graduate student at the University of Illinois at Urbana-Champaign pointed out to me during a 1999 lecture of mine that this reduction could be viewed as the result of ‘peripheralization.’ The B-type ‘vowels’ consist of two central vowels (*t and *R) and a diphthong containing a central vowel (*Ry = *Ri) whereas the remaining vowels (*a, *i, *u, *e, *o) are on the periphery of the vowel triangle (Figure 8.1). The B-type ‘vowels’ (*t, *Ry, *R) seem to have dispersed outward toward the edges of the triangle, merging with their A-type counterparts (*i, *e, *o) in the process (Figure 8.1). These shifts had already begun before the OJ period: (1) Pre-OJ B-type *t and *Ry had shifted to A-type *i and *e after coronal initials (*t, *nd, *n, *r, *s, *nz) and glides (*y and *w). A contrast between

263

OLD JAPANESE

Table 8.49 OJ syllables with C-type ‘vowels’ Type Vowel p b m w t d n r s z y $

*[p] *[mb] *[m] *[w] *[t] *[nd] *[n] *[r] *[s] *[z] *[y]

A yi *[i]

B iy *[t]

A~B yi~iy *[i t]

A ye *[e]

B ey *[Ry]

A~B ye~ey *[e Ry]

A wo *[o]

B o *[R]

A~B wo~o *[o R] po bo (?) mo

wi ti di ni ri si zi

we te de ne re se ze ye

wo

i

e

*i

o

*i

*u

*Cy *e

*C

*o

*a

Figure 8.1 The ‘peripheralization’ of the OJ B-type ‘vowels’

A-type *i, *e and B-type *t, *Ry remained only after grave initials other than *w (*p, *mb, *m, *k, *g). (2) By the eighth century, *R had assimilated to *o after the labial glide *w. *R and *o were no longer phonemically distinct after labial initials (*p, *mb, *m, *w), though the distinction remained after all other initials (*t, *nd, *n, *r, *s, *nz, *y, *k, *g). The above statements may have applied only to native words. The full paradigm of ‘vocalic’ distinctions after all initials could have persisted in OJ period Go-on and Kan-on readings. In my final chapter, I will outline my hypotheses about eighth-century Sino-Japanese phonology after reconstructing the syllabic structure of OJ.

264

CONCLUSION

9 CONCLUSION

In the previous two chapters, I have reconstructed the consonants and vowels of OJ. In (4.34), I provided a crude formulation of OJ syllable structure: (9.1) *C + ‘vowel’ I am now able to supply specific values for the variables ‘C’ and ‘vowel’: C = {*p, *mb, *m, *w, *t, *nd, *n, *r, *s, *nz, *y, *k, *Yg, *Ø} ‘vowel’ = {*a, *i, *t, *u, *e, *Ry, *o, *R} I find the formulation of OJ syllable structure in (9.1) to be somewhat awkward because it treats the diphthong *Ry as a single ‘vowel.’ I cannot reanalyze OJ syllables such as key (*kRy) as two-syllable sequences (i.e., as *kR + *i) because: (1) Most phonograms represented single OJ syllables. If OJ key were composed of two syllables, it might have been written with two phonograms as *
 EMC *ktR qyi. Yet scribes never wrote OJ key or any other ey-‘syllable’ with sequences of phonograms. (2) For the purposes of poetic meter, OJ ey-syllables were treated as single syllables and not as syllabic sequences. Instead of breaking up ey-syllables to force them into a rigid CV structure, I could rewrite (9.1) as (9.2) to accommodate them: (9.2) *CV1(V2 )

V1 = {*a, *i, *t, *u, *e, *o, *R}

V2 = {*i}

(9.2) would appear to account for a handful of native OJ words containing V1V2 combinations other than *Ri such as kai ‘oar’ and kui ‘regret.’ But these apparent CVV monosyllables are always spelled with phonogram sequences implying disyllables:

265

OLD JAPANESE

(9.3) kai ‘oar’


 EMC *kæ qyi
 EMC *®ah (< LOC *gah) qyi)

(MYS 153.9, etc.) (MYS 1052.5)

(9.4) kui ‘regret’
 EMC *kuR qyi / CLMC *k- q(NS 124.7) (cf. SK kwu i [ku i], SV cu y [kuu }ii] ) If these words were truly monosyllabic in the eighth century, we would expect them to have been transcribed with single phonograms such as
EMC *kayh ‘cover’ and
EMC *kuyh ‘valuable.’ These words cannot be used to justify (9.2). I would rather rewrite (9.2) as: (9.5) *C(G)V(G)

G = {*w, *y}

V = {*a, *i, *t, *u, *e, *o, *R}

This formulation might account for OJ period SJ syllables as well as native OJ syllables. (1) Unlike native Japanese, SJ may have allowed initial clusters of the types *Cw and *Cy, represented in post-OJ kana orthography as Cu-w- and Ci-y-. Cy clusters are still found in MSJ (e.g. Tôkyô [tookyoo] < tou kyau (to-u ki-ya-u) <
 EMC *tRwY kiRY ‘eastern capital’), though the Cw clusters are now obsolete. Some modern Japanese dialects still preserve the SJ clusters kw- and gw- (Martin 1987: 72). (2) Post-OJ SJ readings contain the diphthongs in Table 9.1. The diphthong oi occurs only in a Tô-Sô-on reading hoi (p. 231) for
EMC *bwRyh ‘dry over heat.’ This -oi entered the language centuries after the OJ period and was not part of the original system of OJ diphthongs. Did OJ period SJ lack diphthongs such as *-oi or *-Ri which would have merged into post-OJ -oi? This seems unlikely because SK, OSV, and SV have finals resembling those two diphthongs: (9.6) SK

OSV

SV

[oy] -woy
hwoy ‘turn’ -ôi
$ôi ‘double’ -ôi
h4i ‘turn’

[Qy] –

[RRy] –

[Ry] –

-oi
mii ‘every’ –

-o’i
$Ii ‘generation’ –

-ây
tây ‘west’ –

266

[£y] -oy
poy ‘multiply’ –

–

CONCLUSION

Table 9.1 Post-OJ SJ diphthongs V1 \ V2

a

i

u

e

o

a i u e o

– – – – –

ai – ui ei (oi)

au iu uu eu ou

– – – – –

– – – – –

The SX rhymes above correspond to the following OJ ‘vowels’ and SJ finals: (9.7) SK/SV OJ Go Kan

[oy] ey (*Ry) -(w)e, -(w)ai -(w)ai

[Qy] ey (*Ry) -e, -ai -ai

[RRy] ey (*Ry) -e, -ai -ai

[Ry] ey (*Ry) -e, -ai -ai

[£y] ey (*Ry) -e, -ai -ai

I believe that Go-on -e (< OJ -ey = *Ry?) is the ‘missing’ *-Ri. There is no definite evidence for an *-oi in OJ period Go-on or Kan-on. Such an *-oi would have become post-OJ -oi unless it had shifted to -we. (3) I would like to reconstruct OJ -ey = *Ry and the OJ period forms of post-OJ SJ finals such as -ai, -au, etc. as combinations of vowels and glides (/Ry ay aw/) rather than as vocalic sequences (/Ri ai au/). I do not think that the OJ period SJ readings of characters such as
Kan-on kai (cf. SK koy [k£y], SV khai [xaay] ) ‘open’ were bisyllabic like OJ kai ‘oar’ or kui ‘regret.’ Such characters were used to write Japanese monosyllables:
‘open’ was the most frequent phonogram for OJ key (*kRy). Perhaps they were also borrowed as Japanese monosyllables ending in glides (i.e., nonsyllabic vowels): the OJ Kan-on reading of
‘open’ would have been *kay which would have been distinct from disyllabic OJ kai (*kai) ‘oar.’ On p. 238, I noted that *Ry was an “aberrant diphthong” which detracted from an otherwise neat vowel system (Tables 8.32 and 8.48). The glide coda hypothesis would enable me to reanalyze *Ry as the vowel *R followed by the glide *y. *Ry would no longer be a solitary diphthong. Instead, it would be but one of many vowel–glide sequences. All of these sequences with the exception of *Ry (and *t = *[ty]?) would only be in OJ period SJ (Table 9.2). Leon Serafim (p.c.) suggested the possibility of an OJ period SJ *-tw which became postOJ SJ -iu. This final would correspond to SK -wu [u] and SV -eu [tw]. I have placed this final in parentheses since I cannot verify its existence. (4) Vowel-glide sequences may date back to pre-OJ. Pre-OJ diphthongs ending in *-i and *-u might have been VG sequences: *-ay, *-aw, etc. They may have retained their VG structure in OJ, though their vowels sometimes assimilated to the following glide:

267

OLD JAPANESE

Table 9.2 OJ period SJ vowel–glide sequences V\G

OJ *-y

*a *i *t *u *e *o *R

*ay – *ty = *t *uy *ey – *Ry

(9.8) Pre-OJ *ay > *Ry > *uy >

OJ *Ry *t (*[ty]?) *t (*[ty]?)

Post-OJ

*-w

>

-ai

> > >

-i -ui -ei

>

-e

*aw *iw (*tw *uw *ew *ow *Rw

Post-OJ > > > > > > >

-au -iu -iu ?) -uu -eu -ou -ou

(low *a raised to mid *R before *y) (mid *R raised to high *t before *y) (*u lost its rounding before *y)

OJ speakers already had indigenous *Vy sequences (*Ry and perhaps *ty), so they could have borrowed Chinese *Vy sequences such as *ay, *ey, and *uy without much effort. There may not have been any indigenous Vw sequences in OJ, so *Vw sequences were probably restricted to Chinese loans. (5) Even medial glides may have had precedents in pre-OJ. The *i and *u of pre-OJ syllables such as *Cia, *CiR, *Cua, *CuR might have been glides (*y and *w). These glides could have survived in OJ, though the phonogram evidence is ambiguous: (9.9) Pre-OJ OJ *Cya, *CyR > *Ce *Cwa, *CwR > *Co

(*e = *[ye]?) (*o = *[wo]?)

If medial *-y- and *-w- did exist in indigenous OJ words, they were predictable and non-phonemic. OJ did not have any contrasts between *Ce and *Cye or *Co and *Cwo. Nonetheless, it would not be difficult for OJ speakers to borrow Chinese medial *-y- and *-w- and pronounce them before vowels other than *e and *o. The SJ medials *-y- and *-w- must have been phonemic, since post-OJ SJ does differentiate between CV and CGV syllables: (9.10)
Kan-on ka
Kan-on kya
Kan-on kwa

< OJ Kan *ka? < OJ Kan *kya?

‘song’ (cf. SK ka, SV ca [kaa] ) ‘phonogram for Indic g(h)a, g(h)a’ (cf. SK ka, SV già [zaa] < *kyaa) < OJ Kan *kwa? ‘halberd’ (cf. SK kwa, SV qua [kwaa] )

It is even possible that *-y- was phonemic after *e in OJ period SJ: 268

CONCLUSION

(9.11)
Kan-on ken
Kan-on ken

< OJ Kan *k(y)en? ‘erect’ (Grade III) (cf. SK ken [kRn] ) < OJ Kan *kyen? ‘send’ (Grade IV) (cf. SK kyen [kyRn] )

There is no evidence for a *wo: *o distinction in MC, so I doubt that *-wwas phonemic after *o in OJ period SJ. (6) I could go even further by reformulating (9.5) as: (9.12) *C1(G)V(C2) C1 = {*p, *mb, *m, *w, *t, *nd, *n, *r, *s, *nz, *y, *k, *Yg, *Ø} G = {*w, *y} V = {*a, *i, *t, *u, *e, *o, *R} C2 = {*p, *m, *w, *t, *n, *k} (SJ only); {y} (both native words and SJ) The notion of final stops (*p, *t, *k) for OJ period SJ readings may at first seem outlandish because MSJ permits a uvular nasal [N] as a coda only in word-final position. Other codas are possible only in word-medial position: (9.13) -N#: -p-: -m-: -t-: -n-: -s-: -ã-: -k-: -Y-:

MSJ SJ MSJ SJ MSJ SJ MSJ SJ MSJ SJ MSJ SJ MSJ SJ MSJ SJ MSJ SJ












in ippai inbai ittai intai issai inja ikkai ingai

[iN] [ippai] [imbai] [ittai] [intai] [issai] [iãj_9a] [ikkai] [iYgai]

‘stamp’ ‘a cupful’ ‘prostitution’ ‘one body’ ‘seclusion’ ‘all’ ‘hermit’ ‘once’ ‘outside the Diet’

< < < < < < <
> > >

(8th c.) EMJ *-p(u) > *-φu *-k(i/u) > *-ki/ku *-t(i/u) > *-t(i/tu) *-p/k/t > *-q

LMJ (17th c.) > [u] > [ki ku] > [t]~[c\i tsu] > [t]~[c\i tsu]

MSJ (20th c.) > [T] > [ki kT] > [c\i tsT] > [c\i tsT]

I would like to posit a shift of *-p and *-k to *q because it is otherwise hard to imagine that the Japanese would have: 1 2

borrowed Chinese *-p and *-k as *-t(i/u) shifted *-p(u) and *-k(u) to *-t(u).

The issues surrounding the reconstruction of final consonants in OJ period SJ are rather complex, so I will not delve into them any further here. I realize that the proposals presented here for the reconstruction of OJ period SJ phonology are highly speculative. Nevertheless, even if one rejects my proposals for medial glides and final consonants, one must admit that eighth-century Japanese still had a large number of syllables (Table 9.3). OJ speakers could have reproduced Chinese syllables fairly well without medial glides or final consonants by combining their indigenous syllables into sequences whenever necessary: (9.18)
EMC *kiRY ‘capital’ > OJ Go *kiyau > post-OJ Go kyau (or *kyaw?) (MSJ [kyoo] ) (spelled ki-ya-u in kana)

Table 9.3 The syllabic inventory of OJ C \‘V’

*a

*i

*t

*u

*e

*Ry

*o

*R

*p *mb *m *w *t *nd *n *r *s *nz *y *k *Yg *$

*pa *mba *ma *wa *ta *nda *na *ra *sa *nza *ya *ka *Yga *a

*pi *mbi *mi *wi *ti *ndi *ni *ri *si *nzi

*pt *mbt *mt

*pu *mbu *mu

*pRy *mbRy *mRy

*kt *Ygt

*po ~ (*mbo?) *mo ~ *wo *to *ndo *no *ro *so *nzo *yo *ko *Ygo

*pR *mbR *mR

*ki *Ygi *i

*pe *mbe *me *we *te *nde *ne *re *se *nze *ye *ke *Yge

*tu *ndu *nu *ru *su *nzu *yu *ku *Ygu *u

271

*kRy *YgRy *Ry

*tR *ndR *nR *rR *sR *nzR *yR *kR *YgR *R

OLD JAPANESE

Table 9.3 has many gaps because not all of the possible combinations of consonants and ‘vowels’ existed in native words. Some syllables which would fill the gaps in Table 9.3 (e.g., *nt) might have existed only in Chinese loans: (9.19)
EMC *ãt ‘and’ > OJ Go *nt (?) > post-OJ Go ni (MSJ [nyi] ) I have only begun to explore the problem of the SJ component of OJ phonology. There are several other topics related to OJ phonology which I intend to pursue: (1) Non-standard OJ phonology. I have said very little about the Eastern (Azuma) dialects of OJ because I believe that a reconstruction of Central OJ is necessary for the reconstruction of the other dialects which were written in an orthography originally designed for Central OJ. However, now that I have reconstructed the phonetic values of Central OJ phonograms, I will be able to apply this knowledge to the reconstruction of the Eastern OJ dialects. (2) OJ pitch accent orthography. In this study I have restricted myself to reconstructing only the segmental phonology of OJ. In the future I plan to work on the suprasegmental phonology of the language by examining the Chinese tones of OJ phonograms. (3) Pre-eighth century Japanese phonology. I also plan to analyze the phonograms of pre-OJ data in Chinese histories and pre-OJ inscriptions with the aid of post-Karlgrenian Chinese reconstructions and the earlier layers of Sinoxenic. (4) The Korean peninsular origins of Japanese phonogram orthography. I intend to compare the phonograms used in Korean peninsular and Japanese transcriptions to determine the exact nature of the relationship between the writing systems of the two regions. There is almost no doubt that Japanese orthography is related to peninsular orthography, but whether the former is a direct descendant of any specific branch of the latter (e.g., Paekche orthography) or is an amalgam of several Korean orthographic systems is open to question. I will conclude with a poem (KJK 30) in which the legendary hero Yamato Takeru praises his homeland of Yamato shortly before his death. (An English translation is in Keene (1993: 49).) I can only hope that I have succeeded in resurrecting the dead sounds of Yamato:
 *yamatR





 ! pa kuni nR maporRmba



 ! *tatananduku


 ! awokaki


w

 *yamaYgRmRreru yamatR


si


 ! urupasi 272

BIBLIOGRAPHY

BIBLIOGRAPHY

Primary sources Cenwun okphyen (Jewel Book of All Rhymes) (c. 1796). 2 vols. Guang yun (Broad Rhymes) (1008). Modern edn 1972, Zhou Zumo (ed.), Taibei: Hongdao wenhua shiye youxian gongsi. Kojiki (Record of Ancient Matters) (712). Modern edn 1958, Hisamatsu Sen’ichi, et al. (eds), Kojiki taisei (The Kojiki Collection), 8 vols, Tokyo: Heibonsha. Kojiki (Record of Ancient Matters) (712). Shintô Taikei edn 1977, Onoda Mitsuo (ed.), Tokyo: Shintô Taikei Hensankai. Kojiki (Record of Ancient Matters) (712). Iwanami bunko edn 1991, Kurano Kenji, (ed.), Tokyo: Iwanami shoten. Man’yôshû (Collection of Ten Thousand Leaves) (759?). Modern edn 1963, Satake Akihiro, et al. (eds), Tokyo: Hanawa shobô. NgZ thiên t@ (Five Thousand Characters) (n.d.; earliest known edn 1909). Modern edn 1997, VZ V#n Kính and Kh7ng 1Ec (eds), Ho Chi Minh City: Nhà XuPt Bkn V#n Hóa Thông Tin. Nihongi shiki (Private Notes on the Chronicles of Japan) (812). Kokushi taikei edn 1932, Kuroita Katsumi (ed.), Tokyo: Yoshikawa Kôbunkan. Nihon shoki (Chronicles of Japan) (720). Nihon koten bungaku taikei edn 1986, Ienaga Saburô, Inoue Mitsusada, Ôno Susumu and Sakamoto Tarô (eds), Tokyo: Iwanami shoten. Tongkwuk cengwun (Correct Rhymes of the Eastern Nation) (1448). Modern edn 1988, Wenpon Tongkwuk cengwun (Original Correct Rhymes of the Eastern Nation), Seoul: Tayceykak. Tsuchihashi Yutaka and Konishi Jun’ichi (eds) (1957) Kodai kayô shû (Collection of Ancient Songs), vol. 3 of Nihon koten bungaku taikei (Outline of Japanese Classical Literature), Tokyo: Iwanami shoten. Yunjing (Mirror of Rhymes) (n.d.). Modern edn 1971, Tôdô Akiyasu and Kobayashi Hiroshi (eds), Onchû Inkyô kôhon (Perfected Text of the Yunjing with Sound Annotations), Tokyo: Mokujisha.

273

BIBLIOGRAPHY

Secondary sources Alves, Mark J. (1998) “Ruc and Other Minor Vietic Languages: Linguistic Strands between Vietnamese and the Rest of the Mon-Khmer Language Family,” unpublished manuscript. Aoki Haruo (1974) “Reconstruction and Reality: A Case of Japanese Vowels,” Acta Linguistica Hafniensia 15.1: 101–111. Arisaka Hideyo (1932) “Kojiki ni okeru mo no kana no yôhô ni tsuite” (Concerning the Usage of the Kana for mo in the Kojiki), Kokugo to kokubungaku 9.11: 74–93. —— (1934) “Kodai Nihongo ni okeru onsetsu ketsugô no hôsoku” (The Laws for the Combination of Syllables in Ancient Japanese), Kokugo to kokubungaku 11. —— (1955) Jôdai on’in kô (A Study of Ancient Japanese Phonology), Tokyo: Sanseidô. —— (1957) Kokugo on’inshi no kenkyû (Studies on the Phonological History of Japanese), rev. ed., Tokyo: Sanseidô. Asayama Nobuya (1943) “Kokugo no tô’onsetsu ni okeru dakuon ni tsuite” (On Voiced Sounds in the Initial Syllables of Japanese), Kokugo to kokubungaku 20.5: 422–432. Barat, Kahar (n.d.) “A Turkic Chinese Transcription System,” unpublished manuscript. —— (1993) “The Uygur Xuanzang Biography Volume X,” unpublished PhD dissertation, Harvard University. —— (1995) “The Uygur Xuanzang Biography IX and X,” unpublished manuscript. Barker, Muriel A. and Barker, Milton E. (1970) “Proto-Vietnamuong (Annamuong) Final Consonants and Vowels,” Lingua 24: 268–285. Baxter, William H. (1992) A Handbook of Old Chinese Phonology, Berlin and New York: Mouton de Gruyter. Bentley, John R. (1997) “Mo and Po in Old Japanese,” unpublished MA thesis, University of Hawai‘i at Manoa. —— (1998a) “Introduction,” part of unpublished manuscript of a translation of Nihon shoki. —— (1998b) “A New Look at Paekche and Korean: Data from Nihon shoki,” in Byung-Soo Park and James Hye Suk Yoon (eds) Selected Papers from the 11th International Conference on Korean Linguistics, Seoul: International Circle of Korean Linguistics. —— (2001) “The Origin of Man’yôgana,” Bulletin of the School of Oriental and African Studies 64.1: 59–73. Best, Jonathan W. (1983) “Paekche,” Kôdansha Encyclopedia of Japan, Tokyo: Kodansha. Che Qian (1994) “On Certain Tibetan and Chinese Phonological Questions of the Medieval Period,” Linguistics of the Tibeto-Burman Area 14.2: 65–92. Cheng, Robert L. and Miyake, Marc H. (1995) “A Reassessment of Jerry Norman’s Proto-Min Initials,” paper presented at the Fourth International Conference on Chinese Linguistics, Madison, Wisconsin. Choy Namsen (1915) Sin ca cen (New Character Dictionary), Keijô (Seoul): Sinmunkwan. Coblin, W. South (1991) Studies in Old Northwest Chinese, Berkeley: Journal of Chinese Linguistics.

274

BIBLIOGRAPHY

—— (1994) A Compendium of Phonetics in Northwest Chinese, Berkeley: Journal of Chinese Linguistics. Csongor, B. (1949) “Chinese in Uighur Script of the Tang Period,” Acta Orientalia Hungarica 2: 73–127. —— (1952) “Chinese in the Uighur Script of the Tang-Period,” Acta Orientalia Academiae Scientiarum Hungaricae 2: 74–118. —— (1954) “Some More Chinese Glosses in Uighur Script,” Acta Orientalia Academiae Scientiarum Hungaricae 4: 251–257. —— (1962) “Chinese Glosses in Uighur Texts Written in Brahmi,” Acta Orientalia Academiae Scientiarum Hungaricae 15: 49–53. DeFrancis, John (1978) Colonialism and Language Policy in Vietnam, The Hague: Mouton. Downer, Gordon B. (1987) “Rhymes and Rhyming in Vietnamese,” in The Chinese Language Society of Hong Kong (ed.) Wang Li Memorial Volumes: English Volume, Hong Kong: Chinese Language Society of Hong Kong. Eom Ik-sang (1994) “Aspiration and Voicing in Old Sino-Korean Obstruents,” in Young-Key Kim-Renaud (ed.) Theoretical Issues in Korean Linguistics, Stanford: Stanford Linguistics Society. —— (1998) “Different Layers of Borrowing: Sino-Korean Characters with Multiple Readings,” in Byung-Soo Park and James Hye Suk Yoon (eds) Selected Papers from the 11th International Conference on Korean Linguistics, Seoul: International Circle of Korean Linguistics. Erickson, Blaine (1998) “The Origins and Development of Japanese Mora Nasals,” unpublished PhD dissertation, University of Hawai‘i at Manoa. Frellesvig, Bjarke (1995) A Case Study in Diachronic Phonology: The Japanese Onbin Sound Changes, Aarhus: Aarhus University Press. Gage, William W. (1985) “Vietnamese in Mon-Khmer Perspective,” in Ratanakul, Suriya et al. (eds) Southeast Asian Linguistic Studies Presented to Andre-G. Haudricourt, Bangkok: Mahidol University. Gregerson, Kenneth J. (1969) “A Study of Middle Vietnamese Phonology,” Bulletin de la Société des Études Indochinoises 44.2: 131–193. Habein, Yaeko Sato (1984) The History of the Japanese Written Language, Tokyo: University of Tokyo Press. Hamada Atsushi (1952) “Hatsuon to dakuon to no sôkankei no mondai” (The Problem of the Interrelation between the Syllabic Nasal and Voiced Sounds), Kokugo kokubun 21.3: 18–32. Hashimoto, Mantarô J. (1973) The Hakka Dialect: A Linguistic Study of Its Phonology, Syntax, and Lexicon, Cambridge: Cambridge University Press. —— (1978) “Current Developments in Sino-Vietnamese Studies,” Journal of Chinese Linguistics 6: 1–26. Hashimoto, Oi-Kan Yue (1972) Phonology of Cantonese, Cambridge: Cambridge University Press. Hashimoto Shinkichi (1917) “Kokugo kanazukai kenkyûshijô no ichi hakken – Ishizuka Tatsumaro no Kanazukai oku no yama michi ni tsuite” (A Discovery in the Field of Japanese Kana Usage Research Concerning Ishizuka Tatsumaro’s The Mountain Road into the Secrets of Kana Usage), Teikoku bungaku 23.5; reprinted in Hashimoto (1949: 123–163).

275

BIBLIOGRAPHY

—— (1928) “Ha-gyô shion no hensen ni tsuite” (On the Shift of the Ha-Gyô Consonants), in Ichikawa Miyoshi (ed.) Okakura sensei kinen ronbunshû (Essays in Honor of Mr. Okakura), Tokyo: Okakura Sensei kanreki shukugakai. —— (1931) “Jôdai no bunken ni sonsuru tokushu no kanazukai to tôji no gohô” (The Special Kana Usage of Ancient Documents and the Grammar of the Time), Kokugo to kokubungaku 8.9; reprinted in Hashimoto (1949: 164–191). —— (1938) “Kokugo on’in no hensen” (Changes in Japanese Phonology), Kokugo to kokubungaku 10: 3–40. —— (1949) Moji oyobi kanazukai no kenkyû (Studies on Characters and Kana Usage), Tokyo: Iwanami shoten. —— (1950) Kokugo on’in no kenkyû (Studies on Japanese Phonology), Tokyo: Iwanami shoten. Hattori Shirô (1948) “The Relationship of Japanese to the Ryukyu, Korean, and Altaic Languages,” Transactions of the Asiatic Society of Japan (3rd series) 5: 101– 133. —— (1958) “Amami guntô no hôgen ni tsuite – Okinawa, Sakishima sho-hôgen to no hikaku” (Concerning the Dialect of the Amami Islands – A Comparison with the Dialects of Okinawa and Sakishima), Jinrui kagaku 11: 89–90. —— (1976) “Jôdai nippongo no iwayuru ‘hachi boin’ ni tsuite” (Concerning the So-called ‘Eight Vowels’ of Ancient Japanese), Nippon gakushiin kiyô 34.1: 1–16. —— (1978–79) “Nihon sogo ni tsuite” (On the Proto-Language of Japan), Gekkan gengo. Haudricourt, Andre G. (1954a) “De l’origine des tons en vietnamien,” Journal Asiatique, 242: 69–82. —— (1954b) “Comment reconstruire le chinois archaique,” Word 10: 351–364. —— (1965) “Les mutations consonantiques des occlusives initiales,” Bulletin de la Société de Linguistique de Paris 60: 160–172. Henderson, Eugenie (1985) “Feature Shuffling in Southeast Asian Languages,” in Suriya Ratanakul, et al. (eds) Southeast Asian Linguistic Studies Presented to AndreG. Haudricourt, Bangkok: Mahidol University. Igarashi Jin’ichi (1969) Jôdai kanazukai jiten (Dictionary of Ancient Kana Usage), Tokyo: Shôgakukan. Ikegami Teizô (1932) “Kojiki ni okeru kana mo mö ni tsuite” (Concerning the Kana Mo and Mö in the Kojiki), Kokugo kokubun 2.10: 138–159. Inukai Takashi (1978) “Jôdai tokushu kanazukai no hôkai katei to Kojiki no o-shi-ho no kana” (The Course of the Collapse of the Ancient Special Kana Usage and the O, Shi, Ho Kana of Kojiki), Kokugo kokubun 529: 21–38. Ishizuka Tatsumaro (1795) Kanazukai oku no yama michi (The Mountain Road into the Secrets of Kana Usage), 2 vols; reprinted in Nihon koten zenshû (Complete Collection of Japanese Classics) part 3 (1929), Nihon Koten Zenshû Kankôkai. Kamei Takashi (1956) “Ga-gyô no kana” (Kana of the Ga-Row), Kokugo to kokubungaku 39.9: 1–14. Kane, Daniel (1989) The Sino-Jurchen Vocabulary of the Bureau of Interpreters, Indiana University Uralic and Altaic Series 153, Bloomington: Indiana University. Kang Sinhang (1997) “The Vowel System of the Korean Alphabet and Korean Readings of Chinese Characters,” in Young-Key Kim-Renaud (ed.) The Korean Alphabet: Its History and Structure, Honolulu: University of Hawai‘i Press.

276

BIBLIOGRAPHY

Karlgren, Bernhard (1915–26) Études sur la phonologie chinoise, Archives d’études orientales, vol. 15 (in 4 pts), Leiden: E. J. Brill; Uppsala: K. W. Appelberg. —— (1940) Grammatica Serica: Script and Phonetics in Chinese and Sino-Japanese; reprinted from Bulletin of the Museum of Far Eastern Antiquities 12. —— (1954) Compendium of Phonetics in Ancient and Archaic Chinese, reprinted from Bulletin of the Museum of Far Eastern Antiquities 26: 211–367. —— (1957) Grammatica Serica Recensa, reprinted from Bulletin of the Museum of Far Eastern Antiquities 29: 1–332. Keene, Donald (1993) Seeds in the Heart: Japanese Literature from Earliest Times to the Late Sixteenth Century, New York: Henry Holt and Company. Kikusawa Sueo (1935) Kokugo on’in ron (Essays on Japanese Phonology), Tokyo: Kenbunkan. Kim Wancin (1963) “Moum cheykyey uy sin kochal” (A New Study of the [Korean] Vowel System), Cintan hakpo 24: 63–99. Kim-Renaud, Young-Key (ed.) (1997) The Korean Alphabet: Its History and Structure, Honolulu: University of Hawai‘i Press. Kindaichi Kyôsuke (1938) Kokugo on’in ron (Essays on Japanese Phonology), New rev. and expanded edn, Tokyo: Tôkô shoin. Kiyose, Gisaburô N. (1991) Japanese Linguistics and Altaic Linguistics, Tokyo: Meiji shoin. Kobayashi Akemi (1981) “Ennin no kijutsu suru Sansukuritto onsetsu ‘ca’ no onka: kyû-seiki Nihongo suitei no kokoromi” (The Sound Value of the Sanskrit Syllable ‘ca’ as Described by Ennin: An Attempt to Infer Ninth Century Japanese), Ôsaka gaikokugo daigaku gakuhô 52: 63–80. Kôno Rokurô (1939) “Chôsen kanjion no ichi tokushitu” (A Unique Feature of Korean Sinographic Pronunciation), Gengo kenkyu 3. —— (1964–67) “Chôsen kanjion no kenkyû” (Research on Korean Sinographic Pronunication), Chôsen gakuhô 31–44. Koyama Tadashi (1956) Ishizuka Tatsumaro no kenkyû (The Studies of Ishizuka Tatsumaro), Nagoya: Koyama Tadashi Kôen-kai. Lange, Roland A. (1968) “The Phonology of 8th Century Japanese,” unpublished PhD dissertation, University of Michigan. —— (1973) The Phonology of Eighth-Century Japanese: A Reconstruction Based upon Written Records, Tokyo: Sophia University Press. Lee, Haewoo (Yi Haywu) (n.d.) Unpublished computer database of Late Middle Chinese, Mandarin, Literary Taiwanese, Colloquial Taiwanese, Cantonese, Suzhou, Sino-Korean, Kan-on, and Go-on. —— (1994) “The Origin of Sino-Korean,” Korean Linguistics 8: 207–222. Lee, Ki-baik (Yi Kipayk) (1980) “History of Korea” (“The Dawn of History”; “Development of Ancient States”), Encyclopaedia Britannica. Lewin, Bruno (1973) “Japanese and the Language of KoguryÉ,” in Richard B. Mather (ed.) Papers of the C.I.C. Far Eastern Language Institute, Volume IV, Ann Arbor: Panel on Far Eastern Language Institutes of the Committee on Institutional Cooperation. Li Fang-kuei (1971) “Shanggu yin yanjiu” (Studies on Archaic Chinese Phonology), Tsinghua Journal of Chinese Studies 9: 1–61. —— (1977) Handbook of Comparative Tai, Honolulu: University of Hawai‘i Press.

277

BIBLIOGRAPHY

Li Rong (1952) Qieyun yinxi (The Sound System of Qieyun), Beijing: Zhongguo kexueyuan. Liu Guanghe (1984) “Tangdai ba shiji Chang’an yin shengniu” (The Sounds and ‘Knots’ of Tang Dynasty 8th Century Chang’an Pronunciation), Yuwen yanjiu 3: 45–50. Luo Changpei (1933a) Tang Wudai xibei fangyin (The Dialectal Pronunciation of the Northwest in the Tang Dynasty and Five Dynasties Periods), Shanghai: Academia Sinica. —— (1933b) “Shi nei wai zhuan” (Explaining the Inner-Outer Turnings), Bulletin of the Institute of History and Philology, Academia Sinica 4: 209–226. Lyu Lyel (1983) Sey nala siki uy litwu ey tayhan yenkwu: Salam, pyesul, kocang ilum uy pyoki lul thonghaye (Research on the Itwu of the Three Kingdoms Era, Covering the Transcription of Personal Names, Ranks, and Regional Toponyms), Pyongyang: Kwahak paykkwa sacen chwulphansa. Mabuchi Kazuo (1957) “Kojiki no ‘si’ ‘o’ ‘ho’ no kana” (The Kana ‘si,’ ‘o,’ ‘ho’ in Kojiki), Kokugogaku 31: 61–90. —— (1959) “Jôdai, chûko ni okeru sa-gyô tô-on no onka” (The Sound Value of the Sa-gyô Initial in Ancient and Medieval Times) Kokugo to kokubungaku 36.1: 60–67. —— (1983) “On’inshi o dô toraeru ka” (How Do We Deal with Phonological History?), Nihongogaku 2.11: 4–11. Malmqvist, N. G. D. (1973) “The Chieyunn Problem Re-examined,” in Richard B. Mather (ed.), Papers of the C.I.C. Far Eastern Language Institute, Volume IV, 10– 18. Ann Arbor: Panel on Far Eastern Language Institutes of the Committee on International Cooperation. Manomaivibool, Prapin (1976) “Chinese and Thai: Are They Related Genetically?” Computational Analyses of Asian and African Languages 6: 11–32. Martin, Samuel E. (1953) “The Phonemes of Ancient Chinese,” Journal of the American Oriental Society 73.2. —— (1987) The Japanese Language through Time, New Haven: Yale University Press. —— (1991) “Recent Research on the Relationships of Japanese and Korean,” in Sydney M. Lamb and E. Donglas Mitchell (eds), Sprung From Some Comman Source, pp. 269–292, Stanford: Stanford University Press. —— (1992) A Reference Grammar of Korean: A Complete Guide to the Grammar and History of the Korean Language, Rutland: Charles E. Tuttle Company. —— (1997) “Commentary,” in Young-Key Kim-Renaud (ed.), The Korean Alphabet: Its History and Structure, 263–277, Honolulu: University of Hawai‘i Press. —— (1998) “How Has the Korean Vowel System Changed through Time?” E-mail version of paper read at the 11th International Conference on Korean Linguistics. Maruyama Tôru (1981) “Chûsei Nihongo no sa-gyô shiin: Rodorigesu no kijutsu o megutte” (The Sa-Gyô Consonant in Middle Japanese: Concerning Rodrigues’ Orthography), Kokugogaku 124: 10–18. Maspero, Henri (1912) “Études sur la phonétique de la language annamite, les initiales,” Bulletin de l’École Française d’Extrême-Orient 12.1: 1–127. —— (1920) “Le dialecte de Tch’ang-ngan sous les T’ang,” Bulletin de l’École Française d’Extrême-Orient 20: 1–124. Mathias, Gerald B. (1993) Ojlexico, unpublished computer database of the lexicon of Old Japanese.

278

BIBLIOGRAPHY

Matsumura Akira (1972) Kokugoshi gaisetsu (Outline of Japanese Language History), Tokyo: Shûei shuppan. Miller, Roy Andrew (1967) The Japanese Language, Chicago: The University of Chicago Press. Miner, Earl, Odagiri Hiroko, and Morrell, Robert E. (1985) The Princeton Companion to Classical Japanese Literature, Princeton: Princeton University Press. Mineya Tôru (1972) Etsunan kanjion no kenkyuu (Research on Vietnamese Sinographic Readings), Tokyo: Tôyô bunko. Miyake, Marc Hideo (1995) “The Phonetic Values of Old Japanese */s/ and */z/,” unpublished MA thesis, University of Hawai‘i at Manoa. —— (1997) “Early Chinese Loanwords in Korean and Japanese: Reexamining an Old Problem from a Modern Perspective,” in Ho-min Sohn (ed.) Japanese/Korean Linguistics 6, Stanford, California: Center for the Study of Language and Information Publications. —— (1998) “Sino-Korean Evidence against the Korean Vowel Shift,” presentation given at the University of Hawai‘i at Manoa Linguistics Department Tuesday Seminar. —— (1999) “The Phonology of Eighth Century Japanese Revisited: Another Reconstruction Based upon Written Records,” unpublished PhD dissertation, University of Hawai‘i at Manoa. —— (forthcoming) “Philological Evidence for *e and *o in Pre-Old Japanese,” Diachronica. Miyake Takeo (1933) Kamazukai no keukyû (Studies in Kana Usage), Tokyo, Meiji Shoin. Mizutani Shinjô (1956) Title unknown, paper read at the Japanese Sinological Association Conference, quoted in Tôdô (1980: 280–281). —— (1960) “Bongo ‘sorishita’ boin o arawasu kanji” (Sinographs Representing Sanskrit Retroflex Vowels), Gengo kenkyû 37: 45–55. Mori Hiromichi (1991) Kodai no on’in to Nihon shoki no seiritsu (Old Japanese Phonology and the Formation of the Nihon Shoki), Tokyo: Taishûkan shoten. Morimoto Takeyoshi (1933) Man’yôshû yôjihô gaisetsu (Outline of Character Usage in Man’yôshû), vol. 3 of Man’yôshû kôza (Lectures on Man’yôshû), Tokyo: Shun’yôdô. Morohashi Tetsuji (1955–60) Dai kanwa jiten (Great Sino-Japanese Dictionary), 13 vols, Tokyo: Taishûkan shoten. Morohashi Tetsuji, et al. (1981–82) Kô kanwa jiten (Broad Sino-Japanese Dictionary), 4 vols, Tokyo: Taishûkan shoten. —— (1992) Taishûkan shin kanwa jiten (Taishûkan New Sino-Japanese Dictionary), 3rd rev. edn, Tokyo: Taishûkan shoten. Nagata Kichitarô (1932) “Kodaigo no boin” (Vowels of Ancient Japanese), Onsei gakkai kaihô 7.1: 7–8. —— (1934) “Kojiki ni okeru shi-o-ho no mojizukai ni tsuite” (On the Use of Characters for shi, o, ho in Kojiki), Kokugo to kokubungaku 127: 91–102. —— (1935a) “Jôdai kokugo ni okeru o-retsu on no honshitsu” (The True Character of the Sounds of the O-Row Syllables in Ancient Japanese), Kokugo to kokubungaku 129: 94–112; 130: 71–96. —— (1935b) “Kojiki no ho no kana ni kan suru zakkô” (Thoughts on the Kana for ho in Kojiki), Kokugo to kokubungaku.

279

BIBLIOGRAPHY

Nam Kwangwu (1973) Cosen Yico hancaum yenkwu (Studies on the Sinographic Readings of Yi Dynasty ChosÉn), Seoul: Ilcokak. NguyWn Dinh Hòa (1987) “Vietnamese,” in Bernard Comrie (ed.) The World’s Major Languages, 777–796, Oxford: Oxford University Press. —— (1992) “Graphemic Borrowings from Chinese: The Case of Chw Nôm – Vietnam’s Demotic Script,” Bulletin of the Institute of History and Philology, Academia Sinica 61.2: 383–432. —— (1997) Vietnamese: TiUng Viqt Không Son PhPn (The Vietnamese Language Unadorned ), Amsterdam: John Benjamins. NguyWn Ngic San (1987) “Chw nôm và v#n-bkn chw nôm” (Chw nôm and writings in Chw nôm) in Lê Trí ViWn (ed.-in-chief ) Cd-sN Ngw-v#n Hán-Nôm, Tpp IV, PhQn 2 (Basics of Chinese-Nôm Language and Script, Collection IV, Part 2), Hanoi: Giáo-djc. Norman, Jerry (1973) “Tonal Development in Min,” Journal of Chinese Linguistics 1.2: 222–238. —— (1974) “The Initials of Proto-Min,” Journal of Chinese Linguistics 2.1: 27–36. —— (1988) Chinese, Cambridge: Cambridge University Press. —— (1994) “Pharyngealization in Early Chinese,” Journal of the American Oriental Society 114: 397–408. Norman, Jerry L., and Coblin, W. South (1995) “A New Approach to Chinese Historical Linguistics,” Journal of the American Oriental Society 115: 397–408. Oh Sangsuk (O Sangsek) (1998) “The Korean Vowel Shift Revisited,” Ehak yenkwu 34.2: 445– 463. Okimori Takuya (1989) Nihongo shi (History of the Japanese Language), Tokyo: Ôfûsha. Ôkubo Tadashi and Ôno Susumu (eds) (1976) Motoori Norinaga Zenshû (The Complete Motoori Norinaga Collection), 23 vols, Tokyo: Chikuma Shobô. Omodaka Hisataka (1967) Jidai-betsu kokugo daijiten: jôdai hen (Great Japanese Dictionary Divided by Period: Ancient Period ), Tokyo: Sanseidô. Ômori Akihisa, et al. (1982) Man’yôdô shirube (Guide to the Way of the Ten Thousand Leaves), Osaka: Izumi shoin. Ôno Susumu (1953a) Jôdai kanazukai no kenkyû: Nihon shoki no kana o chûshin to shite (Research on Ancient Kana Usage Focusing on the Kana of the Nihon shoki), Tokyo: Iwanami shoten. —— (1953b) Nihongo no Kigen (Origins of the Japanese Language), Tokyo: Iwanami shoten. —— (1975) “Kodai Nihongo to Chôsengo” (Ancient Japanese and Korean), in Ôno Susumu (ed.) Nihon kodaigo to Chôsengo (The Ancient Language of Japan and Korean), Tokyo: Mainichi shinbunsha. —— (1982) Kanazukai to jôdaigo (Kana Usage and Ancient Japanese), Tokyo: Iwanami shoten. Ôno Susumu, Satake Akihiro, and Maeda Kingorô (eds) (1990 [1974]) Iwanami kogo jiten (Iwanami Classical Japanese Dictionary), rev. edn, Tokyo: Iwanami shoten. Ôno Tôru (1962) Man’yôgana no kenkyû (Studies on Man’yôgana), Tokyo: Meiji shoin. Osada Natsuki (1972) Genshi Nihongo kenkyû: Nihongo keitôron e no kokoromi (Research on Primeval Japanese: An Attempt at Discussing the Origins of Japanese), Kôbe: Kôbe gakujutsu shuppan.

280

BIBLIOGRAPHY

Pak Pyengchay (1971) Kotay kwuke uy yenkwu (Studies of Ancient Korean), Seoul: Kolye tayhakkyo chwulphanpu. Pierson, Jan L. (1929) “On the Transliteration and Transcription of the Japanese kana, Archaic, Ancient, and Modern,” Transactions of the Asiatic Society of Japan, second series, VI: 103–144. Pulleyblank, Edwin G. (1962) “The Consonantal System of Old Chinese,” Asia Major 9: 58–144, 206–265. —— (1963) “An Interpretation of the Vowel Systems of Old Chinese and Written Burmese,” Asia Major 10: 200–221. —— (1965) “The Transcription of Sanskrit k and kh in Chinese,” Asia Major 11: 199–210. —— (1970–71) “Late Middle Chinese,” Asia Major 15: 197–239; 16: 121–168. —— (1977–78) “The Final Consonants of Old Chinese,” Monumenta Serica 33: 180–206. —— (1981) “Some Notes on Chinese Historical Phonology,” Bulletin de l’École Française d’Extrême-Orient 69: 277–88. —— (1983) “Stages in the Transcription of Indian Words in Chinese from Han to Tang,” in Klaus Röhrborn and Wolfgang Veenker (eds) Sprachen des Buddhismus in Zentralasien: Vorträge des Hamburger Symposions vom 2. Juli bis 5. Juli 1981, Wiesbaden: In Kommission bei O. Harrassowitz. —— (1984) Middle Chinese: A Study in Historical Phonology, Vancouver: University of British Columbia Press. —— (1991a) Lexicon of Reconstructed Pronunciation in Early Middle Chinese, Late Middle Chinese, and Early Mandarin, Vancouver: University of British Columbia Press. —— (1991b) “The Ganzhi as Phonograms and Their Application to the Calendar,” Early China 16: 39–80. —— (1993) “Old Chinese Phonology: A Review Article,” Journal of Chinese Linguistics 21.2: 337–380. —— (1994) “The Old Chinese Origin of Type A and B Syllables,” Journal of Chinese Linguistics 22.1: 73–99. —— (1998) “Qieyun and Yunjing: The Essential Foundation for Chinese Historical Linguistics,” Journal of the American Oriental Society 118: 200–215. Ramsey, S. Robert (1987) The Languages of China, Princeton: Princeton University Press. Ramsey, S. Robert and J. Marshall Unger (1972) “Evidence for a Consonant Shift in 7th Century Japanese,” Papers in Japanese Linguistics 1.2: 278–295. Ray, David Tryon (1979) “Sources of Middle Chinese Phonology: A Prolegomenon to the Study of Vietnamized Chinese,” unpublished MA thesis, Southern Illinois University at Carbondale. Ryu Min-hwa (Yu Minhwa), 1994, “Nihon shoki kisai no Chôsen koyûmei hyôki – honbun no hyôkiji o taishô to shite” (The Orthography of Korean Proper Nouns Recorded in Nihon Shoki – Concentrating on the Orthography of the Main Text), Chôsen Gakuhô 153: 157–213. Sandness, Karen (1985) “The Pronunciation of the S-Initial Syllables in the Nara Period: A Critical Look at Arisaka’s and Mabuchi’s Theories,” Papers in East Asian Languages 3: 1–16.

281

BIBLIOGRAPHY

Schuessler, Axel (1987) A Dictionary of Early Zhou Chinese, Honolulu: University of Hawai‘i Press. Seeley, Christopher (1979) “Sangumi no kana i-wi, e-we, wo-o ni taisuru Keichû no kannen ni tsuite” (Concerning Keichû’s Notions of the Three Kana Sets i-wi, e-we, wo-o), Gobun 36. —— (1991) A History of Writing in Japan, Leiden: E. J. Brill. Serafim, Leon A. (1994) “A Modification of the Whitman Proto-Koreo-Japonic Vocalic Hypothesis,” Korean Linguistics 8: 181–205. —— (1995) “Evidence for Pre- or Proto-Historic Loans into Japanese from a Related Language of the Korean Peninsula,” unpublished conference paper proposal. —— (1999a) “Why Proto-Japonic Had at Least Six, Not Four Vowels,” presentation given at the University of Hawai‘i at Manoa Linguistics Department Tuesday Seminar. —— (1999b) “Reflexes of Proto-Koreo-Japonic Mid Vowels in Japonic and in Korean,” paper presented at ICHL XIV, Workshop on Korean-Japanese Comparative Linguistics. Shôgaito, Masahiro (1995) “Uiguru moji onsha sareta kango butten danpen ni tsuite (Uiguru kanji on no kenkyuu)” (On the Chinese Buddhist Scriptures Transcribed in Uighur Script (Research on Uighur Sinographic Readings)), Gengo kenkyuu 14: 65–153 + 10 pages of photos without pagination. —— (1997) “Uiguru moji onsha sareta kango butten danpen ni tsuite (Uiguru kanji on no kenkyuu) (Zoku)” (On the Chinese Buddhist Scriptures Transcribed in Uighur Script (Research on Uighur Sinographic Readings) (Continued)), Seinan Ajia Kenkyuu 46: 1–31. Shorto, Harry L. (1976) “The Vocalism of Proto-Mon-Khmer,” in Philip N. Jenner, et al. (eds) Austroasiatic Studies, vol. 2, Honolulu: University of Hawai‘i Press. Sin Yongtay (1982) Wensi Hanile uy yenkwu (Studies on Primeval Korean and Japanese), Seoul: Tongkwuk tayhakkyo chwulphanpwu. Sinor, Denis (1939) “A propos de la biographie ouigoure de Hiuan-tsang,” Journal Asiatique 231: 543–590. Starostin, Sergei (1989) Rekonstrukcija drevnekitajskoj fonologicheskoj sistemy (A Reconstruction of the Old Chinese Phonological System), Moscow: Nauka. Takayama Michiaki (1981) “Gen’on seichô kara mita Nihon shoki ongana hyôki shiron” (An Essay on the Ongana Orthography of Nihon Shoki from the Point of View of the Tones of the Characters’ Original Readings), Kyûshû daigaku gobun kenkyû 51: 13–20 + charts. —— (1982) “Shoki kayô ongana to gen’on seichô” (Shoki Poetry Phonograms and The Tones of Their Original Readings), Kyûshû daigaku bunken tankyû 10: 1–6. —— (1983) “Shoki kayô nionsetsu-meishi no hyôki ni tsuite – akusento gorui to no kanren o megutte” (Concerning the Orthography of Disyllabic Nouns in Shoki Poetry and Its Relationship to Accent Classes), Kyûshû daigaku bunken tankyû 12: 47–55. —— (1984) “Nihon shoki no ongana to sono gen’on seichô ni tsuite – jôdai akusento to no sôkansei o kangaeru” (Concerning the Ongana of Nihon Shoki and the Tones of Their Original Readings – Thinking about Their Mutual Relation to the Accent of Ancient Japanese), Kindaichi Haruhiko hakase koki kinen ronbunshû 1: 45–69.

282

BIBLIOGRAPHY

Thompson, Laurence C. (1976) “Proto-Viet-Muong Phonology,” in Philip N. Jenner, et al. (eds), Austroasiatic Studies, vol. 2, Honolulu: University of Hawai‘i Press. —— (1987) A Vietnamese Reference Grammar, Honolulu: University of Hawai‘i Press. Ting Pang-hsin (1972) “Chinese Phonology of the Wei-Chin Period: Reconstruction of the Finals as Reflected in Poetry,” unpublished PhD dissertation, University of Washington. Tôdô Akiyasu (1957) Chûgokugo on’inron (Essays on Chinese Phonology), Tokyo: Kônan shoin. —— (1978) Gakken Kanwa Daijiten (Gakken Great Sino-Japanese Dictionary), Tokyo: Gakushû kenkyûsha. —— (1980) Chûgokugo on’inron: sono rekishiteki kenkyû (Essays on Chinese Phonology: Its Historical Study), Tokyo: Kôseikan. Toh Soo-hee (To Swuhuy) (1977) Paykceye yenkwu (Research on the Paekche Language), Seoul: Paykcey munhwa kaypal yenkwuwen. —— (1981) “The Paekche Language,” Chôsen gakuhô 98: 21–32. —— (1984) “A Study of Place Names of Paekche,” Chôsen gakuhô 113: 1–55. —— (1985) “Paykcey cenkie wa Kala-e uy kwankyey” (The Relationship between the Early Paekche Language and the Language of Kara), Hankul 187: 49–81. —— (1987) Paykceye yenkwu I (Research on the Paekche Language I ), Seoul: Paykcey munhwa kaypal yenkwuwen. Tsukishima Hiroshi (1977) Kokugo no rekishi (History of the Japanese Language), Tokyo: Tôkyô daigaku shuppan-kai. Unger, J. Marshall (1971) “A Note on Old Japanese Verb Inflection,” Papers from the 7th Regional Meeting of the Chicago Linguistic Society, Chicago: Chicago Linguistic Society. —— (1993 [1977] ) Studies in Early Japanese Morphophonemics, Bloomington: Indiana University Linguistics Club. Vance, Timothy J. (1985) “The Development of Chinese Syllable-Final Stops in Sino-Japanese,” Papers in East Asian Languages 3: 48–62. —— (1987) An Introduction to Japanese Phonology, Albany: State University of New York Press. Varley, H. Paul (1984) Japanese Culture, 3rd edn, Honolulu: University of Hawai‘i Press. Vovin, Alexander (1993) “On the Phonetic Value of the Middle Korean Grapheme [Triangle],” Bulletin of the School of Oriental and African Studies 56.2: 247–259. —— (1996a) “Proto-Japanese-Korean: Fact or Fiction?,” presentation given to the Koreo-Japonic Circle, University of Hawai‘i at Manoa. —— (1996b) “Korean Historical Phonology in a Nutshell,” class handout for EALL 750 class, University of Hawai‘i at Manoa. —— (1996c) “Japanese Historical Phonology in a Nutshell,” class handout for EALL 750 class, University of Hawai‘i at Manoa. —— (1998) “On the Great Vowel Shift in Middle Korean and Position of Stress in Proto-Korean,” paper presented at the 11th International Conference on Korean Linguistics, 1998. Wang Li (1958) Hanyu shi lunwen ji (Collection of Essays on the History of the Chinese Language), Beijing: Kexue chubanshe. Wenck, Günther (1954–57) Japanische Phonetik, 4 vols, Wiesbaden: Otto Harassowitz.

283

BIBLIOGRAPHY

Whitman, John B. (1985) “The Phonological Basis for the Comparison of Japanese and Korean,” unpublished PhD dissertation, Harvard University. Yamada Yukihiro (1983) “Tosa hôgen sa-gyô shiin to jôdai sa-gyô shiin” (The Sa-gyô Consonant in the Tosa Dialect and the Old Japanese Sa-gyô Consonant), Kokugogaku 133: 1–10. Yamaguchi Yoshinori (1971) “Kodai Nihongo ni okeru boin datsuraku no genshô ni tsuite” (Concerning the Phenomenon of Vowel Loss in Ancient Japanese), Kokugogaku 85: 1–17. —— (1974) “Kodai Nihongo ni okeru gotô shi’in no datsuraku” (The Loss of WordInitial Consonants in Ancient Japanese), Kokugogaku 98: 1–15. Yasuda Kiyomon (1934) “Jôdai tokushu kanazukai no on’inteki kiso” (The Phonological Basis for Ancient Special kana Usage), Nihon bungaku 4.8: 1–12. Yi Kimun (Lee Ki-moon) (1961) Kwukesa kaysel (Outline of Korean Language History), Seoul: Mincwung sekwan. Yoshitake Saburô (1934) The Phonetic System of Ancient Japanese, London: Royal Asiatic Society. Yu Changkyun (1991) Samkwuk sitay uy hancaum (The Sinographic Pronunciation of the Three Kingdoms Period ), Seoul: Minumsa. Yuchi Zhiping (1982) “Zhou-Sui Chang’an fangyin chutan” (A First Examination of the Zhou-Sui Chang’an Dialect), Yuyan yanjiu 2: 18–33. —— (1985) “Lun Sui-Tang Chang’an-yin he Luoyang-yin de shengmu xitong – jian da Liu Guanghe he tongzhi” (A Discussion of the Initial Consonantal System of the Sounds of Sui-Tang Chang’an and of Luoyang – and an Answer to Liu Guanghe and Comrades), Yuyan yanjiu 2: 38–48.

284

INDEX

INDEX

Cantonese xv, 113, 127, 162, 224–5, 227–9, 246, 250, 256–7 Central Old Japanese see Old Japanese Cenwun okphyen xviii, 114, 160, 207, 244 Chang’an xvi, 23, 30–1, 91, 99, 105–6, 113, 120, 137, 144, 225, 240 Chang’an Early Middle Chinese 137 Chang’an Late Middle Chinese 30–2, 150–6 Chaozhou 213 Chephay sine 74, 85, 88, 269 chu nôm see nôm Coblin, W. S. 31–2, 93, 97–8, 105–6, 135, 137, 144, 150–6, 166, 171, 224, 231, 257, 261 Csongor, B. 137, 187 C-type ‘vowels’ 53, 81–4, 163, 198, 238–64

Achiki 9, 104, 109 a.chung 155, 185 alpha section 57–61, 106, 160, 165, 167, 181, 190–1, 215, 259–60 Ametsuchi no kotoba 47 Amoghavajra 137, 171, 201–3, 206, 211, 216, 219–20, 226, 230–1, 240–3, 247, 249–50, 257, 261 Aoki H. 65 Arabic 137 Arisaka H. 55, 61–3, 65, 112–13, 157, 258 arithmogram 27 Asayama N. 56 A-type syllable see Type A/B distinction A-type ‘vowels’ 50–7, 61–5, 79–84, 159, 163, 198, 203–7, 221–7, 232–8, 262–4 Azuma Old Japanese 159, 272 Barat, K. 135, 137, 155, 161, 177, 187 Barker, M. A. 122–3 Barker, M. E. 122–3 Baxter, W. H. 7, 63, 92, 135, 138–40, 144–6 Bentley, J. R. 4, 9, 19, 28, 30, 37–8, 40–4, 46–8, 61, 82, 110–11, 114, 180, 182–3, 198, 239, 251, 253–5, 258–61 Best, J. W. 9, 30, 109 beta section 57–60, 106, 160, 165, 167, 215, 229–30, 257–9, 262 Brahmi 137 B-type syllable see Type A/B distinction B-type ‘vowels’ 50–7, 61–5, 79–84, 159, 163, 198, 211–20, 227–31, 262–4 Burmese 135 Bussokuseki no uta 19–20

DeFrancis, J. 118 denasalization 31 Dharmaksema 245–6 Downer, G. B. 122, 132–4, 157 Early Middle Chinese 90–1, 93–5, 98, 143–9; see also Chang’an Early Middle Chinese Early Middle Japanese 67, 74, 76–7, 84–5, 102, 237–8, 271 Eastern Han 90, 109, 131, 141–2, 157, 176 Eastern Old Japanese see Azuma Old Japanese Engishiki 21 Eom I. 110–12, 190 Erickson, B. 74, 88, 193 Eta Funayama inscription 10, 12, 19

285

INDEX

feature shuffling 123 Frellesvig, B. 74, 193 French 257 fudoki 19 Gage, W. W. 119–23, 127, 157 Gishi see Wei zhi gisho 26–7, 159 goals 158 Go-on 12, 14, 21, 36, 40, 44, 89, 95, 100–9, 129–30, 134, 144–8, 171, 189, 193, 197, 205, 215, 218, 224–5, 231, 234, 245, 247, 257, 264, 267, 271 graphemogram 26–7 Gregerson, K. J. 119, 157 Guangyun 93–5, 97–8, 105, 171, 199 Habein, Y. S. 18, 21, 27, 44 Hakka 113, 162, 224–5 Hamada A. 56, 193 Han 105–6 hankul 113–17, 233 Han shu 106, 154 Hashimoto, M. J. 127–8, 162 Hashimoto, O. Y. 162 Hashimoto S. 48–53, 62, 65, 70 Hattori S. 55–6, 62, 67, 68, 79 Haudricourt, A. G. 121–2, 138 Heian 3 Henderson, E. 123 Himiko 8, 11, 41 Hmong-Mien 140 Hua-Yi yiyu 7 hyangchal 111, 129 ideograph 5 Igarashi J. 17, 40, 61, 72 Ikegami T. 61, 258 Ilopha 52 Inariyama inscription 10, 12, 19, 72 Indic 148, 151, 153, 191, 203, 206, 211, 214, 218–20, 222, 224, 226, 228, 230–1, 235, 240–3, 245–7, 249–50, 257, 261; see also Prakrit, Sanskrit Inukai T. 61 Iroha 47 Ishizuka T. 47–50, 62, 258 itwu 111 Iwabuchi E. 49, 51 Japonic 66 Jiankang 104, 120

Jiaozhi 117, 127, 130, 157 Jiaozhou see Jiaozhi jindai moji 8 Jingdian shiwen 95 Jnanagupta 137, 200–1, 205, 208, 214, 218–19, 222, 224, 228 Jôgûki 41 Kaifûsô 21 Kakyô hyôshiki 21 Kamei T. 193 kana 43–4 Kanabukuro 49 Kanazukai oku no yamamichi 47, 258 Kane, D. 7 Kang S. 114–15 Kangxi zidian 114 Kan-on 14, 31–2, 35, 40, 44, 89, 95, 100–9, 112–13, 129–30, 134–5, 137, 150, 155, 171, 186, 189, 202, 206, 211, 215–16, 219, 223–6, 229–31, 234–6, 240, 247–9, 255–7, 259–61, 264, 267–9; see also New Kan-on Karlgren, B. 52–3, 57, 63–5, 90–5, 97, 99, 111–12, 138–9, 143–5, 150–1, 153, 272 Kaya 109 Keene, D. 1, 4, 18–21, 272 Keichû 45–7, 61 Khmer 123 Kikusawa S. 55–6, 62 Kim W. 116 Kim-Renaud, Y. 157 Kindaichi K. 55, 62 Kiyose, G. N. 55–6, 62–3, 74, 111, 157, 193 Kobayashi A. 63 Kobayashi H. 156 Kogen betsuon shô 48 Kogen seidokukô 48 Koguryo xv, 10, 21, 109–11, 233 Kojiki 18–19 Kojiki-den 46, 246 kokugaku 44, 61 Konkômyô-saishôô-kyô ongi 39 Kôno R. 53–4, 112 Korean xv, 10, 14, 100, 104, 110, 151, 201, 236, 241, 250, 259–60, 272; see also Sino-Korean Koryo 134 Kudara see Paekche kungana 25–8

286

INDEX

Kusakado N. 49 kwukyel 111 Lange, R. 20, 24, 43, 52, 55–7, 61–3, 65, 176, 226 Late Middle Chinese 90–1, 95–8, 149–56 Late Middle Japanese 67, 74–6, 85–6, 175–6, 186–8, 196, 211, 216, 271 Late Old Chinese 6–7, 10–16, 22, 24, 37, 40–1, 90, 99, 104, 109–11, 134, 156, 176, 212, 215, 218, 245, 253–5, 257–60, 262; see also Chang’an Late Middle Chinese Lee, H. 109–10, 112–13 Lee, K. 109 Li F. xvii, 92, 139, 224 Li R. 145 Liu G. 137 Luo C. 31, 96–7, 112, 127, 135–6, 150, 154, 161, 177 Lushei 143 Lyu L. 110 Mabuchi K. 61–3, 180, 198, 239, 251–3 Malmqvist 105 Mandarin xv, 5, 6, 41, 91, 256–7 Manomaivibool, P. 224 man’yôgana 27–8 Man’yôshû 1, 4, 6, 10, 17, 20–1, 24–8, 33, 40, 60, 65, 81, 159, 173, 214, 245 Martin, S. E. xv, xvii, 38–9, 42, 45, 50, 66–7, 69–71, 73–7, 80–1, 85–8, 98–9, 102, 114–17, 188, 269–70 Maruyama T. 63 Maspero, H. 112, 126, 137, 145 Matsumura A. 19, 41–2, 74–6, 85 methodology 158–63 Miller, R. A. xvii, 54–5, 62–3, 65, 70, 87, 104, 193, 270 Mimana see Kaya Min 95, 144–5, 213; see also Chaozhou, Proto-Min, Taiwanese Miner, E. 28 Mineya T. 65, 119, 121, 126, 151, 160, 228 Mino census 81 Miyake, M. H. xvii, 61–4, 70, 79, 104, 116, 127, 141 Miyake T. 55–6 Mizutani S. 31, 137

Modern Japanese 67 Modern Standard Japanese 67, 74–7, 85–6, 88, 100–1, 103, 175–6, 188, 193, 196, 211, 216, 238, 266, 269–71 Mori H. 28, 30, 37–9, 42, 57–60, 62–5, 160, 165, 167, 176, 181, 190–1, 215 Morimoto T. 62 Motoori N. 46–7, 50, 61 Morohashi T. xviii, 102, 105, 107, 231 Myôgishô 37 Nagata K. 55, 61–2, 253 Nam K. 114 Nan Qi shu 8 Nara 99 Nara Period 1, 3, 134 New Kan-on 107 Ngu thien tu 119, 132, 162 Nguyen D. 118, 130–1 Nguyen N. S. 131 Nihon ryôiki 21 Nihon shoki 28–30 nôm 3, 4, 90, 118–19, 129–35, 140–1, 157, 162, 200, 205, 212–14, 217, 222, 251 norito 17 Norman, J. 91–3, 95, 98, 125, 138, 143, 148 Oh S. 116 Ôjin 9 Okimori T. 57 Old Chinese 90–3, 138–43; see also Eastern Han, Late Old Chinese, Western Han Old Japanese: coda inventory 266–71; consonant inventory 74, 196, 265–6, 269; consonant reconstructions 164–96; glide inventory 266–9; glide reconstructions 170–1, 186–7; orthography 12–41; previous reconstructions 43–65; syllabic inventory 271–2; syllable structure 265–71; vowel inventory 84, 262–9; vowel reconstructions 198–262 Old Mandarin 112 Old Sino-Korean 110–11 Old Sino-Vietnamese 97, 100, 112, 117–19, 123–9, 131, 134, 141–2, 144–8, 162, 199, 204–5, 207–9, 212–13, 217–18, 221–4, 226–9, 231–3, 239, 245, 251–4, 258–60, 266

287

INDEX

Omodaka H. 7, 24, 40, 42, 178 Ômori A. 20, 26 onbin 74 Ôno S. 43, 49, 51, 53, 56, 65, 81, 160, 197 Ôno T. 13, 21, 23, 32, 41–2, 61, 107, 178–9, 183, 188, 192, 194, 197, 252 onomatogram 27 Osada N. 63 Packard, J. 263 Paekche 9–10, 16, 21, 109–11, 233, 272 Pak P. 112 Period A Shuri 78 peripheralization 263 phonogram 6 Pierson, J. L. 52–3, 61, 63 pitch accent 37–9, 272 Portuguese 74–6, 85, 101, 175, 176, 269–70 Prakrit 137 Pre-Old Japanese 66 Proto-Japanese-Korean see ProtoKoreo-Japonic Proto-Japonic 66, 216, 239 Proto-Japonic-External 67, 78–9 Proto-Japonic-Internal 66, 78–9 Proto-Koreo-Japonic 66 Proto-Min 125, 140 Proto-Tai 141, 225, 246 Proto-Viet-Muong 122 pseudo-Go-on 104–5 pseudograph 184 pseudo-Kan-on 105 Pulleyblank, E. G. xvi, 9, 15, 23, 31, 34, 41–2, 54–5, 59, 62–5, 90–9, 101, 104–6, 112, 121–2, 124–5, 137–56, 160–1, 166, 176, 187, 199, 205, 213, 223, 225, 228, 230, 242–3, 257 Qieyun 93–5, 98, 101, 106, 125–6, 137, 140, 144, 149, 153, 157, 199, 252 quoc ngu 118–20 Ramsey, S. R. 56–7, 62, 64, 68–73, 87, 130, 168, 186 Ray, D. T. 125 Rhodes, A. de 118–19, 130, 132 Ruc 125 Ryu M. 110 Ryukyuan 66, 75, 188

Samghabhara 224, 246 Samkwuk saki 9 Sandness, K. 61, 63, 179 Sanskrit 51, 137–8, 151, 171, 177, 187, 214; see also Amoghavajra, Indic, Jnanagupta Schuessler, A. 138–40 scribal error 160, 184 Seeley, C. 8–10, 28, 45–6 semantogram 6, 24–5, 159 senmyôgaki 17–19, 21 Serafim, L. A. 9, 66, 68, 79, 267 Shiga Island 8 Shijing 92, 138, 140 Shin’yaku Kegon-kyô ongi shiki 21 Shôgaitô, M. 137 Shoku nihongi 21 Shorto, H. L. 123 Siamese see Thai Silla 10, 109–13, 190, 233 Sin cacen 114 sinograph 5 Sino-Japanese xv, xvii–xviii, 3, 60, 98–109, 140, 266–71; see also Go-on, Kan-on Sino-Kaya 109–10 Sino-Koguryo 109–11 Sino-Korean xviii, 3, 9, 30, 35, 51, 53, 60, 64, 87, 89–90, 95, 98–101, 104, 108–17, 127, 129–30, 134, 140, 145, 153, 160, 162, 166, 171–2, 185, 190, 202–3, 206–7, 209–11, 213–16, 219–20, 223, 225–6, 228–31, 233–6, 240–4, 247–50, 255–7, 259–61, 266–70; see also Sino-Kaya, SinoKoguryo, Sino-Paekche, Sino-Silla Sino-Paekche 9, 11–12, 14, 16, 21–2, 30, 35, 41, 89, 99, 103–4, 109–11, 125 Sinor, D. 137 Sino-Silla 109–13, 190 Sino-Tibetan 135–6, 138 Sino-Tibetan Treaty Inscription 136 Sino-Vietnamese 3, 31, 51, 53, 64, 87, 89–90, 95, 98–101, 104, 108, 113, 117–35, 140, 145, 151, 153, 157, 160, 162, 166, 171–2, 174–5, 184–5, 200–3, 206–7, 209–11, 213–16, 219–20, 223–6, 229–30, 234–6, 240–3, 248–50, 255–61, 266–8, 270; see also Old Sino-Vietnamese Sinoxenic 3, 16, 54, 95, 98–135, 139–40, 143, 150, 160–2, 184, 202,

288

INDEX

210, 215, 223, 228–9, 234, 240, 248, 256, 272; see also Old SinoVietnamese, Sino-Japanese, SinoKorean, Sino-Paekche, Sino-Vietnamese Sino-Xenic see Sinoxenic Sin Y. 112 Sizang Chin 143 Song shu 8 Starostin, S. xvi, xvii, 7, 9, 14–16, 41, 63, 90–3, 99, 101, 135, 138–46, 149, 156–7, 218, 225, 232-3 Suda Hachiman inscription 10, 12, 19 Suiko Period transcriptions 12–17, 33, 67, 72, 178–9, 183, 188, 194, 196, 252, 258 System A 11, 40 System B 11, 40, 99, 104; see also Sino-Paekche System-B-based transcription 11–12; see also Eta Funayama inscription, Inariyama inscription, Suda Hachiman inscription System C 12, 40, 104 System-C-based transcription 13–17; see also Suiko Period transcriptions System D 21–2, 40, 103–4 System-D-based transcription 23–8 System E 30–2 System-E-based transcription 32–40 Tai 145–6, 244 Taihô census 245 Taiwanese xv, xviii, 108, 113, 146, 200, 205, 208, 213, 218, 222–6, 230, 232–3, 235, 246–7 Takayama M. 37 Tangut 135 tawamuregaki see gisho Thai 128 Thompson, L. 118–23 Tibetan 135 Tibetan transcription xviii, 113, 126–7, 135, 137, 151–4, 166–7, 171–2, 174, 176, 185, 201–3, 206, 211, 215–16, 219–20, 226, 229–31, 234–6, 241–3, 249, 260 Tibeto-Burman 135, 143 Ting P. 14, 23, 92, 140 Tôdô A. xvii, 31, 53, 62–4, 102, 105–8, 154–6, 171, 176 Toh S. 110

tonal spelling xvi tone 37–9, 117, 126, 128 Tongkwuk cengwun 113–17, 172, 233, 244, 259 Tô-Sô-on 100, 107–8, 231, 266 Tsukishima H. 45 Type A/B distinction 139, 142–3; see also Old Chinese Uighur transcription 136–7, 155, 177, 187 Unger, J. M. 56–7, 62–3, 67–73, 77–8, 80, 87–8, 168, 186, 195–6, 218, 226, 237 University of Illinois at UrbanaChampaign 263 Vance, T. J. 67, 76–7, 270 variant phonograms 159–60 Varley, H. P. 9, 30, 107, 109 Vietnamese xviii, 2, 100, 104, 117–35, 145, 157, 224, 236, 241, 257; see also Sino-Vietnamese Vovin, A. 41, 66, 68, 70, 76, 87, 110, 112–13, 116 Wajinden see Wei zhi Waji shôranshô 45 Wang L. 124, 126, 141 Wani 9, 104, 109 Wei zhi 2, 6–9, 11–12, 29 Wenck, G. 55 Western Han 90, 109, 112, 117, 120, 131, 142 Whitman, J. B. 56–7, 62–3, 66–71, 77–8, 80–2, 87–8, 216, 218, 226, 231 wordprinting 30 Wo ren zhuan see Wei zhi Wu 104, 106, 144, 157 Wylie romanization xviii, 136, 157 Yagi Y. 49 Yale romanization 50–1, 87, 112, 115, 164, 236 Yamada Y. 6 Yamaguchi Y. 69–70, 80–1 Yamato Takeru 163, 272 Yasuda K. 55–6, 62 Yayoi Period 8 Yi K. 111, 113 Yiqiejing yinyi 95 Yôrô census 21

289

INDEX

Yoshitake S. 53, 61–3, 65 Yu C. 60, 109–11 Yuchi Z. 137 Yue 127; see also Cantonese

Yunjing 51, 95–8, 101, 105, 115, 150 Yunying 95, 144, 149, 153 Yupian 95

290

INDEX

291