origin of voicing alternation

10
On the Origin of Voicing Alteration in Japanese Consonants Author(s): Timothy J. Vance Source: Journal of the American Oriental Society, Vol. 102, No. 2 (Apr. - Jun., 1982), pp. 333- 341 Published by: American Oriental Society Stable URL: http://www.jstor.org/stable/602532 . Accessed: 28/03/2013 08:02 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. . American Oriental Society is collaborating with JSTOR to digitize, preserve and extend access to Journal of the American Oriental Society. http://www.jstor.org This content downloaded from 151.100.161.184 on Thu, 28 Mar 2013 08:02:13 AM All use subject to JSTOR Terms and Conditions

Upload: shaina-nordlys

Post on 19-Jul-2016

51 views

Category:

Documents


4 download

DESCRIPTION

Origin of japanese voicing alternation, dakuten by Vance

TRANSCRIPT

Page 1: Origin of Voicing Alternation

On the Origin of Voicing Alteration in Japanese ConsonantsAuthor(s): Timothy J. VanceSource: Journal of the American Oriental Society, Vol. 102, No. 2 (Apr. - Jun., 1982), pp. 333-341Published by: American Oriental SocietyStable URL: http://www.jstor.org/stable/602532 .

Accessed: 28/03/2013 08:02

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

American Oriental Society is collaborating with JSTOR to digitize, preserve and extend access to Journal ofthe American Oriental Society.

http://www.jstor.org

This content downloaded from 151.100.161.184 on Thu, 28 Mar 2013 08:02:13 AMAll use subject to JSTOR Terms and Conditions

Page 2: Origin of Voicing Alternation

ON THE ORIGIN OF VOICING ALTERATION IN JAPANESE CONSONANTS*

TIMOTHY J. VANCE

A largely unpredictable consonant alternation phenomenon known as RENDAKU has been

a feature of Japanese throughout its recorded history. A hypothesis mentioned by Unger (1975) attributes the origin of rendaku to vowel elision and nasal absorption in prehistoric

Japanese. The evidence for this hypothesis is considered in detail, and the unpredictability of rendaku is attributed to irregularity in its origin. The rarity of rendaku in compound

verbs is proposed as corroboration of the hypothesis.

INTRODUCTION

MODERN JAPANESE HAS a non-automatic consonant alternation phenomenon whereby a morpheme with an initial voiceless obstruent in isolation sometimes appears with an initial voiced obstruent as the second element of a compount or stem-and-affix formation. This phenomenon is known in Japanese linguistics as RENDAKU (translated 'sequential voicing alterna- tion' in Martin 1952:48), and the resulting alternations are: /k/(k) /Ig/(g), /t/ (t)'/ d/(d), /sts/(sts) ,/z (z),

/ftf/(sh,ch)<d3/(.j), and /h,o/(hf)/b/(b).' For example:

eda 'branch' kawa 'river' edagaw a 'tributary' raku misfortune' toshi 'year' akudoshi 'unlucky year' take 'bamboo' sao 'pole' takez-ao 'bamboo pole' inochi 'life' tsuna 'rope' inochizuna 'lifeline' Yoko 'side' shima 'stripe' lokojima 'lateral stripe' hana 'nose' chi 'blood' hanaji 'nosebleed' se 'back' hone 'bone' sebone 'back- bone' asa 'morning' furo 'bath' asaburo 'morning bath'

The /h,o/ /b/ alternation is apparently the result of a historical change: /p/ > lol > /h/ (see Hashimoto 1950a). The /ftf/ /d3/ and /s,ts/ /z/ alternations reflect mergers of /d/ and /z/ before high vowels (see Hamada 1955b). In spite of these changes and a

postwar spelling reform, it is still possible to charac- terize rendaku orthographically as a uniform change when the Japanese KANA syllabary is used. Rendaku involves simply adding two small marks known as DAKUTEN to the first letter of the affected morpheme (see Vance 1979:55-56).

Because Japanese has quite simple syllable structure, a segment affected by rendaku is always in the environment V-V or N-V. K. Kindaichi (1938:154) describes rendaku as voicing assimilation to the sur- rounding segments, but, as Sakurai (1972) points out, this is not sufficient to account for the situation in modern Japanese. Voiced and voiceless obstruents contrast both word-initially and word medially in the environments V-V and N-V. (There are no word- final obstruents.) It is generally agreed that there were no word-initial voiced obstruents in native words in eighth-century Japanese, the earliest stage of the lan- guage for which any substantial written records exist (see, e.g., Miller 1967:194-195). It is clear, however, that two obstruent series corresponding to the modern voiced and voiceless series did contrast intervocalically even in the eighth century and that rendaku was already a part of the language. Kamei, Yamada, and Ot6 (1963:298-299) therefore conclude that the origin of rendaku lies hidden in the distant unrecorded past and that a phonological explanation is impossible on the basis of the available information. In any event, there have clearly been two contrasting obstruent series throughout the recorded history of Japanese.

There is little doubt that rendaku is unpredictable in modern standard Japanese. McCawley (1968:87) says, "I am unable to state the environment in which the 'voicing rule' applies. The relevant data are completely bewildering." Martin (1952:48-57) surveys proposed conditioning factors for rendaku, and Okumura (1955)

* I am grateful to Alexis Manaster-Ramer for several helpful suggestions. The shortcomings which remain are entirely my own responsibility.

' The parenthesized letters are the representations of each phoneme in the modified Hepburn Romanization of Masuda (1974). This Romanization is used throughout the paper.

333

This content downloaded from 151.100.161.184 on Thu, 28 Mar 2013 08:02:13 AMAll use subject to JSTOR Terms and Conditions

Page 3: Origin of Voicing Alternation

334 Journal of the American Oriental Societv 102.2 (1982)

and Sakurai (1966) cover much the same ground. Okumura holds out hope that the problem of predicting rendaku may eventually be solved, although he offers no solution himself. Martin, on the other hand, concludes that rendaku is largely unpredictable. Otsu (1980) argues that there is more regularity than previously believed, but his proposals are not con- vincing (see Vance 1 980a). Miller (1 967:194-195) agrees with Martin but speculates that rendaku may have been regular at some time in the distant past:

Probably there were Old Japanese dialects where such voicing was completely regular or at least predictable in terms of pitch, juncture, or other features which the script does not record and which are therefore largely beyond our grasp. Later dialect borrowing between such Old Japanese dialects, and others in which the phenomenon did not occur resulted in the irregular and unpredictable situation which is already present in the older texts and which remains a feature of the modern language.

The various conditioning factors which have been suggested as relevant to rendaku in modern standard Japanese are considered in detail in Vance (1979:59- 79). One of these is a phonological constraint known as LYMAN'S LAW. The strongest version of this con- straint is that rendaku cannot apply in a formation X+Y if Y already contains a voiced obstruent. The available evidence indicates that some version of Lyman's Law has been a constraint on rendaku as far back as records exist. Vance (1980b) treats Lyman's Law in detail.

Two other proposed conditions are of interest here. First, it has been claimed that rendaku is most likely to occur when the first element of the formation involved ends in a nasal. It is unclear to what extent this is true in modern language, but there certainly was a correla- tion between rendaku and preceding nasals in the past. This correlation provides the most important clue to the origin of rendaku. Second, it has been proposed that rendaku does not occur in compound verbs and adjectives. Although there are some exceptions, rendaku is certainly rare in such words. This fact helps to corroborate the hypothesis presented below for the origin of rendaku.

PRECEDING NASALS

Okumura (1955) and Sakurai (1966) claim that rendaku is most likely to occur in modern Japanese when the first element of the formation involved ends

in a nasal, but neither cites any source of evidence for this assertion. There is, however, one small area of vocabulary which provides some support. Martin (1952:49-5 1) discusses a tendency for rendaku to occur in modern compounds of the form Sino-Japanese morpheme plus suru 'to do' when the Sino-Japanese morpheme ends in a nasal.' This tendency reflects a very old pattern, and regardless of the state of affairs in the modern language, there is good evidence for a strong correlation between rendaku and immediately preceding nasals at earlier stages.

According to Okumura (1955), rendaku was almost exceptionless after a nasal in both Sino-Japanese and native words in the Muromachi Period (1333-1573). It is generally agreed, however, that word-final nasals were introduced into Japanese by the large-scale bor- rowing from Chinese that began in the sixth or seventh century, and that they appeared only later in native words (see, e.g., Y. Enda 1955). The various changes in native words which led to syllable-final nasals are known collectively as HATSUONBIN, but most of them did not result in word-final nasals. Even in mod- ern Japanese, it is almost impossible to find a com- pound in which the first element is a native word ending in a nasal. Thus, the evidence for the correla- tion between rendaku and preceding nasals in the Muromachi Period comes from compounds with Sino- Japanese first elements, and since the two elements of a compound usually come from the same stratum of vocabulary, most of these have Sino-Japanese second elements as well. K. End6 (1966), for example, investi- gates the approximately 600 examples of rendaku in the 1603-04 Japanese-Portuguese dictionary and notes a correlation in Sino-Japanese compounds between rendaku and a final nasal in the first element.

Okumura (1952) examines Sino-Japanese words from the eighth century using texts which show how to read Buddhist sutras correctly. These texts distinguish HONDAKU ('original voicing'), which indicates voic- ing in the original Chinese, and SHINDAKU ('new voicing'), which indicates voicing due to rendaku in Japanese. Okumura notes a very strong correlation between rendaku and preceding nasals in this corpus.

The next topic to be considered is the prenasaliza- tion of voiced obstruents in earlier stages of Japanese. This phenomenon helps to provide a plausible expla- nation for the observed correlations between rendaku and preceding nasals.

2 As Martin (1952:52) notes: "Many of the words ending -zuru have colloquial synonyms with the ending -jiru."

This content downloaded from 151.100.161.184 on Thu, 28 Mar 2013 08:02:13 AMAll use subject to JSTOR Terms and Conditions

Page 4: Origin of Voicing Alternation

VANCE: On the Origin of Voicing Alteration in Japanese Consonants 335

PRENASALIZATION OF VOICED OBSTRUENTS

In some modern Japanese dialects vowels are nasal- ized before voiced obstruents which correspond to voiced obstruents in the standard language. In the Tosa dialect of southern Shikoku, this phenomenon is restricted to vowels preceding /d/ or /g/ (see, e.g., Sat6 1977:442). In many dialects of the Thhoku region of northern Honshu, nasalized vowels can precede all voiced obstruents.3 Some writers prefer to describe the phenomenon in question as a nasal onglide preceding voiced obstruents rather than vowel nasalization, but this terminological difference does not appear to reflect any disagreement about the phonetic facts. Phonolog- ically, the nasal feature is best interpreted as a property of the obstruents, and these will be referred to hereafter as prenasalized voiced obstruents. In initial position these segments are realized as simple voiced obstruents.

It is clear from the work of Portuguese missionaries that some medial voiced obstruents were prenasalized in sixteenth-century Kyato Japanese, the standard dialect at the time. Jodo Rodrigues provides a detailed description of contemporary Kyoto pronunciation in his grammar of Japanese, Arte de lingoa de Iapamn (published 1604-08). He notes vowel nasalization before /g! and /Id/ and occasionally before /b/. Hashimoto (1950b) suggests that this nasalization was not very prominent since Rodrigues describes it as a "half til" and warns his readers not to nasalize too much. Hashimoto also notes that the Romanization adopted by the Portuguese Jesuits did not include any special symbol to mark this nasalization. For example, simple h, d, g are used for medial voiced stops in Vocabulario da lingoa de Iapam, the Japanese-Portu- guese dictionary published in 1603-04. As Hashimoto points out, however, other Westerners rendered the prenasalized voiced obstruents as nC. It is generally agreed that prenasalization was widespread in Japan in 1600 and that the modern Tosa and Tahoku dialects are conservative in this respect. As Hashimoto notes, however, Rodrigues mentions the lack of prenasaliza- tion as a well-known characteristic of the dialect of Bizen Province (part of modern Okayama Prefecture).

There are no detailed descriptions of Japanese pro- nunciation before the sixteenth century, but several

sources of evidence indicate that voiced obstruents were prenasalized at earlier stages as well. Hashimoto (1950c) notes that traditional narratives, songs, and chants often preserve older pronunciations better than ordinary spoken language. Such archaic pronuncia- tions are found, for example, in the Japanese Noh drama songs known as YOKYOKU. According to Hashi- moto, evidence for prenasalization comes from the yokyoku pronunciation of borrowings from Chi- nese which contained syllables ending in voiceless stops in the original language.4 When such syllables precede a nasal or a voiced obstruent in yokyoku, the original voiceless stop is pronounced as a kind of nasal stop. According to the Onseigaku Daijiten (Nihon Onseigakkai 1976:625, 680), such nasal stops are stops with a nasal release (also called "faucal plosives"), and English pronunciations such as [tn] in kitten are given as examples. Hamada (1952) describes the same phenomenon and cites work by E. Iwabuchi (see Hamada's note 16), but Hamada says that only syllables which had final / t/ in the original Chinese are involved. In any event, the fact that nasal release appears before voiced obstruents as well as before nasals certainly suggests that voiced obstruents were prenasalized when the yokyoku were written. Ac- cording to Hashimoto, the y6kyoku date from the early Muromachi Period, i.e., the fourteenth century.

Hamada (1952) also discusses transliterations of Japanese by Chinese and Koreans which antedate the Portuguese materials. Of the sources which Hamada mentions, the oldest is the Chinese Hou-lin vii-lu, a collection of miscellaneous articles compiled in the mid-thirteenth century by Lo Ta-ching. The work is known in Japanese as Kakurin Gv okuro (see Sat6 1977:650), and a Japanese word list is included in one edition. The Chinese characters used to transliterate Japanese syllables followed by /b/, /d/, g/&, and /z/ all represent Chinese syllables with final nasals.

Hamada (1952) considers a number of phenomena which are easier to account for on the assumption that voiced obstruents were prenasalized in the Heian Period (794-1185). Some of these phenomena involve the hatsuonbin, i.e., the changes mentioned above which produced syllable-final nasals in native Japanese words and which began to appear in the Heian Period. Chinese borrowings were, of course, the other source of syllable-final nasals. According to Sato (1977:244),

3 As Inoue (1971) notes, voiced velar stops have absorbed the preceding nasalization in many Tohoku dialects, which have [Ve] rather then [Vg]. The only T6hoku dialect on which I have done any fieldwork is that of Aomori City, and my informant has [Vz], [Vb], [Vd], and [Vex

4 In traditional Chinese terminology such syllables are said to carry the ENTERING TONE.

This content downloaded from 151.100.161.184 on Thu, 28 Mar 2013 08:02:13 AMAll use subject to JSTOR Terms and Conditions

Page 5: Origin of Voicing Alternation

336 Journal of the American Oriental Society 102.2 (1982)

many pre-eleventh-century sources distinguish ortho- graphically between syllable-final [m] and syllable- final [n]. In modern Japanese there is only a single syllable-final nasal phoneme, and its place of articula- tion varies widely depending on the following segment. This phoneme is sometimes called the MORA NASAL

because it always constitutes a separate mora, which counts as a unit for poetic meter (see, e.g., McCawley 1968:84-85).5 Early in the Heian Period there was no fixed way of representing syllable-final nasals in the kana syllabary, and their phonemic status was unclear. The kana symbols for mu, ni, i, and u were all used for writing these sounds, and in many cases no symbol was used at all. Sata (1977:244) says that the modern letter for syllable-final nasals appeared in the eleventh century, and Hamada (1955a) says that this symbol became fixed in the twelfth century.

Some of the important examples treated by Hamada (1952) involve hatsuonbin before the verb suffixes -te and -ta. Bloch (1946) labels the modern verb forms with -te GERUNDS and those with -ta PAST INDIC-

ATIVES. For example, the gerunds of shinu 'to die' and Yomnu 'to read' are shinde < shinite and vonde < Yomite. The vowels before the suffixes apparently dropped and left nasals followed by /t/. As in the Sino-Japanese examples mentioned above, preceding nasals apparently triggered rendaku. Hamada argues that this voicing change is easier to explain if voiced obstruents were prenasalized in the Heian Period. He describes it as a king of "analogy": if voiced obstruents were always preceded by a nasal onglide, perhaps an obstruent preceded by a nasal would always be pronounced voiced. He goes on to suggest that one reason that syllable-final nasals often were not repre- sented in kana orthography may have been that they were often absorbed into the nasal onglides of voiced obstruents. Hamada's notation implies that this ab- sorption involved loss of the moras which developed by vowel elision. For example, he says that the gerund of vormu changed from [jo-n-nde] to [jo-nde]. Ex- amples of a similar kind of absorption will be crucial below.

Hamada (1952) also mentions forms of verbs such as tobu 'to fly'and kogu 'to row'. The gerunds are tonde < tobite and koide < kogite. If Heian voiced obstruents were prenasalized, the old form tobite would have been pronounced something like [tobite]. This makes the

change of hi to n easier to understand. By the same token, the old form kogite would have been pro- nounced something like [k6gite], and parallel develop- ment should have resulted in konde. Hamada suggests that this is in fact what happened and that the old kana symbol used for syllable-final nasals in this and similar forms was the letter for i. Later spelling pronunciation then led to the modern form. The usual account of forms like koide is that they developed by simple loss of /g/ (see, e.g., Sat6 1977:242), but this makes it difficult to explain the voicing of the suffix consonant. Another possibility is that the nasal which developed from /g/ was phonetically a palatal. It seems reason- able to assume that the /g/ in kogite was palatalized under the influence of the following high front vowel. If this is correct, the word would have been pronounced something like [kogyite], and a natural development from this would have been [kOpde]. Subsequent weakening of the palatal nasal [p] would have naturally led to [1] and resulted in [koide], for which koide would have been the obvious spelling.

Lange (1973) attempts to reconstruct the phonolog- ical system of eighth-century Japanese by investigating the reconstructed Ancient Chinese values of the Chinese characters used to write parts of the Man 'l shui, a mid-eighth-century poetry anthology. The writing system, known as MAN'YOGANA, in- volved assigning Chinese characters as phonograms for Japanese syllables with similar phonetic values. With regard to prenasalization, Lange says (1973:134):

. .. our data offer no concrete evidence, except pos- sibly in the case of the velar initials. There we see an obvious preference on the part of the Japanese for using syllables with Ancient Chinese initial [p] to represent Japanese syllables with voiced velar initials. But whether one chooses to interpret this as 'pre- nasalization' will depend on the views of the indi- vidual.

As Unger (1975:19) points out, however, Lange's work does not provide any evidence which contradicts the prenasalization hypothesis.

The evidence presented in this section indicates that voiced obstruents were prenasalized at least as far back as the Heian Period. The examples considered in the next section suggest that prenasalization should be assumed for earlier stages as well. Asayama (1943) was apparently the first to propose that prenasalization may have been a prehistoric feature of Japanese. Hamada (1952) says that no definitive answer can be given to the question of whether prenasalization dates

5 Not all writers agree that Japanese actually has syllables distinct from moras. See, for example, H. Kindaichi (1967) and compare McCawley (1968:131-134).

This content downloaded from 151.100.161.184 on Thu, 28 Mar 2013 08:02:13 AMAll use subject to JSTOR Terms and Conditions

Page 6: Origin of Voicing Alternation

VANCE: On the Origin of Voicing Alteration in Japanese Consonants 337

from prehistoric Japanese. He does, however, offer a speculative conjecture that prenasalization was bor- rowed from the pronunciation of Chinese in the T'ang capital of Ch'ang An. The impetus for this, says Hamada, was the KAN'ON pronunciation of Chinese characters, which apparently was based on the Ch'ang An Dialect. The kan'on pronunciation was spread widely as the standard beginning in the eighth century. The explanation proposed below for the origin of rendaku, however, depends crucially on the existence of prenasalization in prehistoric Japanese.

VOWEL ELISION AND NASAL ABSORPTION

Hamada (1952) gives three examples in which an original syllable of the form NV seems to have dis- appeared before a voiceless obstruent. In each case the original voiceless obstruent has become voiced. The examples are:

suzuri 'inkstone' < sumisuri Cf. sumi 'ink', suri nominal form of suru 'to rub'

fude 'writing brush' <furnite Cf. fumi 'writings', te 'hand'

ikaga 'how' < ikanika Cf. ika interrogative stem, ni adverbial particle, ka

interrogative particle

The etymologies are reasonably secure' and are given in the Nihon Kokugo Daijiten (Nihon Daijiten Kan- kokai 1972-76). The earliest attestations7 for the shortened forms range from ca. 900 for ikaga to ca. 1000 for fude. Although the earliest attestation for sumisuri is from the tenth century, ikanika and fumite are attested from the eighth century. Hamada suggests that the voicing in the shortened forms is the result of vowel elision followed by absorption of the nasal into the following consonant. As noted above, the phone- mic status of syllable-final nasals was unclear in the early Heian Period, but since the hatsuonbin were already producing such nasals in verb forms, the result of vowel elision infumite would presumably have been something like [0unte]. The consonant following the nasal would then have become voiced for the same reason proposed above in connection with verb forms

like yonde < yomite. As noted earlier, Hamada sug- gests that the frequent lack of kana representation for syllable-final nasals may have been due to absorption of such nasals into the nasal onglides of following voiced obstruents. This, he proposes is what happened in suzuri, fude, and ikaga. Thus, Hamada cites examples like these in support of the claim that voiced obstruents were prenasalized in the Heian Period.

There is, of course, an obvious problem in account- ing for the different behavior of forms like fumite and verb forms like yomite. If the vowel elision in both cases dates from the early Heian Period (i.e., before the mora nasal was firmly established), the modern reflexes should be fude and i'ode. It is important to note, however, that verb forms like )'omite were not im- mediately displaced but survived through the Heian Period and beyond in texts (see Ikeda 1975:44, note 1). This is probably because the addition of -te to the stem form of a verb was a productive morphological pattern. It is possible, therefore, to attribute verb forms like yonde to a later period than fude, suzuri, and ikaga. It must be noted, however, that the form funde is attested from the twelfth century. Perhaps fumite was still recognized as the source of fude and funde developed in the same way as vonde. Unlike the verb forms, however, fude and not funde has survived.

Hashimoto (1950b) mentions several examples at- tested from the eighth century which he accounts for in terms of vowel elision and nasal absorption. Among these examples are:

Yuge 'bow whittling' < yumike Cf. sumi 'bow', ke < kezuru 'to whittle'

shizue 'bottom branch' < shimotsue Cf. shimo 'bottom', tsu genetive particle, e 'branch'

The etymologies which Hasimoto proposes are all quite plausible, but, with one exception, they are apparently not widely accepted by lexicographers. For example, the Nihon Kokugo Daijiten gives no etymol- ogy at all for i'uge and several alternatives for shizue. In addition, the older forms which Hashimoto posits for some of his examples are not attested. This is true, for example of yumike. At this point, it is important to recall that the hatsuonbin began to produce syllable- final nasals in native words in the Heian Period. If Hashimoto's etymologies are correct, the vowel dele- tion and nasal absorption in his examples predate the hatsuonbin. As noted above, these examples are at- tested from the eighth century, and they may be older since no older records exist. It seems reasonable to

6 See, however, Miller (1967:213-214) on fude. 7The source for all attestations in this paper is Nihon

Daijiten Kank6kai 1972-76.

This content downloaded from 151.100.161.184 on Thu, 28 Mar 2013 08:02:13 AMAll use subject to JSTOR Terms and Conditions

Page 7: Origin of Voicing Alternation

338 Journal of the American Oriental Society 102.2 (1982)

assume, therefore, that syllable-final nasals were not permitted in native words when these examples devel- oped. Vowel elision would have resulted in a nasal followed by a voiceless obstruent. If voiced obstruents were prenasalized at the time, such a sequence would naturally have been interpreted as a voiced obstruent, i.e., absorption would have been necessary since NC sequences were not permissible. This is the evidence which leads Hashimoto to propose that voiced obstru- ents were prenasalized in prehistoric Japanese.

Unger (1975:8-9) cites the rank title muraji, which apparently derives from a compound of mura'village' and nushi 'owner'. The Nihon Kokugo Daijiten tenta- tively accepts this etymology, and the word is attested from the eighth century. It is thus an additional example of the kind treated by Hashimoto (1950b). Unger, however, goes a step farther and suggests that rendaku originated by vowel elision and nasal absorp- tion. This account requires an original syllable of the form NV between the elements of compounds. As Unger notes, the obvious source for such a syllable in compound nouns is the genitive particle no, and the dative/ locative particle ni may have been involved in some other types of compounds. There is no reason to believe, however, that all early compound nouns developed from phrases of the form X no Y, or that all such phrases underwent vowel elision and nasal absorption when the second element began with a voiceless obstruent. Consider the following examples, all attested in the eighth-century Man'v'&sht:

funagi 'wood for boat building' Cf. fune8 'boat', ki 'wood'

funanohe 'bow of a boat' Cf. he 'bow'

funahashi 'pontoon bridge' (modern standard funa- bashi) Cf. hashi 'bridge'

Forms like funahashi presumably derived from simple juxtaposition and did not show rendaku. Forms like funanohe retained the genitive particle and did not show rendaku. Forms like funagi presumably derived from original phrases like funa no ki by vowel elision

and nasal absorption.9 if this is correct, the irregularity of rendaku in the eighth-century texts is due to the fact that compounds were not all derived in the same way. According to this account, rendaku has always been irregular.

RENDAKU IN COMPOUND VERBS

The term INFLECTED WORD (YOGEN) in Japanese grammar refers to verbs and adjectives (KEIYOSHI). Okumura (1955) claims that rendaku does not occur in compounds of inflected word plus inflected word, and he illustrates with the examples wakachikaku 'to write with spaces between the words' and wakachigaki 'writing with spaces between the words'. Both examples derive from the verbs wakatsu 'to divide' and kaku 'to write'. In both cases, as in all compounds with a verb or adjective as first element, the first element appears in its stem form wakachi. In wakachikaku the second element retains its inflectional ending, and the compound is a verb. In wakachigaki, on the other hand, the second element is used in its stem form, and the compound is a noun. Okumura thus appears to be claiming that rendaku will not occur in a compound which consists of two inflected words and is itself an inflected word. Sakurai (1966) makes the same statement as Okumura but then qualifies it by saying that if the first element is used as a noun, rendaku can occur. Sakurai illustrates with the examples ikizumaru 'to reach an impasse'(from iku 'to go' and tsumaru 'to be blocked) and ikidomaru 'to reach an impasse' (from iku 'to go' and tomaru 'to stop (intrans.)'). Since a first element must appear in its stem form, it is apparently a semantic criterion which determines whether it is being used as a noun, and it does not seem unreasonable to interpret ikizumaru as 'going is blocked' and ikidomaru as 'going stops'.

A systematic dictionary search'0 turned up 188 modern examples which are themselves verbs or adjectives and which show rendaku. Of these, 147 have underived nouns as first elements. There are only 11

' The alternant funa is an ablaut form of fune, and as Martin (1952:85) says, "When it does occur, the alternant shape is usually initial in compounds." See also Miller (1967:185-188).

9 It is possible, of course, that funagi was the result of an early analogical change like the one which produced the modern funabashi, but the general argument does not rest on the history of this particular word.

10 The dictionary used was Kenkvusha's New Japanese- English Dictionary (Masuda 1974). Several restrictions were imposed to limit the examples to clear cases of combination and to reduce the amount of labor involved in the search. For details, see Vance (1979:58-59).

This content downloaded from 151.100.161.184 on Thu, 28 Mar 2013 08:02:13 AMAll use subject to JSTOR Terms and Conditions

Page 8: Origin of Voicing Alternation

VANCE: On the Origin of Voicing Alteration in Japanese Consonants 339

examples of rendaku in verb compounds derived from two verbs, and this in only a tiny fraction of the hundreds of compounds of this type listed in any Japanese dictionary. The remainder of this section will concentrate on compounds of this type since relevant examples with adjectival second elements are much less numerous.

Kikuda (1971) analyzes the verb and adjective com- pounds in the French translation of the 1603-04 Japanese-Portuguese dictionary. Kikuda requires that each element of the examples he considers occur as an independent entry with inflectional endings (i.e., as a verb or adjective) in the dictionary. Of the 635 examples which meet this requirement, five have ad- jectival second elements and 630 have verbal second elements. All of the former show rendaku, and according to the discussion of rendaku in Sat6 (1977:246), rendaku is still likely in verb-plus-adjective compounds in modern Japanese. Of the compounds with verbal second elements which Kukuda considers, 30 show rendaku. Only two of these 30 are listed with inflections; the remaining 28 are entered with the second element in its stem form, which implies that the compounds themselves were not verbs. By contrast, of the 600 compounds with verbal second elements which do not show rendaku, only three are listed without inflections. Kikuda quotes H. Kindaichi as saying that speakers in the past felt V+V compound verbs to be sequences of two words, and Kikuda claims that this feeling held true at the time the Japanese-Portuguese dictionary was compiled. Throughout the recorded history of Japanese, the verb stem has been used as a nonfinite form for conjoining clauses. Kuno (1973:195)

labels it the CONTINUATIVE FORM. Kikuda supports his claim with examples like voikuruu 'to go crazy with drink', which derives from You 'to get drunk' and kuruu 'to go crazy'. According to Kikuda, the compound was felt to be two separate words: voi kuruu 'to get drunk and go crazy'. There are two serious problems with this claim. First, modern V+V compound verbs have single-word accent patterns, and Kikuda offers no evidence that this was not the case in the example he considers. Second, the meaning of a compound verb of this kind is generally narrower than the meaning of a two-word sequence of the type required by Kikuda's claim. This is true even of Xoikuruu, which denotes craziness due to drunkenness. The sequence Yoi kuruu does not necessarily mean that the craziness is due to the drunkenness. An even clearer example is the verb kakitoru 'to write down', which derives from kaku 'to write' and toru 'to take'. This word is attested from the eighth century, and it does not mean simply 'to write and take', but rather'to take by writing'. These examples illustrate why dictionary entries are necessary for compound verbs of this kind. Nonetheless, it seems reasonable to assume that V+V compound verbs originated from sequences of the kind Kikuda proposes.

To my knowledge, there has been no research on rendaku in V+V compound verbs from earlier periods. I therefore collected a small corpus of examples from a dictionary of eighth-century Japanese (Maruyama 1967). This corpus consists of the first 25 clear cases of V+V compound verbs in which rendaku could have occurred (i.e., those in which the second element begins with a voiceless obstruent in isolation and does not contain a voiced obstruent'2). Not one of the 25 actually shows rendaku, and this suggests that lack of rendaku in compound verbs of this kind was already the pattern in the eighth century.

As mentioned above, 11 examples of rendaku in V+V compound verbs were found in a search through a dictionary of modern Japanese. One of these, nezamu (modern nezameru) 'to awake', is attested from the eighth century. This word derives from neru 'to sleep' and samu (modern sameru) 'to awake, return to normal'. Since the compound means 'to awake from sleep' and not 'to sleep and awake', one might propose an eighth-century analog to Sakurai's (1966) claim that rendaku is possible if the first element is used as a noun. It is worth mentioning in this connection that compounds of the form N+V often show rendaku in

" The dictionary search turned up seven examples of rendaku in compound adjectives derived from two adjec- tives. Although no actual count was made, the number of A+A compound adjectives listed is very small, and seven examples make a rather high proportion. It might be argued that the first elements in the examples with rendaku are being used as nouns, and some adjective stems are in fact used as independent nouns. For example, both akai 'red(A)' and aka 'red(N)' are independent words, and this could be cited to account for the adjective akaguroi 'dark red' (cf. kuroi 'black'). This reasoning is not applicable, however, to examples like the adjective usugitanai 'dirty, drab'. This word derives from usui 'pale' and kitanai 'dirty', but the stem usu does not occur as an independent word. In any event, it is not at all clear that Okumura's claim about inflected-word compounds holds for adjectives. 1 Recall the discussion of Lyman's Law above.

This content downloaded from 151.100.161.184 on Thu, 28 Mar 2013 08:02:13 AMAll use subject to JSTOR Terms and Conditions

Page 9: Origin of Voicing Alternation

340 Journal of the American Oriental Society 102.2 (1982)

eighth-century Japanese. Further research is necessary to determine the extent to which no and ni are plausible sources for rendaku in such cases, but it should be pointed out that no often marked subjects and sometimes even direct objects in Classical Japa- nese (see Ikeda 1975:194-195).

In any event, the rarity of rendaku in V+V com- pound verbs can be interpreted as corroboration of the hypothesis that rendaku originated by vowel elision and nasal absorption. As mentioned above, such compounds evidently derive from sequences in which the first verb appears in its continuative form, i.e., its stem form. There is no reason to believe that a particle like no or ni would have intervened in such a construc- tion.

CONCLUSION

This paper illustrates an important, though perhaps obvious, methodological point. Recall Miller's (1967: 194-195) conjecture that the irregular character of rendaku is due to dialect mixture and that the phe- nomenon was completely predictable in some Old Japanese dialects. The evidence considered here leads to the inference that rendaku originated in phrases with elements joined by the ubiquitous particle no (or perhaps ni in some cases). It appears that in some phrases of this form, elision of the particle vowel produced phonetic sequences of a nasal followed by a voiceless obstruent. Indications are that voiced ob- struents were prenasalized and that NC sequences were not permissible at the time. Thus, a nasal fol- lowed by a voiceless obstruent would naturally have been interpreted as a voiced obstruent. It appears, however, that not all compounds derive from phrases containing one of these particles, and some which do retain the particle vowel. If this is correct, rendaku has never been regular. In general, one cannot assume that an irregular morphophonemic alternation will always be traceable to an earlier regular stage.

REFERENCES

Asayama, Nobuya 1943 "Kokugo no T6onsetsu ni okeru Dakuon ni

tsuite," Kokugo to Kokubungaku 20, 422-432. Bloch, Bernard

1946 "Studies in Colloquial Japanese I: Inflection," JAOS 66: 97-109.

Enda, Kunio 1966 "Rendakugo no Yure," Kokugo Kokubun 35:5,

68-77.

Endo, Yoshitomo 1955 "Onbin," in Kokugogakkai, ed., Kokugogaku

Jiten, Tokyodo, Tokyo. Hamada, Atsushi

1952 "Hatsuon to Dakuon to no Sokansei no Mon- dai," Kokugo Kokubun 21:4, 18-32.

1955a "Haneru On," in Kokugogakkai, ed., Kokugo- gaku Jiten, Tokyodo, Tokyo.

1955b "Yotsugana," in Kokugogakkai, ed., Kokugo- gaku Jiten, Tokyodo, Tokyo.

Hashimoto, Shinkichi 1950a "Ha-gy5 Shiin no Hensen ni tsuite," in Kokugo

On'in no Kenki'i, Iwanami, Tokyo. 1950b "Kokugo no okeru Biboin," in Kokugo On in

no Kenkvia, Iwanami, Tokyo. 1950c "Kokugoshi Kenkyu Shiryo to shite no Seimei,"

in Kokugo On'in no Kenkyvu, Iwanami, Tokyo. Ikeda, Tadashi

1975 Classical Japanese Grammar Illustrated with

Texts, Toho Gakkai, Tokyo. Inoue, Fumio

1971 "Ga-gyo Shiin no Bunpu to Rekishi," Kokugo- gaku 86, 26-41.

Kamei, Takashi, Yamada, Toshio, and Ot6, Tokihiko 1963 Nihongo no Rekishi 1: Minzoku no Kotoba no

Tanjo, Heibonsha, Tokyo. Kikuda, Norio

1971 "Yogen no Rendaku no Ichiyoin," Kaishaku 17:5, 24-29.

Kindaichi, Haruhiko 1967 "Onsetsu, Mora oyobi Haku," in Nihongo

On'in no Kenkyi, Tokyodo, Tokyo. Kindaichi, Kyosuke

1938 Kokugo On'inron, Toe, Tokyo. Kuno, Susumu

1973 The Structure of the Japanese Language, MIT Press, Cambridge.

Lange, Roland A. 1973 The Phonology of Eighth-Centurv Japanese,

Sophia University, Tokyo. McCawley, James D.

1968 The Phonological Component of a Grammar of Japanese, Mouton, The Hague.

Martin, Samuel E. 1952 Morphophonemics of Standard Colloquial

Japanese, Supplement to Language, Language Dissertation No. 47.

Maruyama, Rinpei, ed. 1967 Jodaigo Jiten, Meiji, Tokyo.

Masuda, Koh, ed. 1974 Kenkyusha 's New Japanese-English Dictionarv,

4th ed., Kenkyusha, Tokyo.

This content downloaded from 151.100.161.184 on Thu, 28 Mar 2013 08:02:13 AMAll use subject to JSTOR Terms and Conditions

Page 10: Origin of Voicing Alternation

VANCE: On the Origin of Voicing Alteration in Japanese Consonants 341

Miller, Roy A. 1967 The Japanese Language, University of Chicago

Press, Chicago. Nihon Daijiten Kankokai, ed.

1972-76 Nihon Kokugo Daijiten, Sh6gakukan, Tokyo. Nihon Onseigakkai, ed.

1976 Onseigaku Daijiten, Sanshusha, Tokyo. Okumura, Mitsuo

1952 "Jion no Rendaku ni tsuite," Kokugo Kokubun 21:6, 8-22.

1955 "Rendaku," in Kokugogakkai, ed., Kokugo- gaku Jiten, Tokyodo, Tokyo.

Otsu, Yukio 1980 "Some Aspects of Rendaku in Japanese and

Related Problems," in Ann Lekach and Yukio Otsu, eds., Theoretical Issues in Japanese Lin- guistics (MIT Working Papers in Lingusitics 2), MIT Department of Linguistics, Cambridge.

Sakurai, Shigeharu 1966 "Kyotsugo no Hatsuon de Chui Subeki Koto-

gara," in Nihon Hoso Kyokai, ed., Nihongo

Hatsuon Akusento Jiten, Nihon Hos6, Tokyo. Sata, Kiyoji, ed.

1977 Kokugogaku Kenklvi Jiten, Meiji, Tokyo. Unger, J. Marshall

1975 "Studies in Early Japanese Morphophonemics," Yale University Doctoral Dissertation. (Distri- buted by the Indiana University Linguistics Club.)

Vance, Timothy J. 1979 "Nonsense-Word Experiments in Phonology

and Their Application to Rendaku in Japanese," University of Chicago Doctoral Dissertation.

1980a "Comments on 'Some Aspects of Rendaku in Japanese and Related Problems' [by Yukio Otsu]," in Ann Lekach and Yukio Otsu, eds., Theoretical Issues in Japanese Lingusitics (MIT Working Papers in Linguistics 2), MIT Depart- ment of Linguistics, Cambridge.

1980b "The Psychological Status of a Constraint on Japanese Consonant Alternation," Linguistics 18.

This content downloaded from 151.100.161.184 on Thu, 28 Mar 2013 08:02:13 AMAll use subject to JSTOR Terms and Conditions