¹ This paper is the fourth in a series of articles under the general title Studies in Colloquial Japanese (abbr. SCJ). The earlier papers in the series are I. Inflection, JAOS 66.97–109; II. Syntax, Lg. 22.200–48; III. Derivation of inflected words, JAOS 66.304–15 (all 1946). For an acknowledgment of the help that I received from the American Council of Learned Societies, from several Japanese informants, and from my colleagues, see SCJ I, fn. 1 (JAOS 66.97), or SCJ II, fn. 1 (Lg. 22.200).

² In SCJ I, JAOS 66.98 fn. 4; in SCJ II, Lg. 22.200–1 fn. 3, 203–4 (§1.4).

³ It was the neat and systematic National Romanization (Kokutei Rōmazi), promulgated by a decree of the Japanese cabinet in 1937, that especially influenced my transcription, rather than the so-called Hepburn Romanization of 1885. Yet the latter—unsystematic and cumbersome as it seems to be—is the one that turns out to be closer to a phonemic notation. (On these two systems of romanization see Denzel Carr in JAOS 59.99–102 [1939] and Edwin O. Reischauer in JAOS 60.82–9 [1940].)

So far as I know, all other attempts by American linguists to work out the phonemic structure of Japanese are open to the same objections as mine. I have seen unpublished analyses by George A. Kennedy (ca. 1942) and A. M. Halpern (ca. 1944), as well as the published account in Joseph K. Yamagiwa, Modern conversational Japanese 1–9 (New York and London, 1942). All these, like my own earlier version, overlook phonemic contrasts and rely heavily on morphophonemic data. But since their main purpose was pedagogical (again like that of my own earlier version), their morphophonemic tinge is a merit rather than a reproach.

⁴ In writing this paper I have had the benefit of several suggestions by Rulon S. Wells and Eleanor Harz Jorden. Masako Yokoyama has kindly checked the examples for me. Most of the examples are drawn from my own files; but where these did not contain illustrations of particular sequences, I consulted Kenkyusha's New Japanese-English dictionary (Tokyo, 1938). All examples taken from this source were submitted to qualified informants before being transcribed.

Several accounts of Japanese pronunciation have been published in English, but none that approaches the subject from a structural (phonemic) point of view, or that presents more than crude phonetic data, undigested and naive. The most useful (or least useless) of the works that have so far appeared in English is Masatoshi Gensen Mori, The pronunciation of Japanese (Tokyo, 1929); others are listed by Borton, Elisséeff, and Reischauer in A selected list of books and articles on Japan 74–5 (Washington, 1940). Mori's treatment is detailed and exhaustive, in the best tradition of the London school of phonetics; but it remains always on the lowest phonetic level. The sounds of Japanese, for Mori (as for practically all the other scholars who have described them), are mere noises, without structural organization or any sort of interrelations except in their physical properties.

⁵ Charles C. Fries and Kenneth L. Pike, Coexistent phonemic systems, Lg. 25.29–50 (1949). The crucial illustration in this article is taken from Mazateco, an Indian language of Mexico. In Mazateco words of native origin there are two dental stops: after [n] only [d], elsewhere only [t]. This situation by itself would call for a single phoneme with two allo-phones. But there is also a word [siento] hundred, a loan from Spanish, with [t] after [n]. Fries and Pike regard this [t] as belonging not only to a different phoneme but to a different phonemic system from that which includes the [t]s and [d]s of native words. But the presence of [siento] necessarily affects the treatment of all dental stops in the language: it contributes no less than other utterance-fractions to the total analysis. The cultural evidence adduced by Fries and Pike against this view, that monolingual speakers of Mazateco learn to read the [nd] of native words more easily in the spelling ‘nt’ than in the spelling ‘nd’, whatever its ultimate explanation may be, is not a cogent reason for setting aside the phonetic and distributional data.

Certain other cases of so-called coexistence are easily accounted for by traditional methods. Thus whispered utterances are utterances without pitch phonemes, singing is a non-linguistic use of linguistic material. Differences in tempo and style must probably be treated as differences in dialect, at least until a more fruitful approach can be worked out. For a phonemic interpretation of tempo differences see now Charles F. Hockett, Peiping morphophonemics §9, Lg. 26.74–7 (1950).

⁶ See the admirable short statement by Charles F. Hockett, A note on ‘structure’, IJAL 14.269–71 (1948).

⁷ This does not mean that every noise made by a speaker in speaking must be accommodated in the phonemic system. Sounds that occur in a completely unrestricted set of environments (before and after all other sound types without exception), such as hiccups, coughs, and sniffles, can be excluded from the relevant material at once. Sounds that never combine with other sound types in an utterance, such as the alveolar ‘click of commiseration’ in English and the polite indrawn hiss in Japanese, can be similarly excluded, or else treated as an organically separate part of the system (though not as part of an independent system coexisting with the main one). Note that the sounds here listed for exclusion are identified by their distribution.

⁸ These definitions are restated, with additions and with a pervading shift of emphasis, from my article A set of postulates for phonemic analysis, Lg. 24.3–46 (1948); they were presented, in a preliminary version, before a meeting of the Linguistic Society in New York, 30 December 1948. The restatement is intended to meet the objections of Martin Joos, Acoustic phonetics 8 fn. 9 and 120 fn. 85 (Language Monograph No. 23, 1948), by establishing the theory of phonemic analysis on an auditory instead of an articulatory basis.

⁹ In SCJ II, Lg. 22.202 ff., this unit is called a pause-group, in order to avoid confusion with the use of the term ‘phrase’ for syntactic units.

It might be argued that pause, the interruption of speech activity, should be distinguished from mere silence, the absence of speech activity that precedes and follows a whole utterance; the two could be distinguished by assigning to pause an arbitrary upper limit of duration—say ten seconds. Such a distinction would complicate the definition of a phrase, and would appear to serve no particular purpose; but any reader who favors it can safely make for himself the small adjustments that would result from its adoption. All the pauses marked in this paper (except in §7) occur internally in utterances.

¹⁰ Whenever two or more phones are mentioned in these definitions, it is understood that they are phones occurring in utterances of a single dialect.

¹¹ This statement is meant to cover both the situation where two phones P and Q have all their environments in common, and the more usual situation where P shares all the environments of Q but occurs also in environments where Q does not occur.

In this paper the term ‘free variation’ is used in a wider sense than is customary. Different phones are said to be in free variation not only if they share environments in the speech of a single person but also if they share environments in the speech of two or more persons who speak the same dialect. Thus, when bilabial and labiodental spirants are said to be in free variation (fn. 13), this means either that both types are used indifferently by the same person, or that each is the only type of labial spirant used by one of two different persons. For the purposes of this article the distinction between the two kinds of free variation is irrelevant.

¹² The term ‘quality’ is used in this paper as the auditory correlate of what I called aspects and components in my article on postulates (cited in fn. 8 above).

¹³ Phonetic transcriptions in this paper, enclosed in square brackets, are allophonic; they have been simplified as much as possible, so as to avoid the use of too many special characters. The hyphen that appears in the transcriptions of §1 is merely a graphic device for showing the number of equivalent fractions in a phrase; it has no phonetic or phonemic significance. For the meaning of the phonetic terms in the following descriptions see Bloch and Trager, Outline of linguistic analysis, ch. 2 (Baltimore, 1942).

Consonant symbols for the most part have their traditional values. [p, b, m] are bilabial, but [f] is bilabial or labiodental in free variation; [t, d, n] are dental, [ts] is dental or dentialveolar, [s, z, r, 1] are alveolar; [tš, š, dž, ž] are prepalatal; [k, g, ŋ] are mediovelar. [y] is a voiced prevelar semivowel, [y] is the corresponding voiceless sound. Italic letters denote palatalized sounds. The term ‘palatalized’ is here used to include on the one hand labial, dental, and glottal sounds pronounced with simultaneous raising of the tongue toward the hard palate (‘palatalized’ in the usual sense), and on the other hand prevelar sounds, pronounced with contact between the tongue and the forward part of the soft palate.

A raised dot after a consonant indicates that the sound has the time value of a full syllable.

Vowel symbols depart in some respects from their traditional meanings. [i] is high front; [e] is mean-mid front, like the vowel in English bet, not upper-mid; [a] is low back (slightly advanced toward central position), not low front; [o] is mean-mid back, approximately like the vowel of German Gott, not upper-mid; [u] is high back, often somewhat advanced toward central position, unrounded or weakly rounded. Capital letters [i, a, O, u] denote voiceless vowels.

Glosses are printed in italics, without quotation marks. A gloss is meant only to identify the cited form, not to provide an exhaustive account of its meaning.

¹⁴ The number of syllables in a phrase is therefore not found by counting peaks of sonority or chest pulses, but only by counting the temporally equal fractions contained in it, or by comparing its duration with that of another phrase in which the number of such fractions is known. In short, the Japanese syllable is a unit of duration. Such a unit is often called a mora; but the term ‘syllable’ is better established in descriptions of Japanese.

It is worth noting, as a cultural correlate of the linguistic structure, that a speaker of Japanese is just as ready as a speaker of English to state the number of syllables in any phrase, though neither one, without special training, can state the number of segments. Since the two speaker's count different kinds of units, they will not always agree on the number of syllables in a given phonetic continuum. In the English word asks, pronounced with a long vowel and distinctly released consonants, a Japanese will hear five syllables; in the Japanese word [gyuunyuu], a speaker of English will hear only two. (The Japanese speaker's ‘feeling’ for the number of syllables in a phrase is no doubt partly due to his knowledge of how the phrase is written in the native syllabary; but the syllabary, in turn, must reflect the speakers' naive structural analysis of their language.)

¹⁵ For a theoretical discussion see Zellig S. Harris, Simultaneous components in phonology, Lg. 20.181–205 (1944). For one type of application see Charles F. Hockett, Componential analysis of Sierra Popoluca, IJAL 13.259–67 (1947), and Peiping phonology, JAOS 67.253–67 (1947).

¹⁶ Morphological and syntactic criteria must not be used to influence the basic analysis leading to the discovery of distinctive qualities and their distribution; but they are often helpful in deciding how to group these qualities, once they have been discovered, into combinations defining phonemes. Cf. Charles F. Hockett, Problems in morphemic analysis §§4–7, Lg. 23.324–7 (1947); Peiping morphophonemics §3, Lg. 26.69–70 and fn. 10 (1950).

An illustration from English. The phones [f] and [ν], [θ] and [ð], [s] and [z] are of course phonemically different. Traditionally, each of them is assigned to a single phoneme contrasting as a unit with all the others; but this interpretation proves to be awkward in dealing with certain morphological relations. In such pairs as belief: believe, safe: save, wreath: wreathe, use (noun): use (verb), it would be convenient to regard the verb as consisting of two morphemes: one that appears by itself as a noun or an adjective, the other a verb-deriving suffix; but if /v, ð, z/ are unitary phonemes, there is no phoneme present in these verbs that could be taken as a suffix. An alternative and equally accurate interpretation of the phones [v, ð, z] would be to regard the quality of voice (at least when accompanied by the quality of constriction) as defining a separate phoneme /V/, so that each of the phones [v, ð, z] would belong simultaneously to two phonemes: to /f, θ, s/ respectively, and to /V/. This treatment, like the traditional one, is based solely on the qualities observed in English utterances; but by providing a phoneme /V/ to serve as a verb-deriving suffix, it greatly simplifies the exposition of the morphology.

¹⁷ In any case, such descriptions are always mechanically convertible into one another, and are therefore scientifically equivalent.

^l7a For an interesting but not (I think) wholly successful attempt to use an auditory terminology see R. Jakobson and J. Lotz, Notes on the French phonemic pattern, Word 5.151–8 (1949). In this article the phonemes of French are defined in terms of six auditory oppositions or contrasts: vowel/consonant, nasal/oral, saturated/diluted, grave/acute, tense/lax, and continuous/intercepted. The following statement (151) is especially to be noted: Our basic assumption is that every language operates with a strictly limited number of underlying ultimate distinctions which form a set of binary oppositions. These opposite features occur either solely as terms of a single relation (pure opposition) or they can occur together as complexes.'

¹⁸ That such a division into qualities of position and qualities of articulation is in accord with acoustic reality has been demonstrated in the laboratory, most plainly in work with the acoustic spectrograph. See Martin Joos, Acoustic phonetics (Language Monograph No. 23, 1948), esp. ch. 4.

^18a The reason for treating affrication as a single quality rather than a succession of two qualities (closure plus constriction) has been stated in §1.

¹⁹ Some of the forms cited in this paper with no pitch higher than /3/ are paralleled by otherwise identical synonymous forms with pitch /2/ on all syllables but the first, when followed without pause by one or more syllables with lower pitch. Thus the word for room is cited in §4.1 as [heya] (with pitch /4/ on both syllables), but occurs with pitch /2/ on the last syllable in the phrase [heyáwa] as for the room.

²⁰ This example is restated from SCJ II, Lg. 22.200–1, §1.1.

²¹ The expression ‘regularly recurring unit’ is meant to exclude the sound types mentioned in fn. 7.

²² In the strictest sense of the word ‘dialect’ (as defined, for instance, in Lg. 24.8, §3.2), the phones listed below represent a mixture of slightly different dialects. But they can just as well be regarded as representing a kind of average among idiolects too similar to merit individual treatment.

²³ Although pause is here called a zero phone, it is not zero in the sense of having no extension in time. Pauses occur, and have duration, like other phones. They are zero only in being characterized by the total absence of qualities rather than by the presence of one or more.

²⁴ In these statements of distribution, the nasalized vowels are not separately enumerated. In general, any phone that occurs before or after an oral vowel occurs also before or after the corresponding nasalized vowel. (But nasalization of vowels is indicated in the transcription of the examples.)

²⁵ The distribution of this phone is unique in type: it occurs only before pause, and only in a few expressions (mostly interjectional).

²⁶ A very few speakers of standard Japanese (but many speakers of other dialects) use a short voiced dental or denti-alveolar affricate [dz] beside or instead of [z], most commonly before [u], as in [midzu] water. In view of its rareness, this phone has been excluded (perhaps arbitrarily) from consideration in the present paper.

²⁷ The reader should bear in mind that ‘long’ has a special meaning in this paper, as already noted in §2.2. A ‘long’ vowel is a vowel of ordinary duration, constituting a syllable alone or in combination with one or two preceding ‘short’ consonants.

²⁸ No instances of voiceless [e] have been observed, but the sound type may nevertheless occur in the dialect, perhaps with a distribution similar to that of voiceless [a].

²⁹ According to this statement voiceless [a] occurs only between voiceless stops; but it perhaps occurs also between voiceless consonants of other types, though I have no record of such instances.

³⁰ Voiceless [O] is here described as having the same distribution as voiceless [a]. Like [a] (see fn. 27), it may occur also between other voiceless consonants.

³¹ Compare the statement concerning nasalized [ĩ, ẽ, õ] in §4.1. What are sometimes described as assimilations—phones partaking of the character of certain neighboring phones —are here treated as distinct phones, with distributions of their own.

³² In this comment no distinction is made between oral and nasalized vowels. But note that in a succession of two or more vowels preceded or followed by a long nasal, only the one vowel contiguous to the nasal is nasalized.

³³ The term ‘overlong’ is used here with the meaning usually attached to ‘long’ (i.e. perceptibly greater in duration than other syllabic vowels) in order to avoid confusion with the special use of ‘long’ in this paper (cf. fn. 27).

³⁴ The diminution in loudness (accompanied by glottal constriction) between certain consecutive vowels in rapid or moderately slow speech is not a phoneme of ‘open juncture’ or the like, but an automatic phenomenon, predictable in terms of the phonetic environment. A similar diminution in loudness, equally predictable, occurs in rather slow speech between a long nasal and an immediately following vowel.

In very slow or very careful speech (excluded from the scope of this paper), such diminutions in loudness between consecutive vowels are not limited to the environments described in the text. In that style of speech, [koo] child (as verbal goal) and [kareeda] dried twigs are often distinguished from [koo] thus and [kareeda] (beside [kareida]) it is splendid by a diminution in loudness between the two vowels, again with glottal constriction. (The diminution occurs only at morpheme boundaries.) In a description of extra-slow or extra-careful speech, it would therefore not be possible to account for the diminution in loudness as an automatic function of the environment; the diminution would have to be referred to a separate phoneme. In view of the glottal constriction that accompanies it, it would be most suitably accommodated in the same phoneme as the phrase-final glottal stop. (I owe this suggestion to Samuel E. Martin.)

³⁵ Successions of four or more like vowels are extremely rare. An example is [oóóo] let's cover it (presumptive form of the verb [oou] covers), which is pronounced as two phonetically overlong vowels with a diminution of loudness between them. It is possible to construct even longer sequences: [oooóóo] let's cover the tail (as of a half-buried fox). Such an oddity is of course not intelligible to Japanese speakers at first hearing; but when it has been explained to them, they can both understand and repeat it.

³⁶ The convention explained in §2.2 concerning the use of the symbols N, V, Q must be modified for the purposes of this list. Here the absence of any of the three symbols N, V, Q implies the presence of the corresponding opposite quality as a necessary part of the definition of a phoneme only if the remaining qualities occur in the list accompanied by the omitted symbol. Thus, in the definition of the phoneme /p/, namely L1, the absence of the symbol V implies the presence of voicelessness as a distinctive (non-determined and non-conditioned) quality because the combination L1 V occurs as the definition of /b/; but in the definition of /y/, namely F5, the absence of V implies nothing, since there is no combination F5 V defining any other phoneme. In short, the absence of N, V, or Q in this list is distinctive only if the combination from which it is absent is otherwise identical with an occurring combination in which it is present.

³⁷ Some light is thrown on the relative initial frequency of certain phonemes by the number of dictionary entries beginning with various letters (interpreted phonemically). Such a count, in which every dictionary word figures once, regardless of its frequency in speech, is of course different in kind from the running count tabulated above. (On the difference between list frequency and text frequency see Martin Joos, Lg. 18.33 [1942].) The following figures denote the number of pages occupied by words with various initial phonemes in Kenkyusha's New Japanese-English dictionary (total 2263 pp.).

The 41 pages shown for /x/ are included among the 225 shown for /h/; that is, there are 41 pages of words that begin with /x/ or /h/ in free alternation.

³⁸ On the logical error involved in saying that phonemes occur, and on the way in which the expression can be justified, see Lg. 24.36, §53.3 and fn. 30.

³⁹ The impression that there is some structural unity in a sequence like /ka/ but not in /ak/ rests in part on the phonetic prominence of syllables, as noted in §1, and in part also on the traditional Japanese orthography, in which /ka/ but not /ak/ is written with a single character.

⁴⁰ Cf. Joos, Acoustic phonetics 80, §3.38.

⁴¹ Each row and each column in Table 2 is a statement, though not explicitly verbalized.

⁴² The text is a new version of the passage transcribed (on the basis of a different analysis) in SCJ II, Lg. 22.238–48. Particulars concerning its provenience, as well as an English translation, will be found in that place.

⁴³ The material of this section was discussed from the diachronic point of view in a paper presented before a meeting of the Linguistic Society in Chicago, 30 December 1946. The same point of view is implicit in Sanki Ichikawa, The pronunciation of English loan-words in Japanese, A grammatical miscellany offered to Otto Jespersen on his seventieth birthday 179–90 (Copenhagen, 1930). Ichikawa has gathered a wealth of material, but makes no attempt to arrive at a phonemic or other structural interpretation.

⁴⁴ /baniči, yuučiričii, skeetoin̄ŋu/ in the conservative dialect.

⁴⁵ /kyaji, han̄jikyappu, jiizeru/ in the conservative dialect.

⁴⁶ Theoretically /mobu, bédo, piramido, ejin̄ŋu, kareji, hógu/ in the conservative dialect; but these forms seem not to occur.

⁴⁷ /ueetaa, uin̄ŋu, uoomin̄nuappu/ in the conservative dialect.

⁴⁸ /hirumu/ or /huirumu/, /heea/ or /hueea/, /hairu/ or /huairu/, /hooku/ or /huooku/, etc., in the conservative dialect.

⁴⁹ /beran̄da, ban̄pu, baniči, baiorin̄, boodobiru, rabu/ in the conservative dialect.

⁵⁰ Leonard Bloomfield, Language 85 (New York, 1933).

⁵¹ The reshaping of a phonemic transcription on the basis of morphological and other criteria, mentioned in §2 and fn. 16—what Hockett calls preliminary normalization of the orthography—does not, or at any rate need not, affect its purely phonemic character.

⁵² Cf. Kenneth L. Pike, Grammatical prerequisites to phonemic analysis, Word 3.155–72 (1947).

⁵³ As defined by Rulon S. Wells, Automatic alternation, Lg. 25.99–116 (1949).

⁵⁴ Even though the normalization was not intended, and is now justified a posteriori.

⁵⁵ But /t/ before /s/ is t.

⁵⁶ These descriptions of the four pitch morphemes, denoted in the orthography by marks of punctuation, do not wholly agree with the statement in SCJ II, Lg. 22.200–1. They are based on the description given by Eleanor H. Jorden in her doctoral dissertation, The syntax of modern colloquial Japanese (Yale University, 1949).