¹ We have made use of the following works, among others, bearing on phonemic theory: Leonard Bloomfield, Language, ch. 5–8 (New York, 1933); id., A Set of Postulates for the Science of Language, Lang. 2.153–64 (1926); Edward Sapir, Sound Patterns in Language, Lang. 1.37–51 (1925); Morris Swadesh, The Phonemic Principle, Lang. 10.117–29 (1934); id., The Phonemic Interpretation of Long Consonants, Lang. 13.1–10 (1937); id., Lang. 11.244–50 (1935); W. F. Twaddell, On Defining the Phoneme, Lang. Mon. No. 16, 1935; id., On Various Phonemes, Lang. 12.53–9 (1936); N. S. Trubetzkoy, Grundzüge der Phonologie, Travaux du Cercle Linguistique de Prague No. 7, 1939. Our indebtedness to these Works, especially to Bloomfield's, will be apparent to the reader.

² A sound-type, as we use the term, is a class of the phonetic events called sounds; each sound is a sum of sound-features (as voicing, aspiration, occlusion, labial position, etc.), which may occur in various combinations. The repetition of what is perceptually the same combination constitutes the sound-type, which is thus an abstraction from a series of utterances clustering about a norm.

³ The reason is that the facts of the utterance give us no clue in any particular instance as to the kind of x we are dealing with; that is, such assignment can be made only on the basis of morphological, lexical, or even more extraneous (e.g. historical) grounds. Apparent instances of complete intersection or overlapping of phonemic classes are therefore always the result of an error in the analysis, though cases of partial intersection (different phonemic interpretations of the same sound-type in different positions) seem to be common enough. See B. Bloch, Phonemic Overlapping, to appear in American Speech 16.3.

⁴ The following studies, dealing in whole or in part with the same subject-matter as this paper, have helped us to arrive at our analysis: Henry Sweet, The Sounds of English², Oxford, 1923; id., A Primer of Spoken English⁴, Oxford, 1932; Leonard Bloomfield, The Stressed Vowels of Chicago English, Lang. 11.97–116 (1935); Morris Swadesh, The Vowels of Chicago English, Lang. 11.148–51 (1935); John S. Kenyon, American Pronunciation⁶, Ann Arbor, 1935 (and later edd.); Daniel Jones, An Outline of English Phonetics³, Cambridge, 1932; Martin Joos, Regional and Personal Variations in General American, Le Maître Phonétique No. 45, 3–6 (1934); id., Stressed Vowels Plus r in General American, ibid. No. 48, 93–7 (1934); Bernard Bloch, Broad Transcription of General American, ibid. No. 49, 7–10 (1935); G. L. Trager, The Transcription of English, ibid. No. 49, 10–3 (1935); A. C. Lawrenson, On the Broad Transcription of Southern English, ibid. No. 50, 22–4 (1935), and many other short articles by various writers in the same journal; Kemp Malone, The Phonemes of Current English, Studies for William A. Read 133–65 (Baton Rouge, 1940).

We are indebted also to several of our colleagues, especially Prof. Leonard Bloomfield, Dr. Charles F. Hockett, and the late Benjamin Whorf, for criticism and for many valuable suggestions on English phonemics. Our statements concerning other dialects than our own are based partly on casual observation, partly on the reports of native speakers, partly on the findings of the Linguistic Atlas of the United States and Canada.

⁵ On such cases as a name: an aim, ceasing: seasick, etc., see Daniel Jones, The ‘Word’ as a Phonetic Entity, Maître Phon. No. 36, 60–5 (1931); G. Dietrich, Das Wort als Phonetische Einheit, ibid. No. 38, 31–3 (1932). Jones assembles an imposing list of words and phrases differing in juncture, and draws attention to most of the phonetic features that distinguish the two kinds of juncture; but he does not systematize his findings, and of course does not use the terminology here proposed.

The characterizing features of open juncture are in many cases akin to (sometimes identical with) the features which Trubetzkoy calls Grenzsignale; see his Grundzüge der Phonologie 241–61. As appears from his discussion (though the English example we learn: will earn on 249 is badly chosen), features of open juncture correspond to boundary signals of the kind that he calls ‘aphonematisch’; but though such features (e.g. the aspiration of voiceless stops in English) do not distinguish segmental phonemes, they serve to characterize one term of a distinctive contrast on the suprasegmental level, and are therefore on that level phonemic.

What Sweet, Sounds of English² 58–65, calls sound-junction is of course not the same thing as juncture.

⁶ Note, however, the contrast between That horse is running and That horse's running (delights me). In the former phrase the rhythmic pattern seems to be that-horse is-running; in the latter, that-horse's running. Both phrases have open juncture before the r of running; but we hear the final consonant of is as weaker than the final consonant of horse's. Prof. Charles C. Fries once suggested that the different aspects of the verbal form in -ing may be distinguished by a difference in rhythm: cf. John-is-going to-the-store (progressive) and John-is going-to-the-dogs (descriptive).

⁷ We owe this example to Dr. Morris Swadesh, who cites it from his own speech; but we do not know whether he would now accept this explanation of it. An instance of a dissyllabic word with inherent weak stress is GLT's pronunciation of twenty. This has, in isolation, not the vowel of ten but the second vowel of handed (see §7 for the relation between these two); the latter vowel is normal for the weak-stressed form of the numeral in the compounds twenty-one, etc., and appears with contrastive stress (restressed) in the simplex.

⁸ In our judgment of stress, more than anywhere else, we feel the lack of relevant experimental data. Although we are fairly sure of the difference between the reduced loud and the medial stress in movie-auditorium and similar compounds, we hope to prove this difference by submitting a series of test words or a prepared text to a laboratory phonetician for recording and electrical measurement.

⁹ Phonemic symbols are enclosed between diagonals to distinguish them from spellings (cited in italics) and from phonetic symbols (enclosed in square brackets). For the consonant phonemes we use Bloomfield's symbols; see his Language 91.

¹⁰ It is esthetically satisfying to find minimally contrasting pairs like vine: thine, and especially pleasant to find a whole series like pill: till: kill, etc.; but such contrasts are by no means necessary to prove a phonemic difference. Thus, it is not easy to find a pair of words exhibiting a minimal contrast between /ž/ and /ŋ/, but the lack of such a pair is easily supplied by the series singer: sitter, letter: leisure or ring: rim, room: rouge. In languages with a more complex morphological structure, even short series like these are often hard to find.

¹¹ Symbols enclosed in square brackets are phonetic, not phonemic. In the phonetic transcriptions, we use (except for a few changes imposed by typographical limitations) the system presented in B. Bloch and G. L. Trager, Tables for a System of Phonetic Description (preliminary edition), New Haven, 1940. Most of the symbols are familiar; [ε] denotes here the mean-mid-front vowel of pet (not, as in that system, the lower-mid-front vowel of French faire); [a] denotes a low-central vowel (not low-front); [α] denotes a low-back unrounded vowel; [] denotes a variety of mid-central retroflex vowels, whether actually articulated with inversion of the tongue tip or with retraction and lateral compression of the body of the tongue; and the like denote nonsyllabic vowels or brief glides. Vowel length is marked by a raised dot; a small arrowhead pointing up is used as a diacritic after vowel letters to denote a somewhat higher vowel than the variety written with the unmodified symbol.

It is to be borne in mind that our phonetic transcriptions are intended to be only approximate, and purposely ignore features irrelevant to the discussion. On the principles underlying our practice see Bloch's statement in Hans Kurath and others, Handbook of the Linguistic Geography of New England, ch. 4, esp. §§1–3, 21–8 (Providence, 1939).

¹² Our judgments of length are admittedly subjective, and lack the precision of mechanical or electrical measurements (cf. fn. 8). Nevertheless we believe that our statements regarding relative length are valid within the limits of this investigation.

¹³ Unless the exclamation oops has [u]; but this is not really part of our active vocabulary. Hermann Michaelis and Daniel Jones, A Phonetic Dict. of the Engl. Lang. 426 (Hannover and Berlin, 1913), list only Ulrich, Ulrica, and Uruguay—besides ugh and an occasional weak form of who—as beginning with [u] in the speech of Southern England.

¹⁴ See G. L. Trager, The Pronunciation of ‘Short a’ in American English, American Speech 5.396–400 (1930); id., One Phonemic Entity Becomes Two: the Case of ‘Short a’, ibid. 15.255–8 (1940). Sapir once remarked on a similar distinction in his own speech: he pronounced all words of this class with a rather low [æ], the length of the vowel being regulated by the voice-term of the following consonant; but in the pair have: halve he pronounced respectively the normal medium-long [æ] and a distinctly overlong vowel of same quality. His own explanation of this anomaly was that the word halve contained, in his speech, a unique phoneme found nowhere else (cf. fn. 33).

¹⁵ See Lang. 11.97–8.

¹⁶ The weak vowel of handed is both higher and somewhat farther front than the weak vowel of sofa, just as the vowel of pet is somewhat higher and of course considerably farther front than the vowel of cut. (The greater frontness of the vowel of handed does not appear from the makeshift transcription here used.)

¹⁷ Our vowel symbols differ somewhat from the corresponding symbols used by Bloomfield in the works cited above (fnn. 1, 4). The letters /i, e, u/ have the same value in both systems; but our /a/ = Bloomfield's /ε/, our /o/ = his /α/, and our /ə/ = his /o/.

¹⁸ See Swadesh, Lang. 11.150.

¹⁹ The articulation of [j] and [w] in normal English may be described as follows. For [j], the tongue moves to the position of a following vowel from any relatively higher and more advanced position, the lips being unrounded at least at the beginning of the glide; for [w], the tongue moves to the position of a following vowel from any relatively higher and more retracted position, the lips being rounded at least at the beginning of the glide. The articulation of the nonsyllabic elements here in question is precisely the reverse of this. For the final elements of the diphthongs in bee, bay, buy, boy, the tongue moves from the position of the preceding vowel to a relatively higher and more advanced position, the lips being unrounded at least at the end of the glide; for the final elements of the diphthongs in beau, boo, bough, the tongue moves from the position of the preceding vowel to a relatively higher and more retracted position, the lips being rounded at least at the end of the glide.

²⁰ That the first vowel of away is not similarly rounded is due to a difference in syllabification; cf. §9, last paragraph.

²¹ In dialects that have monophthongs instead of diphthongs of this type, we encounter four fairly uniform long vowels, [i·, e·, u·, o·], in words like beat, bait, boot, boat; in words like bead, paid, food, load and bee, bay, too, go, the vowels tend to be less uniform; and in words like being, baying, doing, going there is often a clear glide, [j] or [w], between the two vowels. The total pattern is best revealed if we state that the long vowels of beat, bait, boot, boat are respectively /ij, ej, uw, əw/; in such a dialect the allophones of the two phonemes in each of these four combinations can be described by a statement such as this: /i, e, u, ə/ combine with the homorganic semivowel—/i, e/ with /j/, /u, ə/ with /w/—so that the phonetic result is an approximation of the two elements to each other; that is, /i, e/ are raised while /j/ is lowered (in comparison with the allophones in pit, pet, yes), and /u, ə/ are both raised and rounded while /w/ is lowered.

If a language has a pattern for combinations of phonemes such as /Vj/ and /Vw/, and if some of the possible combinations are lacking while at the same time certain phonetically pure long vowels are present which are not paralleled by similar vowels elsewhere in the language, the requirements of phonemic theory (complementary distribution, economy in the total number of units, etc.) force us to analyze these long vowels as the lacking combinations of vowel plus semivowel, the identifications resting on phonetic similarity and Pattern congruity. In the present case, [i·] is obviously more similar to /i/ + /j/ than is [e·], and [e·] is more similar to /e/ + /j/ than it is to /i/ + /j/, and [i·] is higher than [e·] just as /i/ is higher than /e/; the relations among the back vowels are parallel.

The principle of pattern analysis here invoked has never been better stated than in the following passage by Swadesh, Lang. 13.10: ‘The general criterion of phonemics is relativity within the totality of the given language. Sounds must be classified according to similarities of phonetic and permutational characteristics. If in this procedure we seek to find the maximally simple, self-consistent, and complete total formulation, we reduce the subjective element in phonetics [emphasized by Bloomfield, Language 84]. Trying to be objective is not likely to succeed fully without an intelligent understanding of the nature of phonetic systems and a constant effort to see each detail in its relation to every other detail. There is a real danger that pattern-conscious investigators may distort the facts in order to make the pattern seem more symmetrical, but this danger is small in comparison with the danger of distorting or failing to notice facts because of giving no attention to pattern.‘

²² And provided that in those dialects where pot, rod, and the like have a rounded vowel, the beginning of the diphthong in bout is unrounded.

^22a On this diphthong in Virginia speech see E. F. Shewmake, MLN 40.489 ff. (1925); id., English Pronunciation in Virginia 24 (n.p., n.d.); Argus Tresidder, Notes on Virginia Speech, American Speech 16.113–6 (1941). The findings of the Linguistic Atlas of the South Atlantic States were presented in a paper (The Diphthong au in Virginia) by Guy S. Lowman Jr., read before the Practical Phonetics group of the Modern Language Association at the 52d annual meeting, 1935.

²³ The analysis made in §10 is essentially the same as that described by Bloomfield in Lang. 11.101, fn. 8, as ‘the customary alternative statement’. Except for Bloomfield's summary, however, we know of no place where such a statement has appeared. Our analysis differs from Bloomfield's in the following respects: in his treatment, the vowels of balm and law are considered unit phonemes, written /a, ɔ/; then the syllabics of buy, boy are analyzed as /aj, ɔj/ respectively, and the syllabic of cow as /aw/. As appears from §§11, 12 below, we analyze the syllabics in balm and law differently, as non-unitary sequences of phonemes; this enables us to deal with the diphthongs in buy, boy, cow more simply in relation to the six short vowels. Cf. also Bloomfield, ibid. 100, last two sentences and fn. 7.

It may be noted here that we regard /juw/ as a normal sequence of three phonemes, in no way structurally different from the combinations /ruw, wuw, waj/ in prove, woo, wine; this analysis is confirmed by such series as Yale: yowl: yule, yeoman: Yuman, and the like. In addition to the word piano with /pj/ before a syllabic other than /uw/ (Bloomfield, ibid. 101, fn. 8), we may cite the local pronunciation of Pueblo (Colorado) as /pjébləw/, and the Southern pronunciation of car, garden, and the like with /kj, gj/.

²⁴ In addition to the not irrelevant fact that when we have analyzed the other syllabics, there is no other category left for the syllabic of bear.

²⁵ Cases like hay-rick, pay-roll with /ej/ are beside the point here, since these words have open juncture before the /r/. But in New England speech both dairy with /ej/ and bearing with /e·/ have close juncture.

²⁶ See Hans Kurath, Mourning and Morning, Studies for William A. Read 166–73.

²⁷ By manufacturing an example ad hoc, it is possible to adduce evidence also for /ə·/. If the exclamation huh may be used, like hem and haw, as a verb meaning ‘to say huh‘, then its preterit is huh'd. This huh'd, in our pronunciation of it, does not rime with cud, but has a distinctly longer vowel. We conclude that cud has /ə/ but that huh'd has /ə·/.

²⁸ Pronunciations of the Irish names Flaherty and Doherty with a short stressed vowel plus [h] must be regarded as outside the phonemic system of normal English, like the nasal vowels of such French words as fiancée and lingerie.

²⁹ Cf. Kenyon, Amer. Pron.⁶ §§37, 200. Kenyon's description of [h] involves a stress pulse coinciding with the beginning of the sound; but we can see no difference between the stress pulse in ahéad and the one in attáck. No one has ever considered such a pulse essential to the articulation of [t] or any other consonant.

³⁰ Cf. the demonstration of the similarity between /j, w/ and the diphthongal glides in §9, fn. 19. The fact that [h] is voiceless while the lengthening element is voiced need not disturb us. Note that /h/ is the only spirant phoneme in English which does not (in Trubetzkoy's terminology) take part in the correlation of voice—in clearer terms, it is the only voiceless spirant which is not phonetically paired with a voiced counterpart.

³¹ It is scarcely necessary to point out that in a phonemic transcription the symbols need not in every case have the values traditionally assigned to them in conventional spelling or in phonetic notation. Naturally we do not suppose that words like pa, law end in the ‘puff of breath’ with which the letter h is usually associated. There is no pressing reason, really, for using h: the raised dot would serve just as well, provided that we used it also for the initial phoneme of words like hat, hill, hay. It might even be wiser to cut loose from all phonetic association by using a symbol like ¿; in that case pa, law, hill, hall would appear as /pá¿, ló¿, ¿íl, ¿ó¿l/. Readers who are disturbed by the looks of our phonemic transcription—and we admit that it looks unfamiliar—should consider Bloomfield's definitive statement on the choice of symbols, Lang. 11.98, fn. 3.

³² The combinations /ih, eh, oh, uh/ are of course the ‘centring diphthongs’ described by Daniel Jones, Outline of Engl. Phon.³ 108–14. A point which does not appear from Jones's practical treatment is that the vowels in palm and bird (described ibid. 72–5 and 86–9) are structurally parallel to these diphthongs. The relation of the diphthong in course to the long vowel in cause needs to be further studied. If the two are different, it may be that course has /or/ and cause has /oh/; and in that case all the centering diphthongs would be /Vr/, so that starry, dearest, dear old would be /stárri, dírrist, . Jones also mentions a long [æ·] in bad, sad, contrasting with short [æ] in lad, pad (ibid. 218). Since bad is different from bared, the latter may again have /Vr/ while the former has /Vh/: lad, bad, bared would then be /lád, béhd, bérd).

³³ We do not claim that the compartments of this table will accommodate all the syllabic phonemes of all dialects of English, though we believe that the exceptions will be very few and in each dialect statistically unimportant. Thus, BB pronounces gonna (I'm not gonna do it) with a short vowel in the first syllable which is phonetically very close to the vowel of German Sonne. Though it occurs nowhere else in his pronunciation of English, it must perhaps be reckoned an independent phoneme parallel to the six short vowels of §7. Cf. fn. 14.

³⁴ Among them Dr. C. F. Hockett, who has given us our information on this point.

³⁵ Joseph Wright, The English Dialect Grammar 465 (Oxford, 1905), lists the following pronunciations of good which may be analyzed as containing /uj/: [gyd] Sh.I. sn. & nm.Sc., [gy·d] wm.Sc. n.Cum. e.Dev., [guid] sw. & ms.Yks.

³⁶ This is the diphthong transcribed [iu] and described for his own dialect by Kenyon, Amer. Pron.⁶ §§341–55.

³⁷ On the ‘New England short o‘ see Handbook of the Linguistic Geography of New England 3 (and Chart 1), 128 (§17).

³⁸ /ir, er, ar, or, ur/ may also be the correct analyses, instead of /ih, eh, ah, oh, uh/, for the r-less pronunciation of beer, bear, bar, bore, boor; §14, fn. 32.