¹ The present paper derives from the phonemic systems of Bloomfield, Sapir, Trubetzkoy, and their followers. It owes most to Bloomfield, though rather to the methodological rigor of his work than to his phonemic theories. Apart from published material, the writer's chief stimulus has been his correspondence and discussion with George L. Trager and Bernard Bloch. He owes thanks also to Morris Swadesh for many suggestions in the past; to Mary R. Haas for specific criticism of the present paper; and above all to Bloch for advice and active assistance in giving the paper its final shape.

Many of the principles here set forth are illustrated and applied to the problem of English vowels by G. L. Trager and B. Bloch, The Syllabic Phonemes of English, Lang. 17.223–46 (1941). For a general bibliography on phonemic theory, see fn. 1 of that article.

² Albert P. Weiss, A Theoretical Basis of Human Behavior²; Columbus, 1929.

³ The term ‘phonetics’ has two distinct meanings in current usage, unfortunately not always kept clearly apart. To avoid coining one or two completely new terms, this paper uses the expressions α-phonetics and β-phonetics for the two different meanings.

⁴ Kenneth L. Pike, A Reconstruction of Phonetic Theory; University of Michigan diss., 1941. This is the purest treatment of α-phonetics known to the writer; most of the α-phonetic material here presented is taken from Pike's work.

⁵ Phrases in the form 'the initial sound in beat' are to be taken as shorthand for the logically correct but cumbersome 'the first sound of an utterance consisting of the word represented in traditional orthography as beat'.

⁶ It is assumed that all the utterances involved have the same prosodic features; meat? and meat! are of course phonologically different.

⁷ Transcriptions used in this paper are of two kinds, which it will be necessary to keep rigorously distinct. Phonemic transcriptions are enclosed between diagonals; β-phonetic transcriptions are enclosed between square brackets (α-phonetic transcriptions do not occur). This way of distinguishing the two kinds of transcription is adopted from the article by Trager and Bloch (see fn. 1).

⁸ Examples from English in the rest of this paper are from the writer's dialect, which is of the second of these two types.

⁹ On the phonemic interpretation of English stress see Trager and Bloch, The Syllabic Phonemes of English, Lang. 17.226–9.

¹⁰ On the problem of intersection see Bernard Bloch, Phonemic Overlapping, American Speech 16.278–84 (1941).

¹¹ Another problem, that of multiple complementation, has been discussed by W. F. Twaddell, On Defining the Phoneme 30–2 (Lang. Mon. No. 16, 1935) and Lang. 12.54–6 (1936); but this problem is not a real one. If a phone a in a certain position is in complementary distribution with two other phones b and c, and is phonetically similar to both, either (1) both a and b or both a and c can be unambiguously assigned to the same phoneme, or else (2) a, b, and c are members of three separate phonemes.

1. (1)

   (1) The stop in English spin (voiceless, fortis, unaspirated) is in complementary distribution with both the stop in pin (voiceless, fortis, aspirated) and the stop in bin (voiced, lenis, unaspirated). In §7.1 we have defined /p/ as characterized by fortis articulation, voicelessness, and some other features; this definition unambiguously includes the stop in spin.

2. (2)

   (2) To illustrate the other possibility we must invent an example, for no real case is known to the writer. In a certain language, /i/ and /e/ are found in contrasting positions everywhere except before /n/; before /n/ there is a vowel phone which is exactly half-way between /i/ and /e/ and which occurs nowhere else. The latter vowel shares with /i/ and /e/ the distinctive feature of palatal articulation, but does not share the feature of intermediate tongue-height with any other phone. If these statements include all the relevant facts, then the cases of the intermediate vowel (in spite of its limited distribution) constitute a separate phoneme. Certain dialects of English in the South, where pin and pen are homonyms, may require this analysis.

¹² Spanish data are from Navarro-Tomás, Manual de Pronunciación Española⁴, Madrid, 1932; G. L. Trager, The Phonemes of Castillian Spanish, TCLP 8.217–22 (1939); and the writer's observations.

¹³ Variations of sonority may occur α-phonetically; β-phonetically these may be distinctive or not. The terms ‘syllable’, ‘nucleus’, and ‘non-syllabic’ are used here only on the phonemic level, to designate a grouping of phones which is useful in describing the phonemics of a particular language. Where nothing is said to the contrary, it is assumed that points of syllable division are non-distinctive, or that they are definable in terms of the other features involved.

¹⁴ See N. S. Trubetzkoy, Grundzüge der Phonologie 206–18 (Prague, 1939).

¹⁵ Winnebago data are from Amelia Susman.

¹⁶ Kickapoo data are from the writer's notes.

¹⁷ It is possible; see Trager and Bloch, Lang. 17.237–41.

¹⁸ Willem L. Graff, Language and Languages 41 (New York and London, 1932).

¹⁹ Greek and Latin data are from E.H. Sturtevant, The Pronunciation of Greek and Latin²; Linguistic Society of America, 1940. The phonemic analysis of the Greek breath-types is G. L. Trager's.

²⁰ The transcription of vowels represents the phonemic analysis by G. L. Trager and the writer; its correctness is irrelevant to the validity of these examples. The phonemic nature of zeta is not known.

²¹ Our traditional predilection for treating [p] as a unit and [p'] as a cluster can only lead us astray in dealing with Winnebago; for the former here patterns like a cluster, the latter like a unit phone, and the habitual voicing of the second member of clusters explains quite adequately how /pha/ is [pa]. A similar comment applies to the example immediately following.

²² Edward Sapir, Takelma (in Franz Boas, Handbook of American Indian Languages, vol. 2; Washington, 1922).

²³ Potawatomi data are from the writer's notes.

²⁴ The phonemic transcription of English here used, including the marking of stress and juncture, follows the practice described by Trager and Bloch. An acute accent over a vowel symbol denotes the loud stress, a circumflex the reduced loud, a grave the medial; absence of a mark denotes the weak stress.

²⁵ A. Mazon, Grammaire de la langue tchèque; Paris, 1921. The cited phrase occurs in §161; stress is discussed in §6. The transcription given here is certainly erroneous on several points, but the junctural indications are probably correct. The writer has not heard Czech himself.

²⁶ Observed by the writer in a class under C. F. Voegelin at the Linguistic Institute of 1939. Voegelin's orthography for Delaware is quite unambiguous and therefore perfectly satisfactory for the collection of texts, though it does not agree with the present analysis.

²⁷ Whitney, Skt. Gr.⁵ §§125b, 329f, 338e, 353a. Some rule must be establishable, of course, to tell exactly when /u/ or /r/ will be consonantal and when vocalic. If it is impossible to discover such a rule, then the proposed analysis cannot be accepted.

²⁸ See the exhaustive structural classification of English non-syllabics in Bloomfield's Language 131–4 (New York, 1933). Bloomfield establishes 38 overlapping sets defined by the occurrence of consonants singly and in clusters.

²⁹ G. L. Trager, La Systématique des phonèmes du polonais, Acta Linguistica 1.179–88 (1939).