¹ Leonard Bloomfield, A set of postulates for the science of language, Lang. 2.153–64 (1926). In December 1925, a paper of the same title was presented by Bloomfield at the second annual meeting of the Linguistic Society of America in Chicago (Lang. 2.73).

² The following works deal with phonemic theory: Leonard Bloomfield, Language 74–138 = chaps. 5–8 (New York, 1933; reviewed by Kent and Bolling, Lang. 10.40–52); Edward Sapir, La réalité psychologique des phonèmes, Psychologie du langage 247–65 (Paris, 1933); Yuen Ren Chao, The non-uniqueness of phonemic solutions of phonetic systems, Bulletin of the Institute of History and Philology (Academia Sinica) 4.363–97 (1933); Morris Swadesh, The phonemic principle, Lang. 10.117–29 (1934); W. Freeman Twaddell, On defining the phoneme (Language Monograph No. 16, 1935); Swadesh, The phonemic interpretation of long consonants, Lang. 13.1–10 (1937); N. Van Wijk, Phonologie: een hoofdstuk uit de structurele taalwetenschap ('s-Gravenhage, 1939; reviewed by Trager, Lang. 16.247–51); N. S. Trubetzkoy, Grundzüge der phonologie (TCLP 7: Prague, 1939; reviewed by Harris, Lang. 17.345–9); George L. Trager and Bernard Bloch, The syllabic phonemes of English, Lang. 17.223–46 (1941); Bloch and Trager, Outline of linguistic analysis 38–52 = chap. 3 (Baltimore, 1942; reviewed by Sturtevant, Lang. 19.42–4, and Emeneau, JAOS 63.208–9); Charles F. Hockett, A system of descriptive phonology, Lang. 18.3–21 (1942); Kenneth L. Pike, Phonemics (Glendale, Calif., 1943); Zellig S. Harris, Simultaneous components in phonology, Lang. 20.181–205 (1944); Rulon S. Wells, The pitch phonemes of English, Lang. 21.27–39 (1945); Pike, The intonation of American English (Ann Arbor, 1945; reviewed by Wells, Lang. 23.255–73).

³ Lang. 2.143–4.

⁴ Lang. 2.157, §15 and §16.

⁵ Twaddell, op.cit. 17–25.

^5a Compare the statement by Edna Heidbreder, Seven psychologies 368 (New York and London, 1933): ‘Science, of course, is not above muddling through. Scientific practice often runs ahead of theory; it often proceeds unwittingly and unintentionally on assumptions that it discovers only after it has used them. But the justification for this procedure is that in the actual business of acquiring and interpreting its data, those assumptions eventually do come to light. Somewhere along the line there is a point where the scientist encounters recalcitrant facts that force him to examine the assumptions to which they refuse to conform. Such a point is always an important one in the development of science.‘

⁶ See R. K. Potter, Visible patterns of sound, Science 102.463–70 (1945) = Bell Telephone System Technical Publications, Monograph B-1368; R. K. Potter and others, Technical aspects of visible speech, Journal of the Acoustical Society of America 17.1–89 (1946) = BTSTP, Monograph B-1415; R. K. Potter, G. A. Kopp, and H. C. Green, Visible speech (New York, 1947). In an unpublished mimeographed booklet of 58 pages, entitled Basic notions for acoustic phonetics, Martin Joos has summarized a course of lectures given at the 1947 session of the Linguistic Institute, in which he developed some of the linguistic implications of sound spectrography. See now his new book, Acoustic phonetics (Language Monograph No. 23, 1948).

⁷ In arriving at the present formulation of these postulates, I have profited immeasurably from the criticisms and suggestions of many colleagues. I am especially indebted to Yuen Ren Chao, Charles F. Hockett, Henry M. Hoenigswald, Martin Joos, W. Freeman Twaddell, and—above all—Rulon S. Wells.

⁸ Theoretically it would be possible to arrive at the phonemic system of a dialect entirely on the basis of phonetics and distribution, without any appeal to meaning—provided that in the utterances of the dialect not all the possible combinations of phonemes actually occurred. Given a sufficient sample of the dialect—say twenty or thirty hours of connected speech by an informant, recorded either on a high-fidelity machine or in a minutely accurate phonetic transcription—a linguist could probably work out the phonemic system without knowing what any part of the sample meant, or even whether any two parts meant the same thing or different things. (But he would need some kind of guarantee that every part of the sample meant something.) Whether his final description of the sample turned out to be an adequate description of the dialect as a whole would of course depend on how inclusive and how representative the sample was; but this would be true to some extent even if he followed the usual (and more sensible) method of asking what each part of the sample means.

It is worth noting that Daniel Jones's definition of the phoneme, for all its want of precision, is essentially non-semantic; see for instance An outline of English phonetics³ 48, §191 (Cambridge, 1932). Especially striking is this passage, in Le Maître Phonétique 3.7.44 (1929): ‘An important point to notice is that the phoneme is essentially a phonetic conception. The fact that certain sounds are used in a language for distinguishing the meanings of words doesn't enter into the definition of a phoneme. It would indeed be possible to group the sounds of a language into phonemes without knowing the meaning of any words.‘ Cf. Twaddell, op.cit. 25–7 and fn. 11.

⁹ For the larger postulational framework of science in which the present set must be supposed to appear, see Twaddell, op.cit. 36 and 37. As for the precise number of postulates in any set, this is determined largely by stylistic considerations; for whether a given fact or complex of facts is to be covered by one postulate or by more than one depends at least in part on how easy it is to state the matter clearly in a single sentence. The number of our postulates could have been increased by dividing certain of them into two or more separate statements, or reduced by combining certain others into one.

¹⁰ 'An act of speech is an utterance' (Lang. 2.154, §1). Here the term ‘speech’ is used without a definition.

¹¹ See Hockett, Lang. 18.5–6, §4–§4.4.

¹² On components as parts of single sounds, and on their possible value in phonological analysis, see Harris, Lang. 20.186–201, §3.0–§5.5.

¹³ The possibility of different but equally valid formulations of a phonological system is the point of the article by Chao, Bull. Inst. Hist. Phil. (Academia Sinica) 4.363–97.

¹⁴ I take this term from a forthcoming monograph by R. S. Wells, to be entitled History of the phoneme concept.

¹⁵ In an experiment conducted at the 1947 session of the Linguistic Institute (cf. fn. 6), the sound spectrograph produced nearly identical records for two ‘retroflex’ or ‘r-colored’ vowels (as in bird) pronounced by the same speaker—one formed with simultaneous raising of the tongue tip toward the middle of the hard palate, the other formed without such raising but with retraction and lateral spreading of the main body of the tongue. The differences between the two spectrograms were well below the threshold of audibility; and a group of trained observers agreed that they could detect no difference between the two sounds as heard.

¹⁶ The phonetic transcriptions in this paper are not intended to be full or accurate recordings of any real utterances; in other words, they are not really transcriptions at all, but abstractions from possible transcriptions. The phonetic symbols have the values usually assigned to them. We follow in this paper the practice introduced in Lang. 17.229 and now generally accepted, of placing phonetic symbols between square brackets, but phonemic symbols between diagonals.

¹⁷ See Paul Menzerath, Neue Untersuchungen zur Wortartikulation, Actes du IV^e congrès international des linguistes 67–75 (Copenhagen, 1938); and cf. Menzerath and A. de Lacerda, Koartikulation, Steuerung und Lautabgrenzung (Berlin and Bonn, 1933).

¹⁸ See Hockett, Lang. 18.5, §3.3.

¹⁹ Our segments are thus established by a different method from Pike's; see his book Phonetics 107–20 and passim (Ann Arbor, 1943). We define segments by first defining their boundaries; Pike defines them by concentrating on their centers. His definition: ‘A segment is a sound (or lack of sound) having indefinite borders but with a center that is produced by a crest or trough of stricture during the even motion or pressure of an initiator ...‘ (107). The terms in this definition are defined earlier in his book.

²⁰ Harris, Lang. 20.181–205. A description of a phonological system in terms of components is attempted by Hockett, Componential analysis of Sierra Popoluca, IJAL 13.258–67 (1947); but what Hockett there calls ‘components’ are rather what in the present paper we call aspects.

^20a As announced in §0.4, each postulate from this point on is to be read as beginning with the words: ‘In any dialect’; except that each postulate marked with a dagger is to be read as beginning with the words: ‘In any of certain dialects’.

²¹ It would of course be possible to frame special postulates to cover such a choice, but we shall not attempt the task here. See the first sentence in §0.3.

²² This is, of course, a grossly oversimplified statement of the situation, but it will serve as an example. Cf. Bloomfield, Language 381, §21.7 (New York, 1933); W. F. Twaddell, A note on Old High German umlaut, Monatshefte für deutschen Unterricht 30.177–81 (1938).

²³ For instance in Bloch and Trager, Outline of linguistic analysis 42, §3.4.

²⁴ Once again an oversimplified statement will serve as an example. We have not, of course, listed here all the environments in which the two segments [l] and [l] occur.

²⁵ This term is adopted from Hockett, Problems of morphemic analysis, Lang. 23.328, §8 (1947).

²⁶ For instance in Bloch and Trager, op.cit. 38, §3.1.

²⁷ If we had chosen some other segment than [k] for comparison—say [t], which is also homotypical with [p], or [š], which is homotypical with [f]—we might have been able to demonstrate a still closer similarity between the range of the dyad [pf] and that of the chosen segment. In practical work, the investigator must choose the segment that will give him the best results; and this he can sometimes do only after repeated trials.

²⁸ I am indebted for this information to Henry M. Hoenigswald. Cf. his manual Spoken Hindustani 71 (U. S. Armed Forces Institute, EM 544, 1945).

²⁹ The first span is [blæk], not merely [blæ]; for the segment [k], not being open to the feature of pitch, does not mark a boundary of the span.

³⁰ Compare the similar convention (on which ours is modeled) by which Hockett justifies his use of ‘morpheme’ as a variable, Lang. 23.324, fn. 12.

³¹ Cf. Bernard Bloch, Phonemic overlapping, American Speech 16.278–84 (1941)—a statement in some respects now seriously antiquated.

³² Bloomfield's treatment of American English vowels is open to precisely this objection. See for instance his Language 112, §7.4, where in successive lines the sound [ə] is represented by the symbol /o/ (= λ) in concerted and by the symbol /e/ in address.

³³ On the syllable as a phonemic—not a phonetic—term, see Henry M. Hoenigswald, JAOS 64.155 and fn. 5 (1944).

³⁴ On juncture in English see Trager and Bloch, Lang. 17.225–6, §4 and fn. 5; Bloch and Trager, op.cit. 35–6 and 47, §2.14 and §3.7. On juncture in German see William G. Moulton, Juncture in modern standard German, Lang. 23.212–26 (1947). Cf. also Rulon S. Wells, Immediate constituents, Lang. 23.107, §64 (1947): ‘The validity of juncture phonemes is open to grave doubts on phonetic grounds.‘

In his work on American intonation, Pike does not operate with juncture phonemes as such, but treats some of the same phenomena that Trager and Bloch regarded as juncture. See the discussion of this book by Wells, Lang. 23.257–9 and 262–3, §8, §11, and §15 (1947), and the references there given.

³⁵ This is Moulton's solution, Lang. 23.220, §3.14; cf. Wells, Lang. 23.108 fn. 45. The terms ‘pause phoneme’ and ‘zero allophone’ of course call for separate postulates, as yet unformulated:

³⁶ The imaginary situation here described is modeled on a situation said to exist in Bengali.

³⁷ The terms referring to stress (loud, reduced-loud, medial, weak) are taken from Bloch and Trager, op.cit. 47–8, §3.7.

³⁸ The definition of an aspect in relative terms (‘higher and more forward‘) involves an assumption that we have not stated—one of the special assumptions needed to cover the facts of a particular language (cf. §58.1).

³⁹ Distinctive aspects of pitch are denoted by inferior numerals before the letter: [₂] for higher-mid, [₃] for lower-mid, [₄] for low. Distinctive aspects of stress are denoted by superior ticks before the letter: [] for loud, [] for reduced-loud, ['] for medial; weak stress, regarded here as determined, is not marked. Other distinctive aspects are implied by the usual phonetic values of the letters themselves. Segments are separated by hyphens for greater ease in reading; but these have no phonetic or phonemic significance.

⁴⁰ In this section we simplify our transcription by omitting marks of stress and pitch.

⁴¹ There are only fifteen different phonemes in the utterance, but two of them occur twice.

⁴² Cf. Wells, Lang. 23.107 §63, 110 fn. 52, 116 §84, 117 §88.

⁴³ These terms have no meaning beyond what we have just assigned to them. We purposely avoid the terms 'suprasegmental and ‘segmental’, because their traditional meaning is irrelevant here. In the traditional sense, any feature that extends over more than one segment is suprasegmental as compared with any feature that occurs in one segment only. Thus, in our present example, pitch is suprasegmental as compared with vowel color in segment 1, but vowel color is suprasegmental as compared with pitch in segment 10. Such a reversal of status shows that the distinction between suprasegmental and segmental has no useful meaning within the framework of our postulates.