¹ The central importance of the problem of immediate constituents was driven home to me in many valuable conversations with Zellig S. Harris, who also let me read a number of his manuscripts, of which not all have yet been published. Subsequently it was my privilege to acquire a minute acquaintance with Bernard Bloch's fine and well-balanced description of Japanese. Extensive viva voce discussion yields an insight into nuances, the weighing of alternative possibilities, and the far-reaching implications of any assumption that no written account can achieve.

I am indebted to Bloch, to Elizabeth F. Gardner, and to Henry M. Hoenigswald for pointing out ways in which two earlier drafts of this paper needed revision.

The present study is part of a project sponsored by the American Council of Learned Societies, whose support I gratefully acknowledge.

² Lang. 22.161–83 (1946).

³ Farly does not occur in actual utterances. The grammarian must decide whether this non-occurrence is accidental—because farly would have no meaning, or because the meaning which it would have is expressed by some other, current expression, or because (like brown-bird; cf. blackbird, bluebird) it describes an object or a situation which people never have occasion to talk about—or whether, on the other hand, farly is grammatically impossible, like pickly or kindlily. To demarcate what is grammatically impossible (ungrammatical) from what does not occur merely for some stylistic or semantic reason is a difficult problem. We do not deal with it here because it is irrelevant to the theory of ICs, which takes as its data all those utterances known to occur, and analyzes them.

In citing environments, we use a pair of parentheses to indicate the position of morphemes or morpheme-sequences that occur in a given environment. Thus, the parentheses in ()ly, above, indicate the position of such morphemes as slow, near, quaint before the morpheme ly.

⁴ The converse does not hold. For instance, they is not substitutable for Tom and Dick in the sequence I met Tom and Dick downtown.

⁵ Leonard Bloomfield, Language 212 (New York, 1933); Kenneth L. Pike, Taxemes and Immediate Constituents, Lang. 19.65–82, esp. §4.19 (1943).

One particular version of simplicity which someone might propound but which must be rejected would call for the totality of analyses that requires the minimum number of sequence-classes to be defined. The proposal is sound in itself but inadequate to exclude all wrong analyses, such as the king of | England. Note that the king of must be described as a constituent in certain other occurrences, e.g. What country is he the king of?.

⁶ In the sequel, we shall often speak loosely of a sequence as occurring in the zero-environment when we mean that its total environment is the pitch-morpheme of the sentence. Cf. fn. 34.

⁷ This statement is made on the assumption that worked consists morphologically of only the two morphemes work and -ed. If, as some would consider, the proportion they work: John works = they worked: John worked is perfect, so that worked includes also a zero-alternant of the 3rd-singular morpheme -s when it has a singular subject, then John works is indeed shorter than John worked (since it lacks the past-tense morpheme -ed), and furnishes a simpler example of the fundamental sentence type.

⁸ So are Bloomfield's ‘form-classes’; cf. his Language 185 top, 190 top, 194 second paragraph, etc.

⁹ We call an IC-analysis wrong when there is another possible analysis of the same sequence that is better, and right when there is none.

¹⁰ That the king of England is itself an expansion of John is relevant in determining the king of England to be a constituent of any longer sequence in which it occurs, but not relevant in determining its own constituents.

¹¹ Until we admit discontinuous constituents. See Part V of this paper.

¹² The circularity is only apparent. What is assumed as previously defined is the whole phrase ‘preceding modifier’ and the whole phrase ‘following modifier’. A third type of modifier is discussed in §56.

¹³ Cf. Moroccan Arabic n- ‘I will’ and -t ‘I did’, in nmši ‘I will go’ and mšit ‘I went’ (Harris, op.cit. §3.1).

¹⁴ And also to a number of focus-classes in common. But it is always a coincidence when a paradigm is also a focus-class; it may be that some paradigms in some languages have no environment characteristic of all their members.

¹⁵ But a focus-class to which only one of the three sequences am, is, are belongs is characterized by more environments than one to which two or all three of them belong, because of the grammatical agreement which these words require with their subject or actor.

¹⁶ But not, for example, in the environment he () go. Here the expansion of can is wants to. Want to and wants to are paradigmatically related, like want and wants, so that if want to is treated as a constituent, it is desirable to treat wants to in the same way. And if, on other grounds, it has been decided to treat them both as constituents, then they will conjointly support the analysis of any sequence of which they are conjointly either an expansion or a model.

¹⁷ For the purposes of stating this principle, we assume that every sequence has a meaning, though actually it is in general much more difficult to define the meaning of non-constituents—especially when they are discontinuous—than of constituents.

¹⁸ We shall justify in §54 the analysis of a noun phrase A and B into three ICs, A|and|B.

¹⁹ The reader must constantly bear in mind that our definition of this term is not the same as Bloomfield's (Language 169), although both are generalizations of what is meant in traditional grammar by such expressions as ‘the ablative absolute construction’.

²⁰ Cf. Y. R. Chao, The Logical Structure of Chinese Words, Lang. 22.5 (1946).

²¹ Therefore, the statement ‘pages and pages contains three morphemes but five morpheme -occurrences’ is only a loose way of saying ‘pages and pages contains three morphemes and every occurrence of pages and pages contains five morpheme-occurrences’. When we said, in §3, that two members of the same sequence-class must contain the same number of morphemes, we meant that these members, or more strictly any pair of occurrences of them, must contain the same number of morpheme-occurrences.

A sequence might be defined as a class of occurrences; then the statement ‘the sequence A occurs in the environment B’ could be explained as meaning ‘some members of A occur in some members of B’.

²² For instance, if no condition of type (3) is imposed, every sequence-class is also a construction.

²³ The four English pitch phonemes are designated by the numerals from 1 (highest pitch) to 4 (lowest pitch), following Kenneth L. Pike, The Intonation of American English (Ann Arbor, 1946). In a phonemic transcription, each pitch phoneme is written as a superior numeral before the syllabic which it accompanies; when a succession of two pitch phonemes accompanies a single syllabic, the two appropriate numerals are written before the appropriate letter. The absence of a superior numeral before one or more successive syllabics means that those syllabics bear the same pitch phoneme as the last syllabic preceding them on which a pitch is indicated; see fn. 56. But we write, for example, I won't go pronounced emphatically as /³ayw²ôwnt-g²⁴ów/, where the practice described above would lead to confusion with /³ayw²ównt-g⁴ów/; see §79. It is often convenient (e.g. in §54) to write a mixture of standard orthography for the segmental phonemes and phonemic transcription for the pitch phonemes; in this mixed notation, the stress phonemes (see §§66 ff.) are sometimes indicated and sometimes not, depending on their relevance.

Pitch morphemes, like segmental morphemes, are generally identified by citing one of their alternants—the alternant of a pitch morpheme being a sequence of one or more pitch phonemes. The alternant of the pitch morpheme contained in the first pronunciation of I won't go mentioned above would be written either as /32(24)/ (the parentheses indicating that the last two pitch phonemes accompany the same syllabic) or—with a deliberately ambiguous cover notation—as 324, which indicates what pitch phonemes occur and in what order, but not how many syllabics there are in the phrase.

²⁴ More strictly, every occurrence of it that has that meaning.

²⁵ But there is also the case where two sequences, even of three or more morphemes each, have different constructions but the same analysis into ICs; e.g. go | to | | the | | | store (a) as a command and (b) in answer to the question What are you going to do? So also we treat the cases where others would recognize a solitary IC (cf. §52) and the cases of wholly homonymous constructions (fn. 30).

²⁶ Special cases are presented by morphemes that have more than one morpheme-alternant. Thus, the four alternants of the English 3rd-singular morpheme /-z, -s, -ez, -0/, as in reads, writes, teaches, can, are respectively homonymous with four alternants of the plural morphene, as in dogs, cats, horses, sheep, but not with the alternants that appear in oxen, children, geese, etc. But such cases do not concern us here.

²⁷ There are other constructions to which both (A) old | men and women and (B) old men | and | women belong, such as the noun-phrase construction; but these constructions are not responsible for the difference in meaning between A and B, and hence do not interest us here. Both the special construction of A and the special construction of B are compatible with all the more general constructions to which both A and B belong.

²⁸ Assuming these and our other examples to be compounds rather than word-sequences.

²⁹ Unless we choose to assume that there is no such word as lady-killer ‘killer who is a lady’, and conclude therefore that in this respect friend and killer belong to different focus-classes.

³⁰ In this example, as in all wholly homonymous constructions that have come to my notice, the IC-analysis is the same regardless of which construction the sequence belongs to. If a given sequence is to have different analyses into ICs according to the different constructions to which it belongs, it must necessarily contain at least three morphemes.

³¹ Language §11.5 end. Also by Eugene A. Nida, Morphology 149–50 (Ann Arbor, 1946); Bloch and Trager, Outline of Linguistic Analysis 67 bottom (Baltimore, 1942).

³² Chapter 7 of his Morphology (cf. fn. 31). Also noteworthy are Bloch's characterization of the Japanese word (Studies in Colloquial Japanese II: Syntax, Lang. 22.202–6, §§1.3–6 [1946]) and Chao's characterization of the Chinese word (op.cit.).

³³ The term grammar is here used in the narrower sense, as coordinate with phonemics, rather than including it. In this narrower sense, grammar is commonly divided into morphology and syntax: the former treating the internal grammar of words (§15), the latter their external grammar and the sequences of words; but the division does not apply to languages that lack words as a distinct kind of unit. Harris (op.cit.) shows how it is possible to ignore the distinction between morphology and syntax even in other languages (cf. §45). The term taxemics has been suggested to me to cover the narrower sense of grammar; I have not adopted it, because it rests on Bloomfield's unclearly defined term taxeme.

³⁴ See fn. 6 and Bloomfield, Language 168–9. Bloomfield's definition must be suitably modified if the sentence-intonations and certain stresses (§66) of such a language as English are regarded as morpheme-sequences rather than (op.cit. 163, 169) as modulation, one of the four ‘ways of arranging linguistic forms’.

³⁵ Bloch, op.cit. §1.5 and fn. 7. Cases where one would expect this result are rare; and when they do occur (perhaps in personal proper names?) the prior word morphophonemically loses its accent (Bloch, op.cit. fn. 5).

³⁶ Bloch, op.cit. §1.5.

³⁷ Bloomfield recognizes the possibility of solitary constituents (Language, 218) and of multiple constituents (194, lines 10–11; 227, line 6). He is followed by Pike, Taxemes and Immediate Constituents, Lang. 19.70, line 1 and §4.1. Bloch and Trager (op.cit. 67) recognize ‘usually . . . two and only two immediate constituents’. The same view seems to be taken by Nida (Morphology 81; Syntax §2.5.3 et passim [Ann Arbor, 1946]). Pike further admits discontinuous constituents (op.cit. §§4.12–4).

³⁸ Bloch, op.cit. §2.4. For a similar case in morphology, see Bloch, Studies in Colloquial Japanese III: Derivation of Inflected Words, JAOS 66.305, §1.3 (1946).

³⁹ For the factual data, see Bernard Bloch and Eleanor H. Jorden, Spoken Japanese, 135, Note 5.15, and 504, Note 16.18 (New York, 1945–46). Bloch no longer insists on this way of describing the derivation.

⁴⁰ Usages such as [Washington was] first in the hearts of his countrymen are not regular or free.

⁴¹ Note that see through him occurs, often bearing the emphatic morpheme on through; but wake up him does not, no matter how accented.

⁴² Bloomfield, Language 163; George L. Trager, The Theory of Accentual Systems, Language Culture and Personality 131–45 (Sapir Memorial Volume; Menasha, Wis., 1941)-Bloch and Trager, op.cit. 41. For further references see R. S. Wells, The Pitch Phonemes of English, Lang. 21.28 fn. 10 (1945).

⁴³ A morpheme is called segmental if each of its one or more morpheme-alternants consists either of one or more vowels and consonants (with or without stresses /ˆ or /ˋ/), or else of nothing at all (a zero-alternant).

⁴⁴ Bloomfield (Language 113) proposes to phonemicize the contrast between a name and an aim in terms of stress rather than of juncture: /ə ńeym/ vs. /ən éym/ rather than /ənéym/ vs. /ən-éym/. This is a good suggestion, but since the cases of contrast in English between ĆV and CV́ are relatively few, we retain here the more customary notation.

⁴⁵ In order to preclude extravagances, one might adopt the requirement that every phoneme have at least one allophone consisting of an actual segment of sound, or else of a silence between or before or after such segments. This would involve assigning some segments simultaneously to more than one phoneme—e.g. assigning the first i of What is it? simultaneously to the vowel phoneme /i/, to the pitch phoneme /2/, and to the stress phoneme /′/. See R. S. Wells, op.cit. fn. 7. If we could reckon the silence at the beginning and end of an utterance as one allophone of juncture, we would be justified in positing another allophone, a zero-allophone, which occurs for example in night-rate.

⁴⁶ Except perhaps short sentences in which a normal occurrence of /′/ is replaced by an occurrence of the contrastive stress /¡/; see §70.

The system of stress-phonemes assumed in this paper is the one described by Bloch and Trager, Lang. 17.226–9 and Outline of Linguistic Analysis 47–8; but we differ from Bloch and Trager in regarding their phoneme of weak stress as simply the absence of a stress phoneme. The remaining three stress-phonemes are written as accents over vowel letters: /á/ = Bloch and Trager's ‘loud stress’, /â/ = ‘reduced loud stress’, /à/ = ‘medial stress’.

I am especially indebted to Bloch for supplying the phonemic transcriptions of English in Part VI, which reflect his own dialect. To get a system of English phonemes that will apply, with modifications, to as many different dialects as possible, it has seemed wise to take as a descriptive basis a more complicated dialect like Bloch's, rather than a less complicated one like mine, which does not distinguish, for instance, between ‘reduced loud’ stress and ‘medial’ stress.

⁴⁷ Of course it is easy to design a notation that will show which syllabic potentially bears the stress, e.g. /bæ̂ləklævə/. This use of italics, like the leaving of a space between words in most phonemic transcriptions, has no phonemic status whatever; it is purely a mnemonic device equivalent to the statement that the stress /′/ (belonging to the emphatic morpheme) occurs—when it occurs at all in conjunction with a given segmental morpheme—on such-and-such a syllabic of that morpheme. Moreover, simply to cite the phonemic result of juxtaposing the segmental and the stress morpheme (e.g. /bæ̂ləklǽvə/, the result of juxtaposing /bæ̂ləklævə/ and /′/) conveys the same knowledge; and so does a comparison of alternants like /bæ̂ləklævə/ and /bæ̂ləklæ̂və/.

⁴⁸ Exceptions are morphemes like the loom in heirloom /éhr-lûwm/, which accidentally occur only in compound words.

⁴⁹ The verbs alternate /óhltərnèyt/, degenerate /dijénərèyt/, etc. might be regarded morphologically as derived from the adjectives alternate /óhltərnət/, degenerate /dijénərət/, etc. by the addition of a morpheme; or the adjectives might be regarded as derived from the verbs in the same way. If so, the morpheme /èyt/ has an alternant /ət/. Moreover, the nouns alteration, degeneration, etc. exhibit the alternants /êyš/ and /eyš/ of the same morpheme. But even so, the verb as a word by itself never shows an alternant /eyt/ or /ət/ or /eyš/.

⁵⁰ Bloch and Trager posit ‘contrasting intonation /¡/, involving both a distortion of the normal sentence tone and an extra-loud stress’ (op.cit. 52). But we may equally well view the stress, rather than the pitch, as the non-automatic feature. This allows us to group the phoneme /¡/, (now renamed ‘the contrastive stress phoneme‘) with the phoneme /′/, which it resembles (1) in always constituting a morpheme by itself, namely the contrastive morpheme, and (2) in its meaning and grammatical patterning.

⁵¹ According to Bloch's oral statement.

⁵² It would be possible to make emphasis and contrast phonemically as well as morphologically segmental, in the manner suggested in §63: what we now write as /hw³àt-²íz⁴it/ would become, say, /hw3ˋat-2´iz4it/. Applying this procedure to all utterances would result in two types of discontinuity: (1) discontinuous morphemes and (2) discontinuous sequences. The morphemes would be discontinuous not in the sense treated by Harris (Discontinuous Morphemes, Lang. 21.121–7 [1945]), but in the sense in which the Arabic morphemes a...a...a ‘past tense, 3rd person singular’ and k...t...b ‘write’ are each discontinuous in the word kataba ‘he wrote’. The sequences would be discontinuous in the sense discussed in Part V above; cf. §88. When an alternant of one morpheme phonemically intrudes into or interrupts an alternant of another morpheme, it is convenient in morphology to consider that one of the morphemes wholly precedes the other—e.g. that ktb, in the example just given, wholly precedes or wholly follows aaa. The reason is the one already mentioned: that a perfectly linear sequence of units is the easiest to deal with. Accordingly, the sequence /hw3ˋat-2´iz4it/ would be morphologically described, say, as follows—using plus-signs to separate single morphemes in the alternants that occur in the sentence What is it?: hwˋat + juncture + iz 4- ′ 4- it + 324.

⁵³ Though not minimally; for when blackbird is contrastively stressed with /′/, pie appears in the alternant /pây/, unaccompanied by the emphatic morpheme. All the sentences (A), (B), and (C) contain the same pitch-morpheme, though in different altérnants.

⁵⁴ See §11.

⁵⁵ The Intonation of American English; cf. fn. 23.

⁵⁶ An alternative notation would be to ascribe two occurrences of pitch-phonemes to every syllabic. Then the pitch-phonemes of (a) would be written /33333324/, which would indicate four syllabics. This is cumbersome, because each numeral in an even-numbered position would always be identical with the numeral immediately preceding it, unless it accompanies the last syllabic in the scope (the fourth, in our example) and furthermore this last syllabic bears the stress-morpheme /′/ or /¡/.

⁵⁷ How to distinguish whether a given sequence of pause + A + pause + B + pause is one utterance with internal pause or two utterances, is a difficult question. Here we assume outright that the distinction is tenable.

⁵⁸ Two sequences A + B and A + pause + B may be very similar in meaning or quite indistinguishable; Bloomfield (Language §12.3) even calls the difference non-distinctive. However, for reasons stated in connection with juncture, it is unwise to say that pause—supposing that it has the status of a morpheme—exhibits a zero-alternant in the sequence A + B. The two sequences A + B and A + pause + B are close in meaning; but they are no more identical morphemically than I ate dinner there once and I ate a dinner there once. Nevertheless, facultative pause as defined by Bloch (Lang. 22.202) is useful in IC-analysis, because it is more fruitful to define words initially as minimum facultative pause-groups than as minimum free forms.

⁵⁹ Distinct from juncture, because there are environments in which both occur; e.g. after came in the sentence When he came I left.