Published online by Cambridge University Press: 22 April 2026
This paper develops further the theory of morphemic analysis presented by Zellig S. Harris in 1942. Morphemic analysis is the operation by which the analyst isolates minimum meaningful elements in the utterances of a language, and decides which occurrences of such elements shall be regarded as occurrences of ‘the same’ element.
1 Zellig S. Harris, Morpheme Alternants in Linguistic Analysis, Lang. 18.169–80 (1942).
2 Review of Eugene A. Nida, Morphology: the Descriptive Analysis of Words, Lang. 23.273–85 (1947).
3 Leonard Bloomfield, Language §10.4 (New York, 1933).
4 Rulon S. Wells, Immediate Constituents, Lang. 23.81–117, esp. 93 ff. (§§30 ff.) (1947). It would be possible to say that this ambiguity of meaning of old men and women was grammatically irrelevant; but features of order of the type involved in John hit Bill versus Bill hit John cannot be ignored. This being so, Bloomfield's term ‘tagmeme’ for a feature of meaningful arrangement is useful.
4a This is a reasonable assumption because of man's million years or so of natural selection, in which ability in aural memory and oral mimicry has been a factor making for survival.
5 Cf. Zellig S. Harris, From Morpheme to Utterance, Lang. 22.161–83, esp. 162–3 (§2) (1946).
6 Obviously not all of them, but a sampling which we hope will be statistically valid. By working with successively larger samplings, and by predicting on the basis of each what else will occur, we approach, at least asymptotically, a complete description.
7 A convenient term, because it (1) eliminates the lengthy expressions ‘morpheme alternant’ and ‘morpheme unit,‘ and (2) suggests a valid analogy (allo)phone: phoneme = morph: morpheme.
8 The possibilities are investigated by Harris, Discontinuous Morphemes, Lang. 21.121–7 (1945)—but the added complication of this is avoided in the examples of the present paper.
9 Harris, Morpheme Alternants §7.1.
10 The zero element with meaning ‘noun singular’ is one of Harris's parallels (Morpheme Alternants §2.2). Such a morpheme has very dubious status, having no alternant of other than zero shape (see fn. 37 and reference cited there). Harris lists also the parallel -ful; given the modification of criteria proposed in §13 of this paper, one could add also 's 'genitive'.
11 Harris, Morpheme Alternants §4.2.
12 We propose to say both ‘the morph x occurs in such-and-such an utterance’, and ‘the morpheme x occurs in such-and-such an utterance’. By our definition, a morpheme is a class of morphs, so that the latter type of expression, without further qualification, is logically invalid. We render it valid by stating that an expression of the form ‘the morpheme x‘ shall be taken in some cases as a class-name, in other cases as a variable indicating the appropriate though unspecified member of the class, depending on what the context requires. No ambiguity results; this is customary usage in linguistics; but it is a point on which more care is needed than is usual.
13 Stanley S. Newman, The Yokuts Language of California (New York, 1944). The phonemicity of the point of syllable division is my conclusion from the evidence he gives.
13a In this notation the letter k and the apostrophe' are meant to constitute one symbol together. Similarly c and' below.
14 The second notation is that used by Newman. It may be wondered why anyone would be led to investigate the potentialities of our first notation, the one that we decided to reject. But in Southern Athabascan (see citations in fn. 15) an entirely similar problem arises, and Hoijer chooses an orthography comparable to our first Yokuts orthography, not to our second. The complexity of morphophonemic statement which results is considerable, and could be rendered measurably easier if a phonemic notation were used in which syllable division is marked indirectly instead of overtly.
15 Harry Hoijer, Navaho Phonology, University of New Mexico Publications in Anthropology I (Albuquerque, 1945); a similar phenomenon in Chiricahua Apache: Harry Hoijer, Chiricahua Apache, Linguistic Structures of Native America 55–84 (New York, 1946).
16 Hoijer's working notation incorporates this normalization, though he calls the contrast between ‘constant’ pre-pause h and the evanescent type phonemic instead of morphophonemic. A similar normalization leads to the writing, within word borders, of phonemes both of the s-series and of the š-series, although, at least in rapid speech, only those of a single series occur within the stretch bounded by word junctures; see Zellig S. Harris, Navaho Phonology and Hoijer's Analysis, IJAL 11.239–48 (1945).
17 Though it is not clear what word juncture is in Latin: it may be a non-phonemic matter introduced by a previous notational normalization.
18 Full phonemic information is still given, since such a graph as rts before word juncture stands always for phonemic /rs/, never anything else. In the new notation we have multiple writings for certain phonemic sequences, but only one phonemic sequence for each writing.
19 We are forced to use capitals or some other device for evanescent vowels, because other vowels, phonemically the same, are not evanescent. This fact marks these alternations as non-automatic. Where no extra symbols are needed—where the symbols already used phonemically are merely extended to positions in which they do not phonemically occur—the alternations are automatic.
20 In a manuscript not yet published, Harris demonstrates how, at least in theory, this criterion can be eliminated, thus appealing to semantic considerations at only one step of the whole process of descriptive analysis: the step at which one must decide whether two utterances, as historic events, are ‘the same’ or not (Bloomfield's fundamental assumption of linguistics, Language §5.3, §9.5). The first grouping-criterion (same meaning) thus becomes a practical shortcut; as such it is used here.
21 In Yokuts Structure and Newman's Grammar, IJAL 10.196–211 (1944), Harris makes explicit use (§6) of the second grouping-requirement as modified, but without theoretical discussion.
22 Information and forms were kindly supplied by Gordon H. Fairbanks. It does not matter for the present discussion whether the stem differences are submorphemic within the stem or are a part of the suffix. Some other complications, which do not alter the picture materially, are omitted here. In the cited genitive forms, a dot separates stem from ending.
23 Bloomfield, Language §9.5.
24 Harris, Morpheme Alternants §2.2.
25 In his unpublished material Harris shows how this can be handled. His example is English /tuw/ (to, two, too), which in the absence of semantic criteria first appears as a single morph.
26 No defect of many older grammars of less-well-known languages is more marked than the confusion, or at best fuzziness, which results from a neglect of the principle of total accountability. Of course we do not condemn their writers for being ‘men of their times rather than of ours’; for one thing, this doctrine could hardly be stated explicitly until the phonemic principle had been discovered.
27 I choose Fox rather than Menomini because the examples are a bit easier to cite; the same principles apply. The Fox forms are from Leonard Bloomfield, Notes on the Fox Language, IJAL 3.219–32, 4.181–226 (1924–7), and from the same writer's Algonquian, Linguistic Structures of Native America 85–129 (New York, 1946).
28 A point discussed in detail by Benjamin L. Whorf in various unpublished material, and orally.
29 C. Maxwell Churchward, A New Fijian Grammar (1941).
30 This avoids the risky complications which result from calling word juncture a morpheme, as Rulon S. Wells does in his Immediate Constituents §64 (see fn. 4). The semantic contrast between Thank you with word juncture and the same without it means that word juncture is morphemic, but in such cases it might just as well be concluded—I think, a little better so—that absence of word juncture is the morpheme.
31 Edward Sapir and Morris Swadesh, Nootka Texts (Philadelphia, 1939), esp. Part III, The Primary Structural Elements of Nootka; Morris Swadesh, Nootka Internal Syntax [sic], IJAL 9.77–102 (1936–8). The specific examples cited were generously supplied by Swadesh.
32 The alternation among these three shapes of the element will not concern us; it is covered by statements on a lower level of morphophonemic treatment.
33 Following Pike (The Intonation of American English; Ann Arbor, 1946) and Wells (Immediate Constituents §79) in the assignment of figures, and numbering the four levels from top down.
34 See fn. 11. The capital letters at the beginning of cited suffixes are components of the vowels in the part of the word which precedes the specific phonemes of the suffix. Thus the stem me·k'i ‘swallow’ contains two consonants and parts of two vowels: after the first consonant, the vowel components high-front and long, and after the second consonant, the vowel components high-front and short. When this stem occurs with the suffix FRit ‘passive aorist’, the component F merges with the first group of vowel components in the stem to give e·, and the component R merges with the second group of vowel components in the stem to give zero; the resulting form is me·k'it ‘was swallowed’. With a different set of components contributed by the suffix WAˀan ‘durative present’, the resulting form is mik'a·ˀan ‘is swallowing’. For the details of this, see Zellig S. Harris, Yokuts Structure and Newman's Grammar, IJAL 10.196–211 (1944).
35 The ‘irregular’ verbs present more complex cases of both portmanteau and empty morphs, but are tactically quite the same, save where one or another form is missing.
36 Churchward is not entirely clear on the matter: he says (op.cit. I.24.3) that vei koya is ‘unusual’. If it does indeed occur, then the interpretation proposed is wrong; rather vei koya is like English with it and vuaa like therewith.
37 This second alternative is that proposed by Bernard Bloch, English Verb Inflection, Lang. 23.399–418 (1947). Bloch rejects all alternation or subtraction morphs, and interprets all tentative portmanteaus as an alternant of one of the constituent morphemes plus a zero alternant of the other. One special criterion is introduced for dealing with zero alternants: no morpheme is postulated which has only a zero alternant.
38 The unsolved case of children is discussed in detail for a reason. There is no merit in an analytical procedure which ‘eliminates’ all but one of a set of alternative analyses simply by fiat—by saying that when such-and-such types of alternatives present themselves we shall accept the one which has certain characteristics and reject the others. Our aim is to achieve the most accurate and clearest picture possible of the workings of a language, on all levels—phonemic, morphemic, and tactical; in some cases this is attained not by giving a single treatment, but precisely by indicating the alternatives. For in some cases a range of choice is determined not by our approach, but by the nature of the language; and when this is so, the existence of a range of choice in a particular portion of the language is one of the facts about the language that ought to be portrayed in our description. In one sense, any method of description which conforms to the principle of total accountability is correct; if we nevertheless discuss the relative merits of one procedure or another within this fundamental framework, the purpose is to attain greater mutual intelligibility among the writers of grammars and, in terms thereof, more accurate pictures of the languages we describe.
39 We say ‘phonemic’ for simplicity's sake; if our notation has been normalized, then more accurately this should read ‘every bit of orthographic material’.
40 All the empty morphs in a language are in complementary distribution and have the same meaning (none). They could, if there were any advantage in it, be grouped into a single empty morpheme—but one which had the unique characteristic of being tactically irrelevant.
41 For example, that style in which one says me, myself, and I as if the reference were to three people. This is not unrelated to a style which obviously has to be excluded, both here in the discussion of English pronouns and in any other discussion of morpheme alternants: the style of the discussion itself, in which such forms as me and I contrast because they are used as names of particular morphs.
42 We might go further, interpreting we, us as {I} + pluralizing {s}, with a similar treatment for the other plural pronouns. We are deterred from this step not because plural you is identical with singular you (since after all sheep and other nouns manifest this property), but because {he} + {s}, {she} + {s}, and {it} + {s} would all add up to they, them.
