¹ This paper is a radical revision, and adaptation to languages in general, of a paper on Korean morphophonemics written in June 1946 and based on data gathered by myself. I am obliged to C. F. Voegelin, C. F. Hockett, and Bernard Bloch for comments on the earlier paper, and further to Hockett for sending me an abrégé of a talk on automatic morphophonemics delivered by him at the Linguistic Institute in Ann Arbor, in August 1947. I desire to express my gratitude to the American Council of Learned Societies, by a grant from whom I was aided in 1946.

The books and papers referred to are the following; they are cited in the text by author's name only unless more than one paper by the same author is here listed: Jan Baudouin de Courtenay, Versuch einer Theorie der phonetischen Alternationen (1895); B. Bloch, English verb inflection, Lg. 23.399–418 (1947); L. Bloomfield, Language (1933); L. Bloomfield, Menomini morphophonemics, TCLP 8.105–15 (1939); M. B. Emeneau, Kota texts, Part I (1944); H. Grassmann, Wörterbuch zum Rig-Veda (1873); Z. S. Harris, Morpheme alternants in linguistic analysis, Lg. 18.169–80 (1942); Z. S. Harris, Yokuts structure and Newman's Grammar, IJAL 10.196–211 (1944); Z. S. Harris, Review of Emeneau's Kota texts, Part I, Lg. 21.283–9 (1945); C. F. Hockett, Review of Nida's Morphology, Lg. 23.273–85 (1947); C. F. Hockett, Problems of morphemic analysis, Lg. 23.321–43 (1947); H. M. Hoenigswald, Internal reconstruction, Studies in Linguistics 2.78–87 (1944); H. M. Hoenigswald, Sound change and linguistic structure, Lg. 22.138–43 (1946); A. A. Macdonell, Vedic grammar (1910); T. H. Maurer, Unity of the Indo-European ablaut system, Lg. 23.1–22 (1947); E. A. Nida, Morphology: The descriptive analysis of words (1946); E. Sapir, Southern Paiute: A Shoshonean language, Texts of the Kaibab Paiutes and Uintah Utes, Southern Paiute dictionary (Proc. Am. Ac. of Arts and Sciences, Vol. 65, Nos. 1–3, 1930–1); E. Sapir and M. Swadesh, Nootka texts (1938); H. W. Smyth, A Greek grammar for colleges (1920); M. Swadesh and C. F. Voegelin, A problem in phonological alternation, Lg. 15.1–10 (1939); G. L. Trager, A theoretical basis for descriptive phonology (unpublished lecture delivered at Yale University 11 November 1940); G. L. Trager, An outline of Taos grammar, Linguistic structures of native America 184–221 (1946); N. Trubetzkoy, Grundzüge der Phonologie (1939); J. Wackernagel, Altindische Grammatik I (1896); R. S. Wells, Immediate constituents, Lg. 23.81–117 (1947); W. D. Whitney, Sanskrit Grammar (2nd. ed. 1889).

This list is not and does not aim to be a complete bibliography of all significant contributions either to the theory or to the practice of morphophonemic analysis.

² Harris, Morpheme alternants; Hockett, Problems.

³ Rather than processes, or some combination of phoneme-sequence and process. An example in Bloch 400–1 and fn. 6; another in Bloomfield, Language 214, par. 4, lines 1–2.

⁴ Using Hockett's term, but in a sense equivalent to his ‘primary morph’ (Problems 340, par. 3).

⁵ Which is only a part of a more general class of facts, the facts about the phonemic shapes of morphs in general in the language in question. Hockett's notion of canonical forms (Problems §16; cf. his fn. 28, citing discussions of the notion by B. L. Whorf) is a contribution to the study of such facts.

⁶ Sometimes (e.g. Trager, Lecture §2, §2.1) a morph is said to stand in invariant alternation with itself, especially if it is the sole alternant of a morpheme, and more especially if this morpheme is being compared with other morphemes that have more than one alternant. This manner of speaking is a perfectly admissible alternative to ours.

^6a Phonemic symbols are printed in roman type, without slant lines or other distinguishing marks.

⁷ Wells §11.

⁸ Wells §33.

⁹ Bloomfield, Language 211. An example of a non-automatic but regular alternation is the Sanskrit lengthening of root-final i and u to ī and ū before various suffixes and endings beginning with y: Whitney §245a, §770a; Wackernagel §41. Here the conditioning factor is a combination of phonemics and morphology: the lengthening is conditioned mainly by morphs beginning in y, and of these mainly by those whose morphemes are characterizable in terms of their morpheme classes (Wells §2).

Bloomfield (Men. morph. §3), followed by Emeneau 19 fn. 3, speaks of ‘morpholexical variations’. The distinction between morpholexical and morphophonemic alternations, as also any fundamental distinction between internal and external sandhi, is given up by Harris, Morpheme alternants §1.3 and §6.2, and Review of Emeneau 285–6.

Other definitions of automatic: Emeneau §5; Hockett, Problems 327 fn. 19.

¹⁰ This symbolism will be used for any alternation, even irregular, in which the first-mentioned morph or phoneme-sequence is taken as basic to the second-mentioned. The basic alternant need not—cf. Bloomfield, Language 219 top, Emeneau §6—be one that occurs in zero environment (on this term cf. Wells 100 fn. 34; Hockett, Problems §20), or that is more pronounceable in isolation (cf. Bloomfield, Language §12.4), or—contra Bloomfield Language 164 par. 2—that occurs more frequently or in more environments.

¹¹ Similarly Harris, Yokuts §4; Hockett, Review 282.

¹² Smyth §133. Instead of speaking in morphological terms (‘word-final‘), it will be safer to limit our attention to the phonemic (as well as morphological) environment of position immediately before pause.

^12a Any historical (diachronic) validity for these and all other forms discussed in this article is expressly disclaimed (cf. §14). For instance, a synchronic description of ancient Greek would unquestionably take stómat as the basic morph of ‘mouth’ (cf. Smyth §133b, §244, §258), even though historically we see in -mat a ‘formative’ combined in pre-Greek times from the formatives men and t (cf. Brugmann, Gdr.² 2.1.237 §166).

To avoid a cumbersome notation, I do not attach an asterisk to unattested forms. The context will sufficiently indicate for each form whether it is alleged to be attested or only conjectured or imagined.

¹³ E.g. the Latin example of §13.

¹⁴ Search in grammars has turned up at least one exception: trasnu ‘fearful’ (root tras plus suffix nu, Whitney §1162b). In any event, although in general the juxtaposition of as, ās and n is one that occurs only in external sandhi (not stated in Whitney, but implied §§110, 117c, 166, 175a–c), the general pattern of the language indicates that this is an accident, which is due mainly to the rarity of inflectional and suffixal morphemes beginning with n; the pattern also warrants the prediction that if as + n had occurred in internal sandhi, the result would have been simply asn. (The actual form trasnu bears out this prediction.) Quite apart from the actual known exception, then, the imperfection of the example is that it exploits a non-occurrence which may fairly be regarded as quite accidental.

¹⁵ Namely, before pause and before consonant, only the derivative alternants appear.

¹⁶ If, as in Wells §§66–78, the stress be regarded as a separate morpheme, its different placement in θijǽtr and θíjətər is not another difference in the morphs of theater. (Similarly in cylindr-ical ~ cylinder.) The placement of the stress—that is, the fact that it coincides with one syllabic rather than another, in morphs that contain more than one syllabic—is a fact on the morph ‘level’, i.e. a fact about the order-relations of morphs as such. These relations cannot all be reduced to the relations of preceding and following, nor even to these two and simultaneity, because in both θijǽtr and θíjətər the stress morph is simultaneous with the morph that is made up of vowels and consonants. Whether on the morpheme ‘level’ (‘tactics’: Hockett, Review 274 bottom) a similar complexity is required, is a problem which we have no occasion to go into here; cf. Wells 109 fn. 47 and §§71, 84, 88.

¹⁷ A similar treatment could be developed for ‘components’ of phonemes; cf. Hockett, Problems 335, par. 3.

¹⁸ Macdonell §51c.

¹⁹ An alternative treatment of this particular example would be to take b…r…h as the communis and a…zero, zero…a as the propriae of barh and brah respectively.

²⁰ If, contrary to fn. 16, stress be regarded as part of the morph, the following revised account will be required: the morphs are . Then: (a) zero > ə is automatic as before; (b) the stress shifts are non-automatic; (c) the change æ > ə is automatic when ´ or is lacking on the æ. What is new in this account is that the environment (sc. the stress) conditioning a certain alternation in a morph is a part of the morph itself, rather than belonging to a morph of some other morpheme.

²¹ Grassmann 1707 and Macdonell §33.2 so analyze comparable words.

²² E.g. Macdonell §33.2.

²³ E.g. §175a; §203.

²⁴ If we did not specify ‘which is in turn followed by a vowel,’ then mīlē, alternant of mīlit ‘soldier’ in nom. sing. mīlēs, would constitute an exception.

²⁵ We have simplified the actual facts, as recorded in Whitney §§955g–957, by ignoring (a) the form ita, and all participles formed from it such as Epic acita, and (b) the few cases of non-complementation of the forms ta and na. We pass over the question whether ta and na should be analyzed into two morphs each, t-a and n-a.

²⁶ By a method which has been made explicit, and studied, by Hoenigswald in the two papers cited in fn. 1. The principle that the existence of alternants is evidence of historical development out of original unity is elaborately studied, though very diffusely expounded, by Baudouin de Courtenay. But that the dynamically basic alternant does not always represent a historically older form is proved, for example, by an illustration in §5: although, descriptively, Gk. stómat ‘mouth’ is dynamically basic to stóma, there was never, historically, a nominative singular stómat which was later replaced by stóma. Cf. fn. 12a.

²⁷ The possibility of dispensing with fictive base forms is pointed out by Harris, Review of Emeneau 285–6. Occasional fictive forms are to be found in older grammars; e.g. Whitney §765a refers root dīv ‘play’, exhibiting actual morphs dīv and dyū, among others, to a ‘proper’ form dīū, from which the actual morphs could be regularly, though non-auto-matically, derived by his §129. He mentions (§104d, §242a) but does not use the morphophoneme symbolized ṝ (long syllabic r), which was invented by the Hindu grammarians. That the use of fictive base-forms in comparative Indo-European linguistics has led to misconceptions, namely the mistaking of morphophonemic for phonemic formulae, is clearly shown by Maurer; see his last paragraph

²⁸ Swadesh and Voegelin (10) say: ‘If it has been possible, by the recognition of a nonpatent phonology which involves the construction of fictive formulae …, to reduce the apparent irregularity of Tübatulabal phonology to system, this very fact guarantees the truth of our theory.’ However, it is not clear what the ‘theory’ is, as distinct from their construction of formulae. On p. 2 they remark that ‘the process of constructing morphophonemic formulae has some resemblance to that of historico-phonological reconstruction.‘ Is their theory the theory that their formulae represent not only historical realities but synchronic realities of some sort as well?

²⁹ Two other conceivable solutions are sufficiently recherché for a footnote. (a) Both the n of avn and the n of no·čṯ are regarded as empty morphs in Hockett's sense (Problems §15). (b) Hockett's concept (ibid.) of portmanteau morphs could be generalized to provide for the possibility of saying that n, while not a morph at all, is part of two morphs: part of avn and part of no·čṯ. Of course, as with empty and portmanteau morphs, the principle of total accountability (op.cit. 332 top) would have to be respected. We might call n in this example, and other phoneme sequences so treated, ‘partial portmanteaus’; or, more simply, ‘links’. We could say that 1 is a link in the English adverbs fully, amply, etc.: it belongs both to full and to ly, at once. Some zero-morphs would then be rendered superfluous, e.g. the zero alternant of the past tense morpheme assigned by Bloomfield, Language 215 bottom, to cut, hit, etc., when they have past tense meaning. The advantage is that we are spared arbitrary divisions, as of fully fulij into morphs fu and lij or into ful and ij. I have learned orally from F. G. Lounsbury that he has been using the notion of links in his work on Iroquoian languages.

³⁰ It is not perfectly clear that Sapir's letters ‘ŋw’ represent ŋw. See Sapir §§12.2.b, 13.2, 16.1.