² Bernard Bloch, Lg. 23.414: 'Morphophonemics is the study of the alternation between corresponding phonemes in alternant shapes of the same morpheme.' This implies a limitation of the term to cover only alternations, rather than shapes and alternations, and, furthermore, to subsume only alternations between shapes that are more or less comparable phonemically (knife and knive-s), so that one can pick out which phoneme of one shape corresponds to which in another shape. We want also to cover such cases as go and wen-t. Also Rulon S. Wells, Lg. 25.100 (fn. 5): '[the differences between alternant shapes of morphemes are] 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.' Wells limits the term 'morphophonemics' to the narrower class of facts.

³ I had missed this point completely during over six years of work with Chinese, until finally it was called to my attention by Henry C. Fenn Jr.

⁴ Considerable information on intonation is given in Yuenren Chao, Mandarin Primer (Cambridge, 1948).

⁵ There may be other messages as well—tone-quality variations indicating emotional attitudes or identity of speaker or the like. Such matters, however, either stand outside the realm of language, or else have not yet received the kind of study which would reveal their linguistic nature. Kenneth L. Pike is fond of discussing them; see in particular The Intonation of American English, Ch. 7 (Ann Arbor, 1946).

⁶ The notion of structural signals was proposed by David L. Olmsted, specifically for the ‘conjugation vowels’ of Spanish and Portuguese. Instead of treating these as empty morphs (PMA §21), we may take them as morphemes, the meaning of which is something submorphemic in the structure of the language: a particular conjugation vowel tells us ‘the verb stem just spoken belongs to such-and-such a morphophonemic class in its inflection’. Structural signals are then like Bloomfield's substitutes in that their meaning is definable within the structure of language, rather than in terms of the practical world; indeed, while this is by definition true of structural signals, there is some doubt as to whether it really holds for the forms Bloomfield calls ‘substitutes’.

⁷ Leonard Bloomfield, Linguistic Aspects of Science (International Encyclopedia of Unified Science 1.4; Chicago, 1939).

⁸ In this paper, instead of using the term ‘empty morph’ (see PMA), we simply classify certain phonemic material as nonmorphemic or as morphemically irrelevant.

⁹ See PP §4. The differentiation of loud and nuclear stress is meaningful only when a macrosegment includes more than one mesosegment. It is terminologically convenient to say that when a macrosegment includes but one mesosegment, loud and nuclear stress coincide.

¹⁰ This is a statement of historical fact. Since the discovery of the possible rephonemicization obviously came about before this paper was ready for publication, it would have been possible to introduce it before, rather than after, the discussion of morphophonemics. But it seemed worthwhile to retain the ordering of matters as above, as a fairly elaborate demonstration of the way in which grammatical considerations can serve as clues for phonological analysis without implying that the latter is logically built on the former; there are still no grammatical ‘prerequisites’ for phonemic analysis.

Compare the similar view expressed by Bloch, Lg. 26.92–3 and fn. 16, and the argument by which he supports it in §9 of his paper (26.124–5).

¹¹ With the present discussion of stress may profitably be compared that in Yuenren Chao, op.cit. 26, and that in John de Francis, Beginning Chinese 5–6 (New Haven, 1946). De Francis devises a notation for stress very much like our revised notation.

¹² Compare the handling of /#/ in Fijian (PMA §16) and Nootka (PMA §18), and contrast the handling of English /#/ by Rulon S. Wells in Immediate Constituents, Lg. 23. 107–8, §64 (1947).

¹³ Some speakers have no such independent occurrences of /5/. For them, the distinction between [3] and [5] is subphonemic, and the statements of alternation made here can be simplified.

¹⁴ This may depend partly on which tone the morpheme has basically. Loss of tone /3/ seems to be relatively rarer than loss of the other tones.

¹⁵ In PP it was stated that consonants medial in dissyllabic microsegments are voiced. The fast alternant pihu⁴ had not yet been heard when PP was written. If the PP statement holds, the /h/ in this form should be a voiced velar spirant—and so it is. In a small way, this illustrates how one can make an extrapolation beyond one's data, and how verification of this by later observation can test the validity of the analysis. On such matters, see IJAL 14.269–71 (1948).

¹⁶ Admittedly this treatment is circular. Not until after the analysis had been devised did it occur to me, however, that the circularity may be vicious in that we may be introducing a grammatical criterion for the phonemic analysis. If we are, then the treatment certainly has to be rejected. But we can still save the essential procedure of regarding the ‘fast speech’ forms as consisting of the ‘slow speech’ morphemes plus an additional morphemic constituent (or vice versa). We simply recognize the same segmental morphemes in each matching pair, but extract from one set (say the fast forms) an additional morphemic long component not present in the other set, without trying to identify which part of the phonemic material of each fast form constitutes this component. Speed of speech then reverts to nonphonemic status. The procedure is expounded in Zellig S. Harris, Componential Analysis of a Hebrew Paradigm, Lg. 24.87–91 (1948). It might be useful also for the problem solved otherwise in Bernard Bloch, English Verb Inflection, Lg. 23.416–8, §8 (1947).

Bloch has offered the following two highly relevant comments on this section. (1) If we set up tempo phonemes at all, one is enough; it can be present in certain stretches, absent in others. (2) The semantically preferable solution is certainly to set up only one tempo morpheme, the slow one, with the meaning ‘deliberateness, emphasis, etc.‘

¹⁷ A confusing note is introduced by the relative frequency of forms with /e/ and forms with /a/ in such cases as this. The particles ·Ce, ·me, and ·be all have /a/ much more frequently than /e/; the particle ·ne, and terminal vowels of dissyllabic microsegments of the right shape, all have /e/ more commonly than /a/. Yet we cannot set the former morphemes up with basic /a/ instead of /e/, since there are some (e.g. guai¹·da ‘pat, spank lightly‘) which always have /a/. If relative frequency is a criterion, then we must recognize three distinct morphophonemic types.

Chao (op.cit.) eliminates the alternation in initial consonantism by a different phonemicization. This particle is recognized as one of the few forms with initial vowel. Elsewhere where our phonemic analysis recognizes an initial vowel (e.g. ai⁴ ‘to love, like’ or en⁴ 'to press down'), Chao sets up a consonant, phonetically sometimes glottal stop, sometimes weak voiced velar spirant (almost high back unrounded semivowel). The initial consonant of this particle is then simply the final consonant of the preceding microsegment, ambisyllabic when between two vowels. Chao's analysis also eliminates microjuncture, the occurrence of which is rendered automatic and nonphonemic. The difficulty with Chao's analysis is that it does not provide for forms like sreme² and zeme³, where one has an intervocalic consonant with syllabification quite different from that between the second and third syllables of lau³fang²·nge 'Hey, Old Fang!' To take care of all the facts, microjuncture has to be introduced, together with the distinction between syllables and microsegments; and this implies the greater morphophonemic complexity of the particle here under discussion.

¹⁸ Some speakers have far less differentiation phonemically in microsegments with terminal /r/; for their speech the morphophonemic patterns are actually a bit more complex.

¹⁹ Other demonstratives also occur most usually before a measure, but sometimes directly before a noun with no intervening measure: zrei⁴·ge#sr²·heu⁴ or zrei⁴sr²·heu⁴ ‘this time’.

²⁰ Chao (op.cit.) handles some of these forms as instances of the ‘grammatical process’ of reduplication, though noting (39, fn. 7) the possibility of the chameleon treatment. But the Romanization he uses in his course implies the chameleon solution, since ‘x’ is used for ‘preceding syllable repeated’, and ‘vx’ for ‘the two preceding syllables repeated’ (333, 332).

²¹ Chao, op.cit. 41.

²² In the case of alternations from one speaker to another, or within a single speaker's dialect but of the same general nature as those which for the most part are from one speaker to another—e.g. tien¹ versus tian¹ for ‘day’, or iu² versus ieu² for ‘postal’ (see §0)—the same principle ought in theory to apply. If in practice we do not apply it, it is because of the great difficulty in making sufficiently precise observations. We are here at the thin edge between synchronic and diachronic linguistics. If a single speaker says sometimes iu² and sometimes ieu² for ‘postal’, it is ultimately because he has picked up both forms from others, and there has presumably been some first individual to acquire and use both. The two forms then differ at least in the subtle connotations they have due to the specific life-history circumstances accompanying the acquisition of each. To make a blanket statement of this kind is easy; to try to cover all such cases in a description of a language or dialect is totally impossible.

²³ Probably tactical considerations must lead us to take neme⁴, neme³, zeme⁴ or zreme⁴, zeme³, and sreme² as bimorphemic, an alternant of a demonstrative (itself perhaps subject to further analysis as sketched in the text) plus a measure -me, and ner⁴, nar³ and zrer⁴ similarly as an alternant of a demonstrative plus a measure -r. But these forms have somewhat special tactical statuses, and the answers are not clear.

The measure -r, if we so analyze, is a different morpheme from the noun-suffix -r and the other morphemes of that shape discussed in earlier sections. Historically it is probably to be related to the type I tonic (bound) morpheme li³ ‘in, inside’: earlier na⁴·li³, na³·li³ and zre⁴·li³, with loss of the microjuncture in rapid speech and accompanying loss of the terminal /i/, then with the /l/ changed to /r/ ‘because’ /l/ is not habitual in microsegment-final position. But the two-microsegment forms have been reintroduced, analogically or from writing or in both ways, and now stand beside the one-microsegment forms, as more elevated and literary alternants, so that the -r of the short forms can no longer be identified morphemically with any element occurring elsewhere.

²⁴ Chao (op.cit. 40) reaches essentially the same conclusion: 'Although [the examples he gives] are pairs of cognate words (and often written with the same characters), they should, for practical purposes, be learned as separate words.^? 'Cognate', of course, is a historical term, not a descriptive one.