Directions for Historical Linguistics: A Symposium
Winfred P. Lehmann and Yakov Malkiel, eds.
The Notion of Morpho(pho)neme
1. Although much has been written in the last thirty years about morphonology, we still lack a clear definition of its basic concept: the morphoneme. But the concepts of phoneme and morpheme have been thoroughly discussed and defined by their functions: diacritic, and semantic or syntactic, respectively. What is a morphoneme; what is its difference versus phoneme and morpheme; how does change of linguistic status from morphoneme to morpheme take place and vice versa; these are the questions we are interested in.
Besides the terms mentioned, there is another: morph, designating a phonemic complex (structure) with semantic or syntactical function. Without creating misunderstanding one may also speak of the phonemic structure of the morpheme. The morpheme can be represented by several syllables, by one syllable, by a nonsyllabic group of phonemes, by a single phoneme, by one or by several phonemic features, sometimes even by phonetic zero, that is, the absence of phonemic feature. On the other hand, we distinguish compound and disjunct morphemes.
Since morphemes have a semantic or syntactic function, phonemes a diacritic function, the question is whether there are any limitrophous or border problems engaging both domains, phonemics and morphology.
2. Let us start with a few premises. An essential task of phonemics is the exploration of phonemic oppositions or of phonemic features (articulatory or acoustic). Trubetzkoy distinguished several kinds of oppositions based on articulatory features, one of them being especially important, the so-called privative opposition: under certain phonemic conditions (in a given phonemic neighborhood) the marked member is replaced by the unmarked member, the latter (the so-called archiphoneme) being the only one admissible in the position of phonemic neutralization. Consequently, the marked member is perceived by the speaker as containing the surplus of a phonemic feature. Note for example, Polish or Russian t unmarked, neutral, and at the same time negative; d on the other hand is marked by the presence of voice (positive).
From the morphological point of view, however, another classification of phonemic oppositions must be envisaged. They seem to be either morphologically pertinent and conditioned by morphological factors, or morphologically irrelevant. For example:
- (a) Privative opposition — morphologically not relevant; see the above Polish or Russian opposition t:d. The conditioning of this alternation is purely phonemic (a following obstruent or terminal juncture), never morphological.
- (b) Privative opposition — morphologically relevant, for example the alternation of short and long vowels in Semitic (classical Arabic). The alternation is privative, since in closed syllables only short vowels are admissible, thus taquluña but taqŭlna, whereas in open syllables the choice of quantity very often, indeed usually, depends on morphological factors, for example, qatala ‘he killed,’ qātala ‘he tried to kill’ > ‘he fought, attacked (somebody).’
- (c) Nonprivative opposition — morphologically irrelevant. This is probably the most frequent model. In many languages the opposition between front and back vowels (acute and grave vowels) is totally independent of phonemic surroundings and at the same time does not play a role in morphology.
- (d) Nonprivative opposition — morphologically pertinent. In German the contrast between the back vowels a, o, u, au, and the corresponding front vowels, ä, ö, ü, äu (eu) appears in many morphological categories — Bad/Bäder, Wort/Wörter, Gut/Güter, Haus/Häuser; Baum/Bäumlein; kalt/kälter; fanden/fänden — although these two vocalic series are never conditioned by phonemic surroundings.
This transition from diacritic to semantic, as well as the opposite development, deserves special attention. We notice that a phonemic change may be governed either by phonemic or by morphological factors. The replacement of g, d, b, z, ž by k, t, p, s, š in Russian or in Polish is governed exclusively by phonemic factors, the voiced obstruent becoming voiceless before final juncture or before a voiceless obstruent; on the other hand, a voiceless obstruent becomes voiced before a voiced obstruent; for example, Polish grad [grat] ‘hail,’ gen. gradu; żabka [žapka] ‘little frog’ < żaba ‘frog;’ kośba [koźba] ‘mowing’ < kosić ‘to mow.’ The morphological category is not significant, and there are no exceptions. On the contrary, the alternation of front and back vowels in German, as evidenced by the examples quoted, is conditioned morphologically, hence the possibility of “exceptions,” to be explained on semantic and other nonphonemic grounds. See Wiederkäuer < kauen, but Gassenhauer < hauen; behaglich as against kläglich; anmutig but demütig; härter : zarter, and so on.
Accordingly, we may distinguish between a phonological and a morphological alternation of phonemes differing by a distinctive feature. A phenomenon like the inherited Indo-European ablaut (front vowel e: back vowel o; e:ē) seems at first glance a morphological fact. The Germanic umlaut is another example, though its historical origin is quite different. We have seen, on the other hand, that the interchange of short and long vowels in classical Arabic is subject both to phonemic and to morphological factors.
3. Let us, however, continue our analysis of morphologically conditioned alternations. If we consider the Indo-European ablaut e:zero, ei:i, eu:u as morphological, is a contrast like German Tag: plur. Tage also to be regarded as morphological in the same degree? The reduced vowel of Tage is an ending of the plural and has therefore a semantic function. But what about the Indo-European ablaut (apophony) full grade e:zero? What role does e play as against zero in pairs like Skt. rócate : árucat or Gk. ἔχω : ἔσχον (*léuke- : *luké-; *séĝhe- : sĝhé-) and so on?
A careful analysis of such contrasts proves that the morpheme of the aorist sensu stricto is the stressed thematic vowel -é/ó-, whereas the ablaut of the root vocalism represents only an accompanying, redundant morph with semantic zero function. This conclusion results from the following consideration: the zero grade of the aorist does not necessarily contrast with the full grade of the present; see a pair like Skt. riṇákti : áricat without change of the degree of the root vocalism. Generally speaking, the Indo-European ablaut e:zero is morphologically redundant; it only accompanies affixation. Note also the zero grade of the verbal adjectives in -tó-, for example, *lik-tó- < *leiqu̯, *k̑lu-tó- < *k̑leu. Here again the zero grade of the root accompanies the morpheme proper, that is, the stressed suffix -tó-, and is essentially restricted to roots with an internal sonorant or glide. Forms like *pektó-, *settó- show no change of the vocalism of the fundamental form (*pequ̯, *sed). Therefore -tó- is to be considered as the only carrier of the semantic function, whereas the ablaut, limited to roots of a certain phonemic structure, represents from the morphological point of view a redundant feature of the derived form.
We meet the very same situation in the German plural in -er (and other similar formations). The carrier of the semantic function is the ending -er. This ending is governed both by the phonemic form and by the meaning of the root; see, for example, plurals like Bänder, Schilder as against Bande, Schilde. Within this morphological category the umlaut, subordinate to the ending, is conditioned by the phonemic structure of the root (front or back vowel). The umlaut depends not on the phonemic surrounding (see kälter comparative versus nom. sing. masc. kalter), but on the phonemic structure of the morphological surroundings. In roots with front vowels the plural is sufficiently characterized by simple -er. Therefore Brett : Bretter, but Blatt > Blätter.
Incidentally, the same remarks may be applied to the other kinds of Indo-European ablaut, e:o, and e:ē (lengthened grade). The Indo-European o-grade accompanies certain inflectional and derivational processes only in roots with fundamental vocalism e, whereas in roots with fundamental o, affixation by itself is sufficient (cf. ἀπο-τέμνω : ἀποτομή, ἀπο-κόπτω : ἀποκοπή). And the same is true for the lengthened grade. Thus the sigmatic aorist of Skt. bhárati is ábhārṣam (suffix s plus lengthened grade), that of ṡādhati is ásātsam (suffix s with preservation of the root vowel).
A derivational process like *leiqu̯ > *liktó- may be analyzed as (I) *leiqu̯→ (II) *leik-tó-→ (III) *lik-tó-. The absence of any semantic function at the stage (II) → (III) permits us to establish an indirect definition of the morphoneme. We call morphonemic a stage within a morphological (derivational) process or transformation, which is redundant from the morphological (semantic, syntactical) standpoint, but significant from the phonemic (diacritic) point of view. Therefore a plural like *Bander from (das) Band, formed like Bretter from Brett, would be a phonemic, but not a morphological blunder. Lacking is only the last stage (II) → (III), which is phonemic, not morphological.
The stages (I) → (II) and (II) → (III) correspond to the Hjelmslevian terms rection (see the ending of the plural -er) and dominance (see the umlaut). Rection or government exists between two morphemes: see the root Band with its meaning ‘ribbon, band’ and -er as the ending of the plural, not as the ending of the masculine nominative singular of the adjective. On the other hand, dominance exists between a morpheme (for example the ending -er of the plural) and a morph (in our example the phonemic form of the root Band, containing a back vowel).
4. Analysis of the Germanic strong verb offers a more complicated picture. Besides special endings (Gothic zero, -t, zero, -um, -uþ, -un), which have morphological functions, we find, as redundant morphs, ablaut, and reduplication, in a few cases even their cumulation. Compare Goth. laí-lot-um versus haí-hait-um, bit-um, ber-um. The morpheme common to all forms of the strong preterite is the specific ending, whereas the morphonemes ablaut and reduplication show complementary distribution.
Another type of morphoneme is represented by the so-called union-vowels and union-consonants. Compare the union-vowel i of Sanskrit. In Sanskrit future forms like kar-i-ṣyáti, han-i-ṣyáti, stav-i-ṣyáti (but sat-ṣyáti, vak-ṣyati, and so on) the presence of i depends on root structure. It appears after sonorant (r, n, m) or glide (y, v). Therefore the correct analysis of such forms will be kar > *kar-ṣyati > kar-i-ṣyáti. But in the infinitive in -tu(m) the appearance of i is a phonological (diacritic) property of the root; see hán-tum as against bhavi-tum continuing the difference between the so-called aniṭ and seṭ roots.
The original difference between the Greek passive aorists in -ην and in -θην was presumably formal, -θ- being a union-consonant appearing after vowel (whatever the origin of θ): (ποιέω >) ποιε- > *ἐποιή-ην > ἐποιή-θ-ην like kariṣyá- = kar > *kar-ṣyá > kar-i-ṣyá- (and not kar > *kar-i- > kar-i-ṣyá-). The penetration of -θην for -ην after consonant seems to be secondary.
From the descriptive point of view Slavic -x- of the sigmatic aorist is also a union-consonant. For -s- appears both before consonantal and vocalic endings, for example, jȩsъ, jȩste, but -x- only before vowels. Therefore x is to be regarded as a replacement of antevocalic s in certain forms only. Since in a form like rěxъ an s would be admissible (cf. věsъ < vedǫ), this form is to be analyzed as follows: *rě-sъ > *rě-ъ > rě-x-ъ. That is, the replacement of s by x is to be considered as a subtraction of s followed by an insertion of x. Hence the role of x as a union-consonant serving to eliminate a hiatus. This fact has important consequences in word formation, as shown in Onomastica (X, 1-2 , 180-185).
A further, more complicated, example of both union-consonant and of other morphologically conditioned modifications is the Celtic (Old Irish) sandhi. Lenition or nasalization of the following word often depends on double morphological conditioning, for example:
- (a) on the syntactical function of the case form, for example, tuaith, when accusative, nasalizes, tuaith, when dative, lenites;
- (b) at the same time, on the syntactical relation of the affected word to the preceding case form (attribute, genitive, apposition).
Lenition and nasalization after prepositions depend on the preposition as morpheme (phonetic structure plus meaning), for example, fo ‘under,’ air ‘to, for,’ i ‘in’ lenite or nasalize the following noun as prepositions.
The redundancy of the Irish sandhi results from the fact that within certain morphological conditions it can be only partly actualized: no lenition takes place in the case of a vocalic word-initial, and nasalization does not affect word-initial r, l, n, m, s (since n + r, l, n, m, s is phonologically identical with nonlenited initial r, l, n, m, s).
5. Let us now take an example of a morphoprosodeme, that is, the superposition of a redundant prosodic feature (accent, intonation) upon a morphological structure. When comparing a Lithuanian paradigm like acc. sing, bóbą, gen. sing, bóbos, gen. plur. bóbų with a paradigm like nãgą, nagõs, nagų̃ we do not have the right to consider the apparent shift of accent in nagõs, nagų̃ as a (redundant) morphoprosodeme. The oxytone accentuation in nagõs, nagų̃ is a diacritic feature characterizing the feminine ō-stem nag-. In the instr. sing, rankà or acc. plur. rankàs, however, the state of affairs is different. Here the accentuation depends on the structure of the root (shift of accent from a circumflexed or short syllable to certain endings, Saussure's law), whereas in the former case both the phonemic structure and the meaning of the root are relevant, for example, gen. plur. gerų̃ adjective, gẽrų substantive. The nominal stems of Lithuanian have either mobile or immobile accentuation and this accentuation is a diacritic component of the prosodic stem structure. The accentual alternation nãgą : nagõs is therefore a primary prosodic fact, whereas the change rãnką : rankàs is a morphoprosodic phenomenon, conditioned by the intonation of the stem provided with the ending of the accusative plural. From the purely phonemic standpoint *rañkas would also be admissible; compare the nom. sing, lañkas and so on.
6. In order to explain the phenomenon of dominance historically, take as example the Spanish and Italian subjunctive in -ga. In both languages there are a number of subjunctives in -a showing an inorganic (not original nor phonetically developed) -g- before the subjunctive ending proper -a. Thus Italian salga (salire), valga (valere), tenga (tenere), venga (venire); Spanish salga (salir), valga (valer), tenga (tener), venga (venir). We find such an inorganic g also in western French dialects and northern dialects of Old French.
In a remote epoch a structural law with the following content must have acted in these languages: palatal l′, ń at the end of a verbal root were replaced by the groups lg, ng before the suffix of the subj. (-a-). That is, the ideal forms *sal′a, *val′a, *teńa, *veńa (Lat. saliat, valeat, teneat, veniat) were realized as salga, tenga, and so on. Obviously this dominance is the result of certain sound changes. Yet, more important, it is not their immediate result, but so to speak the projection of the sound changes onto the morphological plane. The palatalization of the Latin groups lg, ng giving Romance l′, ń before front vowels was a phonemic change; compare colgit (for colligit) > Ital. coglie, pingit > Ital. pegne. In a number of verbs that change must have been reflected on the morphological plane as a specific relation of the indicative to the subjunctive (colga, pinga). Thus col′-e > *col′-a > col-g-a. The relation between col′-e and col-g-a is perceived as the replacement of the expected l′, ń at the end of the verbal root by lg, ng before the subjunctive suffix -a, that is, as an insertion of a redundant morph g (entailing the depalatalization of the preceding l′, ń). Hence this morphoneme is consistently applied to all verbs having palatal l′ or ń before the a of the subjunctive, *sal′a > salga, *teńa > tenga.
The above example shows us the tendency of redundant morphs to spread in derived or founded forms. The addition of a redundant morph enlarges the distance between the basic and the derived (founded) form and renders the latter expressive.
7. But morphologically redundant morphonemes appear also in basic forms; compare the -s of the nominative singular or the -t of the third person singular in Indo-European. Endings like the -s of the nominative and the -m of the corresponding accusative are not on an equal footing; they do not belong to the same level, neither do the verbal endings -t of the third person and -m, -s of the first and second persons. The nominative singular is the neutral member of the case-number system and so is the third person singular within the verbal paradigm. Therefore the -m of the accusative, the verbal endings -m, -s have to be regarded as morphemes, whereas the -s of the nominative singular as well as the -t of the third person singular are only morphonemes, subordinate to the zero-morpheme of these inflectional forms. Hence an important corollary regarding morphological proportions.
Linguistic proportions are the chief implement for creating new linguistic forms and simultaneously their most abundant source. Loan words and onomatopoeic creations play only a secondary role. The importance of linguistic proportions consists in the fact that they enable us to imitate external reality not in its concrete details, but in its relations. Thus, for example, the linguistic relation count : countess, baron : baroness, lion : lioness, tiger : tigress, and so on, repeats, always under the same form, the sex relation, identical in each pair of the denoted animate beings. Bühler called it relationstreue Abbildung of the external reality.
In morphological proportions the relation between the first and the second member, and similarly between the third and the fourth member, concerns both the phonological structure and the function (meaning). Such proportions reflect the relation between the basic and the founded form, whether it is derivational or inflectional, and constitute the nucleus of the linguistic system, the regular relations between the structure and the function of linguistic forms. For example, German Bild : Bildchen = Brett : Brettchen = Bein : Beinchen, establishing the category of diminutiveness. Within this category we find a morphophonemic proportion, which is obligatory (or at least has been obligatory, cf. the exception Kuh : Kuhchen) and represents a superposition of the redundant umlaut on the semantically pertinent suffixation, thus *Kappchen : Käppchen = *Wortchen : Wörtchen = *Hutchen : Hütchen.
Since the umlaut is predictable, the proportion Brett : Brettchen = Wort : Wörtchen is morphologically (though not phonemically) in order. It shows us that morphophonemic elements, which are not carriers of semantic values, are simply to be disregarded without bearing prejudice about the correctness of the morphological proportion. Thus in derivations like nominative singular : other case forms, present : other tenses, indicative : other moods, third person singular : other persons, the formal surplus of the basic form may be ignored — and the same holds true for the so-called primary derivatives of Indo-European. Examples are as follows:
The sigmatic aorist of Indo-European, which expressed perfectivity, was in semantic contrast with the present-imperfect stem. Now the structure of the Indo-European present was manifold: thematic or athematic, reduplicated, with -i̯e/i̯o- or -sk̑e/sk̑o- suffix, with nasal infix. Confronted with the (sigmatic) aorist all these presents were semantically neutral-negative (nonperfective, imperfective), although they may have served to express different kinds of imperfectivity (simply durative, iterative, inchoative, terminative). Therefore the formation of the aorist consists first in the elimination of these semantically redundant morphs, that is, in the reduction of the present stem to the simple root, then in the addition of the sigmatic suffix (entailing itself the redundant feature of lengthened grade). Thus riṇákti > *rek-ṣ- > raik-ṣ-. Hence the well-known rule of comparative grammar: the sigmatic aorist is built upon the root, not upon the stem of the present. But the latter possibility is not excluded once the suffix of the present blends with the root so as to constitute a new verbal root.
A form like Skt. bhárati is justly regarded by the Hindu grammarians as the basic form of the whole conjugational system, since it is neutral in all respects, as regards person, number, tense, and mood. This choice of the basic form envisages the semantic relations in the first instance. But it also meets to a high degree (though not absolutely) the criterion of predictability, since the rest of the verbal forms (perfect, future, partially also the aorist) may be predicted on the basis of the form of the present. Incidentally, this is not the case everywhere. In Slavic the perfective form is semantically based on the imperfective (the latter being semantically unmarked), whereas predictability generally goes in the opposite direction: it is the form of the imperfective in -ajǫ (plus the redundant change of the root vowel) which is predictable on the basis of the perfective form. See sypljǫ : -sypajǫ, sъxnǫ : -syxajǫ. The same holds for the French adjective in cases like laide : laid, forte : fort, épaisse : épais.
We therefore have the right to consider a proportion like Skt. bhárati : aorist ábhārṣit = riṇakti : áraikṣīt as correct. The characteristic features of the present form (thematic with stress on the root in bhárati, with nasal infix in riṇákti) are redundant morphs, that is, morphonemes with zero value, once the present is confronted with the perfective value of the sigmatic aorist (whose form itself contains the redundant lengthened grade). Similarly, a proportion like Lat. bonus : gen. bon-ī = bona : gen. bona-ī, hence bonae, may be considered as being in order, since the ending -us of the masculine, unmarked as against fem. -a, is to be disregarded. Morphologically this proportion, explaining the innovation bonae for bonās, has to be rewritten under the form bon : bon-i = bona : bona-i. This is the latent form underlying the apparently inaccurate proportion bonus : boni ≠ bona : bonae.
The original secondary ending of the third person singular of the Indo-European mediopassive is -o: see Vedic -a enlarged to -at in Vedic áduhat, áśayat and Hittite -*a (third person singular preterite of the ḫ-conjugation). As a rule this archaic -o is replaced by -to (Indo-Ir. -ta, Gk. -to). The pertinent proportions are:
- (I) -t (active) : -o (mediopassive) = -nt (active) : -nto (mediopassive), the correct morphological proportion being zero (zero number, zero voice) : -o (zero number, voice) = -nt (number, zero voice) : -nto (number, voice)
- (II) -nt : -t = -nto : -to (t negative, not neutral as in I).
Another example illustrating the importance of morphological proportions is the much discussed problem of the Greek desinences -εις and -ει of the second and third person singular of the thematic present.
According to Brugmann we get phonetically *λέγεσι > λέγει, hence λέγεις owing to the secondary addition of -s-; finally, on the model of (ἒ)λεγες : (ἒ)λεγε, a secondary form of the third person singular λέγει (replacing *λέγετι) was created.1
Later Brugmann modified his opinion about the second person singular, assuming its ending to have been inherited and identical with that of Lith. sukì : sukíe-s.2
|series:||3rd. p. sing. *λέγετ||→||2nd. p. sing. λέγες|
|series:||3rd. p. sing. *λέγετι||→||2nd. p. sing. *λεγεσι|
Two phonemic changes take place: the final dental -t and the intervocalic -s are dropped. The latter change is probably older. In the second person the marked ending is formed from the unmarked one by subtraction of -s and addition of -i. Hence, once the final -t of the third person has disappeared, we get proportionally:
- (I) λέγες (nonindicative) [>*λέγε] : *λέγει = λέγε (nonindicative) : λέγει
- (II) λέγε (zero person) : λέγες = λέγει (zero person) : λέγεις
In the so-called primary word formation of Indo-European the basic form is reduced to the simple root before being provided with the necessary derivational suffix. See the derivational types in -o (Gk. τόμος and τομός), for example, Skt. -bhará- < bhárati but also -kará- < kr̥ṇóti. Contrasting with the derived forms the affixes of the basic forms become semantically neutral and are disregarded in the morphological proportions:
In morphological proportions morphophonemes (i.e. redundant morphs) are disregarded, just as in phonological proportions allophones (i.e. phonemic variants) are disregarded. A relation like Gut : Güter, Buch : Bücher is phonemically, though not phonetically correct; see the ach-Laut in Buch, and the ich-Laut in Bücher.
Confusion between the morphological and the morphophonemic approach would make us reject, for example, Saussure's explanation of the rise of mobile accentuation in the Baltic (Lithuanian) vocalic stems. According to Saussure this mobility is to be accounted for by the pressure exerted by the consonantal stems in -er, -en, where the inherited alternation acc. -ér-m, -én-m : gen. -r-és, -n-és, transformed into initial versus desinential accentuation, has been imposed upon the different classes of the vocalic stems. The objection would be of a morphological order, namely, that such an explanation is in contradiction with the fact that the consonantal stems were almost unproductive as against the well-attested productiveness of -o-, -a-, -i- stems. But this semantic factor does not play a role in Saussure's explanation, which belongs to the morphophonemic level and is in the last instance based on the overt distinction of inflectional suffix and desinence in the consonantal stems versus their fusion into a portmanteau morph in the vocalic stems. This is a fact of dominance comparable to that stated in the example Blatt : Blätter, where the morphophonemic umlaut is imposed by the plural morpheme, that is, the marginal morpheme, upon the central morpheme (the root).
In a similar way the fact that verbs with preverbs are derived from simples (this being a derivational, i.e., semanticomorphological fact) does not prevent the recessive accent of compounds from influencing the accentuation of the simples (cf. L'accentuation, p. 152). It is the opposition of compound versus simple morphs, not morphemes, which is envisaged here.
8. The semantic zero value of morphonemes is a counterpart of the semantic function of morphological zero. Thus the ending of a form like Russian slov has the value of the genitive as opposed to the nominative plural slová. The stressed -á of slová is positive or neuter according to whether we envisage the contrast slóvo : slová or slová : slov. Morphoneme with zero function on the one hand, and zero with morphological function on the other hand, depend on actual opposition. The contrast Russian slóvo (nom. sing.) : slová (nom. plur.) has to be reduced to -o with zero value : sign -á of the plural, whereas within the plural the relation slová (nom.) : slov (gen.) must be regarded as -á with zero value : sign zero (gen.).
The difference between morpheme and morphoneme is also important for syntax. For example, the relation between subject and predicate is often analyzed from a twofold point of view: government of (verbal) person by the subject, government of the nominative of the subject by the verb. But both the exponent of the third person of the verb and the ending of the nominative are morphonemes with semantic zero function. Government, occurring only between meaningful elements, is therefore out of the question. As regards number (sometimes also gender) there is agreement of the verb with the subject; see the syncretism of number within the verb, but not within the noun, in constructions like accusative + infinitive.
It would not do to consider morphonemes as a kind of accessory or subordinate morphemes. Morphemes have a semantic or syntactical function, totally absent in morphonemes. Morphonemes serve as a supplementary phonemic characterization of word forms already provided with morphemes. They are not, however, a subject of phonology proper. The latter has to do with phonemes, their features and structures (like stress units, syllables, consonant clusters, and so on), but not with morphs. Morphonology or morphophonemics, on the other hand, is interested in phonemes, phonemic features and structures dominated by morphological, not phonological, factors, and therefore belongs to morphology.
9. In the history of a language morphonemes may become semanticized and raised to the rank of morphemes. This is a well-attested phenomenon. See the umlaut of OHG indicative feris versus subjunctive farês. After the weakening of unstressed vowels (MHG fer[e]st : far[e]st) the back vowel of the subjunctive, contrasting with the front vowel of the indicative, becomes a full-fledged morpheme. The vowel change in English feet, geese, teeth versus foot, goose, tooth is likewise a morpheme.
The change of morphemes into morphonemes is also easily illustrated; see the rise of various union-consonants or union-vowels or the fate of the Indo-European -t- suffix mechanically attached to root nouns in sonorant or semivowel: Skt. kṛ-t-, ci-t-, cyu-t-.
We have seen that morphemes can under circumstances become morphonemes, that is, they can lose their semantic or syntactical function: if a morpheme stands in opposition as a neutral member to a positive one (slová : slov). This occasional zero value of the morpheme is to be distinguished from the zero value of the morphoneme, the latter being a phonological phenomenon dominated by the morphological structure of the word. Within a given morphological category there is a certain phonological dependence between a morpheme and a morphoneme (for example, between a suffix and ablaut or umlaut).
10. To sum up: The morphological analysis of a word yields morphs in the first instance. Some of them are carriers of morphological functions (semantic or syntactic), others are semantically void. The former are the only ones to deserve the name of morphemes. The latter, the morphonemes, may be represented not only by phonemes or phonemic features, but also by phonemic complexes (e.g., Lat. -us of the nominative singular). A morphoneme is therefore not necessarily an elementary phoneme. Taken as an elementary unit the morphoneme is a subject of morphonology, which belongs to morphology. Its phonemic structure is as irrelevant as is the phonemic structure of a simple morpheme. The analysis of the morphoneme into phonemes, that is into diacritic elements, is another matter. But from the point of view of phonemics, which undertakes such an analysis, the difference between morphemes and morphonemes is of no consequence.
The distinction between morphemes and morphonemes is of primary importance also in historical linguistics. The superposition of morphonemes upon the morphological word form, the fact that they represent a superficial layer in the structure of the word, favors the proportional spread of these redundant elements. Phonemically obligatory, they are semantically void. Their function lies in an expressive swelling of the derived and founded word forms, increasing their external difference from the basic forms. In purely morphological proportions, established to account for morphological innovations, morphonemes may be disregarded, just as are allophones in phonemic proportions. This new appreciation of the morphological proportion may open a new vista in historical morphology.
2 “Zur griechischen u. germanischen Präsensflexion,” Indogermanische Forschungen, 15 (1903), 126-128. “Zur Bildung der 2. Pers. Sing, in den idg. insbesondere den baltischen Sprachen,” Indogermanische Forschungen, 17 (1904), 177-185. Followed by Meillet, Introduction, p. 228 (8th ed. Paris: Hachette, 1937).