“Current Approaches to Phonological Theory” in “Current Approaches to Phonogical Theory”
Functionally-constrained Phonology
This paper addresses the question of why some phonological rules are found in some languages but not others. We propose to answer this question in two steps: first, by establishing whether the effect of a particular rule in a particular language is neutralizing or allophonic, and second, by determining the conditions under which a given rule may have a neutralizing or an allophonic effect. We conclude that these conditions are a direct function of the relative typological markedness of phonological systems, and that, at least for the class of general phonologically-conditioned rules, neutralization rules produce relatively unmarked segments while allophonic rules produce relatively marked ones.
Following Kiparsky (1976:169), we define “neutralizing” and “non-neutralizing” as in (1).
(1) A rule of the form
is neutralizing if and only if there are strings of the form CBD in the input to the rule. Otherwise, the rule is non-neutralizing.
The definition in (1) departs somewhat from traditional conceptions of neutralization in at least two ways. First, it considers only those segmental mergers to be neutralizing in which the product is identical to one of the members of the opposition, and it defines all other feature-changing rules (even mergers to some third entity) as non-neutralizing, or, as we will henceforth refer to them, as “allophonic”. While this aspect of the terminology is somewhat at odds with standard practice, it does represent the distinction we will argue is appropriate for explaining the distribution of certain rule types among the world’s languages. The second way in which this definition differs from standard views of neutralization is that it will characterize insertion and deletion rules as either neutralizing or non-neutralizing only if “in the input to the rule” is taken to mean “existing at some level of derivation before the rule applies,” which is apparently what Kiparsky intended, rather than to mean “satisfies the structural description of the rule.” Although preliminary investigations suggest that the constraints developed here are also applicable to insertion and deletion rules, we have limited the scope of this paper to a consideration of feature-changing rules.
The definition in (1) determines the status of a rule by inspecting the rule’s form and its input, but it does not predict which types of rules can be neutralizing and which cannot. However, it can be observed that certain types of rules are always neutralizing and that certain other types are always allophonic. Thus, the rules which devoice final obstruents in German, Russian, and Sanskrit are all neutralization rules, as are the rules which deaspirate final stops in Sanskrit, Korean, and Marathi. Examples illustrating the operation of these types of rules are given in (2) and (3).
(2) Final devoicing of obstruents
a. German (Moulton 1962)
Singular | Plural | |
Ta[t] | Tä[t]e | ‘deed’ |
Ra[t] | Rä[d]er | ‘wheel’ |
b. Russian (Trager 1934:340)
Nom. Sg. | Gen. Sg. | |
[pop] | [papá] | ‘priest’ |
[rap] | [rabá] | ‘slave’ |
c. Sanskrit (Whitney 1879)
Nom Sg. | Acc. Sg. | |
[marut] | [marutam] | ‘wind’ |
[pa:t] | [pa:dam] | ‘foot’ |
(3) Final deaspiration of stops
a. Sanskrit (Whitney 1879)
Nom. Sg. | Acc. Sg. | |
[stup] | [stubham] | ‘praising’ |
[pat] | [patham] | ‘road’ |
b. Korean (Informant)1
Citation | Objective | |
[nat] | [nadʉl] | ‘a grain’ |
[nat] | [nathʉl] | ‘piece’ |
c. Marathi (Informant)
[tap] | ‘fever’ | [tapala] | ‘to the fever’ |
[top] | ‘cannon’ | [tophela] | ‘to the cannon’ |
[vad] | ‘discussion’ | [vadala] | ‘to the discussion’ |
[dud] | ‘milk’ | [dudhala] | ‘to the milk’ |
The devoicing of obstruents in these examples is neutralizing because it eliminates the contrast between voiced and voiceless obstruents, i.e., there are both voiced and voiceless obstruents in the inputs to the devoicing rules but only voiceless ones in their outputs. Similarly, the deaspiration of stops in these examples is neutralizing because there are both aspirated and unaspirated stops in the inputs to the deaspiration rules but only unaspirated ones in their outputs.
Other types of rules, however, are always allophonic. Thus, the rules which voice intervocalic obstruents in Corsican, Korean, and Old English are all allophonic, as are those which aspirate initial voiceless stops in English, German, and Swahili. Examples illustrating the operation of these types of rules are given in (4) and (5).
(4) Intervocalic voicing
a. Corsican (Dinnsen and Eckman 1977)
[peδe] | ‘foot’ | [u beδe] | ‘the foot’ |
[tengu] | ‘I have’ | [u dengu] | ‘I have it’ |
[sakku] | ‘bag’ | [u zakku] | ‘the bag’ |
b. Korean “plain” stops
Citation | objective | |
[tap] | [tabʉl] | ‘answer’ |
[pok] | [pogʉl] | ‘fortune’ |
[nat] | [nasʉl] | ‘sickle’ |
c. Old English fricatives (Campbell 1959:179-81)
[stæf] | [stavas] | ‘staff’ | (sg./pl.) |
[snāe] | [snīδan] | ‘cut’ | (pret./inf.) |
[ċēas] | [ċēozan] | ‘choose’ | (pret./inf.) |
(5) Initial aspiration
a. English b. German
[ph]ort | [th]at | ‘deed’ |
re[ph]ort | ge[th]an | ‘done’ |
s[p]ort | S[t]aat | ‘state’ |
c. Swahili (Tucker and Ashton 1942:89-90)
[khaháwa] | ‘coffee’ |
[msithúni] | ‘in the forest’ |
[msítu] | ‘forest’ |
In these examples, intervocalic voicing never neutralizes a contrast between voiced and voiceless obstruents. In Korean, the only voiced stops in the language are those in intervocalic position, or more precisely, those occurring between voiced segments (Martin 1951:524). They correspond to, and alternate with, lax voiceless unaspirated stops in other positions. Thus, these voiced stops are analyzed as underlying lax voiceless unaspirated ones that are voiced by a rule of intervocalic voicing. This rule is allophonic because it is the only source of voiced stops in the language. The situation in Old English is similar, where the only voiced fricatives in the language are those in voiced environments (Campbell 1959:20). They are derivable from underlying voiceless fricatives by an allophonic intervocalic voicing rule. However, in Corsican, there are voiced stops that contrast with voiceless ones in word-initial position. But these voiced stops are spirantized in exactly the same environment in which the voiceless stops are voiced, as can be seen in (6) below. Thus, the intervocalic voicing rule is allophonic, because at the time that it applies, spirantization has already taken place, and there are no voiced stops in the input to the rule.2
(6) Corsican intervocalic spirantization (Dinnsen and Eckman 1977)
[bokka] | ‘mouth’ | [a βokka] | ‘the mouth’ |
[dente] | ‘tooth’ | [u δente] | ‘the tooth’ |
[gola] | ‘throat’ | [di ƴola] | ‘of throat’ |
Similarly, initial aspiration rules never neutralize a contrast between aspirated and unaspirated stops. In languages with such rules, voiceless aspirated stops are in complementary distribution with corresponding voiceless unaspirated ones and are derived from them by the rule that aspirates stops word-initially and syllable-initially in word-medial stressed syllables, which is the only source of these aspirated stops. Since there are no aspirated stops in the input to the rule, the aspiration rule is allophonic.
Within current phonological theories, however, there is no reason why devoicing and deaspiration rules could not be allophonic. Nothing suggests, for example, that we should not find a language otherwise like German but whose underlying voiced obstruents are devoiced word-finally by an allophonic rule, or a language otherwise like Sanskrit but whose underlying aspirated stops are deaspirated word-finally by an allophonic rule. By the same token, nothing in current phonological theory excludes rules like intervocalic voicing and initial aspiration from the class of neutralization rules. There is no reason to expect, for example, that we should not find a language with a contrast between voiced and voiceless obstruents, like English, that neutralizes this contrast intervocalically by a voicing rule, or a language with a contrast between aspirated and unaspirated stops, like Sanskrit, that neutralizes this contrast word-initially by an aspiration rule. But there are no such languages. In order to explain their absence, we must first turn to the typology of phonological markedness.
The distinction we draw between marked and unmarked segments is a generally accepted one, and is derived from well-known implicational universals of phonological systems. According to these universals, a segment X is considered to be ‘marked’ with respect to a segment Y just in case the presence of X in a given language implies the presence of Y but the presence of Y does not imply the presence of X in that language. Examples of such implicational universals are given in (7).
(7) Selected Implicational Universals
a.The presence of voiced stops implies the presence of voiceless stops, but not vice versa (Jakobson 1969:70).
b.The presence of voiced fricatives implies the presence of voiceless fricatives, but not vice versa (Jakobson 1969:70).
c.The presence of fricatives implies the presence of stops, but not vice versa (Jakobson 1969:51).
d.The presence of aspirated stops implies the presence of unaspirated stops, but not vice versa (Jakobson 1969:14).
e.The presence of a bilabial nasal implies the presence of a dental or alveolar nasal, but not vice versa (Ferguson 1963: 56-7).
f.The presence of mid vowels implies the presence of high vowels, but not vice versa (Jakobson 1969:56).
g.The presence of voiceless vowels implies the presence of voiced vowels, but not vice versa (Greenberg 1969:156).3
Given the implicational universals cited in (7), the relative markedness of the segment types they refer to is determined as in (8).
(8) Relative Markedness
a.Voiced stops are marked with respect to voiceless stops.
b.Voiced fricatives are marked with respect to voiceless fricatives.
c.Fricatives are marked with respect to stops.
d.Aspirated stops are marked with respect to unaspirated stops.
e.The bilabial nasal is marked with respect to the dental or alveolar nasal.
f.Mid vowels are marked with respect to voiced vowels.
g.Voiceless vowels are marked with respect to voiced vowels.
While implicational universals such as those in (7) must obviously refer to inventories of phonetic segments, the assumption that they hold for phonemic inventories as well offers a principled basis for establishing phonological representations. Thus, if a language has corresponding marked and unmarked segments in complementary distribution, e.g., aspirated and unaspirated voiceless stops in English, the question of which should be posited as the underlying phoneme will be resolved in favor of the unmarked member of the pair. In English, then, the unmarked, unaspirated stops are basic, while their marked, aspirated counterparts are derived by rule. If a language has corresponding marked and unmarked segments in contrast, on the other hand, both are accorded phonemic or underlying status.
Since the relative markedness of a pair of segment types can be determined from typologically based implicational universals, and since these universals hold at both the phonemic and the phonetic levels, it follows that any rule which converts marked segments (e.g., voiced or aspirated stops) into unmarked ones (e.g., voiceless or unaspirated stops) must be a neutralization rule. This follows because a language has phonemic marked segments only if they contrast with unmarked ones, and such a rule neutralizes this contrast. Our definition of markedness thus warrants the generalization stated in (9).
(9) Any rule which converts marked segments into unmarked segments must be a neutralization rule.
Hence, rules like final devoicing and final deaspiration, which convert marked into relatively unmarked segments, are necessarily neutralization rules. There are no languages like German in which final devoicing is allophonic or languages like Sanskrit in which final deaspiration is allophonic, nor could there be, because no languages have the kinds of marked underlying systems which would be required in order for these rules to have purely allophonic effects.
Thus, all rules which produce relatively unmarked segments must be neutralization rules. It is not obvious, however, that all neutralization rules produce unmarked segments. Yet this generalization also appears to be true, at least for the class of well-motivated, phonologically-conditioned rules. In addition to the rules of final devoicing and deaspiration discussed above, which clearly do produce unmarked segments, consider the neutralization rules given in (10).
(10) Neutralization Rules
a.Neutralization of mid and high vowels to high vowels in word-final unstressed syllables in Portuguese (Mattoso Câmara 1972:34);
b.Neutralization of /s/,/č/, and /t/ to [t] in syllable-final position in Korean (Moon 1974);
c.Neutralization of /m/ and /n/ to [n] in word-final position in Finnish (Collinder 1969:10); and
d.Neutralization of /l/ and /n/ to [n] word-initially in Korean (Shibatani 1973:98).
In each case noted in (10), contrasts are neutralized in favor of the unmarked member of the opposition. There appear to be no languages in which purely phonologically-conditioned rules always neutralize similar contrasts to the marked member. For example, there are no languages that consistently neutralize voiced and voiceless obstruents to voiced obstruents, or aspirated and unaspirated stops to aspirated stops. Neither are there languages with neutralization rules like those in (11).
(11) Non-occurring Neutralization Rules
a.Neutralization of mid and high vowels to mid vowels.
b.Neutralization of /s/, /č/, and /t/ to [s] or to [č].
c.Neutralization of /m/ and /n/ to [m].
d.Neutralization of /I/ and /n/ to [1].
There are, however, some rules which do appear to neutralize contrasts to a marked member. Examples of such rules are given in (12) and will be discussed in turn.
(12) Rules Producing Marked Segments
a.Merger of /t/ and /d/ to the tap [ɾ] between stressed and unstressed syllabic segments in English;
b.Merger of unstressed vowels to [ə] in English;
c.Merger of /s/ and /t/ to [s] before a morpheme boundary followed by [i] in Finnish (Kiparsky 1976);
d.Merger of plain and palatalized consonants to palatalized consonants before /e/ in Russian (Jones and Ward 1969: 198-99);
e.Voicing assimilation to a following voiced or voiceless obstruent in Russian (Jones and Ward 1969:197-98); and
f.Nasal assimilation to the place of articulation of a following consonant in Spanish (Harris 1969:8-18).
Of these rules, (12a) and (12b) are not regarded as neutralization rules by definition (1) above. Rather, they are considered allophonic rules which merge segments to some new segment not present in the input to the rule. Taps in English can be derived from their ‘unreduced’ counterparts /t/ and /d/, with which they alternate in slow or formal styles of speech in words like latter and ladder. Thus, there is no reason to grant phonemic status to the tap in English, and the tapping rule is allophonic because there are no taps in the input to the rule.4
Similarly, schwas in English can be derived from their ‘unreduced’ counterparts, the range of full vowels. Some schwas alternate with full vowels, as is illustrated by alternations of the type found in pairs like telegraph and telegraphy. Other non-alternating schwas in unstressed syllables are in complementary distribution with the slightly longer and lower vowel that occurs only in stressed syllables and is transcribed with the wedge [Ʌ] in words like cut and rut. Non-alternating schwas in unstressed syllables, then, can be derived from /Ʌ/ by the same rule that derives alternating schwas from their full-vowel counterparts. In this analysis, a word like above is phonemically ɅbΛ́v/, as is shown in (13) below, and its phonetic form [ǝbΛ́v] is derived by the vowel reduction rule that merges most unstressed vowels to schwa. Schwa is therefore not a phoneme of American English, and the vowel reduction rule is allophonic because there are no schwas in the input to the rule.5
Rules (12c) and (12d) are both rules of questionable status. They do meet the definition of neutralization in (1) because there are morpheme-final non-derived /s/’s that occur before /i/ in Finnish and there are non-derived palatalized consonants before front vowels in Russian. However, both rules are severely restricted as to the environments in which they apply, and both rules are opaque in that there are phonetic representations that meet the conditions for the rules but have not undergone them.
Specifically, the Finnish rule converting /t/ into /s/ before morphemeinitial or derived /i/ accounts for alternations such as those in the singular paradigm of ‘hand’, shown in (14a). As Skousen (1975:70-71) makes clear, however, the purported /t/ to /s/ rule is anything but a valid generalization about Finnish surface phonology. Thus, the suffix -i ‘repetitive’ does not induce assibilation, and neither does the suffix -isi ‘conditional’, shown in (14b) and (14c). Similarly exceptional are the suffix -jä ‘agentive’ and the onomatopoeic suffix -ise, shown in (14d) and (14e). A Finnish rule of t → / s / ____ i, if it exists at all, does therefore not quality as purely phonologically-conditioned.
(14) Finnish Assibilation (Skousen 1975:70-71)
The rule in Russian that changes unmarked plain consonants into marked palatalized consonants before the mid front vowel /e/ effects a neutralization of the contrast between plain and palatalized consonants, as can be seen from the forms in (15a). Before the high front vowel /i/, however, the palatalization rule only applies to velar consonants, where its effect is strictly allophonic (Jones and Ward 1969). After non-velar consonants, /i/ is realized as its central allophone [ɨ], as is shown by the alternations of the suffix /-ij/ in (15b). Thus, palatalization of consonants in Russian must be restricted to apply only before the mid front vowel /e/, except in the case of velar consonants, where it also applies before the high front vowel /i/.
Palatalization before /e/ is not a productive phonological rule, however, in that it fails to apply in a large number of foreign words. In his investigation of the rates of assimilation of borrowings in Russian, Holden (1976:143) states that “dental consonants before e in borrowings show little if any movement toward palatalization in the course of assimilation,” although velar and labial consonants do palatalize in some borrowings. Examples of non-palatalized consonants in the palatalization environment in loanwords are given in (15c). Palatalization before /e/ fails to apply in certain types of native words as well. The /e/ in the first syllable of etot ‘this’ does not condition palatalization of preceding consonants in prefixes, as is shown in (15d), where the /v/ in v etom gorod’ e remains non-palatalized. In addition, palatalization does not apply to consonants before /e/ in acronyms, as is shown in (15e). Finally, consonants do not palatalize before /e/ in the names of the letters of the alphabet, as is shown in (15f).
We conclude, then, that palatalization of consonants in Russian is not a thoroughly productive phonological rule because it fails to apply to loanwords and to certain types of native words in the language.
(15) Russian Palatalization (Jones and Ward 1969)6
a.Nom.sg. loc. Sg. Gen. Sg.
[zal] | [zal’𝜖] | [zala] | ‘hall’ |
[st’il’] | [st’il’𝜖] | [st’il’a] | ‘style’ |
b. Masc. Nom. Sg. Fem.Nom. Sg.
[d’ik’ij] | [d’ikaja] | ‘wild’ |
[glupij] | [glupaja] | ‘stupid’ |
[s’in’ij] | [s’in’aja] | ‘blue’ |
c.[konsom𝜖] ‘consommé’
[t𝜖mbr] | ‘timbre’ |
[mᴧdél’] | ‘model’ |
[z𝜖ró] ‘ | zero’ |
d. v etom ([v𝜖təm]) gorod’e ‘in this city’
e. [n𝜖́p] ‘NEP’ (New Economic Policy)
[𝜖s𝜖́r] | ‘SR’ | (Socialist Revolutionary) |
f. [p𝜖] ‘p’ (letter)
[b𝜖] | ‘b’ | (letter) |
The last two rules cited in (12), rules (12e) and (12f), are quite general rules which do change unmarked segments to marked ones. (12e) changes some unmarked voiceless obstruents to marked voiced ones, and (12f) changes some unmarked alveolar nasals to marked labial, alveopalatal, and velar ones. However, both of these rules also have the opposite effect. That is, they both also change marked segments to unmarked ones. Such rules, then, do not falsify the claim that all neutralization rules produce unmarked segments, because they do in fact produce unmarked segments. It just so happens that they produce marked segments as well. Assimilatory rules that would disconfirm this claim would be those that produced only marked segments, but with the improbable exception of rules like (12c) and (12d), there are no such rules that have a neutralizing effect.
More specifically, alongside rules which assimilate voicing in obstruents to the value for voice found in an immediately following obstruent, as in Russian, we find neutralization rules which guarantee a uniform value for voice in obstruent clusters by devoicing the entire cluster, as in Swedish. But we never find the third logical possibility, a neutralization rule which only voices obstruents in a cluster. These three possibilities are schematized in (16), where the non-occurring rule type in (16c) is marked with an asterisk.
(16) Logically Possible Voicing Assimilation Rules
On the basis of the above observations, we propose that phonological rules are constrained in such a way that neutralization rules must produce unmarked segments. We call this constraint the Markedness Constraint, which is stated in (17).
(17) Markedness Constraint: Phonologically-conditioned neutralization rules convert relatively marked segments into relatively unmarked segments.
The Markedness Constraint obviously excludes a host of possible rules from the grammars of natural languages, namely, all rules defined as neutralizing by (1) whose outputs are always relatively more marked than their inputs.
This constraint does not preclude the existence of rules which do produce exclusively marked segments relative to their inputs, however; it just maintains that such rules are not neutralizing, but rather are allophonic. Thus, the cases of intervocalic voicing and initial aspiration cited in (4) and (5) all represent instances of rules which produce exclusively marked segments, and each of them is allophonic. The same is true of the rules given in (18).
(18) Allophonic Rules
a.Devoicing of vowels in voiceless environments in Comanche, Japanese, Serbo-Croatian, and other languages (Greenberg 1969).
b.Devoicing of certain word-final sonorant consonants in Russian, French, and Icelandic (Jones and Ward 1969:189, Deferrari 1954, Einarsson 1945:24).
c.Spirantization of voiced stops post-vocalically and after certain consonants in Spanish (Harris 1969:37-40).
d.Change of /n/ to [ŋ] syllable-finally in certain dialects of Spanish (Canfield 1962:70-71).
e.Change of /s/ to [h] syllable-finally in certain dialects of Spanish (Canfield 1962:83-84).
These phenomena, and countless others like them, are all accounted for by the obvious corollary to the Markedness Constraint stated in (19).
(19)Corollary: Phonologically-conditioned rules which produce exclusively marked segments from relatively unmarked ones are allophonic.
Related to (19) is the corollary to (9) above, which is that any rule which is not a neutralization rule does not convert marked segments into unmarked ones. This corollary, together with (19), will insure that all rules that produce only marked segments are allophonic rules and that all allophonic rules produce only marked segments.
It should be noted that the Markedness Constraint differs from the proposals concerning the relationship between neutralization and markedness made by Trubetzkoy (1969). Trubetzkoy claimed that the relative markedness of a pair of phonemes in a privative opposition could be determined by the representative of the archiphoneme occurring in the position of neutralization. That is, he defined unmarked segments as those that are found in the position of neutralization. By appealing to implicational universals, however, we define relative segmental markedness independently of phonological neutralization and claim that the product of neutralization is unmarked because neutralization rules are constrained so as to produce unmarked segments, not construed so as to define them.
Our understanding of “neutralization,” moreover, is quite different from the sense in which Trubetzkoy used the term. For Trubetzkoy, neutralization included those cases in which a contrast is actually eliminated as well as cases in which a contrast is simply absent. Thus, neutralization could be either dynamic, involving the merger of morphemes (e.g., German Rad ‘wheel’ = Rat ‘advice’ through final devoicing), or static, involving no such merger (e.g., the English contrast between voiced and voiceless stops is suspended after /s/, as in spit, but no ambiguity results since alternations do not support a /sp/ - /sb/ contrast).
The distinction between the two positions is important, because many static phenomena characterized as neutralizing by Trubetzkoy are actually allophonic in our view. Consider the following quotation.
In German the opposition s-š is neutralized before consonants. The archiphoneme is represented root-initially by š, root-medially and finally by s. [And] . . . the opposition between “sharp” s and “soft” z is neutralized root-initially as well as morpheme-finally, the archiphoneme being represented by “soft” z initially, and “sharp” s finally.
(Trubetzkoy 1969:82)
But while alternations exist which justify positing an underlying /s/-/z/-/š/ contrast in medial and final position in German, as can be seen in (20a) below, no such alternations can be found in morpheme-initial position, where only [z] occurs prevocalically and [š] preconsonantally, as can be seen in (20b).
(20) German Sibilant Consonants
This defectiveness of distribution may well justify positing only /s/ morpheme-initially, along with rules which convert /s/ into the relatively more marked [z] and [š], but since no morpheme-initial /z/ or /š/ would then be found in the inputs to these rules, they would be allophonic by definition (1). There is thus neither contrast nor neutralization among these initials in German. (Of course, if one had posited /z/ and /š/ to begin with as the initials of sagen and sprechen, respectively, no rules would be involved at all, neutralizing or otherwise.)
Thus far, we have suggested that there is a direct correlation between the neutralizing status of a rule and the relative markedness of its input and output. Thus for the class of well-motivated phonologically-conditioned rules, two of four logically possible configurations of rule status and output are excluded in principle, namely, neutralization rules which produce exclusively marked segments, and allophonic rules which produce unmarked ones. These exclusions are indicated by an asterisk in (21).
We maintain that the Markedness Constraint is a functional principle of phonology which explains why a particular rule may be found in one language but not in another. Whether a rule consistent with this constraint actually is found in some language is doubtless due to various factors bearing on the language’s history and phonetic naturalness. But phonetic naturalness alone, for example, is not sufficient to explain why one language maintains a rule of intervocalic stop voicing and another does not. Our explanation is that the phenomenon per se is “phonetically natural” for all languages. But the only languages that can exploit the natural phonetic tendency toward intervocalic voicing are languages like Korean, and unlike English, for which inclusion of such a rule does not conflict with the Markedness Constraint. That is, Korean voices intervocalic stops because the production of these relatively marked segments is accomplished through an allophonic rule. English does not because, if it did, the same rule producing relatively marked voiced stops would effect a neutralization with the set of phonemic voiced stops, and this neutralization would violate the Markedness Constraint.
Now let us consider some of the theoretical implications of this hypothesis and its interaction with other theories that have been proposed.
A theory incorporating the Markedness Constraint claims that permissible phonological rules can be distinguished from impermissible ones by considering only the interaction between a rule’s function, i.e., whether it is neutralizing or not, and the relative markedness of the input and output of the rule. The Markedness Constraint itself thus does not consider the positions within a word or syllable in which particular rules apply. Moreover, the relative markedness of segment types is determined only by reference to typological universals governing phonetic and phonemic systems, quite irrespective of the positions in which various segment types are found to occur.
Other recent phonological theories have proposed that phonological rules be constrained in terms of the types of changes they may produce in particular environments. In the theory of Natural Generative Grammar, proposed by Vennemann (1971 and 1972a) and by Hooper (1974), all segments in a language are arranged on a hierarchy according to their relative strength. Furthermore, all rules are constrained in such a way that those which apply in syllable-initial position must be strengthening rules, and those which apply in syllable-final position must be weakening rules. In the Simplex-Feature Hypothesis proposed by Sanders (1974b), distinctive features are unary or simplex, rather than binary or complex, and all phonological rules involve either the insertion of features or segments or the deletion of features or segments. In this theory, all rules which apply in word-initial position must be insertion rules and all rules which apply in word-final position must be deletion rules.
In both of these theories, a particular rule changing a segment type X to a segment type Y is considered either a strengthening or a weakening rule (in Natural Generative Grammar) or an insertion or deletion rule (in the Simplex-Feature Hypothesis) and is restricted accordingly as to the position in which it may apply. Both theories, then, correctly predict that rules like aspiration rules, considered as either strengthening or insertion, may apply only word- or syllable-initially and rules like deaspiration, considered as either weakening or deletion, may apply only word- or syllable-finally. However, neither theory can handle situations in which a rule changing segment type X to Y in some environment exists in one language and the opposite rule, changing segment type Y to X in the same environment, exists in another. Of course, such situations could be accommodated in these theories if relative strength and the choice of distinctive features were a language-specific matter, but then neither theory would have any empirical content.
Situations such as the one just described do exist in natural languages, however. Consider the types of rules described in (22) below:
(22) Rules effecting opposite changes
a. i. Raising of unstressed mid vowels to high vowels in word-final syllables in Portuguese (Mattoso Câmara 1972:34) and raising of word-final /e/ to /i/ in Finnish (Collinder 1969:10).
ii. Lowering of phrase-final high vowels to mid vowels in Tagalog (Bloomfield 1917) and lowering of high vowels to mid vowels word-finally and before uvulars in Western Greenlandic Eskimo (Schultz-Lorentzen 1945).
b. i. Change of word-final /m/ to [n] in Finnish (Collinder 1969:10) and change of root-final /m/ to [n] when word-final in Sanskrit (Whitney 1879:47).
ii. Change of word-final /n/ to [ŋ] syllable-and word-finally in many dialects of Spanish in both America and Spain (Canfield 1962:70-71).
In (22a) and (22b), examples are given of rules in certain languages effecting a change from segment type X to segment type Y and of rules in other languages effecting the opposite change under essentially identical conditions. Pairs of rules like these appear to constitute counterexamples to the two theories of phonology mentioned above, since if one type of change is weakening in word-final position, the opposite type must be strengthening in the same position, or if one type of change involves the deletion of particular features in word-final position, the opposite type must involve the insertion of those features in the same position.
However, a theory incorporating the Markedness Constraint predicts that we should find situations exactly like the ones described in (22). Examination of the languages involved shows that the rules in (22ai) and (22bi) are neutralization rules and that those in (22aii) and 22bii) are allophonic rules. Since we have already seen that high vowels are unmarked with respect to mid vowels and dental nasals are unmarked with respect to labial or velar nasals, the neutralization rules in (22ai) and (22bi) change marked segments to relatively unmarked ones and the allophonic rules in (22aii) and (22bii) change unmarked segments to relatively marked ones. Thus, these examples provide support for the Markedness Constraint over the theory of Natural Generative Grammar and the Simplex-Feature Hypothesis.
While we have argued that neither of these two theories alone, without recognizing some equivalent of the Markedness Constraint, can predict the variety of different types of rules that are found in natural languages, it is also true that the Markedness Constraint alone does not disallow certain types of rules that apparently do not occur in natural languages. With respect to rules of particular types, the Markedness Constraint restricts the function the rule may have, but does not restrict the positions in which it may occur with that particular function. Consider the schema in (23), which gives examples of some types of rules conditioned by position that would be allowed by the Markedness Constraint, where asterisks are used to indicate rules that are not known to occur in natural languages:
(23) Rules allowed by the Markedness Constraint
It can be seen from (23) that the Markedness Constraint, because it does not refer to position in a word or syllable, allows certain types of rules for which there is no empirical support, in particular, rules like initial deaspiration in a language like Hindi, with a contrast between aspirated and unaspirated stops, and intervocalic devoicing in a language like English, with a contrast between voiced and voiceless obstruents. It appears, then, that in addition to incorporating the Markedness Constraint, an adequate theory of phonology must restrict phonological rules by referring to position. Either the theory of Natural Generative Grammar or the Simplex-Feature Hypothesis, appropriately revised and combined with the Markedness Constraint, might provide such a theory. But the type of situation outlined in (23) can also be explained by a principle of language which restricts neutralization rules to final position, of words or perhaps of syllables, and allows allophonic rules to apply in any position, either initially, medially, or finally. It is as yet unclear whether assimilation rules, in general, are subject to the same constraints as are non-assimilatory rules, or what the types of constraints are on rules affecting vowels, where the level of stress appears to be more relevant than the position in a syllable. But, based on the cases we have examined at this point, at least, it appears that allophonic rules apply in all positions within a word or syllable while neutralization rules apply only in word- or syllable-final position.
An additional comment is in order here concerning the types of rules found in the positions listed in (23). According to the Markedness Constraint, any obstruent voicing rule must be an allophonic rule, because it changes relatively unmarked voiceless obstruents into relatively marked voiced ones. Still consistent with the Markedness Constraint, though, as well as with the preceding observation on the positions to which neutralization is restricted, such an allophonic rule could apply anywhere within a word. Voicing of obstruents, in this view, then, would be expected to occur in word-final position as well as in intervocalic position, although most other phonological theories have claimed that only final devoicing of obstruents, and not final voicing, are permissible phonological processes. However, a recent study by Steyaert (1977) provides evidence for just such a rule of final voicing in the Siouan language of Dakota.7 Steyaert shows that voiced and voiceless obstruents contrast only morpheme-initially in Dakota, but that intervocalically within morphemes stops are voiceless and fricatives are voiced, while word-finally stops are voiced and fricatives are voiceless. Even though the facts here regarding stops are contrary to expectations based on notions of phonetic naturalness, it does appear that Dakota must be analyzed as having a rule of word-final voicing of stops. Further, this rule is allophonic, since there are no contrasts involving voicing that are neutralized by application of the rule.8 If this analysis of Dakota is correct, the language provides a counterexample to many theories of phonology which disallow final voicing of obstruents. However, a theory which incorporates the Markedness Constraint disallows final voicing only if it is neutralizing, but allows such a rule if it is allophonic. As uncommon as this situation appears to be, we take the existence of allophonic rule of final voicing of stops in Dakota to provide strong support for the proposals suggested in this paper.
Historical linguistics constitutes another proving ground for the Markedness Constraint. All things being equal, the Markedness Constraint precludes the possibility of any phonologically-conditioned neutralization rule becoming allophonic or of any allophonic rule becoming neutralizing. It is not the case that all things always are equal, however, and one can imagine circumstances in which change in a language’s phonemic system may precipitate change in the status of its phonological rules, or circumstances in which a rule’s output itself changes so as to alter its markedness value, i.e., the marked output of an allophonic rule might simplify somewhat, resulting in neutralization with other segments in the language.
While we know of no clear-cut cases where these kinds of changes have taken place without also (perhaps necessarily) destroying the purely phonological nature of the rules involved, it is certainly true that synchronically non-general, morphologized neutralization rules are found in most if not all natural languages, and such rules typically produce relatively marked segments. Since they are not purely phonologically-conditioned, however, these rules do not fall within the purview of the Markedness Constraint, which would otherwise disallow neutralization to marked segments. Yet such rules are always assumed to have originated as phonetically-conditioned, general rules, and there is no reason to suspect that the relatively marked output of a morphologized process like umlaut in the Germanic languages, for example, was any less marked at an earlier stage. However, there is reason to doubt the neutralizing character of this earlier rule.
In fact, the consensus is universal that, in its earliest manifestations, the Germanic umlaut rule which fronted vowels (without unrounding them) before an i or j in the following syllable was strictly allophonic (cf. Anderson 1974:288; Twaddell 1938). The umlaut of u to ü and o to ö, though producing relatively marked segments, was not neutralizing because umlaut itself was the only source of front rounded vowels in early Germanic. As King (1969a: 92-101) has shown, like Twaddell (1938) before him, the point at which umlaut does become neutralizing is precisely the point at which the rule loses its phonetic conditioning through other developments, e.g., the reduction of umlaut-inducing vowels to schwa or null. Thus, while an Old High German (OHG) alternation between the singular and plural for ‘worm’, shown in (24a) below, was allophonic because [ü] appeared only before [i] or [j] in the following syllable, the corresponding Middle High German (MHG) alternation, shown in (24b), was phonemic, because [ü] had come to appear in a host of unpredictable environments (e.g., in MHG [übǝl]ᐸ[OHG][übil] ‘evil’).
(24) Germanic Umlaut
That is, [ü] had become phonemicized, and a MHG rule accounting for umlaut in the plural of ‘worm’ now resulted in the neutralization of unmarked /u/ with marked /ü/. Such a rule would have to refer to morphosyntactic properties like “PLURAL”, however, because /u/ did not generally front before schwa (cf. MHG [gutǝ]obl. ‘good’).
This sequence of events is well known, of course, but we suggest that it typifies the way in which some neutralization rules which produce marked segments may arise in a language. Indeed, given our assumptions about the markedness of underlying representations, according to which a marked element is posited only if the corresponding unmarked one is also posited, the ONLY way in which such neutralization rules can come into existence is through the morphologization of allophonic rules. For once a phonemic contrast between marked and unmarked segments has been established, the Markedness Constraint will prevent the introduction of any general rule which converts the latter into the former.
The Markedness Constraint is thus fully consistent with these kinds of historical developments, although, like the discipline of historical linguistics in general, it cannot predict when or if they will occur. However, it can explain some heretofore puzzling phenomena. In the case of OHG umlaut, for example, not only was u converted to ü and o converted to ö, which were purely allophonic changes, but a was converted to e, as in gast singular versus gesti plural, ‘guest’. Since an e vowel was already extant in early Germanic, it would seem that the umlaut of a to e a relatively more marked segment, constituted a neutralization even during the period when umlaut was phonetically conditioned.
But the facts are otherwise. Based on meticulous rime and borrowing studies, handbooks such as Braune (1967:27-28) (cf. also Twaddell 1938) are very clear on the point that the e from umlaut was a closer, higher vowel consistently distinguished from original e (normalized as ë), which was a more open, lower vowel. While there is no dispute that the two were different, and that umlaut of a to e hence did not constitute a neutralization in OHG, there is no reason at all, save one, that the umlauted vowel should have risen to a point higher than original ë. That one reason, we suggest, was to preserve the allophonic character of umlaut by avoiding the neutralization, prohibited in principle by the Markedness Constraint, of umlauted a with the relatively more marked ë.9
In summary, we have proposed that the Markedness Constraint and its corollary, together with the consequences of our definition of segmental markedness and its corollary, constrain phonologically-conditioned rules to the effect that (1) all neutralization rules produce unmarked segments, (2) all rules that produce unmarked segments are neutralization rules, (3) all allophonic rules produce exclusively marked segments, and (4) all rules which produce exclusively marked segments are allophonic rules. Given these constraints, it can be determined by examining the structural change of a particular rule and the phonemic system of a given language whether that rule may be a rule of that language. We have further claimed that a theory incorporating the Markedness Constraint is a more adequate theory of phonology than ones which only constrain particular types of rules as to the environments in which they may apply, although we have noted that some type of restriction appears to be needed regarding the positions in a word in which neutralization rules may apply. Finally, we have shown how instances of historical change are consistent with, and governed by, the Markedness Constraint, which also gives a principled account of certain previously unexplained developments in, for example, the history of Germanic umlaut.10
NOTES
We would like to thank Daniel A. Dinnsen, Larry W. Martin, Gerald A. Sanders, and Tom Walsh for their helpful comments during the preparation of this paper. We are also indebted to Ashley Hastings, Hong Im Iverson, Indira Junghare, Anatoly Liberman, and Per Linell for their contributions to the data included here.
1.These data also show the effect of the phonological process in Korean that voices intervocalic stops, discussed further below. The three series of stops in Korean in word-initial position are described as lax voiceless unaspirated, or slightly aspirated (e.g., [p]), tense voiceless unaspirated (e.g., [p’]), and voiceless aspirated (e.g., [ph]), but intervocalically the lax voiceless unaspirated series appears as voiced. Word-finally, and syllable-finally, these contrasts are neutralized and only voiceless unreleased stops occur (Moon 1974).
2.Both of these aspects of consonant gradation in Corsican, the spirantization of voiced stops and the voicing of voiceless stops, can be conflated with the use of angled brackets as follows:
The first expansion of this schema is the spirantization rule, which is allophonic because there are no voiced non-strident fricatives corresponding to those produced by the rule before the rule applies. The second expansion is the voicing rule, which now also is seen to be allophonic, since there are no voiced stops in the input to the rule. This example shows that it is individual subrules of a rule schema that must be considered in determining whether a rule is neutralizing or not.
3.Greenberg (1969:162) further notes, in his synchronic universal (4): “To every voiceless vowel in any language there is a corresponding voiced vowel of the same quality, but not necessarily vice versa.”
4.Moreover, Ladefoged (1975:147) notes that the rule which merges /t/ and /d/ is part of a more general tapping process in English that applies to all alveolar stops, including the nasal /n/. In the case of the nasal, the result is a nasal tap, as in the word tanner, which, like the oral tap, is not present in the input to the rule. Further, as was pointed out to us by Larry W. Martin, tapping may be one aspect of an even more general phenomenon of consonantal “weakening” between stressed and unstressed syllables. This “weakening” results in the shortening of all consonants, and even in the reduction of /b/ to [β] and /g/ to [ƴ] as in neighbor [néyβr] and wagon [wӕƴṇ].
5.In British English, on the other hand, minimal pairs can be found between [ə] and [ʌ], as in [khǝt] ‘curt’ and [kht] ‘cut’ (Ladefoged 1975:71-72). But here too[ə] must be viewed as a derived segment, in this case from / ʌr/ or perhaps from a rhotacized /ʌr/, because stressed [ə] is in complementary distribution with initial and intervocalic [r] and [ɾ] as in [r𝜖d] ‘red’ and [b𝜖ri] ‘berry’. Thus, despite the superficial contrast between [ə] and [ᶺ], it is still the case that every stressed [ə] is uniquely identifiable as phonemic /ʌr/ (or /ʌr/) and is derivable from such a representation through a general rule of preconsonantal and syllable-final /r/-loss (or derhotacization), a rule which is necessary for British English in any event.
6.The allophonic variation of the front mid vowel /e/ in Russian can be seen in the data in (15). According to Jones and Ward (1969:40-44), /e/ in stressed syllables is realized as [e] before palatalized consonants. In other environments, /e/ is realized as a raised [𝜖ˆ] after palatalized consonants and as [𝜖] word-initially and after non-palatalized consonants. In unstressed syllables the /e/ phoneme is “reduced” in various ways, depending on the dialect (Jones and Ward 1969:44-6).
7.The dialect of Dakota investigated in Steyaert (1977) is that spoken on the Santee reservation in northeastern Nebraska.
8.Steyaert claims that the rule of word-final voicing of stops is neutralizing because there are voiced stops in word-final position that do not alternate with intervocalic voiceless ones. However, we maintain that the rule is allophonic and that these non-alternating final stops are phonemically voiceless, since there is no reason for positing a phonemic contrast between voiced and voiceless obstruents in any position other than morpheme-initial. Under our analysis, there are no voiced stops in the input to the word-final voicing rule, and the rule is therefore allophonic.
9.Comparative Germanic linguistics offers another familiar sound change, which might appear to run counter to the Markedness Constraint. This is the case of Verner’s Law, which had the effect of voicing the Proto-Germanic (PGmc) fricatives /f 𝜃 h/ to [β 𝛅 ƴ] whenever the immediately preceding syllable in the same word was unstressed. This resulted in merger with the PGmc reflexes of Proto-Indo-European /bh dh gh/, which, according to the very thorough reconstruction of Moulton (1954), had the allophones [β 𝛅 ƴ] in postvocalic environments, [b d g] elsewhere. But if we follow Moulton’s persuasive phonemic analysis for PGmc, in which /b d g/ are basic and [β 𝛅 ƴ] are derived, Verner’s Law remains consistent with the Markedness Constraint. This is so because, even though a PGmc form like *[fa𝛅ár] ‘father’ is phonemically ambiguous (/fadár/ or /fa𝜃ár/), no neutralization of the form as defined in (1) is involved, i.e., the merger that takes place between /d/ and /𝜃/ is not to either member of the opposition, but to the third elecent, [𝛅]. Verner’s Law is thus “allophonic” for the same reason as the tapping of American English /t/ and /d/ to [ɾ] is.
10.Discussion after the presentation of this paper revealed several potential counter-examples to the predictions of the Markedness Constraint. In particular, Jonathan Kaye cited a variety of apparently quite general yet neutralizing palatalization rules from diverse Amerindian languages, rules which produce only segments that are marked relative to their inputs and which therefore constitute prima facie falsification of the Markedness Constraint
The individual cases cited bear further investigation, of course, although we have no reason to believe the facts are other than what Kaye presented. If these rules are found to be general, phonologically-conditioned neutralization rules, then the status of the Markedness Constraint will be called into question. However, we believe that the Markedness Constraint represents a true generalization about the vast majority of phonological rules and therefore that it cannot be easily abandoned. In order to accommodate these apparent counter-examples, we suggest two directions of further research. First, the interaction between historical change and the Markedness Constraint must be investigated in more detail. In the Germanic umlaut example discussed above, we considered the situation in which the loss of the conditioning environment of an allophonic rule results in the rule becoming a morphologically-conditioned one and simultaneously produces a change in the phonemic system of the language, altering the rule’s status from allophonic to neutralizing. The palatalization rules described by Kaye are examples of independent changes in the phonemic system of a language which turn an allophonic rule into a neutralizing one, apparently without causing the rule to become morphologically conditioned. Additional cases of this type must be investigated before any principles governing this situation can be proposed. Second, the status of assimilation rules with respect to the Markedness Constraint must be further investigated. Non-assimilatory rules, i.e., feature-changing rules conditioned by boundaries, syllable structure, the presence or absence of stress, etc., pose no problems for the Markedness Constraint, while certain types of assimilatory rules, such as the palatalization rules under discussion, apparently do. Again, we believe that the majority of assimilation rules are consistent with the Markedness Constraint, but only further research will reveal whether there are definable conditions under which the Markedness Constraint is systematically suspended with respect to specific types of assimilatory rules.
We use cookies to analyze our traffic. Please decide if you are willing to accept cookies from our website. You can change this setting anytime in Privacy Settings.