Chapter II

Semiotic Foundations for
Distinctive Features

2.1 Chomsky and Halle ended Sound Pattern of English with a chapter titled Epilogue and Prologue: the Intrinsic Content of Features. This chapter contains a mechanism designed to accommodate the primes they proposed to general facts of phonology and to the authors’ notions of naturalness. The primes Chomsky and Halle postulated forced them to propose conditions for marked and unmarked values: positive specifications were considered marked in certain environments and unmarked in others, just as negative specifications were marked values in some environments and unmarked in others. For example, positive specifications for [voice] or [continuant] were unmarked intervocalically because the mechanics of sound production, articulation, seemed to favor them there. Rather than append a mechanism for adjusting my theoretical primes to the reality of language, I shall establish a semiotic theory first, then use this theory to elaborate an inventory of distinctive features.

2.2 An a priori system of language semiotics should begin with a definition of language. For the purposes of this study, a language is a systematic historical-cultural institution of communication developed by human beings over the ages to unify, divide, stratify, and occupy one another. Vocal language potentiates communication through the use of acoustically discernable, systematic modifications of and impediments to the flow of air between lungs, lips, and nostrils. These modifications and impediments produce sounds which are transmitted through the air and apprehended by the human ear and interpreted by the human brain. The gestures which produce these sounds are overlays on the non-communicative activities of the articulatory tract, breathing and eating. Breathing and eating are non-communicative acts and differ from linguistic acts because the noises they produce: coughs, wheezes, smacks, belches, etc., although meaningful, especially when they are voluntary, are symbolic and analogic rather than semiotic and digital. To communicate with language human beings produce temporal artifacts in the form of semi-arbitrary articulatory gestures which in turn create other artifacts—sound wave configuations or systematic silence. As part of their enculturation, human beings, as senders and receivers of messages, learn the code that interprets the systematic production or suspension of sounds in their language, and they learn to associate meanings with the sound patterns of their language.

2.3 The first act of human sonorous communication is infantile crying, the undisciplined production of laryngeal resonance. Crying is a psycho-motor response to basic needs, among these are air and food. Infants learn its efficacy very early because their guardians attend to their needs as a response to their crying. As infants are enculturated, this laryngeal resonance and the unconstrained babble of early infancy are molded into the sound system used in the infant’s culture. This sound system, a cultural product, can be analyzed into D.F.’s which represent either the sound spectra (acoustic data) or the articulatory gestures used to create the sounds. On both a physiological and a semiotic level, distinctive features are accretions, structures superimposed on the basic act of communication, laryngeal resonance, i.e. crying. An articulatory D.F. inventory is a registry of the gross gestures necessary for sonorous communication, a registry of the marks that human beings make on the air stream.

2.4.1 The concept of markedness has intrigued and frustrated scholars since it first appeared in the writings of the Prague School linguists. Baltaxe (1978: 35-47) traces its history and interpretation from the writings of theoreticians who coined the term, up to the works of contemporary linguists. She asserts that “... the concept of markedness, as it exists in its current form . .. should be taken as a proposal that awaits a great deal of further thought and exploration.” Kuipers (1975) refers his readers to Jakobson (1968) for what he feels to be the proper approach to markedness, an approach in which (SPE, 404) “. . . the marked coefficient of a feature was assumed to be + and the unmarked coefficient always –.” A theory of marking in which the marked coefficient is always + is a much simpler (i.e. stronger) theory than the one which allows the conditions and adjustments that Chomsky-Halle proposed. The strong approach should be examined closely to see if it works effectively in, at least, the paradigmatic specification of phonological systems.

2.4.2 Kuipers categorically rejects Chomsky and Halle’s proposals on markedness and optimistically asserts (Kuipers, 43) that

As to phonological markedness, here, too, a simple approach is ... enlightening ... If we write in black on a white background the black ‘stands out’ and is ‘marked.’ That of which there is less, that which is less usual, will be experienced as ‘marked’ ... therefore, that one of a correlated pair of phonemes which occurs more often, will tend to become the ‘background’ against which its correlate stands out.

2.4.3.1 Notwithstanding this optimism, it is not easy to find an articulate formulation of a theory or program for markedness in any school of linguistic description. The closest approximation to a satisfactory program of marking appears in Greenberg’s (1966) Language Universals with Special Reference to Feature Hierarchies¹ He accepts the idea that a mark is a positive something and suggests the following 5 principles for approaching markedness (13-24 and 58-59). 1) The feature which occurs in neutralization is unmarked. When in a particular class of environments no contrast occurs within a set of lexemes or phonemes which differ from each other only in a single feature, it is the unmarked feature which appears in this environment. 2) The unmarked member of a pair of phonemes is more frequent in texts. 3) Unmarked members have greater allophonic variety than marked members. 4) The number of phonemes with the marked feature in the phonemic inventory of a given language is always smaller than or equal to the number with the unmarked feature. 5) The basic allophone, defined in terms of phonologic independence from its environment, is the unmarked one.

2.4.3.2 Principles 2 and 4 are capital to the semiotic foundations of the distinctive feature inventory I shall propose. Principle 3 will have to be substantiated in interlinguistic studies once a satisfactory set of phonological primes has been derived. It seems, however, that it is correct; least marked segments will have relatively few positive specifications, thus they should have greater phonetic flexibility than highly marked members of the same system, provided that there is ample phonological space around the unmarked segment. For example: according to this principle and the system of D.F.’s that I propose, the least marked contoidal phoneme in Portuguese is /r/, the most highly marked, /š/. Using the system of primes that I propose in this study they are specified as follows:

/š/ has, in standard dialects of Portuguese, no significant allophonic variation; and it occurs only syllable initially;² /r/, on the other hand, occurs in the following environments: C__V, V__V, V__C, __#. When it occurs in syllable initial consonant clusters, it often assimilates in voice to the preceding consonant. When syllable final it may have, in Carioca Portuguese, any of the following realizations: [x, , h, r, ɹ, Ø, x̃, ]. This happens partially because of the few intrinsic positive specifications of the phoneme and because none of these sounds represents any phoneme other than /r̃/, which contrasts with /r/ only intervocalically. I am, however, somewhat ill-at-ease with Principle 3. The allophonic variation of a set of segments seems to be language-specific rather than phonetically predetermined by a set of abstract principles. In American English, the voiced stops have virtually no allophonic variation whereas the voiceless ones may be aspirated, flapped,³ glottalized, or articulated with no aspiration. On the other hand, in Portuguese and Spanish, voiceless stops have no significant allophonic variation whereas the voiced stops do. Linguists need to understand the dynamics of phonological systems much better before Principle 3 can be an axiom of phonological markedness.

2.4.3.3 Principle 5 is taken for granted in D.F. approaches to phonemic inventories. Segments are specified to the point that is necessary to distinguish them from their closest coordinates, provided that they are distinguished from all segments. If a particular allophone of a phoneme acquires extra features, i.e. is phonetically more complex than the basic allophone, the redundant features are added in a phonological rule. For example, the voiceless stops of English contrast with all other segments in that language by virtue of their being [+occlusive +surd]. Their aspirated allophones are phonetically more complex than the basic phoneme, but this complexity is not necessary to distinguish them from the other phonemes of English. The aspiration of these consonants is mandatory in certain environments and must be reflected in a rule in the phonological grammar of English, and this rule adds a feature to their specification.

2.4.3.4.1 Principle 1 has its basis in dicta of the Prague School. Trubetzkoy (Introduction to the Principles of Phonological Descriptions, 27-28) maintained:

Two phonemes or, as the case may be, phonemic classes between which a neutralizable contract exists, are said to be especially closely related to one another, and, when this neutralizable contrast is one which can be described as presence or absence of a particular feature, the phonemes in question are termed respectively the ‘marked’ form and the ‘unmarked’ form. The unmarked form here is always that phoneme which, where the contrast in question has been neutralized, appears as the sole representative of the relevant pair of phonemes—provided, of course, that the situation is not obscured by assimilation . . . For determining the phonological content of the individual phonemes and for the understanding of the whole structure of the relevant system . .. any phonological description ... must be written in such a way as to reveal just which phonological contrasts are neutralizable in the language under study and which terms in such contrasts are to be regarded as marked, or, as the case may be, unmarked.

It is remarkable that this principle still is accorded currency today. The statement has led to confusion in the writing of many linguists, including Greenberg. Neither Greenberg nor Trubetzkoy give us any reason why the unmarked phoneme occurs in a position of neutralization. In the Portuguese data, for example, /r/ and /r̃/ contrast only intervocalically: caro ‘dear’ carro ‘car.’ There are environments where only /r̃/ occurs: #__, n__, z__, 1__; where only /r/ appears: p__, t__, k__, (see above); and there are environments where the phoneme /r/ occurs exclusive of /r̃/: __#, __Ȼ. In the last of these three environ ments, prestige dialects of Brazilian Portuguese realize /r/ regularly as [x], the normal realization of /r̃/, yet morphophonemically is it /r/. The phrase ‘... obscured by assimilation ...’ is of supreme importance. While it is difficult but not impossible to make a case for the assimilatory nature of much of the allophony of /r/ in Portuguese, many neutralizations of phonemic contrasts are the result of assimilation of one kind or another. Common examples are: 1) regressive assimilation of point of articulation of nasal consonants contiguous to other consonants in Spanish, Portuguese, and Italian; 2) regressive or progressive assimilation of voice between continguous consonants; 3) vowel harmony.

2.4.3.4.2 The neuturalization of intervocalic /t/ and /d/ in North American English can be seen as assimilation. With the loss of this contrast in words like latter and ladder the /t/ loses its [surd] quality and becomes assimilated to its environment as far as its voicing is concerned.

2.4.3.4.3 The neutralization that researchers refer to most often when dealing with markedness is that which occurs in Germanic and Slavic where intervocalic voiced consonants alternate with surds word finally: Polish [šfábe] ‘Krauts’, [šfáp] ‘Kraut’. Here again, a case can be made for considering the devoicing of the word final stop as an assimilation—an assimilation to the voicelessness of a potential pause which may occur after any word. But linguists have concluded from data similar to the example above that stop consonants are naturally voiceless, that voiced stops are marked because air is impeded at the point of articulation yet the vocal bands continue to vibrate to produce voiced stops. Because of his adoption of this tenet, Greenberg is forced to acquiesce to a position on the markedness of segments which foreshadows Chomsky and Halle’s (24):

It should be noted that in some cases we had what might be called conditional categories for marked and unmarked. For example, whereas for obstruents, voicing seems clearly the marked characteristic, for sonants the unvoiced feature has many of the qualities of a marked category.

In this work I shall challenge this assertion and adopt the feature [surd] (i.e., voicelessness). Thus voiceless segments, be they stops, fricatives, laterals, nasals, or vocoids are marked. That is, [-surd] phonemes will normally exceed the number of [+surd] phonemes in text segmental frequency counts, and, normally, the [-surd] member of a pair will be more proponderous than the [+surd] member.

2.4.4.0 Incorporating Principles 2, 4, and 5 from Greenberg; considering 3 as moot, and rejecting 1 as baseless, I shall proceed to discuss six other parameters of markedness that Greenberg touches on in his treatise. The following parameters are interrelated facets of markedness: exclusion, complication, irregularity, infrequency, implication, and accumulation.

2.4.4.1 A form or category which excludes the other members of a set to which it belongs is the marked category and its markedness should be specified positively. In languages which have grammatical gender, as the Romance languages do, [feminine] is the marked category because its use excludes many members and identifies a culturally determined perceptual minority. On the other hand, the non-feminine (traditionally called the “masculine”) gender is the unmarked category. When we say in Portuguese Tenho 16 filhos, 15 moças e um rapaz, “I have 16 children, 15 girls and one boy” we use the unmarked form for “offspring” because we can include the marked category [feminine] as a subset. On the other hand, were we to say Tenho 16 filhas “I have 16 daughters” we preclude any possibility of there being a son among those 16 offspring. Traditional distinctive features also perform this function: [+nasal] or [+lateral] identify minority subsets in the totality of segments in any phonemic inventory. Negative specifications such as [-nasal] or [-lateral] will characterize majorities. In other words, there will always be more segments not marked for a given feature than segments so marked.⁴

2.4.4.2 A form which is more complex in the number of elements that comprise it usually is the marked category since it designates a subset of the set in which it is included. In the Hebrew utterance /haizraelí tov/ “The Israeli is a good person” the morpheme -i represents all citizens of Israel. On the other hand, if we say /ha izraelít tová/ ‘the Israeli woman is good’ we make no linguistic judgment about the other half of the population because we have used the marked forms and refer to only the perceived minority. The tenses and aspects of Romance languages illustrate this as well: if we say in Spanish Damos dinero a los pobres ‘We give money to the poor’, such an utterance can be used to designate 1) habitual action Todas las navidades damos dinero a los pobres, ‘Every Christmas we ...’; 2) futurity Hoy es sábado, damos dinero a los pobres mañana ‘Today is Saturday, we(‘ll) give ... tomorrow’; 3) historical present (e.i., timeless) Nos levantamos a las siete, vamos a misa a las ocho, damos dinero a los pobres, y finalmente almorzamos ‘We get up at 7, go to mass at 8, give alms to the poor, and finally, we eat.’ 4) Hypothetical action Si damos dinero a los pobres nos sentimos felices ‘If we give ... we feel happy’. In contrast, the morphologically complex, albeit portmanteau and highly irregular, form dimos ‘we gave [perfective]’ can be used only to indicate past action at a certain point without future, present, or hypothetical implications. Complexity in phonology is reflected by the accumulation of marks to specify a given segment: a⁵ is universally accepted as a simple sound in human articulation. The organs of speech are at rest with the larnyx vibrating. A sound such as ü (agreed to be more marked than a) is much more complex since its articulation involves the raising of the tongue blade and the rounding of the lips—additions or marks on the egressive air stream. This complexity is paralleled by ü’s relative infrequency as a systematic sound in the languages of the world and its low frequency relative to other vowels, in languages where it has phonemic status.

2.4.4.3 Irregular or non-general members of a particular set are marked members of whatever set they pertain to. Grammatical gender again serves to illustrate this point. In Portuguese most feminine (marked) entities are designated by the use of the feminine gender suffix, -a.

The marked members are learned by rote rather than rule. In phonology, sounds that do not use the egressive pulmonic air stream are exceptions to the general rule that egressive pulmonic air is the principal source of sound systems, thus glottalic and velaric air stream sounds receive special marking for their special status.

2.4.4.4 Infrequency or rarity is another trait of marked forms. Greenberg showed (43) that singular items outnumber plurals and that cardinal numbers occur more than ordinal numbers in text frequency counts. Articulatorily simple sounds are more frequent than complex ones: a likely outnumbers all other segments in overall frequency both intra- and inter- lingually; a sound like a occurs in all languages, one like ë in relatively few.

2.4.4.5 A mark implies the existence of its negation. That is, a feature such as [+nasal] is only invoked when nasal sounds contrast with non-nasal sounds in a given language. The existence of a first and second person in a grammar implies, through the negation of both, the existence of an unmarked (third) grammatical person.

2.4.4.6.0 The accumulative facet of markedness applies both to segments and to systems. As a segment becomes more and more complex, its complexity is reflected in more and more positive specifications. As a system becomes more complex, more and more primes are necessary to distinguish all the items in it. We may therefore refer to systems as more or less marked according to the number of items and rules needed to describe them. The feature acquisitional hierarchy that Jakobson proposed in Child Language, Aphasia, and Phonological Universals implies a successive acquisition of greater and greater markedness in phonological systems, but late acquisitions in the segmental inventory are not necessarily highly marked. For example, l and r can be shown to be simple and unmarked relative to p’s and k’s yet their existence implies the presence of more complicated segments.

2.4.4.6.1 Jakobson maintains that the acquisition and decay of phonological systems operate as mirror images of one another: the last segments acquired are the first to be lost, the earliest acquisitions, the last to be lost. An examination of his ontogenesis, because of its putative universality, will be part of the formulation of a theory of phonological markedness since we shall see how the addition of distinctive marks conforms to the principles of markedness independently set up here.

2.4.4.6.2 The contrast between contoid and vocoid is held to be universal. It is the first to appear in a child’s phonology and is the last to disappear in speech pathology. The first vocoid is a sound like a, the first contoid, a sound like p. At this point, all the sounds are either [±contoid] with the positive specification indicating the later acquisition.⁶ The next contoid to appear is m [+nasal] after which t [+raised] appears. According to Jakobson (48) these are the minimal ingredients for any phonological system. Vocoidal differentiation proceeds from a to i [+high], then to either u or e which produces either triangular i~a~u or linear i~e~a systems and the minimal number of primes for vowel systems [±raised].

2.4.4.6.3 While Jakobson does not point it out, the development of a minimally differentiated vowel system, which occurs after the minimally differentiated consonant system has developed, pushes the consonant system towards further development with the addition of either [high] or [dorsal] to the acquirer’s repertoire. Now, a stop such as k [+high +dorsal] can appear. After the acquisition of a k, fricatives appear with an s as the basic segment (55). Jakobson is not explicit about the rest of the ontogenesis. He maintains that for any affricate to appear, the system must already contain the homorganic fricative. Once friction is added to segmental inventories, consonants contrast in two primes: [nasal] and [friction]. For [occlusion] to become a distinctive feature, a contoid must appear that is either an affricate or an r. Affricates are both fricative and occlusive, r’s are neither. Sounds such as r and l emerge late in the acquisition schedule, but are the first to be affected in the decay of phonological systems, either through linguistic change involving simplification or through the speech pathology of an individual. After these consonants have been added to repertoires, there may be even further elaboration—the acquisition of tenseness, labiality, and nasality in the vocalic sector, for example.

2.4.4.6.4 If Jakobson is correct, phonological systems seem to develop in a dialectic process—going from diametrical oppositions towards successive approximations. The maximal contrast, a~p, involves complete opening vs. complete closure at the outermost point of the vocal tract. Nasality is a synthesis: m consists of the oral closure of p along with the voicing and constant, albeit rerouted, flow of the egressive air of a. The addition of t [raised] adds a new parameter and is another diametrically opposed expansion because place of articulation or use of tongue has become distinctive in contoidal articulation. This new parameter is put to use in the vocoid inventory: the addition of an i makes tongue position a distinctive category in the vocalic subset of segments. Once e or u is added, either relative tongue height or backness becomes distinctive. The addition of the parameter [high] suggests a synthesis of articulatory properties present since [high] approaches contoidal articulation; the addition of [dorsal] suggests a diametric opposition since articulation in a different part of the mouth has become distinctive.

2.4.4.6.5 The last segments to appear in contoidal inventories, r and l, are, at least aerodynamically, negations of contoidal properties. They are the most vocoidal of the contoids. Conversely, late acquisitions in vocoidal inventories, front rounded vowels, and nasal vowels, have more and more marks, thus are more akin to contoids. Contoids and vocoids begin as polar opposites and gradually approach one another as phonological systems become more and more elaborate.

2.5 What should we ask of a D.F. inventory as a theory of phonological systems? A theory of phonological systems should, in a manner as economical as possible, 1) be able to specify all of the contrasting segments that linguists have found or are likely to find in the languages of the world; 2) be able to describe all of the articulatory changes of distinctive consequence that occur in human speech sounds as defined by the inventory, 3) be able to explain sound changes in a manner consistent with what is known about the mechanics of the articulatory tract, 4) provide a positive mark for each distinctive sound segment type that linguists describe; 5) describe complex sound with more positive specifications than those necessary for simple sounds; 6) reflect unusual or unexpected articulatory gestures and air stream initiators with positive specifications, since these are considered marked; 7) reflect the frequency of common sounds with fewer positive specifications; 8) reflect the greater complexity of phonological systems with a larger number of distinctive parameters specifying the segments; 9) imply the presence of other articulatory gestures which are not marked; 10) establish a neutral segment in the description of each phonological system (This neutral segment should be the a like sound, the segment with no positive specifications. It will be the semiotic backdrop for all the other sounds of the language and all other sounds will be marked in relation to it. This is, of course, not to imply that the a’s of any two languages are phonetically the same, rather that all languages have a backdrop phoneme.); 11) contain no feature which is necessary for the specification of segments in all phonological systems. According to the principles established in this chapter, anything that is universal in language cannot be a mark since that which is universal serves as a backdrop for further modification. For example, [voice] cannot be a mark because all languages have voiced segments. On the other hand, the converse of voice, [surd], is a mark because it implies its converse, there must be voiced sounds for voiceless ones to function; there are languages which contain no voiceless phonemes, and there are languages in which the voicelessness of segments can be shown to be redundant.⁷ A mark is non-universal. By adhering to these principles, a D.F. inventory is a theory of phonological markedness.

2.6. I would like to express caution concerning the implications to be drawn from the apparently diametrical expansion of distinctive feature inventories (phonological systems). While many scholars appear to claim that binarity is intrinsic to the way all human beings think, such a conclusion may be premature. The binary analysis of speech sounds predisposes us to see the development of segmental inventories as diametrical. While the binary approach facilitated alphabetic writing, it is not the only way to study phenomena, and it is not a particularly strong hypothesis about language or human behavior, if we can continue to add binary parameters in order to distinguish or specify all the data we apprehend. (I refer the reader back to the power of the Chomsky-Halle D.F. inventory). I claim no psychological reality for the features I propose. They are analytic tools tailored to the mechanics of the articulatory tract and the principles outlined in this chapter.

Footnotes

¹Even though Language Universal preceded SPE by two years the authors of SPE did not acknowledge Greenberg’s work on markedness.

²The best analysis of syllable final sibilants in Portuguese considers them to be allophones of /z/, realizable as [s, z, š, ž] according to dialect and position (Brakel, 1977: 139-140).

³Some speakers of English flap intervocalic /d/’s: bed [béd] bedding[béƌIn]. This does not occur in my speech. Intervocalic /t/ and /d/ are realized as [d]. The point that the voiceless stops of English have a wider range of allophonic variation than the voiced stops is incontrovertible.

⁴A possible exception to this generalization is the major class feature [contoid]. Phonological systems regularly contain more consonants than vowels, although when all segments of a relatively large text are counted, vowels and consonants occur in equal numbers—roughly 50% of the corpus will be vowels even thought the vowel phonemes number much less than the consonants, Brakel (1979a), Alarcos LLorach, (1968: 198-200).

⁵When referring to a segment with no regard to the linguistic sytem in which it occurs, I italicize. The segment a is a low, central, unrounded, non-nasal, tense, vocoid.

⁶Other features of a contoid sound such as p are redundant at this point. Its voicelessness, labiality, and occlusion play no role. Admittedly, any one of these features could be considered to be the first acquired.

⁷Voicelessness does not play a distinctive role in the phonological systems of Andoa, Gadsup, and Nunggubuyu. Voiceless phonemes do not occur in Dyribal. See Appendix A.