Chapter III

Distinctive Features Revisited:
Towards a Stronger Theory of
Phonology

3.0.1 A D.F. inventory is a hypothesis about all the phonological systems in the world. It says that these and no more than these features are necessary to characterize all the phonological systems produced by the human race. The smaller the number of parameters the easier the theory is to falsify. The following is a list of the phonological primes that Chomsky and Halle proposed for characterizing all the phonological systems of the world:

3.0.2 Chomsky and Halle based many of their features on phonetic phenomena with no regard for the semiotic value of these phenomena. They followed time-honored procedures for the most part, so, in order to implement a strong theory of markedness, it is necessary to question either the place in the system, the definition, or the validity of 23 of the 28 features proposed in SPE. I shall examine those features bearing asterisks; 1 have no comment concerning the rest. My aim is to reduce the number of primes and strengthen the hypothesis at the same time making the features phonetically and physiologically more accurate, thus increasing the observational, descriptive, and explanatory adequacy of the theory.

3.1.1.0 In this section 1 shall discuss the Major Class Features and their relationships to [continuant] and [nasal].

3.1.1.1 My first clue that something was wrong with the SPE D.F.’s came in work on Portuguese. Like other Romance languages, Portugue possesses an archetypical, balanced, apparently totally unremarkable segmental inventory. Such a system should be easily accounted for with a set of features that purports to be universal, yet neither those of SPE nor those suggested by Ladefoged (1971, hereafter PLP) can define a natural class for a morphophonemic rule of Portuguese nor can they define the consonantal syllable codas as a natural class. The canonical pattern of syllables in Portuguese is CV, with ony 4 consonants, /r, n, 1, z/, and the glides, [y] and [w], capable of forming codas.¹ Mira Mateus (1975) specifies the consonants as:

When I reviewed her work I felt that there were two errors in these specifications: 1. nasals specified as [-continuant], 2. the flap /r/, a sound incapable of prolongation, the most momentary sound of the language, specified as [+continuant]. Mira Mateus’ specifications are inspired by Chomsky and Halle’s description of [continuant] (317): “... sounds in which the primary constriction in the vowel tract is not narrowed to the point where the air flow past the constriction is blocked.” As far as they were concerned, occlusion in the oral cavity made a segment [-continuant]. Their reasoning (318) on flaps as continuants is contorted and unconvincing.² The Portuguese segments in question are voiced, anterior, and coronal, yet these features do not define these segments as a class since they include /d/ [+voice,+coronal, +anterior, -continuent] which does not occur syllable finally in Portuguese. The feature [sonorant] (SPE: 302) excludes both /d/ and /z/ forcing syllable-final consonants to be described as:

In order for this specification to work, it would have to be ordered before the rule of voiced consonant lenition whereby /b/, /d/, and /g/ become [], [] and [] since it suggests that [] may be syllable final, which is not the case. The disjunction in this rule suggests an underlying similarity that SPE D.F.’s are incapable of describing.

3.1.1.2.1 Do /r, 1, n. z/ form a natural class? The SPE phonologist would have to say “no”; yet to the question “Is there an exclusive phonetic similarity in these segments?” a phonetician would answer “yes.” The phonetic traits that unite all four consonants is their alveolar articulation and their continuous stream of egressive air. At no point in their articulation is the air stream completely occluded. The flap is a ballistic movement of the tongue against the alveolar ridge; this movement is not dependent on, nor does it occlude, the egressive air stream. We may describe these sounds as a natural class most easily and without disjunction if we discard the feature [continuant] for [occlusion], i.e. the total impedence of the air stream.

Now, syllable codas, the consonants /r, n, 1, z/, and all the vowels can be described in a single statement.³

3.1.1.2.2 Wheeler, (1972: 92) remarks that “[continuant], though a manner of articulation feature, is required to define the basic category of stops.” Indeed, linguists often find it necessary to refer to stops (the segments in which the egressive airstream is totally impeded) as a separate class, yet the only means they have for doing this using SPE D.F.’s is with negative specification of [continuant] and [sonorant]. Semiotically this is wrong since the stops in any language constitute a subset of the entire inventory. Indeed, SPE oriented phonologists often refer, informally, to the class of stops and fricatives using an ad-hoc feature [obstruent] because the phonetic primes they use do not provide a positive identity for these sounds. Stops impede the natural state of the vocal tract (breathing); and in a strong theory of marking, should be positively specified. Nasal consonants do not impede the air stream and should not be classified as stops.

3.1.1.2.3 Jakobson, Fant, and Halle (JFH) distinguish stops as (21) interrupted phones with “abrupt onsets” and “sharp wave front[s] preceded by periods of silence.” Nasal consonants (considered [-continuant] in SPE) are characterized by JFH (39, 40, 52) as having vowel-like qualities, notwithstanding stop-like articulation in the oral cavity. JFH felt that redundant features could be eliminated from D.F. matrices and consequently never specified nasals as [∓interrupted] in their grids nor did they take any rigid stand on the character of these consonants (cf. Anderson, 1976: 328).

3.1.1.2.4.0 Apparent rationale for specifying nasal consonants as [-continuant] is found in Harms (1968: 33): “The nasal (stop) consonants are normally seen as phonetically non-continuant, thus allowing for nasalized (continuant) fricatives as well.” The distinctive quality of nasal fricatives in any language is debatable; Chomsky and Halle (316) state that while:

Ladefoged (1964, p. 24) reports that Twi has nasal affricates which contrast with both nasal and non-nasal plosives, we do not know of any certain examples of nasal continuants such as nasal [s̃] or [z̃].⁴

3.1.1.2.4.1 The most elaborate defense of [continuant] (i.e. of nasal consonants as [-continuant]) appears in Anderson (1976) who acknowledges (328-329) that nasal consonants do not have the acoustic properties of stops. He maintains, however, that often stops and nasals may pattern similarly and may trigger similar operations and should be included in a natural class. He asserts (329):

... it is perfectly coherent ... to define non-continuant as . . . articulated with a complete blockage in the oral cavity. This will delimit a class of sounds including both the oral stops and the primary nasal consonants. It is then an empirical question whether this class is a useful, linguistically significant one, or whether we would rather prefer the class of sounds involving continuous airflow.

3.1.1.2.4.2 Anderson cites two examples in which nasals and stops (and in one case fricatives and sonorants) undergo similar modifications. He feels that a rule for this can only be written if nasals are specified as [-continuant].

A.1. The phoneme /k/ in Finnish supposedly spirantizes before oral stops and nasal consonants, i.e. before /t/ and /n/. He offers the following rule:

Aside from the fact that this rule would spirantize /k/ before many more segments than /t/ and /n/, Anderson leads the reader astray with this datum and his rule. Except for the two verbs he cites, /näk-/and /tek-/,/k/ only spirantizes before /t/. The spirantization of /k/ in these verbs (before /n/ and /k/) is considered an anomaly (cf. Campbell 1975, 15-16), yet Anderson uses them to defend his postulates (330):

Just as this class [presumably /n, t/] is described by the feature [-continuant], defined as ‘complete blockage of the oral cavity’. It would not, of course, be given by the specification ‘blockage of airflow’. Thus we have a case in which the proposed definition of the feature [continuant] allows us to specify the environment of a phonological rule in a much more satisfactory way than would be possible under the alternative proposal.

A.2. Even the case of a /k/ that regularly spirantized before /t, n/ the feature [-continuant] is not necessary for generating this phenomenon. The rule:

produces the desired effect using SPE Features and implies oral closure without commitment on the impedence of the egressive air stream. Of course the disjunction in this rule indicates that /t/ and /n/ are not members of a natural class. Nasal consonants are similar to stops only because egressive oral air is occluded, but nasals share the property of “continuous flow of air” with all other segments.

B. The other example he gives is the “... well known case of consonant lenition between sonorants in the Celtic language ...” where (330):

(a)Voiceless stops become voiced (/p t k/ → [b d g])

(b)Original voiced stops become spirants (/b d g/ → [v ])

(c)Voiceless continuants become voiced (/s / → [z r l])

(d)/m/ becomes a nasalized spirant [ṽ].

Anderson writes no rule in features to describe these changes; he even suggests that the voicing of // and // cannot be included if such a rule were written.

It is clear that is /m/ is treated as a non-continuant, the weakening of consonants other than dental sonorants [my italics] can be described in a unitary fashion as follows: between sonorants, such consonants become voiced; if already voiced, they become continuant.

The formulation of such a rule would be something like:

Anderson’s exclusion of // and // is remarkable inasmuch as the following rule generates all the changes:

Friction is contingent on underlying voice and the only nasal to spirantize is /m/.

3.1.1.3 It is convenient, and often necessary, to refer to stops and nasal consonants as a natural class (segments with oral closure), but this should not be done at the expense of phonetic accuracy. All consonants are united by the feature “oral point of articulation” (cf. SPE, 302-303). In the event that nasal and occlusive consonants occasion similar phenomena in a particular language, the following specification will suffice to distinguish them from all other segments in the particular language:

Any closer association of nasal and stop consonants is misleading. In Jukun (Hill, unpublished), there is a three way contrast between occlusives, nasals and nasal occlusives:

These segments can only be distinguished using the features [occlusion] and [nasal]. In addition, non-occlusive nasals nasalize a following vowel /ma:/→[mã:], /na:/––––[nã:] and /ηà/→[ηã]. This can be regulated with the following rule:

The SPE specification of nasals as [-continuant] which unites them with stops, would force a linguist to consider the nasal stops of Jukun sequences rather than units.

3.1.1.4 By classifying all sounds with oral closure as [-continuant], linguists imply that the only difference between stops and nasals is the lowering of the velum and “sonorance” While these may be the only articulatory differences, the dynamic differences are so great as to indicate a serious shortcoming of a phonetic theory that claims similar air stream properties for stops and nasals. Notwithstanding their consonantal articulation, nasals may be prolonged and may constitute a syllable nucleus. Indeed all [-occlusion] segments share this property (even flaps may be lengthened into trills)⁶; [-occlusion] segments may be used to provide basic resonances on which tunes may be hummed or on which pitch variations may be superimposed. Occlusives, articulated by themselves using egressive pulmonic air, do not share these properties. Nasality may be superimposed on any oral vocoid, and nasal consonants are, like vowels, “naturally” voiced. The only affinities that “normal” nasal consonants have with stops is the major class feature [+consonantal] which implies a point of articulation in the oral cavity.

3.1.1.5 Do we loose anything by discarding [continuant] for [occlusion]? 1 have checked phonological systems of many different languages (see Appendix A) and no distinction has been lost. Nasal fricatives can be indicated as their oral counterparts plus nasality—ṽ or z̃ etc. Pre- or post-nasalized stops (see 3.4) would be specified as both occlusive and nasal with delayed transition of either the primary or secondary articulation if necessary. With [occlusion] as the distinctive feature of stops and affricates, the consonants divide as follows:

3.1.1.6 Seeing that nothing is lost by replacing [continuant] with [occlusion], what is gained? Specification of a natural class of syllable final segments in Portuguese is, perhaps, the least important desideratum of this change. The feature [+occlusion] provides: 1. a positive and unique specification for stops and affricates. 2. a greater level of observational adequacy, since this feature provides a means of recognizing that nasals have articulatory affinities with stops and air stream affinities with all other segments. 3. a strengthening of the D.F. hypothesis because by replacing [continuant] with [occlusion], the feature [sonorant] can be eliminated.

3.1.2 Chomsky and Halle define sonorants as (302) “. . . sounds . . . in which spontaneous voicing is possible . . . Sound[s] . . . with more radical constrictions than the glides, i.e. stops, fricatives, and affricates, are nonsonorant, whereas vowels, glides, nasal consonants, and liquids are sonorant.” Chomsky and Halle contradict themselves because nasals and liquids have consonantal articulation; their oral articulation is more radical than that of glides. In the system advocated here, the class of vowels, glides, nasals, and liquids and approximants may be united as follows:

The specification of a set of segments using independently motivated primes is more highly valued than the postulation of a prime to define a class of segments, especially in this case since [sonorant] includes more segments than it excludes, being a specification for both vowels and consonants. The elimination of [sonorant] from the D.F. inventory strengthens the hypothesis by a factor of 1 and it obviates the counter-intuitive specification of the glottal stop as a [sonorant] (SPE 303). The glottal stop can be specified as [-consonantal +occlusion].

3.1.3.0 Since nearly all phonological systems include the feature [nasal] and since nasals represent a synthesis of what has been traditionally referred to as “consonantal” and “vocalic” articulation, I consider them more basic to a D.F. hypothesis than the heading “secondary apertures” (see 1.1) suggests. Nasality is one of the major class features.

3.1.3.1 Grappling with consonants that can only be interpreted as units and which involve either prenasalization e.g.mb, post-nasalization bm, or mesonasalization dnd, Anderson⁸ concludes (343):

. . . nasality occupies a peculiar status in the inventory of features. Though generally considered on a par with other features of manner of articulation, it is in some ways a suprasegmental, on a par with features of pitch. As a result the nasal consonants most generally found in the languages of the world are most naturally treated as oral stops on which a nasal pattern is realized: if the stop is nasal throughout, we get the common primary nasals, while ‘contour’ nasality patterns give rise to pre- and post-nasalized stops. The result is a theory of phonological and phonetic representation in which the segmental idealization is no longer quite so pervasive, since segments have internal structure which may be manipulated by rules.

The ability of the feature [nasal] to co-occur with (SPE) features such as [vocalic], [consonantal], etc. gives it a pervasiveness reminiscent of suprasegmentals, but some well-known suprasegmentals also involve laryngeal activity while nasality need not. Anderson recognizes (313) that the two types of nasal consonants (common primary nasals: m, n, ñ, etc. and pre- and post- as well as mesonasal stops) could be easily described using a feature such as [occlusion], but he argues in favor of [continuant].

3.1.3.2 A theory which includes [occlusion] can specify these unusual segments by combining the features: [occlusion, nasal, primary delayed transition, secondary delayed transition].

Nasality is considered to be the secondary articulation here since [+nasal] implies the presence of [-nasal] and not vice-versa. The mb order is the natural or unmarked combination of nasal and oral contoidal articulation, therefore neither D.T. 1° nor D.T. 2° is marked.

3.1.4.1.1 Chomsky and Halle’s purportedly phonetic definitions of [vocalic] and [consonantal] do not help phonological theory. They define the class of [consonantal] segments as (302-303) “... sounds ... produced with a radical obstruction in the midsagittal region of the vocal tract; . . . not every sound produced with a raised tongue is consonantal. The so-called retroflex vowels ... are ... non-consonantal.” Rigorous work in phonetics has used the terms “consonant” and “vowel” to indicate patterning properties and the terms “contoid” and “vocoid” to indicate phonetic properties of segments. Pike (1947: 13-14) defines vocoids as “... any sound which has air escaping (1) from the mouth (2) over the center of the tongue (that is, not laterally) (3) without friction in the mouth (but friction elsewhere does not prevent the sound from being a vocoid).” He defines non-vocoids (i.e. contoids) as (24): “. . . any sound in which the air stream escapes from the nose but not the mouth; sounds in which air escapes from the mouth but over the side of the tongue [laterals]; sounds in which air escapes from the mouth but with friction in the mouth; and sounds during which the air stream has no escape.” Abercrombie (1967: 170) states that “Their [contoid and vocoid’s] rigorous definition is their great advantage; without it their usefulness is destroyed.” Vocoid (the lack of oral) articulation corresponds to the “natural” state of the vocal tract in communication and is the backdrop for segmentation. Contoidal articulation involves impedence and should be considered the marked, positively specified, value. On a systematic level there is no need for both contoid and vocoid since vocoid is the negation of contoid. The distinctive feature hypothesis can be strengthened by the elimination of the SPE feature [vocalic] and hereinafter +C indicates [+contoid] (c.f. SPE: 302).

3.1.4.1.2 The class of sounds known as approximants (Catford: 119-127) is useful and, while not considered by Pike in his definitions of [vocoid] and [contoid], can be identified using the features ad-vocated here. Approximants are ambivalent sounds; they may be vocoidal when voiced, but contoidal when voiceless because of the greater turbulence of air at the point of articulation caused by the increased velocity of the surd air stream (Catford: 120). Sounds such as [y], [w], [ɹ], [l], [i], and [u] are approximants. The common denominator of all of these sounds is the close approximation of the tongue to a point of articulation. The vocoids are [+high]. This close approximation creates redundant friction when the air is not subjected to laryngeal vibrations.

3.1.4.2.1 One of the uses of the feature [vocalic] has been the specification of semivowels y and w as [-cons] [ -voc]. Chomsky and Halle suggest that the feature [syllabic] might.be an apt substitute for [vocalic]; it (354) “... would characterize all segments constituting a syllable peak ... vowels would normally be syllable peaks, liquids, glides, nasal consonants could become syllabic under specific circumstances.” The latter set of segments constitutes the SPE class [sonorant] which has been eliminated from this inventory.

3.1.4.2.2 The syllable is a controversial linguistic unit. Chomsky and Halle’s elevation of the concept to the level of a phonological prime has enabled linguists to attribute a distinctive function to syllable division. For example, a contrast such as Portugues rio [xí.u]’river’vs. riu[xíw] ‘he laughed’ could be accounted for by considering the final segment of riu to be [-syllabic] whereas the final segment of rio could be characterized as [+syllabic]. Pulgram (1970: 65) does to believe the syllable to be a contrastive unit. He says: “The syllable is a unit ... whose phonological boundaries,... are determined by a general set of phonological Rules [sic] ... of a given language.” Being rule-produced it is non-distinctive. Catford (90) does not consider the contrastive power of the syllable, but defines it as “... a minimal ‘chunk’, or stretch of initiator activity, bordered by either minor, intra-foot, retardents, or by foot divisions themselves ...” Typical of initiator activity is a “power curve” which may be created by pulmonic, glottalic, and velaric air streams. Segments may be part of or may actually constitute a power curve.

3.1.4.2.3 The syllable is not really as difficult to define as Pulgram and his predecessors imagine, nor as primal as Chomsky and Halle seem to suggest. The traditional categories “consonant” and “vowel” have hindered rather than helped our understanding of the phenomenon known as syllabicity. Syllabicity is, essentially, the ability that a segment has to be articulated alone and be heard. Catford’s definition suggests that movement of air initiated in the vocal tract is essential to syllabicity. Air stream initiation may be pulmonic, glottalic, or velaric. Using the pulmonic air stream, any non-occlusive segment can be articulated alone and be heard, thus any non-occlusive segment may be syllabic. Those segments we commonly call vowels are simply non-contoidal, non-occlusive, non-lateral, and non-fricative. Many languages allow syllabic [n], [r], [z], [š], [ž], etc. Beyond this, the glottalic and velaric air streams are supraglottal and require pressure, either positive or negative, to make ejectives, implosives, and clicks—all of which are audible when articulated in isolation. This pressure is achieved by contoidal articulation, thus the “vowel” or the vocoid is not an essential ingredient of the syllable nor is it a phonological prime. For languages which only use the pulmonic air stream the essence of the syllable is expressed in the following rule:

Potential Syllable Boundary Insertion-Pulmonic Air Stream

The universal essence of syllables is:

Universal Potential Syllable Boundary Insertion

These rules state the minimal ingredients for syllables. Specific syllables in specific languages will have different and more complex configurations, but every syllable must have air stream movement, be it glottalic, velaric, or pulmonic.

3.1.4.2.4 I have claimed that all segments other than pulmonic occlusives can potentially constitute a syllable nucleus using the egressive air stream, that they may be articulated with no support and still be heard. Of course, the likelihood of their doing so in a particular language is inversely proportional to their degree of closure: a is the most likely candidate for a syllable nucleus, i less so, /less so, s idem, with virtually no likelihood that t or any other voiceless stop, which depend on support from a contiguous segment for their identity, can form a syllable peak. Certainly there are universal principles which determine the syllable shapes of strings of segments according to a sonority hierarchy which will be the converse of the strength hierarchy illustrated in Chapter VI of this work. In specific languages, rules may be written to account for the syllabation of strings of segments. Consider the following data.

Syllabation in Portuguese

The following rules account for the correct syllabation:

Rule 1.

This rule says A) no contoid shall have an independent air stream initiation when followed by a non-contoid, Ȼ → Ø/ [+contoid] _______[-contoid]; B) when a fricative (only /z/ in Portuguese) is preceded by /r/ or /n/ and followed by another contoid, there shall be no air stream initiation solely with either segment:

and C) when /f/ and /v/, the only slit fricatives in Portuguese, occur before /r/ or /1/, a syllable boundary is deleted:

Rule 1 eliminates the following potential boundaries:

Rule 2.

Rule 2 says A) delete a syllable boundary between a non-contoid followed by one or two contoids after which there is a syllable boundary; B)1) delete a syllable boundary either when a syllable initial non-stressed high vowel is followed by a non-high vowel that is syllable final or that precedes a syllable final contoid; or B)2) when a non-high vowel is followed by a non-stressed high vowel that has a syllable boundary or a contoid and a syllable boundary after it. Part B of Rule 2 is a mirror image rule that is right-to-left iterative. That is, it would first scan a world like saia /sa.i.a./ and eliminate the second syllable boundary /sa.ia/ which would make this word exempt from the mirror image of the rule that eliminates syllable boundaries between a non-stressed high vowel preceded by a non high vowel. The structural description is not met.⁹ Rule 2 finishes the syllabation of the Portuguese data:

These rules create syllables such as CV, CCV, XVCC and are probably fairly universal, since syllable structure in Portuguese is simple.

3.1.4.2.5 A problem may appear to exist concerning non-syllabic vocoids such as y or w, since the inventory of D.F.’s proposed here seems to have no way of distinguishing them from their syllabic counterparts, i or u. When their syllabic quality can be accounted for in a rule, as in the Portuguese data above, there is no problem: the sounds in question are distributional allophones of /i/and /u/. On the other hand, a language may have a contrast between syllabic and non syllabic vocoids: [yá] ~ [i.á] ~ [ža] or [wé] ~ [u.é] ~ [ǥé]. I suggest that such contrasts, if they exist, may be accounted for with the following specifications:

The feature [widened glottis] corresponds to fortis articulation in contoids and lenis articulation in vocoids. (see 3.2.3., 1-4). The use of [+widened glottis] to distinguish [y] from [i] claims that glide articulation will be the same as that of lax vocoids. I am also claiming that y and w, when phonemic, are more highly marked than i and u, and that no language will contrast yV with IV or wV with UV. That is, no language will contrast a glide with a lax vowel followed or preceded by a non-lax vowel. This solution comes from Catford (165-6) who says: “... both the ‘vowel’ [i] and the ‘semi-vowel’ [j] involve an identical type of articulatory structure but ... [i] always has a noticeable duration whereas [j] is ... a rapid glide away from (or to) an approximant type structure.”

3.2.0 In this section I examine SPE Manner Features.

3.2.1 The feature [continuant] has been replaced by [occlusion].

3.2.2.1 Chomsky and Halle associate the features [delayed release] 1° and 2° with the latter part of the articulation of a segment (318-319) yet nasals are initiated by a velic opening, itself a release, and are terminated by a velic closing. The term “delayed release” (primary and secondary) is misleading and inaccurate, since in the production of speech nothing is “released” in the stream of speech sounds which linguists analyse into segments. It is often objectively difficult to tell where each segment begins and ends; concatenations of articulatory gestures produce compromises in the vocal tract which cause contiguous sounds to round, to palatalize, velarize, etc. In addition, some sounds are so phonetically complex that the analyst must appeal to the dominant patterns in the language in order to decide whether they are units or sequences of units. Affricates, labialized, velarized, palatalized, rounded consonants, diphthongs, and nasal vowels often lend themselves to interpretation as sequences of segments rather than as units. When phonological analysis indicates that sounds such as these are units, a prime should indicate their unusual status through positive specification. The SPE features fail in this sector because, as Ladefoged points out (1971: 106): all continuants have a positive specification for [delayed release] in SPE.

3.2.2.2 Earlier (3.1.3.2 and 3.1.1.4) I suggested that the positive value which distinguishes complex units such as č, ǰ, k^x, mb, etc. from sequences is “transition” rather than release. This term will facilitate specification of units such as mb, bm, bmb as well as those which might be transcribed as ši (cf. Catford, 1977: 212). The term “transition” identifies the subphonemic affrication which occurs in many languages when alveolar stops occur before high front vowels. These sounds do not make a clean break from occlusive to vocoidal articulation ([ti], [di]) when affricated ([tši], [dži]) rather there is an intermediate stage [+contoid], [+high], [+friction] which is halfway between vocoidal and occlusive articulation.

3.2.2.3 Another benefit of the feature [delayed transition] is the elimination of the feature [long]. Since we consider segments which could possibly be analysed as sequences of segments as possessing [delayed transition], long segments (often analysed as gemminates) will receive a positive specification for [delayed transition]. Both primary and secondary [delayed transition] will be necessary to distinguish between long and non-long affricates in languages such as Amharic. A complex contoid is considered to have a delayed release feature when both of its parts are contoidal: b, p, is, bm, etc.. When contoids are labialized k^w, t’^w etc., palatalized p’ t’, k’, aspirated, or glottalized, p^?, t^?, c^? they shall not be considered as having a positive specification for [delayed transition] since the added feature is non-contoidal. This solution allows segments which are, it is necessary to repeat, theoretical constructs rather than absolute articulatory reality, to remain completely unitary, contrary to the conclusions of Campbell (1974) and Anderson (1976).

3.2.3.1 Lass (1976, 39-50) argues that the qualities [tense] and [lax] have been defined impressionistically by theoreticians trying to maintain the three vowel height hypothesis. He asserts that the motor correlates for distinctions such as /i ~ I, e ~ ę, u~U, o~ǫ/ in English are articulatory and can be described in terms of tongue position. Catford (1977: 199-208) has a similar opinion but suggests that there is (208) “... some justification for the retention of a parameter of tense/lax for the description of consonants, [but] For vowels such as parameter is dubious,...” He indicates (203) that the only “tenseness characteristic” is a difference in intra-oral pressure which is achieved by glottal adjustment. In the production of contoids, the close approximation of articulator to the point of articulation, plus the greater volume of egressive pulmonic air thanks to a widened glottis which allows more air to escape produces what we call articulatory tension. Tension, then, is a feature concommitant with the feature [surd] (see 3.4.1 below) since the vocal bands are open. On the other hand, non-contoidal articulation is naturally voiced, that is, the vocal bands are close together. The “lax” vowels are articulated with a wider opening than their “non-lax” counterparts (cf. Catford, 1977: 204). This widening produces an inevitable lowering of intra-oral pressure. At the same time the glottis is widened, the subglottal pressure and the frequency of laryngeal vibrations is lowered.¹⁰

3.2.3.2 Glottal widening produces opposite and similar effects on contoids and non-contoids. Contoids with a widened glottal opening will have greater supraglottal pressure and will be “tense” whereas vocoids with a widened glottis will be “lax” owing to reduced laryngeal vibration caused by the greater opening. In both cases, a widened glottis contributes to a tendency that both types of segments share, the tendency towards voicelessness, e.g. the lenition of vocoids in languages such as Portuguese and English is a necessary precondition to their being devoiced.

3.2.3.3 The features [tense] and [lax] may be accounted for by the feature [(relatively) widened glottis] when such distinction as /i ~ I, e ~ ę, p ~ p^h/ occur in languages. The feature replaces SPE [tense] and combines with the feature [glottal constriction] (SPE 315-316) which the authors use to account for the strong relatively unaspirated stops of Korean (Catford, 202) and for “creaky voice.” See also Kim (1967, 1970).

3.2.3.4 It seems to be contradictory to consider a sound as having both a widened glottis and a constricted one. The glottis should be thought of according to its possible states. In the production of speech it is either involved in phonation or it is not. In the non-phonatatory state it is either closed or it is so open that no vibration occurs. In the phonatory state it may be widened slightly or the folds may be relaxed in such a way that it vibrates more slowly allowing more air to pass. A strongly aspirated segment such as a p^h has maximal glottal opening. A b has the glottis in a phonatory state. A strong but unaspirated stop such as *p has the vocal bands wider than for b but not so wide as for p^h. For a b^h their width and/or tension is between that of b and *p.Sounds articulated with laryngeal creak have the vocal bands in a quasi phonatory state. “Creak” is handled by the combination [+widened glottis], [+surd], [+glottal constriction]. The strong relatively unaspirated stops of Korean contrast with the aspirated and unaspirated voiceless stops as follows: (see also 3.4.3)

3.2.4 The SPE feature [pressure] which the authors associate with velaric or glottalic air streams can be eliminated. Since we have assumed egressive air to be the unmarked state of the air stream, potential pressure is the natural state of phonetic production, since air is moved into the articulatory aparatus from an originator then moved outward, thus always creating potential pressure if the vocal tract is constricted. Pressure adds nothing to the distinctive capabilities of the inventory as we can see from the following specification:

In other words, common stops, ingressives, glottalics, egressive velarics, clicks, and glottalized ingressives can be distinguished with no need for the feature [pressure], and thus the inventory is strengthened by the elimination of yet another feature.

3.3.0 In this section I shall deal with SPE features which the authors claim involve secondary apertures.

3.3.1 [Nasal] has been dealt with (3.1.1.3.2.2-3.3.2) and is located in the section Major Class Features.

3.3.2 The appropriate place for [lateral]“...sounds... produced by lowering the mid section of the tongue at both sides or at only one side...” (SPE: 317) (cf. 3.4.1) is with the Manner Features. The Secondary Aperture class may be eliminated.

3.4.0 In this section I shall deal with the source features, [voice] and [strident].

3.4.1 Laryngeal vibration [voice] is a parameter common to all phonological systems. It is the backdrop of communication, essential for the production of suprasegmentals. In most phonological systems voiced segments are more frequent than voiceless ones,¹¹ indeed some phonological systems, e.g. those of Dyirbal, Gadsup, or Nunggabuyu, have no voiceless sounds as systematic phonemes, all phonemes may be contrasted without invoking [voice]. (See Appendix A). I suggest that the marked value in any phonological system is voicelessness i.e. [surd] which shall be positively specified.¹²

3.4.2 Chomsky and Halle define stridents as having (329):‘... greater noisiness than their nonstrident counterparts.’ They claim that the sounds /ø/ and /f/; // and /v/ in Ewe, as well as the sounds /ϴ/ and /s/ or // and /z/ of English are distinguished by stridency:

Harris (1969a) questions the status of [strident] as a feature and hints that the phenomenon may be a redundant consequence of points and manners of articulation. The SPE features do not unite fricatives into a positively specified natural class. The feature [sonorant] is necessary to distinguish them from sounds such as w, r, l . Ladefoged (1971: 105-108) concludes that SPE D.F.’s do not handle fricatives adequately, having already stated (49) that “. . . the best solution is to distinguish between grooved and slit fricatives.” Slit fricatives have (48) “. . . a wider articulatory channel and more doming (as opposed to hollowing) of the fore part of the tongue.” The relative width of articulatory area accounts for relative presence of channel turbulence and high velocity of the air stream (cf. Catford, 1977: 24-46, 121-127, and 153-159). Fricatives may be distinguished from all other segments by the feature [friction] and among themselves, by the feature [slit], these two features replace SPE [strident], [distributed], ([continuant] having already been replaced by [occlusion]), and [sonorant] having been deleted. These features, [friction] and [slit], are manner features rather than source and cavity features.

3.4.3 The authors of SPE propose the feature [heightened subglottal pressure] (326) to specify the voiced aspirated stops of Hindi which are produced “without tenseness.” They mention the possibility of setting up a “hyper feature” called “strength of articulation” alleging that “... certain facts such as the treatment of Spanish consonants in different contexts (see Harris, J., 1967) make this suggestion quite attractive.” The feature [widened glottis] that I have proposed for replacement of tense-lax along with [surd] and [glottal constriction] account for the stops of Korean (see 3.2.3) as well as those of Hindi;

This obviates any need for a feature as unevinced (Lisker and Abrahamson 1971) as [h sg. pr.] which Chomsky and Halle use to specify these segments. The features [surd], [widened glottis], and [glottal constriction] suffice in describing the contrastive elements of the occlusive consonants of Spanish, English, Korean, and Hindi.

3.5.0 In this section I shall deal with the cavity features [coronal, anterior, round, covered] and the tongue body features collapsing them into a series called articulation features.

3.5.1 The SPE features mentioned above are used for specifying points of articulation:

The distinctions between labials, apico-alveolars, palatals, velars, retroflexes, and pharnygeals can be made without [coronal] and [anterior] by substituting [labial] for [coronal].

Studies of phonological systems and of phonological theory concur that apico-alveolar consonants are the most frequent; hence in the approach I advocate they should be the least marked. Consequently, they receive the fewest positive specifications which I accomplish by eschewing [coronal] for [labial]. The features [labial] and [dorsal] apply to both contoids and vocoids; the former indicates point of articulation for fricatives and stops without delayed transition, it may also indicate rounding of contoids as well as the rounding of vocoids.

3.5.2 The contrast between the vowels of map and mop [mäp], [map] in many dialects of American English suggests that a feature like [anterior] be maintained. I prefer to use [palatal], to specify both vocoids and contoids; indeed, I maintain that, whenever possible, distinctive features should specify both vocoids and contoids. (See Wang, 1968 and Naro, 1973, pp. 124-134.) Front vowels are [+palatal]. Palatal contoids ñ, ṥ, Ĩ, etc. are also [+palatal] and may be distinguished from retroflex contoids by this feature. The use of [palatal] and [dorsal] provides a negative specification for central vowels [-palatal, -dorsal].

3.5.3.1 The SPE tongue body feature [low] was used to distinguish vocoids ([±high], [±low]) in a three level system of vowel height. Since we regard the vowel a to be the segmental primative, the unmarked value par excellence, and since this is traditionally regarded as a low vowel, we replace [low] with [raised] which indicates any movement of the tongue higher than the teeth.

3.5.3.2 The feature [raised] is useful in phonological descriptions. It is often necessary to distinguish alveolars and palatals from both velars and labials (see Pagliuca and Mowrey 1980). The specification [+raised, -dorsal] accomplishes this. In Baule, a language of the Ivory Coast, the phoneme /w/ will front to [ẅ] after an alveolar contoid and a front vowel: /tẅi/ ‘gun’[tẅi], /swi/ ‘elephant’[sẅi], /lwi/ ‘fat’[lẅi], /mucwe/ ‘eight’ [mucẅe] (Pagliuca and Mowry, 503). After labials and velars it does not change. Vago (1976) writes the following rule in SPE features to describe the process:

Pagliuca and Mowrey illustrate Vago’s rewriting of this rule using the JFH feature [grave], and then they use Wang’s feature [labial], identifying the class of alveolars and palatals as [-labial, -back], which they are. Their rule is (504):

They object to this rule stating (504-505) that it “... obscures the true motivation of w-fronting ... /w/ fronts to [ẅ] when surrounded by articulations which ... involve an active pulling forward of the tongue, which is the articulatory characteristic common to dental, alveolar, and palatal... segments.” Pagliuca and Mowrey do not recognize that a specification such as [-labial, -dorsal] identifies the common contoidal environment of the rule by negation of the other distinctive parameters in the Baule system. However, using the feature [raised] the fronting of /w/ after alveolars and palatals can also be regulated:

Also, in Baule /1/ is realized as [r] after alveolar and palatal contoids. While Pagliuca and Mowrey offer no rule in SPE features, an SPE rule would be:

Which can be written using the features advocated here as:

The other data that Pagliuca and Mowrey illustrate comes from Fe?Fe? where reduplication of syllables creates a high vowel before any alveolar or palatal consonant followed by a front vowel or when any consonant is followed by a high front vowel. The rule they offer (508) for this is, in SPE features:

Using the features advocated here, the same phenomenon can be described as follows:

This rule will front [ї] in the environment described for FE?Fe? If the class of segments that fronts vowels includes dentals (i.e. dentals, alveolars, palatals), the feature [raised] cannot be used. These articulations are identifed, and properly so, as [-labial, -dorsal, +contoid].

3.5.3.3 The feature [covered], (SPE 315) “... sounds produced with a pharynx in which the walls are narrowed and tensed and the larynx raised.” has been used to indicate pharyngealized vocoids and contoids and to distinguish pharyngeal fricatives from other segments. Jakobson and Waugh (1979, hereafter J & W) have claimed (116, et passim) that no language distinguishes labialized from pharyngealized contoids or vocoids because both labial constriction or pharyngeal constriction create flat spectra and are indistinguishable. Iraqui Arabic does contrast /f/, /x/, //, and /h/ but these can be distinguished without a feature such as [covered]:

This suggests that [covered] may be deleted from the D.F. inventory. Further evidence for the dispensibility of a special feature to indicate pharyngeal articulation comes from Egyptian Arabic which contrasts a set of plain alveolar stops and fricatives with pharyngealized ones /t, d, s, z/ ~ /t°, d°, s°, z°/ etc. The auditory impression that the pharyngealized segments give to learners unfamiliar with pharyngeal articulation is one of labiality (J&W, 116) and suggests that the alveolar contoids can be contrasted using this feature, [labial], and a rule of Egyptian Arabic will realize specifications such as [+occlusion, +raised, +labial] as pharyngeal in timbre rather than labial. The pharynx and the lips are opposite analogues of one another, just as the tongue tip and the uvula are. The latter make non-slit articulations for fricatives, flap articulations, and stops; the former may constrict the flow of air either as it enters or leaves the oral cavity. The pharynx is a set of internal, pre-oral lips. Using the features [contoid] and [labial] judiciously, contrasts can be made between labial, glottal, and pharyngeal fricatives, stops, and approximants, as well as between them and rounded and unrounded back and front vocoids:

Thus, [labial] indicates contraction of an extremity conduit, contoidal labiality indicates articulation at the lips, non-contoidal labiality may indicate rounding or pharyngealization of vocoids or contoids, depending on the language and the specific articulatory parameters.

3.6.1 I have finished my discussion of phonological primes without mentioning the segments referred to as ‘r-sounds’ (cf. Brakel, 1974). Because of the western European tradition of using the Roman Alphabet, we have come to associate a rather wide gamut of sounds with the grapheme ‘r’ They range from the vocoids [ɹ] and [ǝ] to flap r, trills, r͂ and R, to x, x̃, , , and h. The most unusual r is the Czech fricative trill // about which Ladefoged (1971: 49) says: “What characterizes the Czech variant of the trill manner of articulation is that it is a laminal (and not an apical) trill, and the stricture is held for longer [sic] (but probably with a shorter onset and offglide).” In the system advocated here r sounds can be specified as follows:

3.6.2 R-sounds represent a synthesis of contoid and non-contoid articulation. Retroflex ɹ is negatively specified for all parameters other than [raised] and [high], for the flap, r, [contoid] articulation is added; for the trill, delayed transition, for [] [friction]; R, [dorsal]; ə is only [raised]. The similarity of r, ɹ, and ə accounts for the allophonic variation that often occurs among these sounds within a single language. Since they are the ultimate synthesis of [contoid] and non-contoid, they are the last to appear, and, within the domain of [+contoid], the fact that r and l only differ by the feature [+lateral] helps explain why these sounds are in complementary distribution in many languages, and why they often develop into one another in many languages.

3.7 The D.F. inventory used in this approach, in order to facilitate the strong theory of marking is:

3.8 The strongest possible hypothesis about the sound systems of languages is that there is only one distinctive parameter. This hardly plausible hypothesis is easily reflected. The strength of a theory can be judged, however, as a function of this strongest hypothesis. The strength of a D.F. inventory as a linguistic hypothesis may be measured by the number of primes actually used divided into one. As the quotient increases, so does the strength of the hypothesis. The SPE inventory contained 25 features commensurate with those included here and, through the fraction ¹/₂₅ those features give it a strength quotient of .04. The system I have indicated here has only 19 features giving it a strength quotient ^l/ ₁₉ = .153. Dividing my quotient by the SPE quotient produces a quotient of .76, i.e., which means that my D.F. hypothesis is 24% stronger. In addition, I have made it more adequate as an instrument of observation and description and have brought it into line with a strong theory of marking. The comparison that I have just made between two hypothetical inventories suggests, as well, as means of comparing the complexity of phonological systems: one simply compares the number of features necessary to specify the systematic phonemes of both systems. Phonological complexity, or the markedness of a particular system can be measured by the quotient of the smallest number of D.F.’s needed to specify a phonological system, divided into the number needed for the system in question. Beyond this, the generative power of this theory, the inventory proposed here, 2¹⁹ or 524,288 separate distinctive segments, is considerably less, 1.6% of that of the SPE D.F.’s: 2²⁵ or 33,554,432 potentially distinctive segments. While the reduction of the number of phonological primes in the revised inventory makes it a stronger hypothesis and explanatorily more adequate, the descriptive and observational adequacy of these primes must be examined. To some extent this has been done in this chapter. Chapter V and Appendix A contain further examination of the descriptive and observational adequacy of the revised phonological primes.

Footnotes

¹One could unite these with a language-specific feature as Lass does for English (168-212), but a proliferation of ingredients weakens the hypothesis contained in a distinctive feature inventory.

²Harris (1969: 46) is very reluctant to accept this specification for the flap /r/ of Spanish.

³Syllable division in Portuguese is accomplished by a rule and [y] and [w] need not enter the segmental phonerpic inventory.

⁴Ladefoged’s examples are of prenasalized affricates. Catford (1977: 147) identifies an “endolabiodental nasalized fricative [ṽ]” for English in words like triumph and triumvirate.

⁵In these rules 1 use SPE features to show that they can be written in the framework Anderson used. Using the features advanced in this section, these phenomena may be expressed in the following rule:

⁶Catford (1977: 130) maintains that trills are not prolonged flaps, although later (196) he admits [length] as a phonological distinguishing factor between sounds such as Portuguese /r/ and /r͂/.

⁷See section 3.4.2.

⁸Anderson could have done his readers a great service by identifying languages in which mesonasalization occurs. In the 700 phonological systems described by Ruhlen, not a single case of mesonasalization appears.

⁹I owe this formulation to Ken Hill of the Linguistics Department at the University of Michigan.

¹⁰Catford (205) criticizes the assertion that “lax” vowels with a more open glottis have a lower subglottal pressure as an account for their shorter duration than that of “tense” vowels. He says: “The English lax vowel /l/, for instance, under certain conditions of intonation can be extremely long: for example “he did”‘? [hi dl: : :d], with a risingfalling tone on ‘did’, expressing astonished inquiry.” In my speech, this rising-fallingrising tone occurs with a laryngeal creak produced by a widened glottis. The prolongation of a “lax” vowel is appropriate for an utterance which expresses “astonished inquiry.” It is a phonatory-gesticulatory analog of an unexpected event. Catford’s point is, simply, that “lax” vowels can be lengthened.

¹¹Catford (1977: 107) supports this assertion although he indicates that voiceless segments outnumbered voiced in a phoneme count of Cantonese. He neglected to mention whether tonemes were included; certainly voiced segments and tonemes which entail laryngeal resonance would have outnumbered voiceless segments.

¹²English speakers may be reluctant to accept voicelessness as the designator of a marked feature because of the cumbrous nature of the word. They should note, however, that voicelessness is the marked form, it implies voice in other segments. I have adopted [surd] here because of its simplicity and convenience.

¹³This sound is a pharyngeal stop.

¹⁴This sound is a pharyngeal approximant.