Download Clement e Hume (1995) - The Internal Organization of Speech Sounds PDF

TitleClement e Hume (1995) - The Internal Organization of Speech Sounds
File Size807.2 KB
Total Pages47
Document Text Contents
Page 1

7. The Internal Organization of Speech Sounds7. The Internal Organization of Speech Sounds7. The Internal Organization of Speech Sounds7. The Internal Organization of Speech Sounds


0 Introduction0 Introduction0 Introduction0 Introduction

In recent years it has become widely accepted that the basic units of phonological representation are

not segments but features, the members of a small set of elementary categories which combine in

various ways to form the speech sounds of human languages. While features are normally construed

as psychological entities, they are defined in terms of specific patterns of acoustic and articulatory

realization which provide the crucial link between the cognitive representation of speech and its

physical manifestation.

The wide acceptance of feature theory results from the fact that it offers straightforward explanations

for many potentially unrelated observations. For example, since features are universal, feature theory

explains the fact that all languages draw on a similar, small set of speech properties in constructing

their phonological systems. Since features are typically binary or one-valued, it also explains the fact

that speech sounds are perceived and stored in memory in a predominantly categorial fashion.

Moreover, since phonological rules apply to feature representations, it accounts for the observation

that phonological rules typically involve “natural classes” of sounds, that is, classes that can be

uniquely defined in terms of a single conjunction of features. It also offers explanations for many

generalizations in the domains of language acquisition, language disorders, and historical change,

among others. Feature theory has emerged as one of the major results of linguistic science in this

century, and has provided strong confirmation for the view that languages do not vary without limit,

but reflect a single general pattern which is rooted in the physical and cognitive capacities of the

human species.

But while much research has been devoted to the questions, What are the features, and how are they

defined?, it is only recently that linguists have begun to address a third and equally important

question, How are features organized in phonological representations? Earlier theoreticians tended to

think of phonemes as unstructured sets of features, or “feature bundles” in Bloomfield's well-known

characterization. In accordance with this view, later work in the Jakobsonian and generative traditions

treated segments as feature columns with no internal structure. In this approach, phonological

sequences were typically characterized as two-dimensional feature matrices, as we illustrate below for

the word sun:


Theoretical Linguistics » Pholonogy




Sayfa 1 / 477. The Internal Organization of Speech Sounds : The Handbook of Phonological Theor...


Page 2

In this view, a phoneme (or phonemic unit) is simply a column of features. Since phonemes follow

each other in strict succession, such models can be regarded as linear.

The matrix formalism has strong arguments in its favor: it is conceptually simple, it is mathematically

tractable, and it imposes powerful constraints on the way features can be organized in

representations. In spite of its advantages, however, it has become apparent that this model (as well

as other models in which phonemes are viewed as strictly sequential feature bundles) has two

important inadequacies.

First, in such models all features defining a phoneme stand in a bijective (one-to-one) relation; thus,

each feature value characterizes just one phoneme, and each phoneme is characterized by just one

value from each category. It follows as a strict prediction that features cannot extend over domains

greater or lesser than a single phoneme. However, there is considerable evidence that this prediction

is incorrect. Simple and dramatic examples demonstrating “nonlinear” - i.e., nonbijective - relations

among features can be drawn from tone languages. For example, in some tone languages, two or

more tones may “crowd” onto a single syllable, forming contour tones (i.e., rising and falling tones). In

many tone languages, single tones “stretch” or extend over several syllables, and in some, tones

“float” in the sense that they are not associated with any particular tone-bearing unit in the

representation. Tones are also found to constitute independent “tone melodies” in abstraction from

the consonant and vowel sequences on which they are realized. (For discussion of these and other

properties, see, e.g., Pike 1948, Welmers 1962, Goldsmith 1976, and Pulleyblank 1986.)

It was earlier thought that nonlinear relations among features of this sort are restricted to a small set

of prosodic or suprasegmental speech properties, including tone, stress, and intonation. However, it

has been convincingly demonstrated that segmental properties, too, show comparable behavior, if on

a more limited scale. For example, in many languages the feature [nasal] may take up only part of a

segment, giving rise to pre- and post-nasalized stops such as [
d] and [d

]; and in some languages it

regularly spreads across more than one segment or syllable, establishing domains of nasal harmony

(see, e.g., Bendor-Samuel 1970; Lunt 1973; Anderson 1976). Similarly, in languages with vowel

harmony, features such as [back], [round] and [ATR] (advanced tongue root) have the ability to extend

across many syllables at a time (see, e.g., Welmers and Harris 1942; Carnochan 1970; Vago 1980).

Other segmental features also show nonlinear properties, as we shall see in the later discussion.

Problems such as these offered a direct challenge to linear theories of phonological representation,

and led to the development of alternative, nonlinear frameworks.
The earliest of these were the

theory of long components developed by Harris (1944) (see also Hockett 1942, 1947 for a similar

approach) and the theory of prosodic analysis developed by J. R. Firth and his collaborators after

World War II (see, e.g., Firth 1948, the Philological Society 1957, and Palmer 1970). A more recent and

still evolving approach is the theory of dependency phonology developed by J. Anderson, C. Ewen, and

their associates (for a general overview and fuller discussion, see Anderson and Ewen 1987 and also

chap. 17, this volume).

Perhaps the most influential of these frameworks at the present time - and the one we will be

primarily concerned with here - is an approach emanating from the theory of autosegmental

phonology developed in the 1970s and early 1980s. In autosegmental phonology, as first presented

by Goldsmith (1976, 1979a, 1979b), features that are observed to extend over domains greater or

lesser than the single segment are extracted from feature matrices and placed on separate “channels”

Sayfa 2 / 477. The Internal Organization of Speech Sounds : The Handbook of Phonological Theor...


Page 23

dependent features on the one hand, and the tongue root articulator and its dependent features [ATR]

(Advanced Tongue Root) and [RTR] (Retracted Tongue Root) on the other. In this model, the Tigre

assimilation rule can be expressed as ordinary spreading of the place node, and the oral node is not


While both of these conceptions are consistent with the guttural transparency phenomenon, they

make substantially different predictions in other respects. Given that laryngeal features are sufficient

to characterize the laryngeals [h ?], McCarthy's model does not straightforwardly predict that these

segments pattern with the gutturals, unless, following McCarthy, we allow them to bear the redundant

specification [pharyngeal]. Halle's model predicts that obstruents with distinctive laryngeal features

such as [+voiced] or [+spread glottis] can potentially pattern with the gutturals by virtue of their

guttural node. Perhaps the central difference, however, regards their claims concerning possible

spreading and delinking rules. McCarthy's model predicts that we should find rules spreading or

delinking [pharyngeal] together with the oral tract place features, while Halle's predicts rules that

spread or delink laryngeal and tongue root features as a unit. To date, no fully conclusive evidence

has been brought to bear on these predictions.

3.4 The Feature Organization of Vocoids3.4 The Feature Organization of Vocoids3.4 The Feature Organization of Vocoids3.4 The Feature Organization of Vocoids

We now consider the feature organization of vocoids, that is, vowels and glides. A long-standing

issue in phonological theory has been the extent to which consonants and vocoids are classified by

the same set of features. While most linguists agree that they share such features as [sonorant],

[nasal], and [voiced], at least at the level at which nondistinctive feature values are specified, there has

been much less agreement regarding the extent to which features of place of articulation and stricture

are shared. The articulator-based framework of feature representation, as described above, has made

it possible to offer a more integrated approach to this problem. In this section we first outline two

approaches inspired by this general framework, and then consider some of the differences between


3.4.1 An Articulator3.4.1 An Articulator3.4.1 An Articulator3.4.1 An Articulator----based Modelbased Modelbased Modelbased Model

In the earlier of these approaches, Sagey (1986) retains the SPE features [high], [low], [back], and

[round]. She integrates them within the articulator-based framework by treating them as articulator-

bound features, linked under the appropriate articulator node. Thus [back], [high], and [low], as

features executed by the tongue body, are linked under the dorsal node, and [round], as a feature

executed by the lips, is assigned to the labial node, as shown below:


In this model, all consonants and vocoids formed in the oral tract are characterized in terms of an

appropriate selection from the set of articulator nodes and their dependents, although coronal,

reserved for retroflex vowels, is usually nondistinctive in vocoids. One of the central predictions of

this model is that the set [back], [high], and [low], as features of the dorsal node, has a privileged

status among subsets of vowel features, in that it alone can function as a single phonological unit.

3.4.2 A Constriction3.4.2 A Constriction3.4.2 A Constriction3.4.2 A Constriction----based Modelbased Modelbased Modelbased Model

A second approach, emanating from work by Clements (1989a, 1991, 1993), Herzallah (1990), and

Hume (1992), proposes to unify the description of consonants and vocoids in a somewhat different

Sayfa 23 / 477. The Internal Organization of Speech Sounds : The Handbook of Phonological Th...


Page 24

way. This model is based on the preliminary observation that any segment produced in the oral tract

has a characteristic constriction, defined by two principal parameters, constriction degree and

constriction location. Since vocal tract constrictions determine the shape of the acoustic signal and

thus constribute directly to the way in which speech is preceived, they can be regarded as constituting

the effective goal of articulatory activity.

Given their centrality in speech communication, it would not be surprising to find that constrictions

play a direct role in phonological representation itself. This is the view adopted by the model under

discussion, which proposes to represent constrictions by a separate node of their own in the feature

hierarchy. The parameters of constriction degree and location are also represented as separate nodes,

which link under the constriction node. This type of organization was already proposed for

consonants above, in which the constriction itself is represented by the oral cavity node, constriction

degree by the [±continuant] node, and constriction location by the place node; this conception is

summarized in (40a). A parallel structure can be assigned to vocoids, as shown in (40b). In this figure,

the constriction of a vocoid is represented by its vocalic node, its constriction degree by an aperture

node, and its constriction location by a place node. As in the case of consonantal constrictions, these

nodes have no intrinsic content, and receive their intepetation by virtue of the feature values they

dominate. In these figures, place nodes of consonants and vocoids, which occur on different tiers, are

designated as “C-place” and “V-place,” respectively.


The aperture node dominates vowel height features, represented by the ellipsis, which are discussed

further in section 3.4.5 below.

A further innovation of this model is that the features [labial], [coronal], and [dorsal], occuring under

the V-place node in vocoids, are sufficient, by themselves, to distinguish place of articulation in

vowels, and replace the traditional features [back] and [round]. In order to fulfill this new and

expanded role in the theory, they must be redefined in terms of constrictions rather than articulator

movements as such. This can be done as follows, to a first approximation (compare the definitions

given earlier in (7)):

(41) Liabial: incolving a constriction formed by the lower lip

Coronal: involving a constriction formed by the front of the tongue

Dorsal: involving a constriction formed by the back of the tongue (= the dorsum, cf. Ladefoged

1982, p. 281)

These statements, valid for consonants and vocoids alike, define constriction location in terms of the

active articulator involved. Since all segments with oral tract constrictions are formed by the lips or

the tongue body, all are characterized by at least one of these three features. As far as vocoids are

concerned, rounded vocoids are [labial] by these definitions, front vocoids are [coronal], and back

vocoids are [dorsal]. Central vocoids satisfy none of the definitions in (41), and are thus treated as

Sayfa 24 / 477. The Internal Organization of Speech Sounds : The Handbook of Phonological Th...


Page 46

concatenative and fusional morphologies.

32 See Maddieson (1990) for discussion (and rejection) of a proposed exception to this generalization in

Shona. Note also that Sagey's definition allows for the possibility of segments with two major articulations.

She exploits this possibility in her analysis of the surface contrast between [p
] and [kp

] in Nupe, treating

the latter but not the former as having two major articulations, [dorsal] and [labial]. If this contrast can be

reanalyzed in other terms, as seems possible (for instance, we might suppose that [p
] does not have a

[dorsal] component), then crucial cases of double major articulations, in Sagey's sense, appear to be rare

and perhaps nonexistant, and a maximally constrained phonological theory would exclude them in

principle, by appropriate constraints on representations.

33 Ní Chiosáin's description implies that either rule may apply in such examples; the choice of rule is not

predictable from other phonological factors.

34 Instead, the second dorsal node must be delinked from the nasal, after triggering the loss of the coronal

node; but no general principle predicts this delinking.

35 We cannot assume that the oral cavity (or root) node spreads, since [continuant] does not spread, as

shown by our examples.

36 Like Sagey's model, the constriction-based model makes no formal claims regarding the phonetic degree

of stricture of a minor articulation. It thus allows for the possibility of languages, like those discussed

above, in which a minor articulation has the same degree of closure (or narrower) closure than a

simultaneous major articulation. See Hume (1992) for further discussion of this point. Further evidence for

the linking of minor articulations as a sister rather than daughter of the major articulation node can be cited

from opacity patterns in Chilcotin (Clements 1990b, 1993). See also Goodman (1991) for comparison with

the dependency-based model of Selkirk (1988), in which minor articulations are treated as daughters of

major articulations.

37 It is not inconsistent to link vowels under the C-place node, since this node has no phonetic content. We

may consider the C- and V-place nodes as in fact the same category of place, the terminological distinction

between them being merely conventional.

38 The spreading of single C-place features (major articulations) to nonadjacent consonants appears to be

restricted to [coronal], and in all known cases of [coronal] spreading, the target must also be [coronal]. We

speculate that a more general constraint is at work, restricting long-distance C-place spreading to cases in

which an OCP violation is involved. In effect, since spreading of [labial] or [dorsal] onto [dorsal] would be

vacuous, since these features do not usually have dependents, such a constraint would limit long-distance

spreading just to the observed cases. Such cases would then be motivated in a manner similar to rules of

long-distance dissimilation which, as was discussed in section 2.2, are also OCP-driven.

39 As David Odden points out to us, if [n] can be regarded as [-distributed], an alternative analysis is

possible in which only the coronal dependent feature [-anterior] spreads. For other, less controversial

examples of long-distance coronal node spreading, see Poser (1982), Hualde (1988b), and Shaw (1991).

40 A further prediction of this model is that a vowel's vocalic node may not spread across a consonant

bearing a minor articulation. This prediction is supported by the rule of vowel copy in Barra Isle Gaelic

(Clements 1986) in which the epenthetic vowel is realized as a full copy of the preceding vowel across a

palatalized or velarized consonant, except that the vowel is always front if the consonant is palatalized and

back if it is nonpalatalized. In addition, the epenthetic vowel is always unrounded, even though rounding is

distinctive. To account for these facts, we must assume that the vowel, but that it spreads the aperture node

of the vowel and the V-place node of the consonant separately. If the vocalic node were not linked to the C-

place node in vowels, we would expect the vocalic node of the vowel to be able to spread, since it would not

violate the No-Crossing Constraint, incorrectly resulting in complete vowel copy.

41 Other examples of the spreading of both values of vocalic place features have been cited in Gaelic

(Clements 1986) and Chilcotin (Clements 1993, p. 139), and can be treated in a similar way. Note that a

further prediction of this approach is that languages may have harmony rules spreading just the [dorsal] or

[coronal] node, instead of the lingual node. In such cases, it should be possible for [dorsal] to spread across

[coronal] vowels, and vice versa.

42 While phonetic lateral velars have been reported in a number of languages, there is no evidence that any

Sayfa 46 / 477. The Internal Organization of Speech Sounds : The Handbook of Phonological Th...


Page 47

Bibliographic DetailsBibliographic DetailsBibliographic DetailsBibliographic Details

TheTheTheThe Handbook of Phonological Theory Handbook of Phonological Theory Handbook of Phonological Theory Handbook of Phonological Theory

Edited by: Edited by: Edited by: Edited by: John A. Glodsmith

eISBN:eISBN:eISBN:eISBN: 9780631201267

Print publication date:Print publication date:Print publication date:Print publication date: 1996

of these sounds are both [dorsal] and [+lateral] at the phonological level; see Levin (1987) for careful

discussion of this issue.

43 The low front vowel /æ/ is backed to [a] by an independent rule after nonlabial consonants. Thus, the

diminutive suffix that surfaces as [a] in vnútf + a is the same suffix occuring in chláp + æ “man” (dim).

44 In this analysis, the major articulator feature [coronal] of the consonant links under the V-place node of

the vowel, creating the unmarked vowel structure. We assume this is the normal mode of operation. Given

our previous analysis of V-to-C place assimilation, however, it is natural to ask whether there are also two

types of C-to-V spreading: one in which the consonant's major articulator feature links under the vowel's V-

place node, as above, and another in which it links directly under the vowel's C-place node. These two

analyses make subtly different predictions, as discussed by Hume (1992); we leave the question open here.

45 This conclusion does not of course follow from the alternative proposed by Halle (1989, 1992), in which

[pharyngeal] is not an articulator feature but a class node, renamed “guttural.”

Cite this articleCite this articleCite this articleCite this article

CLEMENTS, G. N. and ELIZABETH V. HUME. "The Internal Organization of Speech Sounds." The Handbook of

Phonological Theory. Glodsmith, John A. Blackwell Publishing, 1996. Blackwell Reference Online. 31

December 2007 <


Sayfa 47 / 477. The Internal Organization of Speech Sounds : The Handbook of Phonological Th...


Similer Documents