How Powerful Are Elements ?

An Evaluation of the Adequacy of Element Theory in

Phonological Representations
Examination Number: 1563847
MSc English Language
The University of Edinburgh
2010


Table of contents

Abstract

1 Introduction ..

2 Theoretical background ....
2.1 Internal structure of segments .......
2.2 Segmental representation ..
2.2.1Basic elements I, U, A ...
2.3 Important aspects of element theory .
2.3.1 Privativity (Monovalency) ...
2.3.1.1 Privative vs. Equipollent ...
2.3.1.2 Asymmetry between feature values ..
2.3.1.3 Markedness and Underspecification .
2.3.1.4 Arguments against privativity ...
2.3.2 A small inventory of elements ..
2.3.3 Acoustic basis ...
2.4 Element-based theories .
2.4.1 Particle Phonology ...
2.4.2 Dependency Phonology ...
2.4.3 Government Phonology ...
2.5 Summary ...

3 Case studies ..
3.1 Case I: Vowel shift
3.1.1 The English Great Vowel Shift .
3.2 Case II: Vowel harmony
3.2.1 Luganda lowering harmony .
3.2.2 Pasiego Spanish raising harmony
3.3 Case III: Lenition ..
3.3.1 English /t/-lenition
3.4 Summary ...

4 Conclusion ...

References
1

2

3
3
7
7
10
10
10
11
12
14
15
17
20
20
24
28
32

33
33
34
42
43
48
54
56
63

65

68
1
Abstract

In this dissertation, I resume the discussion of privative features as a notational device
in segmental representation. I argue from both theoretical and empirical perspectives
that element theory is a better theory of phonological representation than the
binary-feature system. First, I argue that element theory is a more constrained and
thus preferable theory since, with the proposal of single-valued features and a small
element inventory, it is exempted from overgeneration of natural classes and
phonological processes. Next, in case studies, I weigh the element-based
representations of vowel shift, vowel harmony and consonantal lenition against the
feature-based representations. The result of the evaluation shows that, compared to
binary approaches, element theory is better in capturing the nature of various
phonological processes. It generally provides non-arbitrary representations that can
mirror how the processes occur, though it also has its limit in the characterization of
Pasiego height harmony and consonantal affrication as well as providing a
non-arbitrary account for occurrence of certain types of lenition in particular
environments.









2
1 Introduction

Phonological theories generally concern with two main issues the representations
and derivations of phonological processes. Throughout the history of phonological
theory, they have received different degrees of attention from phonologists. Ever since
the publication of Chomsky and Halles (1968) The Sound Pattern of English
(henceforth SPE), the focus of phonological theory has been centred on phonological
derivations, and the concept of binary features, which underlies the phonological
formulations proposed in SPE, has been regarded as the orthodox way of representing
segments. However, in the 1980s and early 1990s, some phonologists noticed the
limitations of binary features as a notational tool in phonological representation and
thus proposed several theories of representation based on single-valued features, as in
Dependency Phonology (e.g. Anderson & Jones 1974), Particle Phonology (e.g.
Schane 1984a) and Government Phonology (e.g. Kaye et al. 1985; Harris 1990).
Nevertheless, with the proposal of Optimal Theory in the mid 1990s, the focus of
phonology has returned to derivations again.
However, phonological representation is the basis of derivation and should have
received more attention, as McCarthy (1988:84) claims, the emphasis of phonology
should be placed on studying phonological representation rather than rules, because
if representations are right, then the rules will follow. As a result, this dissertation
aims to resume the discussion of segmental representation and re-evaluate the
single-valued feature system (i.e. element theory
1
) as a representational device. The
organization of this dissertation is as follow. In 2, I introduce the basic assumptions
of element theory and argue from a theoretical perspective that element theory is a

1
The term element theory is mainly associated with Government Phonology, yet I will follow Harris
and Lindsey (1995) to use the term in a more general way, as a cover term of any segmental
representation approaches based on privative features.
3
more constrained and thus preferred theory of representation. I will begin my
argument with the motivation for segmental representation (2.1) and then introduce
the three basic vocalic elements [I], [U] and [A] (2.2). Next, I will discuss in
details three important aspects of element theory and compare them with
corresponding aspects of the binary-feature system (2.3). Then, I introduce three
theories based on single-valued features ( 2.4), including Particle Phonology,
Dependency Phonology and Government Phonology. However, I do not intend to
offer a full description of these theories given the limited space. Instead, I will only
introduce the fundamental assumptions and some devices that will be relevant to the
case studies in 3. Three kinds of phonological processes will be discussed in details
in 3, including vowel shift, vowel harmony and consonantal lenition. I will look at
how different processes are accommodated within element theory and traditional
binary approaches. As the discussion goes, I wish to show that element theory is not
only preferred on a theoretical basis but also provides better representations for a wide
range of phonological processes.

2 Theoretical background

2.1 Internal structure of segments
The whole business of segmental representation is developed out of the idea that
segments are composed of smaller units just as words are composed of phonemes.
These smaller units serve to express contrasts between segments, and the evidence of
their existence can be found in the recurrent phonological processes throughout the
history of languages in the world. A common example is the place assimilation of
nasals, where a nasal assimilates itself to the following consonant with regard to place
of articulation, as can be seen in Table 1. I base the following discussion of this
4
phenomenon on Ewen and van der Hulst (2001).

Spanish



u[l] pato
u[n] topo
u[M] gato

a duck
a mole
a cat

(labial)
(alveolar)
(velar)
(Clements 2006)
Japanese


ho[m]pou
ho[n]tou
ho[M]kou
basic salary
really
main school
(labial)
(alveolar)
(velar)
Table 1 Place of assimilation of /n/ in Spanish and Japanese

Table 1 can be simplified in the form of a linear rule as follows:

(1)
[m] / ___ /p/
/n/ [n] / ___ /t/
[M] / ___ /k/

The above representation shows the inputs and outputs of the operations and the
environment where the changes occur. We can see that these assimilation processes
occur in the same environment cross-linguistically, that is, in front of /p, t, k/. The fact
that it is /p, t, k/ instead of, say, /p, t, l/ that recurrently trigger assimilations in
different languages indicates that the former set form a group that can be addressed by
phonological processes. A group of segments that participates recurrently in
phonological processes are identified by phonologist as a natural class, the members
of which have at least one property in common. In the present case, /p, t, k/ are all
voiceless stops, while /l/ does not belong to this category.
5
However, a representation like (1) is not very enlightening because it does not
show the correlation between the outputs and the environments. We cannot see, for
example, why /n/ would become [m] instead of other segments when followed by a
/p/. From the fact that the resulting pronunciations are still regarded by native
speakers of the languages as realizations of the same phoneme /n/, and that /n/ only
changes slightly and is not turned completely into the following segment, we can
deduce that merely part of the nasal and part of its following consonant participate in
the assimilation processes, which can only be achieved if segments are decomposable
into smaller units. We are then faced with a question: What is it, if not the segment as
a whole, that actually participates in the operation and is shared by the resulting nasal
and following consonant?
In order to gain insight to the components that are active in the processes, we
need to re-express the segments in terms of their internal structure, to describe the
processes in a more revealing way. Thus, the phenomenon in Table 1 can be
formulated as follows:

(2)
[labial] / ___ [labial]
[nasal] [alveolar] / ___ [alveolar]
[velar] / ___ [velar]
(Ewen & van der Hulst 2001:7)

In (2), the correlation between the post-change nasals and the environments as well as
the components that take part in the process are clearly shown. The nasal is labialized
under the influence of a following labial consonant, same with the other two processes.
However, a linear representation as (2) gives rise to a theoretical problem. That is, it is
equally easy to form some crazy rules (Bach & Harms 1972) that could not possibly
6
happen in any language simply by substituting the formulations with other
components, as in (3).

(3)
[labial] / ___ [alveolar]
[nasal] [alveolar] / ___ [velar]
[velar] / ___ [labial]
(Ewen & van der Hulst 2001:8)

While rules expressed in formulation (2) are natural in that they are recurrent
phenomena cross-linguistically, processes in (3) lack such universality (Schane
1972). However, both (2) and (3) are equally easy to formulate in spite of the
difference in the degree of naturalness, which implies that such type of representation
is unable to distinguish between recurrent and non-recurrent phonological processes,
a fundamental theoretical inadequacy of linear representation recognized by
Chomsky and Halle (1968:400) themselves. In other words, though such type of
representation seems to have achieved some degree of descriptive adequacy, it
suffers from the problem of overgenerating phonological processes (Rennison
1984:281). That is, many processes that are expressible in this framework, such as
those in (3), are not even attested in the history of languages. Yet a powerful
phonological theory should not only reach the level of descriptive adequacy but more
importantly that of explanatory adequacy (SPE:335). It should be able to distinguish
recurrent processes from non-recurrent ones by allowing the former to be formulated
in a simpler way than the latter in its representation and therefore make prediction
about possible phonological processes. The above representation, however, lacks such
capability.

7
2.2 Segmental representation
From the previous section we can derive several basic requirements for an adequate
phonological representation theory: (i) A set of segments that recurrently participate in
phonological processes should be definable as a natural class in terms of the
components shared by the segments (Lass 1984). (ii) The system should allow
recurrent and natural phonological processes to be expressed in an easier way than
non-recurrent and unnatural ones (SPE; Anderson & Ewen 1987). In this section, I
will introduce element theory, a representation-based theory that is different from the
traditional linear approach based on binary features, and in the next section I will
discuss how well the above two requirements are embodied in this theory.

2.2.1 Basic elements I, U, A
Phonological theories based on single-valued features generally establish their
proposals on three basic elements: [I], [U] and [A]. Within this framework are Particle
Phonology (Schane 1984a), Dependency Phonology (Anderson & Jones 1974;
Anderson & Ewen 1987; van der Hulst 1989) and Government Phonology (Kaye et al.
1985; 1990; Harris 1994a; Harris & Lindsey 1995). Although there are certainly
differences among these theories with respect to how the three basic elements are
utilized to characterize various phonological processes, which I will elaborated in
2.4, some fundamental assumptions of these theories are the same.
We have seen in 2.1 that segments are composed of smaller units, but what are
these units? It turns out that different theories employ different sets of features,
especially for the representation of consonants, yet the proposals of vocalic features
can be generally divided into two types. SPE-type approaches define vowels in terms
of two bidirectional contrasts, high-low and front-back, which together define a
rectangular vowel space, while in element theory, the three elements [I], [U], [A] form
8
a tridirectional relationship, defining a triangular vowel space with three
cornershigh front, high back and low respectively (Rennison 1984), as shown in
(4).

(4)
/i/ /u/ /i/ /u/
front back high front high back
high


low
low
/a/ /@/ /a/
(a) Vowel space in a binary feature system (b) Vowel space in element theory
(Giegerich 1992:15)

The two systems differ not only in the shape of vowel space they define and the
parameters they utilize but also the phonetic interpretability of the features. In the
binary feature system, the definition of any given vowel requires the specification of
both height and frontness/backness; that is, a feature on its own cannot define any
segment. Yet in element theory, each element in isolation is phonetically interpretable.
The three corners of the triangle are defined by the three basic elements [I], [U], [A],
manifesting themselves as /i/, /u/, /a/ respectively.
Elements can also combine with each other to form compound segments (Kaye
et al. 1985) to represent other vowels, which have more complex internal structures
than the vowels represented by the three basic elements but are equally phonetic
interpretable. Thus, exploiting the combinatorial possibilities, we can derive a list of
possible vowels as follows:

9
(5)




(Anderson & Ewen 1987; Harris 1994a)

The combinatory possibilities can be further extended in the mechanism of
dependency relations between elements, as proposed in Dependency Phonology and
Government Phonology. In this mechanism, elements can enter into a relationship
where one element is more prominent than the other. The more prominent element
(head) contributes more of its property while the less prominent (dependent) one
contributes less. For example, /e/ and /z. in a vowel system that contains both of
them would be represented as [I, A] and [I, A] respectively, where the underlined
elements are the head. Notice that the representation of /e/ here is different from that
in (5), a system that lacks /D/ and /z., because phonetic interpretation of elements in
element theory is system-dependent (van der Hulst 1992:123); that is, it is adjustable
to the phoneme system of a language.
There are several pieces of evidence in support of the proposal of [I], [U], [A] as
the basic elements. First, the primary vowels /i/, /u/ and /a/ each occupies a corner of
the vowel triangle of human oral cavity and thus are maximally distinct with regard to
articulation (Dikken & van der Hulst 1988). Hence, all other vowels should be
definable as they fall within the triangle.
Second, the three vowels are recognized by Stevens (1972; 1989) as the quantal
vowels (Anderson & Durand 1986; Dikken & van der Hulst 1988). According to his
observation, these vowels are relatively stable because they can be realized through a
wide range of articulatory configurations while causing little change in acoustic
/i/ [I] /e/ [I, A] /1. [I, U, A]
/u/ [U] /o/ [U, A]
/a/ [A] /y/ [I, U]
10
effects. Other vowels, by contrast, require more precise configurations. This
corresponds to the proposal that representations of the primary vowels require less
specification.
Third, phonological processes addressing the three basic elements individually
are well-attested across languages (Harris & Lindsey 1995), several of which are
mentioned in Jones (1989). For example, in the Anglian dialects of Old English, short
front vowels [z], [e] and [i] in a stressed syllable acquire backness/labiality when
followed by labial sonorants, which can be represented as the spreading of [U], while
in West Saxon Old English, it is the [I] that spreads from /i/ and /e/ to the vowel in a
preceding syllable, making [u] [y] and [o] to [8]. Harmony processes involving
element [A] can be found in Kinande (Clements 1991), a Bantu language, where
vowels raise one degree before the root vowels /i, u/, which can be characterized as
the loss of [A].

2.3 Important aspects of element theory
In this section, I discuss several important assumptions underlying element theory,
including the use of single-valued element, the preference for a small feature
inventory and the acoustic basis of features. I will pay attention to arguments in
support of these assumptions while also making reference to some against them.

2.3.1 Privativity (Monovalency)

2.3.1.1 Privative vs. Equipollent
The notion of privativity in phonology can be traced back to Trubetzkoy (1969:75),
where he defined a privative opposition as oppositions in which one member is
characterized by the presence, the other by the absence, of a mark, while in an
11
equipollent oppositions, both members are logically equivalent; that is, both
members are existent. These two types of relation represent a fundamental difference
between the element-based approach and the traditional binary-feature approach to
segmental representation. In a binary-feature system, each feature is specified with
either value + or -, both of which are accessible to phonological processes, while
within element theory, all elements are single-valued, and thus only the presence of an
element is addressable by phonological processes. Take the feature [nasal] for
example, in binary framework, it is assumed that there are processes addressing a
natural class defined by [+nasal] as well as processes addressing a class defined by
[-nasal], yet in the privative framework, only the natural class defined by the presence
of [nasal] (i.e. [+nasal] in binary terms) are allowed in the formalism and thus there
are no processes addressing a class defined by the absence of [nasal]. As a
consequence, a binary-feature approach essentially defines more natural classes and
allows more possible processes (Harris 1994a). The choice between the binary and the
privative approach then depends on: (i) whether both natural classes defined by the
two feature values have equal chance to participate in phonological processes, and (ii)
whether both values of a feature are equally active in phonological processes.

2.3.1.2 Asymmetry between feature values
Whether both values of all features are equally accessible to phonological
generalisation has long been an issue under debate. Dependency Phonology (e.g.
Anderson & Ewen 1987) claims that all features are privative, while SPE, by contrast,
follows a strict binary-feature approach. To answer the first question raised in the
above paragraph, we pick up the [nasal] example again. Though the natural class
defined by [+nasal] participate recurrently in phonological processes
cross-linguistically, processes addressing a class defined by [-nasal] are not attested
12
(Ewen & van der Hulst 2001). Thus, it has been widely accepted that the binary
approach suffers from overgeneration of natural classes.
As for question (ii) raised above, it is found in many studies that the two
feature-values are not equally active in phonological processes. For example, in the
case of vowel harmony in Khalkha Mongolian, Steriade (1979) recognizes an
asymmetry between the distributions of the two values of [round]. There are two
harmony processes in this language, Backness Harmony and Rounding Harmony. The
Backness Harmony requires that all vowels agree in backness, yet only a front vowel
[i] which appears in a non-initial position of a stem that otherwise contains all back
vowels can be exempt from the harmony. The Rounding Harmony states that a
non-high vowel will be rounded immediately following a non-high round vowel while
neglecting any intervening [i] (Goldsmith 1985:256). According to Steriades (1987;
also reported in Archangeli 1988) analysis, only [+round] is active in the process,
since it is [+round] that spreads in the harmony and the condition under which a
non-high vowel can escape being rounded is either following word-initial [+high]
vowels or immediately after a [+high, +round] vowel.

2.3.1.3 Markedness and Underspecification
The asymmetrical status between the two feature-values, as in the case of [round]
above, is treated within the framework of underspecification (e.g. Kiparsky 1985;
PulleyBlank 1988), where the predictable and thus unmarked feature-value is omitted
in the underlying representation, and is only specified during the derivation of
phonetic implementation by a default rule, which fills in the unmarked value of a
feature. However, though the binary-feature formalism is preserved in such
framework, the fact that only one value is preferred by phonological processes seems
to imply a privative opposition (Dikken & van der Hulst 1988:35). A question then
13
arises: whether underlying unspecified values are ever specified?(Archangeli 1988:
189, italics original) Not much work on underspecification has considered seriously
the necessity to fill in the unmarked values, yet if evidence of the necessity cannot be
provided, any argument for underspecification would turn out be one in favour of the
privative approach (Lombardi 1996).
Taking into consideration the cross-linguistic phenomenon of redundant
feature-values (Steriade 1987), the privative approach simply lets the unmarked value
be characterized by the absence of an element, which renders default rules
superfluous (Kaye et al. 1990). Thus, in the Khalkha Mongolian case, there will no
longer be a rule which fills in [-round] as the default value during the derivation, since
all resulting vowels of the harmony will automatically be unround except for those
specified as [round]. As a result, van der Hulst (van der Hulst 1989:199) regards the
privative approach as the logical end point of Radical Underspecification.
Nevertheless, it is not merely an alternative to underspecification but a solution to the
problem of overgenerating phonological processes mentioned at the end of 2.1. From
the fact that only the marked value of a given feature is recurrently addressed by
phonological processes, we can derive that those addressing the unmarked value are
non-recurrent, and thus the reason why binary approaches suffer from overgeneration
of phonological processes is that processes addressing an unmarked value are equally
expressible in the representation as those addressing the marked value. Yet in the
element framework, non-recurrent processes simply cannot be formulated since the
representation cannot refer to the absence of an element, which makes element theory
more constrained and thus more highly valued than binary approaches in that
non-recurrent processes are systematically excluded from the representation as
inexpressible and thus impossible phonological processes. As a result, what SPE
refers to as the intrinsic content of features (1968:400), namely the relative
14
markedness of feature-values, is directly built into the use of single-valued features
(Harris 1994a). In other words, the marked value is characterized by the presence of a
feature whereas the unmarked value by the absence of it. Therefore, no extra
markedness conventions are in need within the element framework.

2.3.1.4 Arguments against privativity
However, although many features have been shown to act in a privative way, such as
[round] mentioned above, [low] (Goldsmith 1985), [back] (Steriade 1987), [nasal]
(Steriade 1995) etc., arguments in support of the binary-feature system is also worth
considering. For example, though it has been found in some languages that only
[+voice] but not [-voice] can spread to other segments, as in Japanese (It & Mester
1988), Ukrainian (Danyenko & Vakulenko 1995; cited in Wetzels & Mascar 2001),
which leads to the conclusion that voicing is a privative feature, Wetzels and Mascaro
(2001) reject this proposal with evidence from other languages where only [-voice]
spreads in an assimilation process. They argue that in Yorkshire English and Parisian
French, voiced obstruents become voiceless when followed by a voiceless obstruent
but not vice versa, which can only be explained by the spreading of [-voice] because
there is no rule of final devoicing in these two languages. Evidence from other
languages is also mentioned in the same article to support the need for the
specification of [-voice].
It is not the intention of this dissertation to examine all the arguments for and
against privative features, but suppose the evidence proposed in favour of binarity is
valid; that is, the processes cannot be reanalyzed in terms of privative opposition, it
may be the case that some features are privative while some are binary, or even that
the unmarked value of a feature is language-specific, as concluded in Steriade (1987).
However, it is worth noticing that when comparing privative approaches with binary
15
approaches, we should make sure that the evaluation is made on a level playing field,
since the sets of features proposed by different representation-based theories may not
be identical (Harris & Lindsey 1995). Evidence in support of the binarity of a feature
only used in binary approaches does not entail abandoning the privative approach. If a
theory can show that a wide range of phonological processes can be explained in
terms of the privative features it employs, it would be naturally considered a preferred
approach due to its more restrictive generative capacity (Harris & Lindsey 1995).

2.3.2 A small inventory of elements
Compared to the binary approaches, element theory is a more constrained theory not
only in that it utilizes single-valued features but it also has a smaller feature inventory.
It is obvious that, other things being equal, a smaller number of features would reduce
the combinatory possibilities and thus generate fewer possible segments. In other
words, the more features we propose, the more likely we may encounter combinations
that generate universally impossible sounds due to natural constraints on human
speech organs. An example commonly referred to is the universally incompatible
combination of [+high, +low] in binary approaches. More often is the case where
feature combinations formulate possible but unattested sounds on a language-specific
basis, such as voiceless nasals in English, which would require some arbitrary
statements to exclude unattested sounds from the inventory of the language (van der
Hulst 1989). The excessive power of a large feature inventory provides a reason to
reject the proposal of what Clements (2009:19) terms as a direct-access theory of
inventory structure (e.g. Browman & Goldstein 1989; Ladefoged & Maddieson 1996),
which intends to provide finer-grained categories of speech sounds based on the
articulatory gestures and results in a large feature inventory.
Contrary to the proposal of phonetics-centred features is the idea of distinctive
16
features proposed by Jakobson et al. (1952, henceforth PSA), where a feature can
only be proposed if it can specify a phonological contrast, i.e. if there is a pair of
phonemes that differ only in the specification of the feature. For example, it is argued
in PSA that, since no languages contrast between labialized, velarized and
pharygealized consonants but only between either one of them with a plain consonant,
there should be only one feature covering all three types of consonants, and thus
Jakobson et al. propose a feature [flat] as the cover feature (reported in Hyman 1975).
In a sense, the proposal of distinctive features implies a parsimonious view of feature
inventory. However, Lass (1984:97) argues that such an reductionist strategy may
abstract phonological representations too far way from phonetics and then require
complementary descriptions to specify which of the three articulatory gestures are
actually involved in the processes. Therefore, though a small feature inventory is
generally preferable as it is exempt from overgeneration of possible sounds, phonetics
should also be taken into consideration in the proposal of feature set.
A similar concept to parsimony is feature economy (Clements 2003a; 2003b;
2009), which proposes that languages maximize the use of features. Therefore, it
predicts that phoneme systems tend to have more than one segment characterized by
the same feature. Clements (2003b) supports this hypothesis with his research on the
phoneme systems of the worlds languages, which shows that feature economy is
indeed a cross-linguistic principle of the organization of feature inventories. Thus he
proposes an economy index to measure the degree of feature economy, as
formulated in (6), where E stands for the degree of economy, S for the number of
sounds, and F for the number of features.

(6) E = S/F

17
According to the formulation, there are essentially two ways to achieve a greater
economy: either by adding new sounds to the phoneme system (i.e. increasing S) or
reducing the size of feature inventory (i.e. decreasing F). Though Clements rejects the
equation of feature economy to parsimony because the latter is against an increase in
the number of sounds, at least both prefer a smaller feature inventory.
As a result, from the perspectives of distinctive features and feature economy,
element theory seems a preferred feature theory over binary feature systems, for it
utilizes a smaller number of features to represent the same amount phonological
contrasts. This result can be attributed to two innovations of element theory. First,
elements are allowed to represent more than one property. For example, element [I]
specifies frontness and palatality, and [U] represents roundness and labiality. Each of
these properties requires a separate feature in the traditional binary system. The
characteristics of multiple identities not only reduces the number of elements needed
but also enables processes such as palatalization of consonants adjacent to a high front
vowel to be represented in a more revealing way, which the SPE system fails to
achieve, as pointed out by Campbell (1974). Another innovation leading to a smaller
inventory is the proposal of dependency relationship (Van der Hulst 2006), which
exploits the combinatory possibilities of a limited number of elements, as has been
mentioned in 2.2.1. Together with the assumption that each element alone can
specify a segment, which would require more than one feature in binary approaches,
element theory uses its features in a more economical way.

2.3.3 Acoustic basis
Unlike the articulation-centred SPE feature system, element theory can be regarded in
a sense as a revival of Jakobsonian acoustic features (Harris & Lindsey 1995).
However, element theory does not abandon the articulatory dimension but adopts a
18
neutral position in that elements are defined in terms of both articulatory gestures and
acoustic properties, as in (7). Acuteness is defined by the predominance of the upper
side of the sound spectrum and gravity by the predominance of the lower side of the
spectrum. Namely, an acute segment has an F
2
closer to F
1
than F
3
, while in an grave
segment, F
2
is closer to F
3
. Compactness, on the other hand, is defined by a high
phonetic power, which is often associated with sonority (Anderson & Ewen 1987).

(7)



(Anderson & Ewen 1987:28)

It has now been recognized in many discussions that both articulatory and
acoustic dimensions need to be taken into consideration in the proposal of features.
On one hand, Lass (1984) notices that labials and velars have often been addressed as
a group by phonological events cross-linguistically, such as in Old English, Uralic
languages and Middle Korean. The SPE system fails to present these two types of
sound as a natural class, whereas PSA successfully characterizes them with a single
feature [+grave]. On the other hand, Halle (2009:71), one of the authors of PSA,
recognizes the difficulty of representing the palatalization of consonants before front
vowels in Slavic languages in a revealing way with Jakobsonian features, as
formulated below:

(8)
[-palatalized] [+palatalized] / _____ [-grave]
I frontness (or acuteness and sharpness)
U roundness (or gravity and flatness)
A lowness (or compactness)
19
Halle further points out that (8) fails to reflect the fact that such process is the
spreading of frontness from the following vowels to the consonants, which can be
expressed simply with the feature [back] in the SPE system.
The reason for the failure of articulation-based features to represent some natural
classes and phonological processes in a natural way is that, though some processes
result from ease of articulation, there are also some that originate from the perception
side of speech transmission. Evidence from acoustic phonology has shown that
articulation and acoustics do not always form a one-to-one relationship. For example,
Stevens (1972) found that there are some conditions where acoustic signals are
relatively insensitive to change in articulation, and there are also conditions where a
slight adjustment of articulation gives rise to significant change in acoustic effects. It
is also mentioned in many publications (e.g. Lindau 1975; Halle 1983; Lass 1984)
that there are cases where several articulatory gestures can be utilized to achieve the
same acoustic effects.
Ohala (1981:178) claims that such inherent ambiguity of the speech signal is a
source of phonological change, such as the different realizations of the English with
which range from [vHS] to [vHe] in various dialects, since [S] and [f] are similar
acoustically. This kind of process would be better characterized by features based on
acoustics than on articulation, since it results from the listeners misperception of the
acoustic signals. And as Ohala (1996) also argues, speech perception is a process
where a listener perceives sounds in the form of acoustic signals instead of
articulatory gestures as proposed in the motor theory (Liberman & Mattingly 1985).
This statement can be further supported by psycholinguistic evidence
(Dehaene-Lambertz & Pena 2001) which shows that 2-6-day neonates, though not
capable of replicating what they hear, can discriminate sounds of different categories.
As a result, since acoustic signals are the medium of speech transmission,
20
element theory integrates the notion of Jakobsonian acoustic features into the
definitions of elements. In so doing, it should be able to formulate more natural
classes and phonological processes in a more revealing way than any feature theory
with a strong bias towards either articulation or acoustics.

2.4 Element-based theories
In this section, I introduce three element-based theories Particle Phonology (PP),
Dependency Phonology (DP) and Government Phonology (GP).They generally share
the aspects discussed in the previous section, yet there are still differences in their
element inventories and the formal apparatus they utilize. Thus, in order to support
the case studies in the following chapter, I will look at some fundamental assumptions
of each theory and introduce the descriptive devices relevant to the phenomena to be
discussed in the next chapter.

2.4.1 Particle Phonology
Particle phonology is mainly discussed in the work of Schane (1984a; 1984b; 1995;
2005). Thus, in the following, I will provide a summary of his proposal. As Schane
claims, PP is a theory motivated by the failure of traditional binary systems to
characterize the internal structure of vowels and the relationship between vowels and
diphthongs. As a result, it is a theory proposed to deal with the representation of
vowels.
Schane divides the three elements (particles in Schanes term) into two
categories on the basis of formant frequencies: i and u are tonality particles and a is
the aperture particle. In terms of articulation, i stands for palatality, u for labiality, and
a for openness. Schane illustrates the definitions of the three elementary particles as
follows:
21
(9)
tonality (F2)
palatality |i| |u| labiality
(elevated F2) (lowered F2)


|a|
aperture (elevated F1)
SONORITY
(Schane 2005:338)

As in DP and GP, the three particles in PP individually manifest themselves as
the three primary vowels, but PP is different from the other two theories in the
representation of complex segments. The representation of vowel height has been
pointed out as a problem of traditional binary system, which with two features [high]
and [low] can only derive three vowel heights. Hence, a mid-low vowel in a system
with four vowel heights would be classified arbitrarily as either [-high, -low] or [-high,
+low] (Durand 1990:291). Some linguists (Ladefoged 1971; Sommerstein 1977;
Lindau 1978) propose a multi-valued feature to represent the scalar property of vowel
heights, marking the feature [high] with natural numbers to characterize various
heights. Thus, the lowest vowel /a/ would be specified as [1 high], which is lower than
the second lowest vowel, specified as [2 high], while the total number of heights
needed depends on the system to be described. However, such proposal would
undermine the radical position of binarism. PP, on the other hand, allows multiple
occurrence of the aperture particle in the description of different vowel heights. In this
way, PP is able to accommodate a scalar property within a privative system. Therefore,
a vowel system with four front unrounded vowels is represented as in (10).


22
(10) Short vowels

/i/ i /u/ u
/e/ ai /o/ au
/2/ aai /N/ aau
/z/ aaai /a/ aaa

To represent long vowels and diphthongs, Schane (1984a) proposes two punctuators:
space and half-moon. The extra length of long vowels is characterized by the
addition of an extra tonality particle, following a space, to the end of corresponding
short vowels, but for non-high central vowels, where no tonality particles are involved,
an extra aperture is added instead. Thus, long vowels corresponding to (10) are
represented below:

(11) Long vowels

/i9/ i i /u9/ u u
/e9/ ai i /o9/ au u
/29/ aai i /N9/ aau u
/z9/ aaai i /a9/ aaa a

Representation of diphthongs contains specifications of the two parts separated by a
space, with a half-moon marked under the glide. The specification at the two sides
of the space each occupies a skeletal position. Some examples are shown in (12).

(12) Diphthongs




Schane argues that, compared to traditional binary system, the notational devices
/ii/ i i /u/ u
/ei/ ai i /o/ au
/ai/ a i /a/ a
23
of PP can better characterize processes involving alternations between monothongs
and diphthongs. In Schane (1984a), he compares the representations of two
phonological operations fusion and fission under PP and traditional binary system.
Fusion, as defined by Schane, is the monothongization of a diphthong, whereas
fission is the opposite. For a fusion process such as /ai/ /e9/, PP and traditional
binary approach offer the following representations respectively.

(13)
(a) (b)
V V V V V V
+low +high 1
a i a +back - back - low 1
i - round - round - back
1 2
(Schane 1995:587) (Schane 1984a:140)

Schane argues that formulation (13b) represents the process as a conversion of
the first half of the diphthong into /e/ plus the loss of the second half instead of a
fusion of the two segments of the diphthong as in (13a). He continues to point out that
there is no correlation between the input and output of the process in (13b). We cannot
see, for example, why [+back] becomes [-back] or why [+low] becomes [-low]. The
specification of feature values seems to be arbitrary in this sense. By contrast, in (13a),
the output is clearly shown as an amalgam of the particles present in the input, and
vowels containing particles other than a and i would be excluded from possible output,
which presents PP as a more restrictive theory than the binary system in the
generation of possible outputs.
Furthermore, Schane (1984a) proposes three laws to constrain representations
under PP. First, the law of mora conservation requires that the number of skeletal
24
position be preserved during fusion and fission. Thus, a diphthong occupying two
morae could not be fused into a short monothong. Second, the law of diphthongal
differentiation demands that the two halves of a diphthong not be identical. Hence the
two parts should differ at least in height or tonality. Third, the law of maximum
aperture requires that the representation of /a/ should not have fewer aperture
particles than the lowest tonality vowels (Schane 1984a:139). Therefore, in a vowel
system as the one in (10), since the lowest tonality vowel /z/ contains three aperture
particles, the vowel /a/ should be specified with three aperture particles as well.

2.4.2 Dependency Phonology
As indicated by its name, the most salient innovation of DP is the use of dependency
relationships among elements, or components in DPs terminology. I have already
mentioned that in DP and GP, elements in a combination can enter into various
head-dependent relations. With regard to the notation, DP uses a semi-colon or an
arrow to represent dependency relationships. Dominating components are placed on
the left side of a semi-colon or the end of an arrow, while a mutually dependent
relationship, where none of the components are more preponderant than the others, is
represented with a colon or double-headed arrow, as illustrated in (14a-c) (Dikken &
van der Hulst 1988:8). When presented in the form of dependency trees, the head
component resides in the top of the branch. Thus, (14d-f) correspond to (14a-c)
respectively.






25
(14)

Thus, exploiting all combinatory possibilities, we can represent a system with four
front unrounded vowels and four back rounded vowels as follows:

(15)
(Anderson & Ewen 1987:31)

As a result, DP is more constrained than PP in the number of vowel heights it can
generate (Durand 1990). In the latter system, there is no intrinsic limitation on the
number of aperture particle a segment can contain, though Schane (1995:fn.17) does
note that the aperture particle can generally occur at most three times since vowel
systems with more than five degrees of heights are rare.
Another important aspect of DP is the grouping of gestures. It is motivated by the
observation that features often participate in phonological processes in groups. For
example, the place assimilation of nasals mentioned in 2.1 is not just several
unrelated processes that happen to share the same input but a unified phonological
event addressing place features as a whole (Clements 2006). Though SPE (1968:300)
also uses terms such as major class features and cavity features to refer to classes
a. {|X;Y|} or {|X => Y|} Y is dependent on X
b. {|Y;X|} or {|Y => X|} X is dependent on Y
c. {|X:Y|} or {|X <=> Y|} X and Y are mutually dependent

d.

X

Y

e.

Y

X

f.

X:Y
{|i|} = /i/ {|u|} = /u/ {|a|} = .@.
{|i;a|} = /e/ {|u;a|} = /o/
{|i:a|} = /D/ {|u:a|} = /N/
{|a;i|} = /z/ {|a;u|} = /P/
26
of features, they are proposed merely for expository purposes but are not
incorporated into formal representations (Anderson & Durand 1986:20).
DP, by contrast, classifies components into groups of gesture. Evidence
supporting such proposal can be found in phonological processes. Lass (1976) argues
that the reduction of English /p, t, k/ to the glottal stop [>] and the voiceless fricatives
to [h] indicates that gestures should be divided into at least two categories, since both
processes involve the deletion of all articulatory gestures but leave the other features
unaffected (Dikken & van der Hulst 1988). However, this does not implies that there
can only be a maximum of two categories, as argued by Anderson and Ewen (1987).
In fact, different approaches under DP may set up slightly different categories. In
Anderson and Ewens (1987) proposal, components are categorized into three main
groups: the categorial gesture, the articulatory gesture and the tonological gesture.
The first group is further divided into phonatory sub-gesture and initiatory sub-gesture
and the second into locational sub-gesture and oro-nasal sub-gesture, as illustrated in
(16).

(16)
segment


categorial gesture articulatory gesture tonological gesture

phonatory initiatory locational oro-nasal
sub-gesture sub-gesture sub-gesture sub-gesture

(Dikken & van der Hulst 1988:8)

Since it is impossible to explore the full range of components proposed in DP
given the limited space, here I will only introduce the three vocalic components |i|, |u|,
27
|a| and the phonatory sub-gesture, which will figure in some discussions in 3.
According to Anderson and Ewens (1987) proposal, the three basic elements are
posited under locational sub-gesture. As already mentioned, they can represent a wide
range of vowels with different dependency relations, yet the three basic elements are
not only used to characterize vowels but also consonants, where |i|, |u| and |a| specify
the palatality/frontness, labiality/roundness and lowness/openness respectively.
The phonatory sub-gesture contains only two components, |V| and |C|, which
specify the consonantality, voice, continuancy and sonorance of a segment. |V| is
defined as relatively periodic and independently represents the class of vowels,
whereas |C| is defined as periodic energy reduction which on its own characterizes
voiceless plosives. These two components generally correspond to the features
[vocalic] and [consonantal] in the PSA system, but the difference is that the presence
of one phonatory sub-gesture does not necessarily entails the absence of the other;
instead, |V| and |C| can enter into various dependency relations just as |i|, |u|, |a| in the
representation of vowels. However, unlike the three basic elements, each of which can
only occur at most once in a segment, phonatory components are allowed to appear up
to twice in the representation of a segment type, but the two instances of the same
component should not enter into a mutually dependent relation (Anderson & Durand
1986). Thus, some natural classes of segments can be represented in the form of
dependency trees as follows:








28
(17)

(Anderson & Ewen 1987:158)

2.4.3 Government Phonology
GP shares with DP in the use of head-dependent relationship to characterize various
complex segments, only in GP the headedness is expressed through an underlined
element in the notation. According to Kaye et al.s (1985) proposal, an element is an
independently pronounceable constituent defined by a fully specified matrix of
phonological features, one of which is the salient feature of the element, termed as the
hot feature. For example, the three basic elements are defined as follows, with the
salient feature underlined.

(18)

- ROUND + ROUND - ROUND
- BACK + BACK + BACK
I = + HIGH U = + HIGH A = - HIGH
- ATR - ATR - ATR
- LOW - LOW + LOW
(Kaye et al. 1985:306)

V


vowel
V
|
V:C
approximant
V
|
C
nasal
V:C
|
V
voiced
fricative
V:C


voiceless
fricative
C
|
V
voiced
plosive
C


voiceless
plosive

29
In the fusion of two elements, the hot feature of the dependent (operator in Kaye et
als terminology) substitutes for the corresponding feature value in the head, giving
rise to a new matrix that defines a complex segment. For example, when [U] fuses
with [A], with the former as the head, the feature [+ HIGH] in [U] would be replaced
by the hot feature [- HIGH] in [A], which then forms a matrix specifying the vowel
/o/. As in DP, [I], [U], [A] are also available for the characterization of consonants,
representing palatality, labiality and lowness respectively. In addition, Kaye et al.
(1990) propose another three elements, [R], [?] and [h], defined respectively as
follows. In acoustic terms, [R] is recognized for a second-formant transition
characteristic of a coronal gesture. Independently, it represents a coronal tap [3]. The
element [?] is characterized by an abrupt decrease in overall amplitude and is
interpreted as a glottal stop />/ when occurring alone. The element [h] represents a
noise with high-frequency aperiodic energy that features the turbulent airflow in
fricatives and affricates and the noise burst in plosives. In isolation, it is realized as a
glottal fricative /h/ (Harris 1990:263). Incorporated into the mechanism of
head-dependent relation, these elements can represent a wide range of consonants.
Some examples are provided in (19) (where x represents a skeletal position):

(19)







s
x
|
h
|
R
S
w
{
h
{
R
t
x
|
h
|
R
|
?
R
w
{
h
{
R
|
I
f
x
h
|
U
30
Harris (1990:256) argues that the set of privative elements enables GP to explain
phonological events in terms of either composition or decomposition; the former is
the fusion of a segment with other elements spreading from an adjacent segment, and
the latter refers to the loss of elements from the internal structure of a segment. Recall
the discussion about the monothongization of diphthongs above, we have already seen
in both PP and DP how element-based approaches are able to offer a non-arbitrary
account to the output of a fusion process. GP not only shares with these two theories
the ability to predict possible outcome of composition but also attempts to provide a
reason for the loss of elements from a segment occupying a particular position in a
syllable. In what follows, I will introduce two important assumption of GP:
phonological licensing and segmental complexity, which will figure in the discussion
in 3.
In GP, an important principle is the notion of phonological licensing, which
states that within a domain, all phonological units must be licensed save one, the
head of the domain (Harris 1994a:156), while possible domains include the foot,
syllabic constituents, skeletal positions and so on, with the former placed higher on
the prosodic hierarchy than the latter two. Thus, the presence of a phonological unit
has to be sanctioned by the presence of its licensor. Some licensing relations that will
become relevant in latter discussions are illustrated in (20) (where O stands for onset,
N for nucleus and R for rhyme).








31
(20)
(a) Onset Licensing (b) Coda Licensing







(c) Nuclear projection

O N O N

x x x x

O N O N

x x x x


Worth noticing is that, in languages such as English, where a closed-syllable as
(C)VC is allowed, an empty final nucleus is sanctioned by the phonological system
of the language, which Harris refers to as the final-empty-nucleus parameter,
whereas in some languages, the system does not sanction an empty nucleus and thus a
final consonant is not allowed since its licensor, the final empty nucleus, is not
licensed.
Among the licensing relations illustrated above, (20b) further belongs to a
special case of licensing the government relation, defined by Kaye et al. (1990) as an
asymmetric relation holding between two skeletal positions in a phonological string.
Harris (1990) recognizes three types of governing relations, that is, between the two
skeletal positions of a branching onset or nucleus as well as between a coda and a
following onset. According to Kaye et al. (1990:198), government relation is
characterised as strictly local and strictly directional. The condition of strict
locality demands that the governor and the governee be adjacent to each other, and the
strict directionality assumes that the projection of governing relation initiates from the
O N


x x


R O

N

x x x

32
head of a constituent. Thus, the governing relations are rightwards within a branching
onset or nucleus and leftwards between a coda and the following onset. The governing
relation plays an important role in determining what segments are allowed to occupy a
particular skeletal position within a syllable. A universal phonotactic principle
underlying this mechanism is the Complexity Condition, which states that a governed
segment should not be more complex than its governor with respect to the internal
structure (Harris 1994a:170). That is, the number of elements in the representation of
a governee should not be more than that of its governor. The concept is also crucial to
the loss of element in lenition processes and will figure in the discussion in 3.

2.4 Summary
In this chapter I have introduced a segmental representation theory, Element Theory,
which bases its notation on privative phonological constituents. Theories within this
framework, including Particle Phonology, Dependency Phonology and Government
Phonology, differ with regard to the sets of elements and the representational devices
they utilize, yet in general they share the following aspects. First, they all establish
their proposal on three basic vocalic elements [I], [U] and [A]. Second, the elements
proposed in these theories, however different, are all single-valued. Third, they
generally prefer a small feature inventory. Fourth, the elements are defined in both
articulatory and acoustic terms. These characteristics lead to a theory that is more
constrained in the generation of natural classes and possible phonological processes
and thus is preferable to the traditional binary system which suffers from excessive
generative capacity. Having argued on a theoretical basis for element theory, I
summarize some important proposals and the formal apparatus of PP, DP and GP at
the end of this chapter, which will serve the basis of discussion in the case studies in
the chapter below.
33
3 Case studies
Having introduced the notational devices of three element-based theories above, in
this chapter, I will evaluate the adequacy of these theories in their representations of
various phonological events. Here I invoke two criteria for the evaluation. First, an
adequate theory of representation should be able to mirror phonological processes;
that is, the representation should not only present the input and output of a process but
also explain how the change happens where it does (Schane 1984a). For instance, if a
consonant is palatalized under the influence of a following high vowel, an adequate
representation would not simply show that the consonant transforms into a
palatalized consonant before a high vowel, but it is supposed to characterize the
process as one where the palatality spreads from the high vowel to the preceding
consonant. Second, an adequate notational theory should avoid arbitrariness in its
representation. That is, any addition of features or elements to the representation
should have a local source. Taking for example again the consonantal palatalization
mentioned immediately above, the account is non-arbitrary since the spreading
palatality is contributed by the following high vowel, yet it would be regarded as
arbitrary if it is labiality that spreads, which does not exist in the internal composition
of the following high vowel.
In what follows, I will test element theory with three important phonological
processes: vowel shift, vowel harmony and consonantal lenition. The representation
of element theory will be weighed against that of the binary approaches to see which
approach can offer a non-arbitrary account that mirrors these processes.

3.1 Case I: Vowel shift
Vowel shifting is a process where vowels move up or down along the scale of vowel
34
height or become more to the front or the back in the vowel space. Chain shifts,
however, is a special case of vowel shifting in that the place vacated by the input of a
vowel shift is filled up by the output of another shift. Three general principles that
concern the direction of chain shifts are provided by Labov (1994:116): (i) Long
vowels rise, (ii) short nuclei of upgliding diphthongs fall, and (iii) back vowels move
to the front. Labov also finds that, among the three principles, the first two apply most
generally to chain shifts in the worlds languages. These two principles happen to
characterise a series of sound changes in the English Great Vowel Shift (henceforth
GVS). Thus I consider the GVS a representative case that can serve as an appropriate
subject for the analysis of vowel shifting. If a notational theory can represent the GVS
appropriately, it would be applicable to a wide range of vowel shifts in the worlds
languages.

3.1.1 The English Great Vowel Shift
The GVS is a series of interlocked vowel shifts that occurred between the 14
th
and the
17
th
centuries. During this period, all non-high long vowels climbed up gradually in
the vowel space and ended up one degree higher than their original places. The high
vowels /i:/ and /u:/, unable to move up further, diphthongized and eventually reached
the position of /ai/ and /au/ in Modern English, as illustrated in (21) by Jespersen
(1909:232).








35
(21) The English Great Vowel Shift






Though it is generally agreed that (21) represents the endpoint of the GVS, there
has been much debate on how this final situation was reached. One issue concerns the
first impulse that sets off the chain reaction. Some scholars (e.g. Jespersen 1909;
Stockwell 1964; Stampe 1972) propose that the GVS started with the
diphthongization of /i:/ and /u:/, which left the space previously occupied by the two
high vowels empty and thus dragged the mid vowels up to fill the places. Others (e.g.
Luick 1964; Ogura 1990; Lass 1999) argue that it was the mid vowels /e:/ and /o:/ that
moved upwards first and pushed the high vowels out of place, forcing them to
diphthongize. Another problem centres around the path through which the high
vowels diphthongized to their present positions. One assumption shared by Jespersen
(1909), Chomsky and Halle (1968) and Wolfe (1972) is /i://iy//dh//?h//`h/,
which is paralleled by the back high vowel. In their proposal, the nuclei of the
diphthongized high vowels first lowered, then centralized, and then lowered again.
Another view is that the centralization of the nuclei took place last in the development,
following the path /i://iy//ei//zi//ai/. It is not the purpose of this dissertation
to argue for or against any of these positions, but we need to bear in mind that
analyses by different scholars may be based on different assumptions about the
development of the GVS.
The GVS has long posed problems for many theories of representation. One
reason is that, unlike in assimilation, where the source of spreading properties can be
ah i:

e:

D9.z9

a:
u: au

o:

N9
36
found in adjacent segments, the GVS is widely regarded as a context-free process, for
there does not seem to be any phonetic motivation from other segments in the same
phonological string (Jones 1989:204). Another problem is that the GVS is not just a
bundle of unrelated vowel shifts but a chain reaction composed of two different
operations diphthongization and vowel raising. Thus it presents a challenge for any
notational theory that attempts to characterize the whole event as a unified
phenomenon. SPE (1968:264-5) proposes a sequence of rules to describe the GVS,
reproduced as (22).

(22)
(a) Diphthongization






(b) Vowel shift






(c) Diphthong Laxing




(d) Vowel raising





1



- vocalic
- consonantal
back
+ vocalic
- consonantal
+ tense ______
+ high
back

high
- low
[- high]

_____
+ tense
+ stress

[ - low ]

[ - tense ] _____
- vocalic
- consonantal
back
round
[ - low ]
37
According to rule (22a), non-syllabic glides /y/ and /w/ are inserted at the end of
/i:/ and /u:/ respectively, yielding two diphthongs /i:y/ and /u:w/ with long high
vowels as nuclei. Next, rule (22b) exchanges high vowels with mid vowels. Hence /e:/
and /o:/ interchange with the two nuclei of the diphthongs produced by rule (22a).
Rule (22c) then shortens the nuclei of the diphthongs. Finally, rule (22d) makes all
vowels that agree in backness and roundness non-low, which, according to SPE,
affects two low vowels /z/ and /N/. As it turns out, these rules successfully derive the
correct results: the diphthongization of high vowels and the raising of all non-high
long vowels by one degree. However, such representation is problematic in several
aspects.
First, diphthongization is not characterized as a change in the internal structure
of a monothong but an insertion of a glide from nowhere, which constitutes an
arbitrary representation since the additional glide lacks any local source. It is common
for a long vowel to development into a corresponding short vowel plus a homorganic
glide, yet given that the length of the high vowel remains the same, it is unlikely that
the glide originates from the high vowel it follows (Schane 1984b). We are therefore
unable to account for the unexpected appearance of the glides. This problem is
pointed out by Schane (1984a) as a failure of traditional binary approach to capture
the relationship between the internal structure of diphthongs and monothongs, as
mentioend in 2.4.1. The crux of this problem lies in the use of a feature [tense] to
characterize vowel length, a quantitative property, as if it is simply a qualitative one as
roundness or backness. Hence a long vowel occupying two skeletal slots is
represented with a single matrix just as its corresponding short vowel, only the former
is specified with [+ tense] and the latter with [- tense]. Therefore when it comes to the
fission of a monothong into a diphthong composed of two different halves, SPE
system is forced to insert an additional matrix to the representation. The second
38
problem is the proposal of two different mechanisms to deal with a chain shift which
is supposed to be a unified event (Schane 1984b). The lowering of diphthongs and the
raising of mid vowels are achieved not by a vowel raising rule but by an exchange
rule (SPE:256) which switches the mid vowels with the nuclei of the diphthongs
resulting from rule (22a), and only the raising of low vowels is the result of vowel
raising. Such representation of a unidirectional raising process indicates the incapacity
of SPE system to reflect the scalar nature of vowel heights and generalize the chain
shift to a single operation (Fox 1976; Anderson & Ewen 1987).
Both the above mentioned problems are absent from an element-based
representation. The following is a non-linear representation of the GVS proposed by
Anderson and Jones (1977) on the basis of DP. They assume that the diphthongization
of high vowels are triggered by the raising of mid vowels, so diachronically the
processes follow the path (23a) (24a) (23b) but are presented in the following order
to make the representation more revealing. Notice that long vowels in DPs notation
are supposed to be represented as diphthongs with two identical halves, as in (23), to
indicate the length but are presented in (24) as their corresponding short vowels just
for the reason of simplicity. A detailed explanation of this representation is in the
following.








39
(23) Vowel raising


(a)
i
|
a
i
|
a
|
i

i

/e:/ /i:/


(b)
a
|
i
a
|
i
|
i

i
|
a


/z:/ /e:/


(c)
u
|
a
u
|
a
|
u

u

/o:/ /u:/


(d)
a
|
u
a
|
u
|
u
u
|
a


/N9/ /o:/

(24) Lowering of diphthongs



(a)
i


i
i


i i
i


a i

/ii/ /ei/




(b)
u


u
u


u u
u


a u

/uu/ /ou/
40
In the above representation, diphthongization of high vowels and the raising of
long vowels are implemented through the same mechanism, the addition of a tonality
element (as shown in bold type above), which implies that the GVS is a unidirectional
process. In (23), the series of vowel shifts receive a uniform treatment. For vowels
containing |i|, an additional |i| is added to the internal structure, and for those
containing |u|, a |u| is added, both representing a step upwards the height scale. The
scalar relationship between the inputs and the outputs of the raising processes can be
seen in the change in dependency relations. In (24a), paralleled by (24b), the tonality
element |i| is subjoined to the first half of the diphthong resulting from the fission of
the long high vowel /i:/, yet since a segment characterized by a single tonality element
is already placed at the top of the height scale and thus cannot be raised further,
Anderson and Jones (1977:82), following the suggestion from Foley (1977), propose
that the surplus element, which has been strengthened out of the vowel system,
re-enters the system as the element furthest from the direction of promotion.
Namely, the extra element |i|, having been squeezed out in the raising process,
re-enters to the system as element |a|, which then combines with the existing tonality
element in the nucleus to form a mid vowel. In so doing, they manage to provide a
non-arbitrary account for the addition of |a| in the last stage in (24a) and (24b). Also,
compared to (22), where no indication of a unidirectional phenomenon is shown,
formulation (23) and (24) successfully characterize the GVS as a chain shift in that
the space previously occupied by the input of (24a) is now taken by the output of
(23a), and the vacancy left by the input of (23a) is in turn filled by the output of (23b).
The same logic can be applied to (23c-d) and (24b) as well. In what follows, I present
another element-based analysis that is similar to DP analysis discussed above but
further attempts offer an account for the impulse of the GVS.
Though at the beginning of the section I have mentioned that the GVS is widely
41
regarded as a sound change without phonetic motivation, Schane (1984a; 1984b) in
his analysis of the GVS not only tries to represent it as a unitary process but also
attempts to provide an explanation for the impulse of mid vowel raising, which he
regards as the first movement of the chain. His representation based on PP is as
follows.

(25)



As in Anderson and Jones formulation, all vowel raisings are implementations
of the same mechanism, only this time they are characterized unitarily as the loss of
an aperture particle instead of the addition of a tonality particle. The high vowels,
with no aperture particle to lose, diphthongize in response to the force of mid vowel
raising. Again, (25) successfully presents the GVS as a chain shift in that the vacancy
left by the input of one shift is filled by the output of another shift. However, a major
difference between DPs and PPs formulation lies in how they specify vowel length.
While DP represents a long vowel by repeating the segmental structure of a
corresponding short vowel, PP indicates vowel length with an extra tonality particle.
Schane claims that PPs representation enables us to account for the raising of mid
vowels in the GVS by analogy to the height assimilation of a short mid vowel to a
following high vowel. He draws a parallel between the raising of mid vowels in the
GVS with two height assimilation processes in early Germanic (Schane 1984a;
1984b): (i) the raising of /e/ to /i/ when the former is followed by a high vowel, and (ii)
(a) /e:/ /i:/ ai i i i
(b) /i:/ /ei/ i i i i ai i
(c) /D9/ /e:/ aai i ai i
(d) /o:/ /u:/ au u u u
(e) /u:/ /ou/ u u u au
(f) /N9/ /o:/ aau u au u
42
the monothongization of /ei/ to /i:/. The two processes are formulated respectively as
follows.

(26) (a) /e/ /i/ ai i / ___ i
(b) /ei/ /i:/ ai i i i i i

Schane proposes that the raising of /e/ in (26a) is a process whereby a mid vowel
assimilates itself to the following vowel in height. Likewise, in (26b), the raising of
the nucleus in /ei/ also results from the influence of particle |i| in the following glide.
Given the similarity between (25a) and (26), he argues that mid vowel raising in the
GVS is also a height assimilation where the first part of the internal structure of the
long vowel (as shown in bold in (25a,d)) assimilates itself in height to the following
tonality particle that indicates vowel length in the representation. Thus, within the
framework of PP, the initial movement of the GVS does not occur without reason but
is accounted for by an intrinsic impulse that lies in the internal structure of mid
vowels. Moreover, PPs notation also provides an explanation for the cross-linguistic
tendency of long vowel raising in chain shifts, the first principle proposed by Labov
introduced at the beginning of this section. Therefore, though both element-based
analyses can characterize the GVS as a unidirectional process, I consider Schanes
proposal an even better analysis of the phenomenon.

3.2 Case II: Vowel harmony
Vowel harmony is a process where vowels within a word agree with respect to certain
property. In some cases, a property spreads from the stem to the affixes, while in other
cases it is the other way round. Thus, unlike vowel shifting, which is often regarded as
a context-free sound change, vowel harmony is a type of assimilation in that the
43
property involved in the process spreads from one vowel to the others. Various types
of harmony processes have been found in different languages, such as labiality
harmony in Turkish, palatality harmony in Finnish, height harmony in Bantu
languages, and so on (van der Hulst & van der Weijer 1995). Given the limited space,
I will only discuss two types of vowel harmony herethe lowering harmony in
Luganda and the raising harmony in Pasiego Spanish. Both these two processes have
been analysed within the binary-feature and the element frameworks in the literature
and thus can serve a good foundation for the comparison of the two notational
systems in their representation of vowel harmony.

3.2.1 Luganda Lowering Harmony
Luganda is a Bantu language with a five-vowel system containing /i, u, e, o, a/. It has
a height harmony process operating in verb suffixes, which is similar to Chichewa,
another Bantu language. In this section, I will base the discussion of Luganda height
harmony on Harris (1994b), where Chichewa height harmony is discussed in details.
According to Katamba (1984), Luganda vowel harmony requires that all vowels
in verb suffixes agree in height with the vowel in the root, as can be seen in (27a).
However, the low vowel a behaves differently in that a low vowel root selects high
vowel suffixes, as shown in (27b), and the verb final suffix a does not participate in
the harmony process.
(27)
(a) Passive Conversive Reciprocal
simb-ibw-a simb-ul-a simb-agan-a plant
fumb-ibw-a fumb-ul-a fumb-agan-a cook
tem-ebw-a tem-ol-a tem-agan-a cut
gob-ebw a gob-ol-a gob-agan-a chase
(b) lab-ibw-a lab-ul-a lab-agan-a see
(Katamba 1984)
44
To characterize the process, Katamba divides Luganda vowels into two harmonic
categories: mid vowels and non-mid vowels. The mid vowels are specified as [-high,
-low], while non-mid vowels may be [+high, -low] or [-high, +low], which can be
collapsed into [ high, - low]. In Katambas analysis, Luganda height harmony is
characterized as the spreading of height specification from the root to the following
suffixes, as illustrated below.

(28)

However, the above representation has a problem. Though [ high, - low]
captures the distinction between mid and non-mid vowels, the use of variable implies
the failure in predicting the exact height of the vowel following a non-mid root or
suffix. Take (28c) for example, the minus-alpha notation might predict that the verb
root lab spreads [-high, +low] to the following suffix, yielding incorrect form
*lab-al-a in replace of lab-ul-a (Harris 1994b).
Harris (1994b) argues that the problem of Katambas representation described
above is solved in Scullens (1992) proposal, where Bantu height harmony is analysed
within the framework of underspecification. In Scullens proposal, only [-high] is
(a) high
-low


simb ibw a


(b) - high
- low


tem ebw a

(c) high
-low


lab ul a
45
active in the process since if we assume [+ high] as the lexically specified feature that
triggers the process, suffixes following a [- high, + low] root would fail to undergo
height harmony and then be specified with the default value [-high] in a later stage,
giving rise to incorrect results such as *lab-ebw-a instead of lab-ibw-a. As a result,
following Harris (1994b) analysis of Chichewa, we can formulate Luganda height
harmony as follows.

(29)

(30)
(a) temebwa

(b) simbibwa
Harmony (29a) [-high]


tem ebw a
| |
- low - low
- back - back


Redundancy
rules (29b)
[+high] [+high]


sim ibw a
| |
-low -low
-back -back

However, things get a bit more complicated when it comes to low vowel roots,
for the non-high specification would trigger the harmony process, yielding incorrect
(a) Harmonic category: [-high]
(b) Redundant rules: [+ low] [- high]
[ ] [+ high]
46
forms such as *lab-el-a instead of lab-ul-a, as in Katambas analysis (Scullen 1992).
As a result, a constraint is needed to prevent the harmony process from being applied
to low vowel roots before the redundancy rule in (29b) fills in the default value
[+high]. Therefore, Harris discusses a feature-conditioned rule to amend the problem.
This rule states that [-high] can only initiate the process when it is linked to [-low].
Under such condition, a low vowel, which is specified as [-high, +low], will not
trigger the harmony process.
With the supplement of a feature-conditioned rule, we are able to predict correct
outputs for Luganda height harmony within the framework of underspecification.
Nevertheless, Harris argues that the choice of [-low] as a condition for the spreading
of [-high] is arbitrary since it is equally possible for us to choose any other features as
the condition for a harmony process. In the binary framework, there is essentially no
constraint on a principle basis to reject, for example, [-sonorant] or [-round] as the
conditioning feature though such harmony processes may not be attested. It just
happens that in the case of Luganda height harmony, the spreading of [-high] is
conditioned by [-low] and we are unable to explain why [-high] must cooperate with
[-low] to trigger the harmony process (Harris 1994b:521)
Though the binary system is criticized for the arbitrary condition rule, the
underspecification analysis succeed in predicting the outputs of Luganda height
harmony, yet it also indicates that an adequate formulation can be achieved within the
privative framework since only one feature value of [high] is active in the process.
Harris thus proposes an element-based analysis to characterize the process, and he
argues that element theory provides a more adequate representation since the problem
of arbitrary condition rule is absent within such framework. In Harris analysis,
suffixes are assumed to be lexically [A]-less; that is, they are either specified as [U] or
[I], and the height harmony is characterized as the spreading of element [A]. Thus, we
47
can formulate Luganda height harmony as follows.

(31)

Yet again, as in the underspecification analysis, we are now facing the problem of
wrong prediction in the case of low vowel roots, where [A] spreads from a low vowel
to the following suffix, rendering incorrect forms as *lab-el-a in place of lab-il-a.
Thus, Harris proposes a mechanism to constrain the low vowel from initiating the
process, which he terms the tier dependency.
Since in element theory, one of the differences between mid vowels /e, o/ and the
low vowel /a/ is the headedness of element [A], which is a dependent in the former
but a head in the latter, if we want to suppress the spreading of [A] from a low vowel,
we need to narrow down the harmonic category to only dependent [A] (Harris
1994b:530). In order to achieve this effect, Harris proposes that in a five-vowel
system as Luganda and Chichewa, there exists a hierarchical relationship between the
tiers where elements reside, and he makes the following assumption (Harris
1994b:531):

(32) An element on a dominant tier is always the head of a melodic expression.

He claims that in a vowel system containing /i,e,o,u,a/, the [I]/[U] tiers are the
dominant tiers while the [A] tier is the dependent tier, which explains why the mid
(a) [A]


tem ebw a
| |
[I] [I]
(b)





simb ibw a
| |
[I] [I]
48
vowels in Luganda are /e/and /o/ instead of /D/ and /N/. Since the above assumption
states that elements on dependent tiers can never be the head when fused with an
element on a dominant tier, only [I, A] and [U, A], which defines /e/ and /o/
respectively, are allowed. This assumption is further supported by the absence of
segments containing both [I] and [U] in this vowel system since, under assumption
(32), it is also impossible for elements both residing in dominant tiers to fuse with
each other. As a result, Luganda height harmony is reformulated as follows.

(33)

As shown in (33), the assumption of tier dependency achieves the same result as the
feature-conditioned rule proposed in the underspecification analysis above, but as
Harris argues, the former does not suffer from the problem of arbitrariness because
the constraint it provides is derived from the structure of the vowel system.

3.2.2 Pasiego Spanish raising harmony
In the above section, we have seen that the element-based analysis can better capture
the nature of Luganda lowering harmony. In the following, I will discuss a case of
(a) tem ebw a
| |
[I] [I]
|
[A]


(b) sim ibw a
| |
[I] [I]
(c) lab ul a
|
[U]
[A]
49
raising harmony, Pasiego height harmony, to see if element theory is equally adequate
in characterizing such process. Pasiego is a dialect of Montanes Spanish with a
nine-vowel system as follows (Vago 1988:344).

(34) Vowel system of Pasiego Spanish





Since its height harmony does not refer to the distinction between [+tense] and
[-tense], I will simply based my argument on the tense vowel set for its symmetrical
organization. In Pasiego, the height harmony requires that all non-low vowels agree
with the stressed vowel in height (McCarthy 1984; Vago 1988). Hence, within a word,
all vowels preceding a stressed high vowel would become high, while those preceding
a stressed mid vowel are required to be mid. Again, the low vowel a exhibits an
exceptional behaviour. It is transparent in that it occurs freely in the harmonic span of
a high or mid stressed vowel without blocking the process. Also, a stressed low vowel
does not initiate the harmony, so vowels preceding a stressed low vowel would remain
their underlying specification of height. Some examples of Pasiego height harmony
are listed below.

(35)

tense vowel lax vowel
i u
e o
a
H T
N
`
to feel to take
(a) sint:s kux:s 2pl. pres. ind.
sinta kuxa 1sg. imp. ind.
(b) sentmus koxmus 1pl. pres. ind.
(c) sintmus koxmus 1pl. pres. sub.
50





(data from McCarthy 1984; Vago 1988)

From the description above, we can derive that there are two harmonic sets in
Pasiego height harmony, high vowels /i,u/ and mid vowels /e,o/. We can thus
formulate the process as follows.
(36)

Since within the framework of traditional binary system, both feature values are
assumed to be active in phonological processes, the height harmony is characterized
as the spreading of [high] from a stressed vowel to its preceding vowels. However,
problems arise when it comes to the case of stressed low vowel. Under the assumption
that both [+high] and [-high] can trigger the harmony, the current analysis would
predict that a stressed low vowel is always preceded by mid vowels as the result of
[-high]-spreading. Though the prediction conforms to cases like koxmus, where the
underlying height specification of the root is [-high, -low], it will generate incorrect
results as *sentmus (with the definition of to feel) instead of sintmus, where the
preceding vowel is underlyingly [+high], as shown in (37)

abidl birch tree
arna flour
legatrna lizard
(d)
kalr heat
(a) [+high]


sin t:s
| |
-low -low
-back -back
(b) [-high]


sen t mus
| |
-low -low
-back -back
51
(37)







This problem can be amended if we assumes that only [+high] is active in
Pasiego height harmony, as proposed by Vago (1988) within the framework of
underspecification theory. In Vagos treatment of this process, only the positive value
of [high] is lexically specified and thus accessible to the harmony process, whereas
[-high] is filled in by default rules after the harmonizing operation and therefore
cannot trigger the process. In so doing, we can prevent the spreading of [-high] from a
stressed low vowel to its preceding vowels and maintain the underlying height
specification of the latter. The formulation in the underspecification framework is
presented as follows.

(38)







*sentmus

[+high] [-high]
=
sen t mus
| |
-low +low
-back +back
(a) Harmonic category:
Direction:
[+high]
right to left
(b) Default rules: [+low] [-high]
[ ] [-high]
52
(39)

In the underspecification analysis above, we successfully prevent the stressed low
vowel from initiating the height harmony. Moreover, /a/ is not only excluded from the
harmonic sets but is also prevented from being the target of spreading [+high] since
there is a universal incompatibility between [+high] and [+low]. As a result, the
transparency of the low vowel is accounted for by an innate constraint of the binary
feature system.
However, the success of the underspecification treatment of Pasiego height
harmony seems to indicate that the process can be reanalysed in terms of
single-valued elements since only one feature value is active in the operation. Since
the spreading of [+high] is characterized in element theory as loss of element [A],
Harris and Lindsey (1995) proposes the following formulation for Pasiego height
harmony.
(a) sint:s (b) koxmus (c) sintmus
Harmony
(38a)
[+high]


sin t:s
| |
-lo -lo
-bk -bk






[+high]


sin t mus
| |
-lo +lo
-bk +bk
Default
rules
(38b)
[-high] [-high]


ko x mus
| |
-lo -lo
+bk -bk
[+high] [-high]


sin t mus
| |
-lo +lo
-bk +bk
53
(40)

In representation (40), the height harmony is not characterized as feature-spreading
but as the delinking of a certain element. According to Harris and Lindseys proposal,
vowels within the harmonic span have to be licensed by the stressed vowel. They
assume that all vowels to the left of a stressed vowel underlyingly contain an element
[A], yet under the harmony process, any [A] in the preceding vowels have to be
sanctioned by the [A] in the stressed vowel, or else it will be delinked. At first sight,
this treatment seems to indicate that the element analysis is unable characterize the
lowering harmony, i.e. a assimilation process, as feature spreading since the
representation cannot refer to the absence of [high], which is specified as [-high] in
the binary framework. However, the fact that the process can be reanalysed in terms
of privative features also implies that only one feature value is actually needed in the
representation.
However, as can be seen in (40c), the element-based analysis presented by Harris
sint:s koxmus (a)

sin t:s
| |
[I] [I]
=
[A]
(b)

ko x mus
| |
[U] [I]
| |
[A] [A]


*sentmus (c)

sen t mus
| |
[I] [ A]
|
[A]
54
and Lindsey suffers from a problem which is not present in the underspecification
analysis. It erroneously predicts that a stressed low vowel is preceded by mid vowels,
for the [A] in /a/ would sanction the [A]s in the vowels within the harmonic domain,
yielding incorrect forms as *sentmus instead of sintmus. This problem is absent
from the underspecification approach because the low vowel does not contain the
feature value [+high] which is accessible to the harmonizing operation, while in
element theory, the low vowel can initiate the process since it contains an [A] which is
addressed by the licensing relation.
In order to fix the problem in cases like *sentmus, we may invoke the
mechanism of tier dependency proposed by Harris for Chichewa height harmony, as
reported in 3.2.1, to deprive the low vowel of its legitimacy to license. However, we
would instead derive incorrect results in cases where vowels in the harmonic domain
are underlyingly mid. That is, by narrowing down the elements addressable by the
licensing relation to only dependent [A], we make correct prediction in the case of
sintmus but in turn produce incorrect form as *kuxmus instead of koxmus since in
the tier dependency framework, the head [A] in the stressed low vowel would not
sanction a dependent [A] in its harmonic domain and thus gives rise to preceding high
vowels. This case, however, was correctly derived before the mechanism of tier
dependency is invoked. Either way, we are unable to make correct prediction in both
cases. It seems that, though element theory provides a better account for Luganda
harmony, it fails to capture the nature of Pasiego height harmony.

3.3 Case III: Lenition
In the previous two case studies, I have discussed two phonological processes that
concern with vowels. Yet an adequate theory of phonological representation should be
able to characterize not only vocalic but also consonantal change. Thus in this section
55
I turn to consonantal lenition to see if element theory can better characterize
consonantal change than traditional binary approaches.
The concept of lenition is based upon the assumption that some segments are
stronger than others. For such relative strength, Vennemann gives the following
definition: A segment X is said to be weaker than a segment Y if Y goes through an
X stage on its way to zero (cited in Hyman1975:165). Also, Lass and Anderson (1975)
observe that, in lenition processes throughout the history of languages, segments tend
to become more open or more sonorant. A preferred route of lenition is thus illustrated
as follows (Lass 1984:178)

(41)

Voiceless
stop
5a
Aspirate

Affricate
4a

Oral
fricative
3a

Glottal
fricative
2a





Voiced
stop
5b

Affricate
4b

Fricative
3b

Approximant
2b





sonorization

opening





|
1

English t-lenition is a phenomenon that includes a series of synchronic phonological
processes embodying many of the progressions presented in (41), such as
spirantization (from stage 5 to stage 3) and debuccalization (progression towards
stage 2). Therefore I consider it a suitable case for the evaluation of feature-based and
element-based representations for consonantal lenition, which I present in the
following.
56
3.3.1 English t-lenition
In different varieties of English, the alveolar plosive /t/ may have different realizations
depending on its position in a phonological string. For example, Harris mentions
(1990:266) that in Liverpool English, /t/ may spirantize into an /s/ after stressed nuclei
or debuccalize into an /h/ at the end of function words, as shown in (42). However,
Harris description of Liverpool t-lenition is somewhat simplified. In fact, Watson
(2006:60) found that /t/ /h/ does not only occur at the end of function words but
also polysyllabic lexical items such as biscuit, chocolate and ticket, where /t/ appears
in an unstressed final syllable, yet since the focus of the current discussion is on
phonological representation, here I do not intend to give a detailed description of the
phenomenon itself.

(42)

In some other varieties of English, including American English and Australian English,
/t/ is realized as an alveolar tap in the intervocalic position, as in better [bD] and pity
[p

], while other varieties, such as Scottish English and London English, have a
glottal stop instead, as in water [wN?] and city [sH>h]. Moreover, in many varieties, a
syllable-final t is often realized as an unreleased stop.
In the framework of traditional binary system, lenition processes are represented
as the changing of feature values (Harris 1990). For example, spirantization of /t/
listed in (42a) involves a shift from [-continuant] to [+continuant], while process (42b)
is further characterized by the shifts from both [+anterior] and [+coronal] to their
a. [bes]
[bD s?]
bet
better
b. [zh]
[Czh]
at
that
57
negative value. Though the traditional binary system correctly describes the input and
output of a lenition process, it is inadequate as a notational system in several aspects.
First, nothing in such type of representation implies that the output of the process
is weaker than the input. This problem results from the very principle of binary
feature system that every segment has to be fully specified with features in order to be
phonetically interpretable, and the only difference between any two segments lies in
the specification of feature values. Therefore, the internal structure of the input of a
lenition process is no more complex than that of the output, and thus there is no
telling which segment is weaker than the other. The second problem is that, in the
binary framework, various types of lenition processes do not receive a uniform
treatment. As mentioned above, process (42a) involves a shift from the negative to the
positive feature value, while in process (42b) it is the opposite case. It seems that the
binary system provides no constraint on the direction of feature-value changing, and
whether the shift is from minus to plus or plus to minus is completely arbitrary (Harris
1990). Another related problem is the overgeneration of possible lenition processes.
Not only is the direction of feature-value changing unconstrained, there is also no
restriction on the features that are addressable by lenition processes. Thus in principle,
a given segment can weaken into any other segment within such framework, which
does not reflect the fact that lenition processes tend to favour certain trajectories
(Harris 1990).
Taking into account the above mentioned problems of traditional binary
representation, Harris (1990; 1994a) proposes an analysis on the basis of GP, which,
as he argues, not only provides lenition processes with a uniform treatment but also
explains why /t/ tends to weaken in some environments. In Harriss analysis, various
types of lenition are characterized unitarily as processes of decomposition, that is, loss
of elements from the internal structure of segments. For example, a favoured
58
trajectory of lenition that follows plosive fricative h (i.e. 5a3a2a1
in (41)) is represented as follows (Harris 1990:269).

(43)







Following Vennemanns definition of consonantal weakening, Harris
characterizes the lenition trajectory as progressive decomplexification (Harris
1990:268). That is, at each stage of the trajectory, an element drops from the
segmental structure, yielding an output that is less complex in its composition than the
segment in the previous stage. Thus, representation (43) clearly shows that segmental
complexity decreases step by step on its way to zero, and therefore segments with less
complex internal structure are weaker than those with more complex composition.
Harris argues that the reason why lenition can be characterized as loss of element
from segmental structure is that, in GP, every element is assumed to be phonetically
interpretable, and thus the difference between any two segments lies in their internal
compositions instead of feature-values as in traditional binary framework.
Worth noticing is that (43) does not imply that the elements can only drop in that
order. In addition to spirantization and debuccalization, which are presented in (43),
other types of lenition processes are also predicted under the element-based analysis.
For the tapping, glottalling and unreleased /t/ as mentioned at the beginning of this
x
|
R
|
h
|
>

t
x
|
R
|
h



s
x



h



h
(x)








59
section, Harris and Lindsey (1995:125) offers the following representation.

(44)







In (44), the unreleased /t/ is presented as the loss of element [h] from the internal
composition of /t/, and tapping is captured by a further loss of [>\ while glottalling by
the loss of [R] instead. One of the biggest advantages of this analysis is that, unlike in
binary approaches where many unattested processes are allowed in the formalism,
only a limited number of lenition types are predicted in the element framework. For
example, in the binary-feature framework, an unattested lenition from /t/ to /p/ can be
easily represented as a shift from [+coronal] to [-coronal], while in an element-based
representation, it would involve an arbitrary replacement of [R] by [U], yet since the
addition of [U] lacks any local motivation, such process is not predicted in the
element analysis (Harris 1990:268).
From the above representation, we can see that a wide range of lenition processes
are given a uniform treatment, namely, the loss of element from the internal
composition of a segment. However, such analysis is not without its limit. In dealing
with the affrication, Harris provides the following representation (1990:270).


a. unreleased b. tapped c. glottalled

x
{
R
{
>

s

x
|
R





x



>

>
60
(45)










In the above analysis, affrication is characterized as breaking of the internal
composition of a plosive instead of loss of element. However, though Harris argues
that breaking is also a process of decomplexification in that both parts of the
resulting affrication are less complex than the original plosive, such treatment seems
to undermine the strong claim that lenition is a process of element loss (Honeybone
1999). However, in spite of the problem caused by affrication, the nature of lenition
processes is still better captured in GP than in the binary framework since the former
further provides an account for possible lenition sites.
I have mentioned in 2.4.3 that the principle of Phonological Licensing requires
all phonological units except for the head of the domain be licensed. Harris
(1994a:206) also proposes the concept of Licensing Inheritance Principle which states
the licensing potential will be transmitted from a licensor to its licensee, which helps
to compile phonological units into a phonological string, yet during the process of
transmission the licensing potential will be diluted. Thus, an intervocalic consonant is
subject to lenition because the position it occupies is licensed by a following nucleus
which is in turn governed by the nucleus to its left, as illustrated below.


x
|
R
|
h
|
>

t
x

>





ts


R
|
h
61
(46)





The above illustration shows that the licensing potential of the onset should be
weak since it has already been transmitted through two stages from N
1
, that is, the
head of the domain. As a result, an intervocalic segment tends to lenite since the
skeletal position is not capable of licensing complex internal composition. Following
the same line of reasoning, a word-final /t/ is subject to lenition since the position it
occupies inherits it licensing capacity from a following empty nucleus which is
licensed by the system at the same level, as in (47a), whereas a word-initial /t/ does
not tend to weaken because the position it occupies inherits its licensing potential
directly from a nucleus that is the head of the domain and thus has a greater capacity
to support segmental complexity, as in (47b).

(47) Final-empty-nucleus parameter

The characterization of lenition as element loss is further supported by a context
where no varieties of English seem to display t-lenition, that is, when /t/ is preceded
by an obstruent such as /s/, which is represented below (Harris 1990:290).

N
1

x
O
x
|
t
N
2

x

(a) (b)
N O N
x x x
| |
z t
O N O N
x x x x
| | | |
t p
62
(48)









According to Harris analysis, the coda /s/ is governed by the following /t/, and
since the Complexity Condition requires that the governor be more complex than its
governee with respect to internal composition, the /t/ in such context is prevented
from lenition, which is indeed the case found in all the varieties of English in Harris
(1990) study. However, though this analysis generally makes some correct predictions
about possible lenition sites, it does not explain why certain lenition type occurs in
particular environment. For example, though Harris provides an account for the
absence of lenition in the word-initial position, his analysis does not seem to explain
why unreleased /t/ never occurs in an intervocalic position or why tapping never
occurs word-finally before a consonant or pause. In this sense, which type of lenition
will occur in which environment seems to be arbitrary.
Alternatively, Ewen and van der Hulst (2001) argue that not all lenition processes
have the same cause as proposed by Harris. According to their analysis, though
lenition to /h/ and />/ are processes of decomplexification, intervocalic lenition is a
process where a consonant assimilates itself to the surrounding vowels with respect to
the category gesture and thus should be characterized as a progression where
R O

N

x x x
| |
R R
| |
h h
|
?

s t
63
consonants become more vowel-like. For example, based on Anderson and Ewens
(1987:176) notation, a lenition trajectory such as 5a 5b 3b 2b in (41) can be
represented in DP framework as in (49):

(49)

{|C|} {|C:V|} {|V:C => V|} {|V:C|}
voiceless voiced voiced approximant
plosive plosive fricative

According to Ewen and van der Hulsts characterization of intervocalic lenition, in the
above representation. the phonatory component V becomes more prominent along the
trajectory, indicating that the segment becomes more vowel-like. However, this
analysis also suffers from the problem of arbitrariness as Harris analysis discussed
above. Though debuccalization to [h] does not occur in intervocalic position and thus
is not a process of assimilation to surrounding vowels, t-glottaling is mainly found in
intervocalic position but is not a result of categorial assimilation, which runs counter
to Ewen and van der Hulsts prediction. Hence, whether an intervocalic consonant
would undergo decomplexification or categorial assimilation seems to be arbitrary in
the DP analysis. It seems that, though both element-based analyses can better mirror
lenition processes than binary approaches, they are unable explain why different
lenition types occur in different environments.

3.4 Summary
In this chapter, I have discussed feature-based and element-based representations for
three important phonological processes. Generally speaking, element theory seems to
be a more adequate notational theory based on the discussions in this chapter since it
64
meets the two criteria introduced at the beginning of this chapter.
In the case of the GVS, with the proposal of four different phonological rules,
SPE representation fails to characterize the phenomenon as a series of unidirectional
and interlocked vowel shifts. By contrast, the element-based analyses successfully
capture the nature of the GVS in that the output of one shift replaces the input of
another shift, and all vowel shifts, including diphthongization and vowel raising, are
results of the same operation. In DP analysis, vowel shifts are uniformly presented as
the addition of |i| and |u| whereas in PP as loss of an aperture particle, both
representing a step upwards the scale of vowel height. However, PP seems to be more
explanatory in that it provides the first movement of the chain shift with a
non-arbitrary account by drawing a parallel between the height assimilation of a mid
vowel followed by a high vowel and the raising of mid vowels in the GVS, which
results from the height assimilation of the first part of the internal structure to a
following tonality particle.
In the case of Luganda height harmony, the traditional binary approach fails to
make correct predictions about the height of vowels within the harmonic domain of a
transparent low vowel. This problem, however, is solved in the framework of
underspecification where only one feature value is assumed to be active in the process
and a condition rule which states that [-high] can only trigger the process when linked
to [-low] is applied to prevent the transparent a from triggering the harmonizing
process. However, the proposed condition rule seems to suffer from the problem of
arbitrariness since it is equally possible for us to propose other features as the
condition. Thus, to solve the problem caused by the transparent a, Harris (1994b)
proposes the mechanism of tier dependency to further restrict the harmonic category.
Unfortunately, this mechanism fails to make correct predictions in Pasiego raising
harmony since it cannot generate correct results for both koxmus and sintmus. In the
65
former case, vowels preceding a stressed low vowel are underlying mid while in the
latter case, they are underlyingly high. In this sense, the underspecification approach
seems to be more powerful with regard to the characterization of different harmony
types in spite of the arbitrariness of the condition rule.
For the consonantal lenition, the traditional binary approach fails to provide
various types of lenition processes with a uniform treatment and nothing in its
representation implies the relative strength of the input and output segments. By
contrast, in element-based analyses, we can see the relative strength of sounds from
their segmental representation and thus predict possible lenition trajectories, though
DP analysis characterizes intervocalic lenition as the assimilation of consonants to
surrounding vowels while GP equals all types of lenition to element loss. Both
analyses reflect the manner by which lenition processes occur, but they also have their
limits. In GP analysis, where lenition is assumed to be a process of element loss,
affrication is characterized as breaking instead of loss of element. Moreover, both
DP and GP analysis do not seem to provide a completely non-arbitrary account for the
occurrence of different types of lenition processes in certain positions.

4 Conclusion

To sum up, in this dissertation, I have argued from both theoretical and empirical
perspectives that element theory is a better theory of phonological representation than
the binary-feature system.
In 2, I discuss three important aspects of element theory. The proposal of
single-valued element largely reduces the amount of possible natural classes and
phonological processes since the representation cannot refer to the absence of an
element. Thus, unlike in traditional binary approaches where crazy rules can be
66
easily formulated, only the feature-value that is recurrently addressed by phonological
processes is accessible to the representation in the element framework. Moreover, the
proposal of a small inventory of elements also places a constraint on the generative
capacity of element theory since a small number of elements will essentially allow
fewer combinatory possibilities. These two proposals lead to a more constrained and
thus preferable theory that is exempted from overgeneration of natural classes and
phonological processes, which is a major problem of binary approaches. However,
this does not indicates that element theory is a less powerful theory of representation
than binary approaches since the proposal of dependency relations between elements
enables element theory to utilize a small amount of elements in a more economic way.
Furthermore, since element theory not only defines its phonological components in
terms of articulation but also acoustics, it can capture the nature of more phonological
processes than traditional binary system, which shows a bias towards the articulation,
since some processes refer to the articulatory aspect of segments but others address
the acoustic aspect.
In 3, I weigh the element-based representations of vowel shift, vowel harmony
and consonantal lenition against the feature-based representations. The result of the
evaluation shows that element theory has its limit in the characterization of Pasiego
height harmony and consonantal affrication and it is unable to account for the
tendency of certain types of lenition to occur in particular environments, as already
summarized in the previous section. However, compared to binary approaches,
element theory is still better in capturing the nature of various phonological processes
in that it generally provides non-arbitrary representations that can mirror how the
processes occur. Also, the comparison between element-based and feature-based
representations in the case studies corresponds to the conclusion from 2 that element
theory is a more constrained theory since fewer possible phonological processes are
67
predicted in this framework, which, for example, enables the prediction of several
preferable lenition trajectories. This kind of prediction is not possible in a binary
approach since it suffers from excessive generative capacity. As a result, I conclude
that element theory should be regarded as a more successful approach to segmental
representation than the orthodox binary feature system.































68
References

Anderson, J. & C. Jones (1977). Phonological Structure and the History of English.
Amsterdam: North-Holland Publishing Company.

Anderson, J. M. & J. Durand (1986). Dependency Phonology. In Durand, J. (ed.)
Dependency and Non-Linear Phonology. London: Croom Helm.

Anderson, J. M. & C. J. Ewen (1987). Principles of dependency phonology.
Cambridge: Cambridge University Press.

Anderson, J. M. & C. Jones (1974). Three theses concerning phonological
representations. Journal of Linguistics 10, 1-26.

Archangeli, D. (1988). Aspects of Underspecification Theory. Phonology 5, 183-207.

Bach, E. & R. T. Harms (1972). How Do Languages Get Crazy Rules? In Stockwell,
R. P. & R. K. S. Macaulay (eds.) Linguistic change and generative theory.
Bloomington; London: Indiana University Press.

Browman, C. P. & L. M. Goldstein (1989). Articulatory Gestures as Phonological
Units. Phonology 6, 201-251.

Campbell, L. (1974). Phonological Features: Problems and Proposals. Language 50,
52-65.

Chomsky, N. & M. Halle (1968). The Sound Pattern of English. New York: Harper
& Row.

Clements, G. N. (1991). Vowel height assimilation in Bantu languages. Working
Papers of the Cornell Phonetics Laboratory 5, 37-76.

Clements, G. N. (2003a). Feature Economy as a Phonological Universal. In Sol, M.
J., D. Recasens & J. Romero (eds.) Proceedings of the 15th International Congress of
Phonetic Sciences. Barcelona: Futurgraphic.

Clements, G. N. (2003b). Feature economy in sound systems. Phonology 20, 287-333.

69
Clements, G. N. (2006). Feature Organization. In Brown, K. (ed.) The Encyclopedia
of Language and Linguistics 2th Edition. Oxford: Elsevier Limited, 433-440.

Clements, G. N. (2009). The role of features in speech sound inventories. In Raimy, E.
& C. Cairns (eds.) Contemporary views on architecture and representations in
phonology. Cambridege: Mass: MIT Press.

Danyenko, A. & S. Vakulenko (1995). Ukrainian. Munich: Lincom Europa.

Dehaene-Lambertz, G. & M. Pena (2001). Electrophysiological evidence for
automatic phonetic processing in neonates. Neuroreport 12, 3155-3158.

Dikken, M. d. & H. G. van der Hulst (1988). Segmental hierarchitecture. In van der
Hulst, H. G. & N. Smith (eds.) Features, segmental structure and harmony processes.
Dordrecht: Foris, 1-59.

Durand, J. (1990). Generative and Non-Linear Phonology. London: Longman.

Ewen, C. J. & H. G. van der Hulst (2001). The Phonological Structure of Words: An
Introduction. Cambridge: Cambridge University Press.

Foley, J. (1977). Foundations of theoretical phonology. Cambridge: Cambridge
University Press.

Fox, A. (1976). Problems with phonological chains. Journal of Linguistics 12,
289-310.

Giegerich, H. J. (1992). English Phonology: An Introduction. Cambridge:
Cambridge University Press.

Goldsmith, J. (1985). Vowel Harmony in Khalkha Mongolian, Yaka, Finnish and Hungarian
Phonology Yearbook 2, 253-275.

Halle, M. (1983). On Distinctive Features and Their Articulatory Implementation.
Natural Language and Linguistic Theory 1, 91-105.

Halle, M. (2009). Two Comments on "The Role of Features in Phonological
Inventories". In Raimy, E. & C. Cairns (eds.) Contemporary Views on Architecture
70
and Representations in Phonology. Cambridge: Mass: MIT Press.

Harris, J. (1990). Segmental complexity and phonological government. Phonology 7,
255-300.

Harris, J. (1994a). English sound structure. Oxford ; Cambridge, Mass.: Blackwell.

Harris, J. (1994b). Monovalency and opacity: Chichewa height harmony. UCL
Working Papers in Linguistics 6, 509-547.

Harris, J. & G. Lindsey (1995). The elements of phonological representation. In
Durand, J. & F. Katamba (eds.) Frontiers of Phonology. New York: Longman.

Honeybone, P. (1999). I blame the government. Language Sciences 21, 177-221.

Hyman, L. M. (1975). Phonology : theory and analysis. New York: Holt, Rinehart
and Winston.

It, J. & R. Mester (1988). The Phonology of Voicing in Japanese: Theoretical
Consequences for Morphological Accessibility. Linguistic Inquiry 17, 49-73.

Jakobson, R., G. Fant & M. Halle (1952). Preliminaries to speech analysis.
Cambridge, Mass: MIT Press.

Jespersen, O. (1909). A Modern English Grammar on Historical Principles. Part I:
Sounds and Spellings. London: George Allen & Unwin.

Jones, C. (1989). A History of English Phonology. New York: Longman.

Katamba, F. (1984). A nonlinear analysis of vowel harmony in Luganda. Journal of
Linguistics 20, 257-275.

Kaye, J., J. Lowenstamm & J. Vergnaud (1985). The internal structure of phonological
ements: a theory of charm and government. Phonology Yearbook 2, 305-328.

Kaye, J., J. Lowenstamm & J. Vergnaud (1990). Constituent structure and government
in phonology. Phonology 7, 193-231.

71
Kiparsky, P. (1985). Some Consequences of Lexical Phonology. Phonology Yearbook
2, 85-138.

Labov, W. (1994). Principles of Linguistic Change 1: Internal Factors. Oxford:
Blackwell.

Ladefoged, P. (1971). Preliminaries to Linguistic Phonetics. Chicago: University of
Chicago Press.

Ladefoged, P. & I. Maddieson (1996). The Sounds if the World's Languages. Oxford:
Blackwell.

Lass, R. (1976). English phonology and phonological theory. Cambridge:
Cambridge University Press.

Lass, R. (1984). Phonology: An Introduction to Basic Concepts. Cambridge:
Cambridge University Press.

Lass, R. (1999). Phonology and morphology. In Lass, R. (ed.) The Cambridge History
of the English Language Vol. 3 1476-1776. Cambridge: Cambridge University Press.

Lass, R. & J. Anderson (1975). Old English Phonology. Cambridge: Cambridge
University Press.

Liberman, A. M. & I. G. Mattingly (1985). The motor theory of speech perception
revised. Cognition 21, 1-36.

Lindau, M. (1975). Features for vowels. UCLA Working Papers in Phonetics 30.

Lindau, M. (1978). Vowel features. Language 54, 541-563.

Lombardi, L. (1996). Postlexical rules and the status of privative features. Phonology
13, 1-38.

Luick, K. (1964). Studien Zur Englishen Lautgeschichte. New York: Johnson
Reprint Corp.

McCarthy, J. J. (1984). Theoretical Consequences of Montaes Vowel Harmony.
72
Linguistic Inquiry 15, 291-318.

McCarthy, J. J. (1988). Feature geometry and Dependency: A Review. Phonetica 43,
84-108.

Ogura, M. (1990). Dynamic Dialectology. Tokyo: Kenkyusha.

Ohala, J. J. (1981). The listener as a source of sound change. In Masek, C. S., R. A.
Hendrick & M. F. Miller (eds.) Papers from the Parasession on Language and
Behavior. Chicago: Chicago Linguistic Society.

Ohala, J. J. (1996). Speech perception is hearing sounds, not tongues. Journal of the
Acoustical Society of America 99, 1718-1725.

PulleyBlank, D. (1988). Vocalic underspecification in Yoruba. Linguistic Inquiry 19,
233-270.

Rennison, J. (1984). On tridirectional feature systems for vowels. In Durand, J. (ed.)
Dependency and Non-linear Phonology. London: Croom Helm, 281-303.

Schane, S. A. (1972). Natural Rules in Phonology. In Stockwell, R. P. & R. K. S.
Macaulay (eds.) Linguistic Change and Generative Theory. Bloomington; London:
Indiana University Press, 199-229.

Schane, S. A. (1984a). The fundamentals of particle phonology. Phonology Yearbook
1, 129-155.

Schane, S. A. (1984b). Two English Vowel Movements: A Particle Analysis. In
Aronoff, M. & R. T. Oehrle (eds.) Language Sound Structure. Cambridge, Mass.: MIT
Press, 32-51.

Schane, S. A. (1995). Diphthongization in Particle Phonology. In Goldsmith, J. (ed.)
The Handbook of Phonological Theory. Cambridge, Mass.: Blackwell, 586-608.

Schane, S. A. (2005). The Aperture Particle |a| : Its Role and Functions. In Carr, P., J.
Durand & C. Ewen (eds.) Headhood, Elements, Specification and Contrastivity.
Amsterdam: John Benjamins B.V., 313-338.

73
Scullen, M. E. (1992). Chichewa vowel harmony and Underspecification Theory.
Linguistic Analysis 22, 218-245.

Sommerstein, A. H. (1977). Modern Phonology. London: Edward Arnold.

Stampe, D. (1972). On the natural history of diphthongs. CLS 8. Chicago: Chicago
Linguistic Society, 578-590.

Steriade, D. (1979). Vowel harmony in Khalkha Mongolian. MIT Working Papers in
Linguistics 1, 25-50.

Steriade, D. (1987). Redundant values. In Bosch, A., B. Need & E. Schiller (eds.)
Papers from the parasession on autosegmental and metrical phonology, CLS 23.
Chicago: Chicago Linguistic Society.

Steriade, D. (1995). Underspecification and markedness. In Goldsmith, J. (ed.)
Handbook of Phonological Theory. Cambridge, Mass: Blackwell, 114-174.

Stevens, K. N. (1972). The quantal nature of speech. In David, E. E. & P. B. Denes
(eds.) Human communication: a unified view. New York: McGraw Hill, 51-66.

Stevens, K. N. (1989). On the quantal nature of speech. Journal of Phonetics 17, 3-45.

Stockwell, R. P. (1964). On the utility of an overall pattern in historical English
phonology. In Lunt, H. (ed.) Proceedings of the Ninth International Congress of
Linguists. Hague: Mouton, 663-671.

Vago, R. M. (1988). Underspecification in the height harmony system of Pasiego.
Phonology 5, 343-362.

van der Hulst, H. (1992). The phonetic and phonological basis of the Simplex Feature
Hypothesis. In Dressler, W. U., H. C. Luschtzky, O. E. Pfeiffer & J. Rennison (eds.)
Phonologica 1988. Cambridge: Cambridge University Press.

Van der Hulst, H. (2006). Dependency Phonology. In Brown, K. (ed.) The
encyclopedia of language and linguistics. 2nd edition. Oxford: Elsevier, 451-458.

van der Hulst, H. & J. van der Weijer (1995). Vowel Harmony. In Goldsmith, J. (ed.)
74
The Handbook of Phonological Theory. Cambridge, Mass.: Blackwell.

van der Hulst, H. G. (1989). Atoms of segmental structure: components, gestures and
dependency. Phonology 6, 253-284.

Watson, K. (2006). Phonological resistance and innovation in the North-West of
England. English Today 86, 55-61.

Wetzels, W. L. & J. Mascar (2001). The Typology of Voicing and Devoicing.
Language 77, 207-244.

Wolfe, P. (1972). Linguistic Change and the Great Vowel Shift in English. Berkeley,
California: University of California Press.