EPENTHESIS IN CHILDREN'S CONSONANT

CLUSTER PRODUCTIONS: A PERCEPTUAL

AND ACOUSTICAL STUDY

\

by

MARTA KELCEY EVESON

B.Sc, The University of Victoria, 1991

A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF

THE REQUIREMENTS FOR THE DEGREE OF

MASTER OF SCIENCE

in

THE FACULTY OF GRADUATE STUDIES

(School of Audiology and Speech Sciences)

We accept this thesis as conforming to the required standard

THE UNIVERSITY OF BRITISH COLUMBIA

April 1996

© Marta Kelcey Eveson, 1996

In presenting this thesis in partial fulfilment of the requirements for an advanced

degree at the University of British Columbia, I agree that the Library shall make it

freely available for reference and study. I further agree that permission for extensive

copying of this thesis for scholarly purposes may be granted by the head of my

department or by his or her representatives. It is understood that copying or

publication of this thesis for financial gain shall not be allowed without my written

permission.

Department of A u d ^ n q ^ qp A ^Sy^fj^V) Soe^ceS

The University of British Columbia Vancouver, Canada

DE-6 (2/88)

ABSTRACT

The purpose of the present study was to examine epenthesis in children's

consonant cluster productions from phonological and phonetic perspectives. The

following questions were investigated: (1) Do consonant clusters produced with an

epenthetic vowel differ in duration from those without? (2) Is the epenthetic vowel in the

consonant cluster consistent in length and quality, or do co-articulatory effects occur?

(3) Is the epenthetic vowel dependent in terms of duration on the phrasal context or the

duration of the syllable nucleus? The subjects, S_i (Charles) and S2 (Blair), were two of

six subjects in a doctoral research study investigating the application of a nonlinear

phonological framework to the assessment and remediation of phonological disorders.

Consonant cluster data were transcribed from the original data. Acoustic measurements

included the duration of consonant clusters with and without epenthesis and the duration

of the epenthetic vowel. Results of the investigation show that consonant clusters with an

epenthetic vowel are significantly longer in duration than those without. No coarticulatory

effects were seen between the epenthetic vowel and the syllable nucleus suggesting that

the epenthetic vowel is part of the consonant cluster unit which is governed by its own

timing system. Prosodically, syllabification of the word occurs as a result of epenthesis

in the consonant cluster. The implication of these results appears to be that the consonant

cluster containing the epenthetic vowel needs to be considered as a separate timing unit

and representationally attributed unitary status.

ii

TABLE OF CONTENTS

ABSTRACT II

TABLE OF CONTENTS HI

LIST OF TABLES VI

LIST OF FIGURES V H

ACKNOWLEDGEMENT VIII

CHAPTER ONE INTRODUCTION 1

OVERVIEW 1

PHONOLOGY VERSUS PHONETICS 2 Phonology 3 Implications of phonological environment 4 Phonemes 4 Allophones 5 Phonetics 5

GENERAL IMPLICATIONS OF PHONOLOGICAL THEORY 6

NONLINEAR PHONOLOGY 7 Representation Versus Rules 7 Underlying Representation of Syllable Structure 8 Ffierarchical Representation of the Prosodic Tier 8 Theories of the Structure of the Prosodic Hierarchy 10 Tier Association 11 The Segmental Tier 11 Markedness 11 Developmental Implications of a Feature Hierarchy 12

CONSONANT CLUSTER DEVELOPMENT 13 Delay Versus Deviation 15 Pattern of Acquisition 16

IMPACT OF DIFFERENT METHODOLOGIES: UR VERSUS REALIZATIONS

iii

UNDERLYING REPRESENTATION OF CONSONANT CLUSTERS 18

TEMPORAL CO-ORDINATION IN CONSONANT CLUSTER PRODUCTION 21

COARTICULATION 22

METHODOLOGICAL IMPLICATIONS 23

TIMING CONSTRAINTS 24 Duration and Temporal Variability 25

IMPLICATIONS OF THEORETICAL MODELS 26

MODELS OF LANGUAGE PROCESSING 27 Serial Model 27 Parallel Interactive Model 28 Theoretical Assumptions 28

SUMMARY 29

CHAPTER TWO METHOD 32

SUBJECTS 32 51 Summary 33 52 Summary 34

APPARATUS AND PROCEDURES 34

MEASUREMENTS 36 Measurement Reliability 38

CHAPTER THREE RESULTS AND DISCUSSION 40

SUMMARY OF RESULTS 41 Occurrence of Epenthesis 41 Consonant Cluster Duration 42 Epenthesis as a Strategy for Overcoming Timing Demands 44 Effect of Phonological Context on the Epenthetic Vowel 44 Coarticulatory Effects 45 Epenthetic Vowel Duration 46

IMPACT OF EPENTHESIS ON PROSODIC STRUCTURE 48 Representation of Consonant Clusters 48

iv

APPLICATION OF NONLINEAR PHONOLOGICAL FRAMEWORK 50 Epenthesis in a Serial Model of Language Processing 51 Epenthesis in a Parallel Interactive Model 52

STUDY LIMITATIONS 54

FUTURE RESEARCH 55

CONCLUSION 56

REFERENCES 58

APPENDIX ONE SI DATA 62

APPENDIX TWO S2 DATA 67

v

LIST OF TABLES

TABLE 1: Summary of Mam-Whitney test for difference in duration between consonant clusters without and with epenthesis 42

vi

LIST OF FIGURES

FIGURE 1: Hierarchical representation of prosodic structure 9

FIGURE 2: Alternative representation of CV syllables in onset-rime

and moraic theories 10

FIGURE 3: Specified feature geometry for English 12

FIGURE 4: Representation of epenthesis and deletion processes 15 FIGURE 5: Duration of consonant clusters without and with epenthesis

for Subjects 1 and 2 43

FIGURE 6: Relationship between syllable nucleus duration and epenthetic

vowel duration for Subjects 1 and 2 47

FIGURE 7: Representation of epenthesis in moraic and onset-rime theories" 49

FIGURE 8: Representation of ClV eC2 as one unit 49

vii

ACKNOWLEDGEMENT

I would like to express my sincere gratitude to my supervisors, Dr. John H. V.

Gilbert and Dr. Barbara Bernhardt, for their guidance and patience as I have made my way

through this process and for the large amount of time that they have spent reading and

revising numerous copies of my drafts. Thank you for the encouragement when it was

needed the most.

I would also like to thank my Mom and Dad for their endless support and

encouragement throughout this endeavor. I could not have gotten to where I am without

you. Special thanks need to go to my sister, Paige, for her statistical knowledge and help

and for putting up with me during the final stages.

Lastly, I would like to acknowledge the support and encouragement of all my

friends and colleagues. Thanks to Eva Major for her time, resources and crisis

management skills (coffee and a sense of humour) as I neared the end. Special thanks to

NJG, MM, NT, ST, and CM as they were there along the way.

viii

CHAPTER ONE

INTRODUCTION

OVERVIEW

When examining children's productions of initial consonant clusters it is frequently

observed that correct production of clusters occurs at a late stage in phonological

development. Factors such as timing and lack of neuromotor maturation contribute to a

child's difficulties in combining consonants into clusters. To overcome these difficulties

children use a variety of transformation processes. It has been reported (Ingram, 1976;

Stoel-Gammon and Dunn, 1985) that one of these processes, epenthesis, does not occur

as frequently as others (e.g. deletion of one consonant). Therefore, little attention has

been paid to the effect of epenthesis on the consonant cluster unit. However, because

epenthesis affects timing of the word and the cluster unit itself, it raises questions with

respect to representation and syllable structure. This paper will focus on epenthesis in

consonant cluster development from phonological and phonetic perspectives. To do this,

it is necessary to discuss current theories of consonant cluster representation in the

underlying phonology and to examine acoustic aspects of them. This paper will examine

how consonant clusters are represented using a nonlinear phonological framework. A

review of literature on consonant cluster development will provide the necessary

background information to discuss the relevant timing issues. In addition, two current

models of language processing (Serial Model, Garrett, 1994; Parallel Interactive Model,

Stemberger, 1985a and b) will be reviewed to examine how they might deal with the issue

1

of epenthesis and the relationship between the underlying representation and phonetic level

of these consonant clusters.

Children's consonant cluster productions have been examined at the surface or

realization level. Detailed explanations exist of the varying processes (e.g. deletion of one

consonant, substitutions) that seem to occur before a child's consonant cluster productions

match an adult model. However, very few studies have looked at consonant cluster

acquisition while considering the underlying representation level of a child's phonology. In

recent years, emphasis has shifted to investigating child phonological development with

respect to a specific phonological theory, allowing for the consideration of the relationship

between the underlying phonology and the output (i.e. the surface level).

In one of the few studies which considered an underlying representation of a child's

phonological system, Chin and Dinnsen (1992), using a two-level generative phonology

framework, found that a systematic relationship could be described between adult

representations and children's underlying representations as well as between children's

underlying and phonetic representations of consonant clusters. Since then, more advanced

versions of nonlinear frameworks have evolved which have been shown to account better

for many phonological occurrences (Bernhardt, 1992). One major purpose of this paper is

to examine the consonant cluster productions of two children using a current nonlinear

phonology framework.

PHONOLOGY VERSUS PHONETICS

In the field of speech/language pathology, it is the practice of many to distinguish

between "phonological" disorders and "phonetic" disorders. This application of linguistics

2

to the field of speech/language pathology has proven useful but has led to some confusion

and oversirnplification (Grunwell, 1985; Hewlett, 1985). Therefore, when differentiating

between levels of analysis of speech production, it is necessary to define the concepts

discussed.

Phonology

In his discussion of the relationship between phonetics and phonology, Laver

(1994) outlined that the function of phonology is to relate the phonetic events of speech to

other areas such as the morphological, lexical, syntactic and semantic levels. Thus,

phonology is directly tied to phonetics. Laver states, "at the phonological level of

analysis, two utterances are held to be different if the phonetic differences between them

serve to identify the two utterances as representing different grammatical units of a given

language" (1994, p.30). Further to this, a phonological system can be defined as a set of

consonants which exist in potential distinctive opposition to each other. More simply,

phonology involves the manner in which speech sounds function to contrast or distinguish

different words (e.g. /p/ in 'pan' versus /k/ in 'can' versus Iml in 'man'), (Hewlett, 1985;

Laver, 1994). The presence or absence of such a contrast can be used as the criterion for

assigning a given sound to a specific phonological category. It is also frequently used as

the criterion for distinguishing between phonological and phonetic disorders. If a

phonological contrast is lost it is considered a phonological disorder, whereas if the

contrast is maintained, the disorder is phonetic or at least, lower-level (Hewlett, 1985).

Lower-level refers to a late occurring process as defined with respect to models of

language processing. Such models assume that language is composed of a hierarchical set

of interconnected levels of processes (Stemberger, 1985a; Stemberger, 1985b).

3

Implications of phonological environment

It is important, when examining the characteristics of mdividual phonological units,

to consider the specific phonological structure and phonological context in which the unit

is found. Phonological structure can be defined as the sequence of consonants

(represented as C) and vowels (represented as V) that constitute syllables and feet (e.g.

the structural formula for the word 'cup' > /kAp/ is CVC, or a strong monosyllabic foot

[see Figure 1]). This type of sequencing formula is commonly used to describe syllable

structure (Laver, 1994). Related to this, the phonological context refers to the actual

identity of the C's and Vs adjacent to the given phonological unit. The relevance of

considering phonological context lies in the influence that it exerts on the phonetic

realization of the individual phonological unit being examined.

Phonemes

In considering the notions of contrastiveness, structure, and context, it is important

to discuss a widely used phonological concept, the phoneme. Laver (1994) states that the

notion of the phoneme is based on the alphabetic tradition of writing and is not

theoretically strict. In the simplest of terms, each individual segment represented in an

underlying form is called a phoneme (Sloat, Taylor and Hoard, 1978). More explicitly,

two speech sounds are to be considered as different phonemes in a given language when

they act in contrastive opposition to distinguish two words with identical phonological

structure. That is, in the context in question, the two speech sounds must exhibit parallel

distribution by having the potential to occupy the same place in a given phonological

structure and form a minimal pair (Laver, 1994).

4

Allophones

Each given phoneme can include a set of members called allophones. Two speech

sounds are classed as allophones of a specific phoneme if they are found to occur "in

complementary distribution, and i f they display sufficient phonetic similarity to make it

plausible to class them together as members of a common set" (Laver, 1994, p.42). The

concept of an allophone is an abstract description and should not be confused with the

definition of a phone which will be discussed later.

In summary, a distinct phoneme is identified when a given speech sound occurs in

contrastive distribution to other speech sounds in words with identical phonological

structure. Different pronunciations of a speech sound which show phonetic similarity are

said to be allophones of a given phoneme when they are complementary in distribution and

their phonetic differences are a result of the phonological structure or environmental

context (Laver, 1994).

Phonetics

This then leads us to the area of phonetics. Whereas phonology deals with

underlying representations of individual units, phonetics deals with speech sounds as

physical entities (Hewlett, 1985). Phonetics is the description of the "learnable aspects of

use of the vocal apparatus" (Laver, 1994, p. 28). Individuals acquire the specific phonetic

behaviours common to the language in their environments in a social context. The

phonetic level of description is considered to be abstract. Therefore, a statement of

phonetic sameness is based on the comparison of abstract features, not acoustic identity.

This is based on the assumption that two different speakers should be capable of

producing phonetically identical utterances (although acoustically different). A speech

event which is considered phonetically equivalent between speakers is called a phone

5

(Laver, 1994). Phonetic description is theoretically held as independent of phonological

description in that knowledge of the linguistic value that the speech event may have as a

communicative element in a given language is not required. A phonetic "distortion"

occurs when the underlying representation is correct but the phoneme is incorrectly

realized. For example, if a person is trying to produce the word 'gum' (/gAm/), the fg/

could be "distorted" as a result of loss of some of its voicing during actual articulation. In

the most extreme case /g/ may be retrieved from storage but the loss of voicing is

complete and /g/ is produced as Ikl (Hewlett, 1985).

GENERAL IMPLICATIONS OF PHONOLOGICAL THEORY

In the past, the assessment and intervention of speech disorders was strictly

constrained to description at the phonetic level (Grunwell, 1985). In the past 25 years,

researchers have presented different theories of phonology and have applied them in the

assessment and intervention of speech disorders (Edwards and Shriberg, 1983; Hewlett,

1985; Bernhardt, 1992). The application of phonological theory has had direct impact on

the field of speech/language pathology by allowing generalizations and descriptions to be

made about a child's speech output with reference to an underlying phonological system.

Nonlinear phonological frameworks have been shown capable of describing and explaining

observed phonological events of a given language (Bernhardt, 1992).

6

NONLINEAR PHONOLOGY

Major purposes of this paper are to examine epenthetic vowel insertion in

consonant cluster productions from phonetic and phonological (nonlinear) perspectives. It

is necessary to outline first the major aspects of nonlinear phonology. (For a more

complete overview of nonlinear phonology, please refer to Bernhardt, 1992.)

Nonlinear phonology has risen from generative phonology and is based on many of

the same tenets (Bernhardt, 1992). Two main assumptions of generative phonology which

also act as the basis of a nonlinear phonological theory are as follows:

1. Each linguistic component, although interactive for language processing, can

be studied as an independent system.

2. Within a linguistic component, there exists an 'unmarked' versus 'marked'

parameter option where it is usually assumed that the 'unmarked' value of a

variable is the one available innately from the 'universal grammar' (UG).

Representation Versus Rules

One major difference between nonlinear and classical generative phonology is the

emphasis that nonlinear phonology places on representation. As a result, within a

nonlinear phonological framework, rules and processes become more limited and

generalized. Rather than being sequential, as in classical generative phonology, the

nonlinear representation is multi-tiered and involves prosodic information. The limited set

of phonological rules or processes results from association between the separate tiers and

is restricted in function. Phonological processes either add content, through information

linking ('spreading') or 'insertion' of new information, or subtract content, through

information delinking between domains.

7

Traditionally, as a child's phonological system develops, comparisons are made

with an adult system. Any differences between the child and adult systems have been

described in terms of rules and processes. Bernhardt (1992) outlines two basic theoretical

weaknesses with this approach:

1. It may suggest more 'neurological activity" than is actually happening.

2. It explains acquisition in terms of a negative 'progression' whereby the child has

to learn to undo a specific process (e.g. un-front or un-delete).

If it is assumed that children have an intact representational framework which acts

more as a passive 'filter' when they begin to learn language, as it might be assumed with

nonlinear phonology, phonological development can be seen more as a building process.

If the information from the adult form matches the child's present perceptual and

production level, it will pass through and be accepted. Any mismatched information will

be ignored until the child's system matures and incorporates it or until the mismatched

information is received enough times to cause recognition and change (Bernhardt, 1992).

Underlying Representation of Syllable Structure

In addition to assuming that a child comes to the language learning process with an

underlying segmental feature inventory, a nonlinear phonological approach assumes that

the child also comes with an expected underlying framework for syllable structure. Thus,

although parameters need to be set for both features and for syllable structure, tier

autonomy implies independence in phonological learning for each tier.

Hierarchical Representation of the Prosodic Tier

In a nonlinear framework, representation occurs as a hierarchically linked set of

tiers rather than as a sequential string with little internal structure (Chin and Dinnsen,

8

1992). Therefore, prominent units of the system dominate other more embedded units. In

looking at the prosodic level, the word dominates feet which in turn dominate syllables

(see Figure 1). By incorporating prosodic information in the hierarchical representations,

several differing conditions can be accounted for including stress patterns, segmental

processes involving feature-spreading, and the phonotactics of a language (Bernhardt,

1992). This then implies developmentally, that if a child comes to the language learning

process with a determined underlying phonological framework, that child also comes with

an expected prosodic structure.

WORD (CV.CV)

FOOT

ONSET RIME ONSET RIME

C V C V

Figure 1: Hierarchical Representation of Prosodic Structure

9

Theories of the Structure of the Prosodic Hierarchy

Two major theories of the structure of the prosodic hierarchy are onset-rime

theory and moraic theory. With onset-rime theory, in the hierarchical structure, the onset

node (O) dominates the prevocalic consonants and the rime (R) dominates the nucleus (N)

(see Figure 2). The nucleus is a node which dominates the most sonorant segment

(usually a V). As the hierarchy is represented with a branching tree structure, more

complex syllable shapes (e.g. those containing consonant clusters) have more branches. In

addition to complexity being related to branching, in many languages, stress assignment is

related to branching in the rime (Bernhardt, 1992).

In moraic theory, it is suggested that prosodic 'weight units' or 'morae' (M) are the

important components of syllables for stress assignment. Moras are realized through

vowels (see Figure 2). Prevocalic and postvocalic consonants not contributing to syllable

weight are adjoined to the mora or syllable node and have no particular structural

function. Moras combine together to form syllables, (usually a maximum of two moras

per syllable), which are grouped into feet. In moraic theory, onsets by representation do

not affect stress assignment whereas in onset-rime theory, onsets although branching, by

stipulation are assumed not to attract stress (Bernhardt, 1992).

O R a

N

V

M

C V

Onset-rime Moraic

Figure 2: Alternative representations of CV syllables in onset-rime and moraic theories.

10

Tier Association

Although tier autonomy allows for independent phonological learning at the

prosodic level, the prosodic tier is linked to the segmental tier. Linking of tiers occurs

according to principles of association (Bernhardt, 1992). Bernhardt and Stemberger (in

preparation), state that association lines can be present in the underlying representation,

created by a mapping rule, or added by an assimilation rule.

The Segmental Tier

By representing segments as geometrically organized sets of features, a single

feature in a dominating node can be shown as affecting other segments within a set

domain. Each feature is autonomous (i.e. on its own tier), but can be influenced by its

dominating features (see Figure 3). The dominating node may only influence

neighbouring segments, or may involve the entire word (Bernhardt, 1992; Chin and

Dinnsen, 1992). Segments exist as distinct phonological entities but in reality, are co-

articulated to such an extent that information from several phonological segments is

perceived at any given time (Bernhardt and Stemberger, in preparation).

Markedness

A hierarchical representation of features also allows for the notion of markedness

('unmarked' versus 'marked' features) in that higher level or dominating features could be

seen as being more 'marked' than lower level, deeply embedded features (Bernhardt,

1992). Different proposals exist as to a set of universal features and their relative

markedness.

11

ROOT

LARYNGEAL

[+voice] [+spread glottis]

PLACE

LABIAL CORONAL DORSAL

[+distributed] [-anterior]

Figure 3: Specified feature geometry for English

Developmental Implications of a Feature Hierarchy

The developmental implications of this type of a feature hierarchy include the

notion of developmental progression. Features which appear more deeply embedded may

appear later developmentally that those features which are found higher in the hierarchy

(Bernhardt, 1992).

12

CONSONANT CLUSTER DEVELOPMENT

As stated earlier, the main objective of this paper is to look at the development of

consonant clusters in child phonology with reference to the nonlinear phonology

framework just outlined. More specifically, epenthetic vowel insertion in consonant

clusters will be examined. Therefore, it is now necessary to review what is known about

consonant cluster development.

One of the major controversies that still exists among researchers interested in the

study of phonology is the twofold issue of how children acquire the necessary phonemes

of their language and how they learn to combine these sound segments into speech.

Children do not simply need to learn the specified features of their language. They are

also required to learn the complex patterns of sound combinations for their language.

Research shows that this process begins with children's earliest babblings of a labial

consonant plus a vowel (e.g. /ba/) and continues until approximately the age of six when

children finally become adept at producing word-final consonant clusters (Stoel-Gammon

and Cooper, 1984; Stoel-Gammon and Dunn, 1985).

As discussed earlier, one major difference between nonlinear and classical

generative phonology is that nonlinear phonology places emphasis on representation, thus

limiting the number of rules and processes that need to be applied. However, many prior

studies into consonant cluster development used a process or rules-based approach.

Therefore, to look at the findings of these studies, it is necessary to discuss the types of

processes to which they refer. Regardless of the chosen theory, it has been recognized

that children use phonological strategies to help them organize and simplify the processes

of acquiring and combining speech sounds, especially with consonant clusters. It is also

evident that children exhibit great variability in the use of these processes and strategies.

13

There are three general categories into which these processes can be grouped: syllabic

structure processes, substitution processes and assimilation processes. Syllabic structure

processes involve modification of the syllable structure of the target word by means such

as deleting an unstressed syllable (e.g. "telephone" /tolafoun/ > /telfoun/), final consonant

deletion, or cluster reduction. Substitution processes allow for the replacement of one

sound with another that the child has already mastered (e.g. using a stop /p/ for a fricative

/f/). Assimilation processes are identified as occurring when one sound becomes more

similar to another (e.g. fronting: "cat" /kaet/ > /taet/) (Roach, 1983; Stoel-Gammon and

Dunn, 1985). It is important to note that there is large individual variation in the use of

such strategies and processes as well as in their outcomes. While some children may never

use a certain strategy, others may rely solely on that same strategy and apply it to all

utterances that they produce. In addition, target words can be affected by different

processes simultaneously which can account for some of the odd productions made by

children (e.g. /kik/ for "stick" involves consonant cluster reduction and assimilation).

Children usually do not apply these strategies or processes randomly. Rather, each child

has an internal system or filter which helps them to work towards adult-like speech sound

combinations. According to Stemberger (1985a) a given rule is represented in two parts:

1) a structural condition, a pattern governing where the rule applies, and 2) a structural

change, a pattern signifying what changes occur to the underlying representation. Each

rule is filtered through to see if the information matches with the child's current

representation. As stated earlier, according to nonlinear phonology, any mismatched

information will simply be ignored until the child's system is ready to incorporate it

(Bernhardt, 1992). Figure 4 illustrates an example of how representationally the effect of

different processes on syllable structure can be shown.

14

Adult Representation

Epenthesis Deletion

onset

A C C

onset nucleus onset

V e

onset

C C

t r t a r

Figure 4: Representation of epenthesis and deletion processes

Delay Versus Deviation

When examining child speech, it is necessary to have a thorough understanding of

the basic sound modification possibilities. Understanding these basic processes has led to

the reanalysis of the nature of functional (i.e. nonorganic in nature) speech disorders

(Ingram, 1976; Dinnsen, Chin, Elbert and Powell, 1990). In examining data from children

with functional speech disorders, it is necessary to consider the properties of such

disordered systems. If it is accepted that functional speech disorders represent a delay in

speech acquisition, then the information gained can also be applied to normal language

development. However, if functional speech disorders represent a deviation from the

normal pattern of acquisition, the information cannot be applied in the same manner.

In an attempt to clarify this issue, Dinnsen, Chin, Elbert, and Powell (1990) looked

at the phonological systems of 40 children with functional speech disorders and identified

the different properties and constraints which appeared to be acting. One of the

15

difficulties with this type of characterization is that although there are fundamental

acquisition processes which occur, as mentioned earlier, there also appear to be

widespread individual differences in the properties and in the specific order of the normal

acquisition of the different speech sounds. For the purpose of their study, a variety of

language tests were administered to each child and a single-word spontaneous speech

sample was elicited. According to the researchers, analyses indicated that the properties

of the speech systems of the children with functional speech disorders closely paralleled

the principles applied during normal language acquisition. Therefore, it would seem that

these speech disorders could be classified as delays rather than deviations: the children

with phonological disorders appeared to be using the same processes as the children with

'normal' speech but applying them at a different rate.

Pattern of Acquisition

Although children use a variety of different processes in learning speech, a general

pattern of sound acquisition can still be outlined. Furthermore, a general pattern of

acquisition can be described within individual speech sound categories. For example,

children tend to acquire front stops before back ones (labials and alveolars before velars).

As stated earlier, children first begin producing words by combining these early acquired

consonants with vowels in a simple CV syllabic shape. As they acquire more phonemes,

the variety of syllabic shapes that they are able to produce concurrently increases.

However, there is much evidence to support the claim that children do not seem to be able

to combine consonants in the form of a consonant cluster (as required for a more complex

CCV syllabic unit) until quite late. Even if a child has mastered the two single phonemes,

it cannot be assumed that s/he will be able to combine them to produce the form of a

consonant cluster. Templin (1957) found that children did not produce initial consonant

16

clusters correctly until approximately age 4. She claimed that final nasal clusters were

mastered at the same time but [liquid + stop], [liquid + fricative], and [fricative + stop]

clusters in final position did not appear until between the ages of 6 and 7. At this stage,

phonological development is very transitional. Much of the child's early development is

set and the use of processes begins to occur less frequently (Ingram, 1976).

IMPACT OF DIFFERENT METHODOLOGIES: UR VERSUS REALIZATION

Although early research has provided some answers about the acquisition of

consonant clusters into children's phonologies, there are still many questions to be

answered. Two different approaches have been taken in studying consonant clusters: 1)

from the perspective of normal, non-disordered acquisition, and 2) within the scope of

treatment studies with speech-disordered children. Chin and Dinnsen (1990) point out

that although different, both approaches mainly focus solely on the realization level. Thus,

the representational level, or the underlying segment representation, has not been

thoroughly investigated, especially with regard to a specific theoretical framework.

"There seems to be no principled, structural explanation available of why target clusters

come out the way they do in children's productions, either normal or disordered" (Chin

and Dinnsen, 1990, p.2).

17

UNDERLYING REPRESENTATION OF CONSONANT CLUSTERS

In the few earlier studies done which examine children's underlying representation

of clusters, it has been hypothesized that clusters may first be represented in children's

repertoire as a single unit which later is separated into its distinct components. This

hypothesis has been supported by acoustic evidence reported by Menyuk (1972), which

indicated that consonant clusters may be represented as a single consonant in some

children's underlying phonology. Further evidence can be found in a study conducted by

Barton, Miller, and Macken (1980), in which 24 children between the ages of 4;0 and 5;0

completed three experimental tasks which examined their ability to segment initial

consonant clusters into distinct phones. The data support the theory that children treat

clusters as a single phonological unit before they are capable of distmguishing separate

segments. As children acquire the metalinguistic skills (i.e. the explicit awareness and

understanding of language forms or structures) necessary to separate a sound into its

distinct phoneme components, different processes, such as deletion of one component, or

weakening or substitution of one consonant in the cluster are seen in their cluster

realizations. Other processes such as apparent epenthesis (inserting a vowel or consonant)

or apparent metathesis (a reversal of adjacent segments) are also seen although most

researchers claim that they are much more uncommon (Ingram, 1976; Stoel-Gammon and

Dunn, 1985).

Other researchers (Stemberger and Treiman, 1986) have upheld the claim that

children use specific strategies as evidence that consonant clusters are not first represented

as a single unit. They believe that the fact that a child uses epenthesis may support the

hypothesis that both consonants are represented. Furthermore, although children often

use reduction to avoid producing the actual consonant cluster, a general pattern of

18

deletion does exist (/s/ deletion in /s/-clusters and liquid deletion in liquid clusters) and is

somewhat predictable, depending on the target cluster. The one-unit stage, hypothesized

by Greenlee (1974), can be viewed as a lower-level application of the process of reduction

on a two-element adult-like underlying representation (Chin and Dinnsen, 1992).

In support of this notion that children have a two-element underlying

representation for consonant clusters, some researchers (Menyuk and Klatt, 1975;

Kornfeld, 1976) claim that although in early stages of cluster acquisition adults perceive

children's consonant clusters as a singleton consonant, these children are making acoustic

distinctions between singleton consonants and clusters. It has been suggested that in the

early stages of cluster development, specific consonants (i.e. phonemes) may be less

accessible to children for their consonant cluster productions, even though both consonant

slots are present in the underlying representation (Stemberger and Treiman, 1986).

Stemberger and Treiman (1986) looked at whether word-initial consonants are more

accessible, by examining loss, addition and substitution errors made in cluster productions,

occurring naturally and induced experimentally. Both sets of data indicated that the

position of a consonant in an initial cluster affects the rate of error occurrence. For

example, for addition, shift and loss errors, far more errors occur involving C2 than CI.

Stemberger and Treiman state that to account for their data, CI and C2 must be present in

the underlying representation but are represented differently, with CI having a greater

level of activation. Additional evidence supporting the hypothesis that both consonants

are represented in the underlying representation is that, in speech errors, consonant

clusters do not act like single phonemes. Rather, individual segments are affected

(Stemberger and Treiman, 1986).

As stated above, even after children are able to differentiate perceptually between

the component parts of a consonant cluster, their cluster realizations may not accurately

19

reflect this knowledge. Research has shown that although children are able to perceive

and produce acoustic distinctions between single consonants and consonant clusters

reliably, these differences are not necessarily noticed by adult listeners. Evidence

supporting the claim that children make acoustic distinctions between a consonant as a

singleton and in a cluster can be seen when studying voice onset time (VOT)

characteristics. Menyuk and Klatt (1975) examined spectrograms of words in isolation

and in sentences produced by eleven children and two adults and measured VOT. The

results showed that when the average VOT of singleton stops and stops in clusters were

compared, the VOT values were longer for stops in clusters for both children and adults.

Menyuk and Klatt noted that even though the VOT values of stops produced by the

children did not match those of the adults, few of the children's cluster productions were

perceived as incorrect in the sentence context. SHghtly more of the consonants in the

children's consonant cluster productions were perceived as incorrect when produced in

isolation. The most common errors involved the incorrect production of a liquid in a

cluster and substitutions of stops in clusters. Another error which was noted involved the

introduction of a schwa between the consonants in [stop + liquid] clusters. This

epenthesis occurred most frequently with labial and dental stops and occasionally with

velar stops. According to Kornfeld (1976), these findings theoretically suggest that

children may use principles different from adults for classifying and representing their

underlying phonology. In other words, children may be responding to different acoustic

cues and relying more on phonetic information for speech recognition than adults do.

Evidence to support this can be seen in that children, both normal and language

disordered, can produce a consonant correctly as a singleton but distort it in the context of

a reduced consonant cluster. Furthermore, children frequently can differentiate more

20

phonetic contrasts than they can produce, and will not accept adult imitations of then-

productions as correct (Kornfeld, 1976).

TEMPORAL CO-ORDINATION IN CONSONANT CLUSTER PRODUCTION

Further to Menyuk and Klatt's (1975) study, Gilbert and Purves (1977) examined

several hypotheses regarding the development of temporal coordination involved in the

production of consonant clusters. Typically most research that has looked into temporal

organization in children's speech has narrowly focused on variability in durations.

Comparatively, the main issue in studies of temporal organization of adult speech has been

to identify what triggers the initiation of speech gestures. The question that has been

asked is, is speech sound initiation governed by a higher level timing programme over a

given unit or is each individual phone produced separately, i.e. does the completion of one

phone trigger the production of the next. In exarnining this issue, Kozhevnikov and

Chistovich (1965) concluded that some type of higher order timing programme seems to

be in effect. According to Ohala (1970), this is the result of a timing dominant system,

which he distinguished from an articulation dominant system. In a tirning dominant

system, the speaker must perform the necessary specified articulatory gestures within a

limited and rigid time requirement. In contrast, in an articulatory dominant system, cluster

production is governed by a chain reaction of articulatory gestures.

21

COARTICULATION

Evidence supporting an articulation dominant system can be found by examining

the results of coarticulation studies. Coarticulation is defined as the influence of one

speech segment upon a neighbouring segment (Daniloff and Hammarberg, 1973).

Although it is assumed that at some level, for example, in the underlying representation,

speech sounds exist in invariant uncoarticulated forms, in running speech, sounds seem to

overlap and there is a "spreading of features." It has been hypothesized that the

uncoarticulated forms in the underlying representation are encoded when entered into the

articulatory mechanism. However, controversy exists because it is impossible to identify

the encoding unit and the contribution of the encoding mechanism. The question also

arises whether coarticulation is a deliberately applied process or whether it is a result of

the articulatory mechanism. Ohman's (1967) study of the coarticulation of [+voiced] stops

in VCV type utterances provides evidence which is suggestive of some degree of

articulatory preprograrmning. Some degree of preparation for the following speech

gesture appeared to occur simultaneously with production of the preceding segment.

Peterson and Lehiste (1960) examined the influence of preceding and following

consonants on the duration of stressed vowel and diphthongs. They point out that

changes in the duration of a sound may be related to or determined by the linguistic

environment. These durational changes may become cues for the identification of

associated phonemes (i.e. the preceding or following segmental sounds). Peterson and

Lehiste determined that the influence of an initial consonant on the syllable nucleus

duration appeared negligible. Rather, the duration of the syllable nucleus was significantly

affected by the nature of the following consonants. As Peterson and Lehiste only used

words containing a singleton word-initial consonant, direct conclusions cannot be made

22

about consonant clusters. However, the study raises the idea that if it is the following

consonant that has the effect on the duration of the syllable nucleus, the consonant cluster

unit may be separate from the word in its timing. In addition, if the syllable nucleus is

subject to anticipatory lengthening, can this also be expected in the epenthetic vowel?

Ohala (1970) does not completely support the notion of an exclusive tirning

dominant system. Rather, he proposed that a co-occurrence of the two systems could

exist with each operating on separate levels. Allen (1973) furthered this idea by proposing

a model which involves the notion of three separate factors interacting to deteirnine the

phonetic duration of a given segment. The three factors he outlined are: 1) the speaker's

intended speech rate; 2) the underlying phonological length of the segment; and 3) the

peripheral effects of transmission time. According to Purves (1976), although Allen's

(1973) model "clarifies the role of various factors in determining segment durations, it

does not provide any information about the way in which speech gestures may be

integrated into higher-level units with a cohesive temporal program" (p. 10).

In summary, the phenomenon of coarticulation and the process of consonant

cluster reduction are upheld as evidence of preprograrriming of higher level temporally

organized speech units. It is also accepted that many factors such as speech rate and the

nature of the underlying segment and its neighbouring segments interact to determine the

produced duration of a given segment within an utterance.

METHODOLOGICAL IMPLICATIONS

However, Purves(1976) notes that different methodological approaches can be

suggestive of differing results and need to be further evaluated before stronger claims can

23

be made. A primary example can be found in the fact that differences in durational aspects

of clusters can be seen by examining results from child data rather than adult data.

Hawkins (1973) found that children tend to lengthen fricatives in the CI position of a

cluster. Furthermore, in child speech, Is/ + stop + prevocalic Ixl clusters were lengthened.

The data indicated that specific aspects of the timing relationship within a cluster differ

fairly consistently between children's and adults' speech. Hawkins (1973) suggests these

differences may reflect specific articulatory difficulties on the part of the children.

However, she does note that specific methodological aspects also need to be taken into

consideration. Primarily, a consistent criterion for the segmentation of an acoustic

waveform needs to be determined.

If the durational differences discussed by Hawkins (1973) were significant, it could

be concluded that the control of segmental duration in clusters is mastered over time. In

support of this, Tingley and Allen (1974) examined the extent to which control of timing

improves over time and found that timing accuracy increased with age. This finding is

consistent with the fact that a child's motor skill development continues to improve as the

child grows older.

TIMING CONSTRAINTS

According to Gilbert and Purves (1977), findings which suggest that children's

timing control may be less accurate than that of adults', could also indicate the presence of

timing constraints in child speech production. Many different factors, including the effects

of surrounding phonological segments, are involved in controlling duration of given

speech segments and need to be taken into consideration. The results of the Gilbert and

24

Purves (1977) experiment indicated that consonant cluster productions of young children

can be differentiated from those of older children and adults on the basis of absolute

duration of consonants. In addition, it was found that children use strategies of reduction

and sphtting (schwa insertion) during the acquisition of consonant clusters. They suggest

that the splitting process may be a means for a child to overcome rigid timing demands

and to attempt to match the adult model. That is, during the earlier stages of consonant

cluster acquisition, the child's timing control has not yet sufficiently developed to allow the

correct production of both consonants within the restricted time frame. Thus, at first, the

child omits specific features and produces reduced consonant clusters. Then later, the

child establishes an individual timing system and applies the process of sphtting where

segmentation allows for the individual phonemes to be accurately articulated. Over time,

the child's consonant cluster productions become more refined and the overall absolute

duration becomes less until eventually it matches that of adult speakers. This description

of how children overcome timing constraints to approximate consonant clusters is

suggestive of more frequent use of epenthesis than most studies report. This may be the

result of the epenthetic vowel being considered as part of the following consonant and

measured as such.

Duration and Temporal Variability

As noted from the above studies, it is generally accepted that duration and

temporal variability tend to decrease as children grow older and become better speakers.

Researchers are interested in these two parameters because they are generally viewed as

indicators of neuromotor maturation of speech skills. As there is an observed tendency for

duration and variability to decrease with increased age, many researchers assume that

these two measures are closely related. However, Smith, Sugarman and Long (1983)

25

observed that although children speaking at an increased rate had durations similar to

adults, variability was still greater. The question arises whether these two measures can be

viewed as separate indicators of neuromotor maturation or whether variability is a by­

product of duration. In an attempt to address this, Smith (1992) examined temporal data

from children ranging between 2;6 and 9;6 and found that duration and variability were

not closely correlated. Thus similar conclusions cannot be drawn about speech motor

control. Therefore, it is necessary to consider carefully which variable is being discussed

when examining children's speech productions. Smith concluded that it can generally be

assumed that duration and variability will decrease with increased age but that conclusions

cannot be made as to how these measures pertain to mdividual children. Duration and

variability do not appear to be congruent indicators of individual speech maturation. From

Smith's data, it appears that measures of duration become more adult-like earlier than

measures of variability. Smith also noted that his results indicated that variability should

not be considered a function of duration. He suggested that one of the measures (perhaps

duration) may be a better reflection of lower-level articulation skills whereas variability

may be associated more with higher-level organizational skills.

IMPLICATIONS OF THEORETICAL MODELS

Although different theories exist, it is clear that the exact relationship between

children's early consonant cluster productions and those of adults is still not completely

understood. There is no convincing explanation detailing the process and stages by which

children learn to coordinate consonants into clusters. This is complicated by the fact that

concrete and accurate knowledge regarding the nature of the underlying representation of

26

the phonological system is not available. Different methodologies lend themselves to

distinct theories and measures of the processes or levels that are applied to the underlying

representation before it is perceived by a listener. That is, without an existing model of

the nature of the underlying representation, it is virtually impossible to measure the effect

of the hypothesized processing levels.

MODELS OF LANGUAGE PROCESSING

Serial Model

One current model of language processing is Garrett's (1980, 1984) model which

consists of three distinct processing levels: a message level, a linguistic sentence level

(consisting of functional, positional and phonetic levels of representation) and a motor

articulatory level. If the different processing levels are serial and are affected separately,

as Garrett suggests, then it would be possible to hypothesize that differences in the

breakdown of speech may also be seen. In this model, the message level is where the

general concept is first processed into an approximate sentence construction. Then at the

sentence level, the general concept undergoes the application of logical and phonological

rules and is represented with specific linguistic structures. That is, the phonological form

of the lexical item is retrieved. Finally, at the articulatory level, the initial representation of

the signal undergoes phonetic and prosodic encoding and is then translated into the

correct instructions necessary for articulatory sequencing. This is the final output form.

There has been some criticism of Garrett's proposed model, however, because of

its rigid linear nature. Some researchers claim that although there do seem to be different

levels involved in the encoding and processing of language, they do not act entirely

27

independently of each other. Rather, the possibility of interaction exists between levels of

processing.

Parallel Interactive Model

In a parallel interactive model "several levels of the language system are being

processed at a given moment and are interacting" (Stemberger, 1985b, p. 148). A speaker

accesses language production when an intent to speak is formulated and pragmatic and

semantic information leads to the partial activation of all words that represent that

irrformation. The activation of these units causes a cascade of activation to all associated

units. In addition, lower-level units that are associated, such as phonemes, are also

activated. Lower-level activation then spreads back up to cause activation of other words

with those particular phonemes. Eventually, the word that is most highly activated will be

articulated (Stemberger, 1985a; Stemberger, 1985b). Any given unit is not considered to

be on or off but rather, on a continuum of activation. The unit's activation threshold needs

to be exceeded for it to be selected. According to Bates and MacWhinney (1989), this

type of competition model assumes the dynamic control of the mapping of form onto

function for comprehension and of function onto form for production. It is not to be

assumed the relationship between form and function is one to one. Rather, it is a many to

many relationship. It is understood that this mapping is governed by parallel activation

with strength level resolution.

Theoretical Assumptions

For this paper, the premise was made that children typically have a two-element

adult-like underlying representation of consonant clusters as discussed by Chin and

Dinnsen (1992). Therefore, any discrepancies between the child's production of a

28

consonant cluster and the adult version can be explained as the application of

transformation processes at lower levels. For example, as outlined earlier, the one-

element stage in consonant cluster production would be viewed as a lower-level

application of a reduction process on the two-element underlying representation. In

Garrett's model, this would occur at the articulatory level. In an interactionist model, the

reduction process could occur at different points, although it would still be a lower-level

application.

In summary, this uncertainty about the underlying representation is only one of

many unanswered questions about consonant cluster development. There is no concrete

explanation about the relationship between the underlying representation and the output

and why the different stages of consonant cluster development occur. It is accepted that

timing constraints and neuromotor maturation may play a role. However, even this idea is

complicated by the fact that the measures of duration and variability are not correlated.

As stated earlier, it is unclear how these two measures are related and how they pertain to

individual children.

SUMMARY

Although children seem to follow a normal developmental sequence in the

acquisition of consonant clusters, it is known that as a result of such factors as timing and

lack of neuromotor maturation, children have difficulty combining consonants into

clusters. To help deal with these difficulties, they often use a variety of transformation

processes. However, not all children use the same processes in mastering consonant

cluster production. Some researchers claim that one of these processes, sphtting or

29

insertion of an epenthetic vowel, does not seem to occur as frequently as the other

processes discussed (e.g. reduction, assimilation) (Ingram, 1976; Stoel-Gammon and

Dunn, 1985). However, in an intervention study performed by Bernhardt (1990),

epenthetic vowel insertion was seen in consonant cluster productions of four of six

subjects. The objective of this paper is to examine the consonant cluster productions of

two of these subjects and compare the clusters with and without an epenthetic vowel in

order to determine further information about phonology and phonetics of cluster

development. The following questions will be investigated:

1. Do consonant clusters with an epenthetic vowel differ in duration than those

without? This is related to the issue of timing constraints: if as Gilbert and

Purves (1977) suggest, epenthesis is a means of overcoming the tirning

demands of consonant cluster production, clusters without an epenthetic vowel

should be shorter in duration.

2. In consonant clusters with an epenthetic vowel, is the epenthetic vowel

consistent in length and quality or is it affected by the surrounding consonants

of the cluster? In other words, is the epenthetic vowel a constant or are there

coarticulafion effects?

3. Where an epenthetic vowel exists within a consonant cluster, is it dependent in

terms of length on the phrasal context or the duration of the syllable nucleus?

If such outside factors influence the epenthetic vowel, then the epenthesis

needs to occur at a higher processing level because these factors are

determined prior to the final articulatory level.

Whether epenthetic vowel insertion is a late or low-level process will also be

discussed. It is proposed that children have an existing adult-like underlying

representation of consonant clusters which is altered by the application of a lower-level

30

process. Following the theoretical models outlined earlier, the last level of transformation

of the underlying representation would be the pronunciation of the consonant cluster. If

children were purely having articulatory difficulty in producing consonant clusters, it

would seem that epenthesis would occur frequently. For instance, adult speakers of

languages which prohibit initial consonant clusters often insert an epenthetic vowel

between adjacent consonants of clusters in their second language or in 'borrowed' words

from languages with clusters, even though they perceive and spell it as a cluster (Greenlee,

1974). This therefore raises the question of whether or not the epenthetic vowel is part of

the consonant cluster itself (i.e. does the consonant cluster have its own timing rules

independent of the word) or whether it is a separate entity. In an attempt to look at some

of these issues, the focus of this paper will be restricted to epenthesis in consonant clusters

and how it relates to timing issues. The nonlinear phonology framework outlined will be

used to discuss the theoretical underlying representation. The relationship between the

underlying representation and the output will also be examined with reference to the

language processing models discussed.

31

CHAPTER TWO

METHOD

In order to examine epenthetic vowel insertion in children's consonant cluster

productions, consonant cluster data produced by two children in Bernhardt's (1990) study

were examined acoustically.

SUBJECTS

For this study, consonant cluster data were examined from Sl (Charles) and S2

(Blair) from Bernhardt's (1990) study of six children with phonological disorders. In the

original study, all subjects were children who had moderately severe or severe

phonological disorders. Criteria for inclusion were as follows: 1) Subjects were between

the ages of 3 and 7 years. 2) English was the only input language. 3) There was an

absence of any other major impairment excluding a language production disorder, mild

impairment of language comprehension, cognition, or motor development, or controlled

otitis media. 4) Parents participated over the course of therapy. The subjects took part in

therapy sessions three times per week for 18 weeks. An initial assessment consisted of an

audio-recorded phonological and language sample, a battery of standardized tests of

language comprehension and production (Preschool Language Scale. Revised

(Zimmerman et al., 1979), Peabody Picture Vocabulary Test. Revised. Form M (Dunn and

Dunn, 1981)), a hearing screening, oral mechanism examination and case history.

32

SI Summary

Over the period of the original study, SI was 5;10 to 6;4. During the initial

assessment it was noted that he had a tongue thrust during speech and swallowing, and a

finger sucking habit. SJ_ also had mild phonemic discrimination difficulties. SJ_ had

received varying dialectal input as a result of differences in parental dialect (Mother -

English, Father - Australian) and changing residency (Vancouver and Australia). S_l was

the middle of three boys, with the youngest having a mild phonological disorder. Hearing

was screened within normal limits at the beginning of the original project.

SI was initially assessed at the age of 2;9 for phonological difficulties. He

received intermittent therapy in Vancouver and Australia between the ages of 2;9 and 5;0.

Consonant clusters were not targeted during this time. At the beginning of Bernhardt's

(1990) study, SI evidenced frequent consonant cluster reduction. As word length

increased, the frequency of reduction increased. Consonant cluster matches for 'skeletal

slots' ranged from 42% for monosyllabic words to 0% for four syllable words. In cases

where a cluster match occurred for 'skeletal slots' for word-initial consonant clusters, no

segment matches were recorded for both consonants (i.e. there were no instances where

both phones were produced correctly). In several instances where two consonant

elements were realized (in terms of skeletal slots), an epenthetic vowel was transcribed.

33

S2 Summary

During the original study, S2 was 4;2 to 4;6 years of age. Facts associated with

his phonological disorder include: 1) a history of chronic otitis media from age 1 year

with a myringotomy for "glue ear" three months before the study (Hearing was screened

regularly throughout the study and was consistently found to be within normal limits.), 2)

a mild language production delay evidenced by copula/auxiliary BE omissions, pronominal

case errors and sibilant morphophonemes, and 3) a mild attention deficit which was

possibly related to his middle ear history.

S2 had partaken in three months of a group articulation therapy program prior to

the study. The main goal of these sessions had been stimulation of fricatives.

During the initial assessment for Bernhardt's (1990) study, S2's consonant cluster

development was not yet established at the syllabic level. Most consonant clusters were

realized as one element only. Those clusters produced with two elements in syllable initial

position included reduced or full epenthetic vowels.

APPARATUS AND PROCEDURES

For the original study, the children underwent an initial assessment protocol,

outlined above. The experimenter (E) used a standard set of objects and pictures as

stimuli for phonological sampling. Most responses were spontaneous single-word

elicitations. General conversation was also recorded during the session to provide other

words, some in connected speech contexts. Sarnpling probes were administered at the end

of each of the three six-week therapy cycles. The subjects' responses were recorded using

34

a Nagra IV reel-to-reel tape recorder and Ampex 631 tapes with an AKG D202

microphone in a speech/language therapy clinic.

The assessment and major probes were then transcribed phonetically (International

Phonetic Alphabet, 1979, plus diacritics) and orthographically by Bernhardt using a Revox

taperecorder and Videoconcepts F700 dynamic earphones.

For this study, the experimenter (E) transcribed stimulus words containing

consonant clusters from the original data of SI and S2. For SI, data was used from

probes administered at ages 5;10, 6;0, 6;2, and 6;4. Probes for S2 were administered at

the ages 4;2, 4;4, 4;6, and 4;9. Transcription was done independently by the E using a

Revox taperecorder and AKG K240 earphones. This transcription was then compared to

Bernhardt's (1990) original transcription. Discrepancies were resolved by E and

Bernhardt jointly hstening to the tape. Solely exarnining the consonant cluster unit, the

two transcribers were in agreement 88% prior to discussion and 98% after. Overall,

discrepancies in transcription were related to differences in the narrowness of

transcription, involving voicelessness and aspiration.

The consonant cluster data was then dubbed onto a Fuji FR-LlxPro60

audiocassette tape using a quality stereo audiocassette recorder (Marantz PMD 420) from

the original reel-to-reel tapes and a Revox taperecorder. When a stimulus word was

embedded in running speech, the entire utterance was dubbed onto the audiocassette tape.

The data were then converted from analog to digital. This high quality audio bandwidth

analog-to-digital signal conversion was done by hooking the audiocassette recorder

(Marantz PMD 420) up to a stereo audio/DSP port interface (Proport Model 656 dual-

channel analog 1/0 module) which was connected to a NeXT station monochrome

computer. The speech stimuli were recorded onto the NeXT station computer using

Soundworks 3.0 (version 2) application at a sampling rate of 22 kHz. Several stimulus

35

words were saved into one file. In total, there were 171 stimulus words for Sl and 119

stimulus words for S2. For both subjects, data from all four probes were pooled due to

the limited number of words containing word-initial consonant clusters in each sample. As

only one trend was observed over time for S2, (which was the beginning of a

coarticulatory effect between the glide (C2) and the epenthetic vowel), it was felt that the

decision to pool the stimuli was justified. During analysis, four words for Sl and eight

words for S2 were discarded because of distortion.

MEASUREMENTS

Using the visual and auditory playback of the digital signal in Soundworks 3.0

(version 2), (off the NeXT station computer), three separate measurements of duration

were made for each stimulus word: 1) the consonant cluster and syllable nucleus (CCV),

2) the consonant cluster (CC), and 3) the epenthetic vowel, when it occurred. Visual

playback included a signal waveform and a sound spectrum of the stimulus word. A

sound spectrum involves a description of the different frequencies found in a given sound

and is represented graphically with the vertical axis representing the amplitude of the

sound signal and the horizontal axis representing the component frequencies (Fry, 1979;

Borden and Harris, 1984).

As most of the stimulus words were in isolation, the onset of the C C V and CC

units was taken as the beginning of the speech waveform. Occasionally, during the initial

auditory and visual playback, extraneous noise was observed to precede the stimulus word

(e.g. click of the taperecorder being activated). The noise was represented as an

aperiodic, low amplitude waveform. In these few cases, onset of measurement was taken

36

to be where a marked rise in waveform amplitude occurred. When the stimulus word was

embedded in running speech, and the initial consonant of the cluster (Cl) was a voiceless

fricative, the onset of measurement was placed where a rise in ampHtude occurred in

conjunction with an marked increase in intensity of the waveform. When C1 was a

plosive, onset of measurement was placed at the point where the waveform rose in

amplitude, with the second wave peak being sUghtly higher than the first.

For the CCV unit, the offset of measurement was taken at the point where the

periodicity of the waveform changed from that of the vowel to the following consonant.

For the consonant cluster, C2 was either a semivowel (/r,l,w/), or a voiceless plosive. In

cases with a semivowel, the offset of measurement was signaled by a rise in amplitude and

a change in the periodicity or shape of the waveform. In instances where C2 was a

voiceless plosive, the measurement offset was placed at the end of the plosive release.

In cases where the consonant cluster was judged auditorily to contain an

epenthetic vowel, the vowel duration was measured. In general, the signal waveform of

the epenthetic vowel was distinct from the signal representation of the surrounding

consonants. When C l was a voiceless fricative, onset of the epenthetic vowel was marked

at the point where the waveform amplitude began to rise after decreasing at the end of

ffication. When C l was a stop, the onset of measurement was set at the end of the plosive

release. Determining the offset of measurement for the epenthetic vowel was the most

difficult. If the waveform was magnified, a distinct change in the character (amplitude,

intensity and periodicity) of the waveform could be observed. Offset of measurement was

placed at the point where the change occurred. It was observed that determining the

transition between the back rounded vowel (lul) and /w/ was the most difficult. These

criteria for measurement are consistent with those described elsewhere by Haggard (1973)

and Hawkins (1973).

37

While analyzing the data, E checked the accuracy of each of the duration

measurements made on the signal waveform by measuring the given segment on the

produced sound spectrum. E also checked the accuracy of duration measurements by

producing spectrograms of the consonant cluster and syllable nucleus portion of the

stimulus words using Sonogram 0.9, on the NeXT station computer. A spectrogram is a

graphic display of a speech event with frequency represented on the vertical axis and time

on the horizontal axis. Intensity is represented as relative darkness of the display (Fry,

1979; Borden and Harris, 1984). After launching Sonogram 0.9, a spectrogram of poor

quality was first produced, based on the program's default options. To produce

spectrograms of good quality with the same scales throughout the study, the spectrogram

parameters were consistently set each time Sonogram 0.9 was run. First, analysis method

was set to FFT, window shape was set to Hanning and peak picking was set to sideband

1000 Hz. Next, for the analysis resolution, window size was set at 128 points

(corresponding to a frequency resolution of 172.26 kHz), time increment was set to 62

points (corresponding to a time resolution of 2.90ms), and sampling frequency indicated

22.05 kHz. Occasionally, window size was set at 512 points so that a narrow band

spectrograph was produced. A narrow bandwidth produces a more sharply tuned

spectrogram. Generally, wide band spectrographic analysis is more useful than narrow

band analysis when studying speech, as a result of the short durations of many speech

events (Fry, 1979). For the display options, the upper limit of the frequency range was set

at 5 kHz and the dynamic range was set at -7.777dB to -40.806dB.

Measurement Reliability

To ensure that measurement criteria did not change over time, E remeasured the

segment durations of the first thirty-five stimulus words once data measurements were

38

complete, and found measurements to be in agreement within 5ms. In addition, random

measurements were made independently by a second person on fifteen percent of Si's

stimulus words, with measurements in agreement within 5ms.

39

CHAPTER THREE

RESULTS AND DISCUSSION

As discussed in Chapter One, there is no definite description or explanation of the

relationship between the underlying phonological representation and the final phonetic

representation of children's consonant clusters. The purpose of this paper has been to

look at one transformational process, epenthesis, and to examine its use in children's

consonant clusters from phonological and phonetic perspectives. Little information exists

on the effect of epenthesis on a consonant cluster unit. It is generally accepted that timing

constraints and neuromotor maturation may play a role in consonant cluster development

and impact on the occurrence of epenthesis (Gilbert and Purves, 1977; Smith, 1992).

However, because epenthesis affects the timing of the word and the consonant cluster

unit, it raises some interesting questions with respect to representation and syllable

structure. This study investigated the following questions:

1. Do consonant clusters produced with an epenthetic vowel differ in duration

from those without?

2. Is the epenthetic vowel in the consonant cluster consistent in length and

quality, or do co-articulatory effects occur?

3. Is the epenthetic vowel dependent in terms of duration on the phrasal context

or the duration of the syllable nucleus?

In an attempt to answer these questions, this study examined how consonant

clusters are represented using a nonlinear phonological framework and how two models of

language processing (Serial Model, Garrett, 1984; Parallel Interactive Model, Stemberger,

40

1985a and b) may account for the relationship between the underlying representation and

the final output.

The results of measurements of consonant cluster duration, epenthetic vowel

duration, and syllable nucleus duration will be reported. Their significance and what they

tell us about the relationship between the underlying phonology and the phonetic level will

be discussed below.

SUMMARY OF RESULTS

Occurrence of Epenthesis

During Bernhardt's original study, it was found that Sl used epenthesis in 77/166

(46.4%) of his consonant cluster productions while S2 used epenthesis in 67/111 (60.%)

of his consonant cluster productions. Bernhardt notes that prior to participating in the

original phonological intervention study, both subjects used epenthesis as a strategy for

producing consonant clusters. When consonant clusters were set as an intervention target,

Bernhardt (1990) used epenthesis as a means to create pronounceable onset conditions

(i.e. CVC for CC) which resulted in an increase in the use of epenthesis in consonant

cluster productions for both subjects.

41

Consonant Cluster Duration

Do consonant clusters with an epenthetic vowel differ in duration from those without?

One aim of this study was to look at epenthesis in consonant cluster productions to

see how it related to the timing of the consonant cluster unit. The duration of consonant

cluster units with and without an epenthetic vowel was measured. For Si, the duration of

consonant clusters without epenthesis ranged from 46ms to 696ms (x =223ms, 5=145ms)

while the duration of consonant clusters with an epenthetic vowel ranged from 61ms to

1017ms ( x =332ms, s=163ms). For S2, the duration of consonant clusters without

epenthesis ranged from 300ms to 2704ms (x =712ms, 5=611ms) while the duration of

consonant clusters with an epenthetic vowel ranged from 190ms to 1040ms ( x =348ms,

s= 129ms) (see Figure 5). It must be noted that S2 had few consonant clusters without an

epenthetic vowel where both elements were realized. A Mann-Whitney test was used to

determine that a significant difference in duration exists between those consonant clusters

with an epenthetic vowel and those without for both subjects. Test results are

summarized in Table 1.

Table 1: Summary of Mann-Whitney test for difference in duration between consonant clusters without and with epenthesis

Subject Mann-Whitney Number of CC Number of CC p-value Statistic (U) without epenthesis with epenthesis

1 1475.5 72 77 .0000

2 193.0 18 67 .0000

42

S 1 : consonant clusters without epenthesis

0 I 1 1 1 1 1

0 200 400 600 800 1000

duration (ms)

S 2 : consonant clusters without epenthesis

00 -

I 1 1 1 1 1

0 200 400 600 800 1000

duration (ms)

S 1 : consonant clusters with epenthesis

1 1 1 1 1 1

0 200 400 600 800 1000

duration (ms)

S 2 : consonant clusters with epenthesis

° 1 1 1 1 1 1

0 200 400 600 800 1000

duration (ms)

Figure 5: Duration of consonant clusters without and with epenthesis for Subjects 1 and 2

(Note: For "S2: consonant clusters without epenthesis", two outlying durations of 1844ms and 2704ms, where S2 purposefully lengthened the CC, were omitted to improve scale comparison.)

43

Epenthesis as a Strategy for Overcoming Timing Demands

If, as Gilbert and Purves (1977) suggest, epenthesis is used as a strategy for

applying appropriate segmentation and overcoming timing demands, those consonant

clusters containing an epenthetic vowel should be longer in duration, (assuming that both

consonants are represented at the underlying level). As reported above, in the case of Sl

and S2, clusters containing an epenthetic vowel were longer in duration than consonant

clusters produced without epenthesis. Thus, epenthesis appears to allow for the

exaggeration of clustered features, so that both phonemes can be realized at the

production stage. Gilbert and Purves (1977) suggest that at this stage, Ohala's (1970)

articulation dorninant system might better explain consonant cluster production. An

alternate explanation would be if the timing system and articulatory system were co-

occurring at separate levels of production (Ohala, 1970). Sl and S2 could be using

epenthesis as a strategy to achieve correct articulatory segmentation while working within

the constraints of their own timing systems (i.e. the timing systems may not yet match that

of the adult model). Although one system may be more dominant than the other, at

differing stages of the consonant cluster development, the second system is still active.

Effect of Phonological Context on the Epenthetic Vowel

In consonant clusters with an epenthetic vowel, is the epenthetic vowel consistent in

length and quality or is it affected by the surrounding consonants of the cluster?

When discussing the relevance of the application of phonological theory in the first

chapter, the importance of exarnining the phonological structure and context of a given

unit was raised. It was stated that when examining a specific phonological unit it is

44

necessary to consider the influence that the phonological context may have on the unit's

phonetic realization. In looking at consonant cluster productions of both subjects, it was

noted that epenthesis occurred only with consonant clusters where C2 was either /Lwj/.

However, in looking closely at the data, an interesting difference was seen between the

two subjects. For SI, when epenthesis occurred, if C2 was HI, the epenthetic vowel was

usually IQI, if C2 was /w/, it was /u/, and if C2 was /j/, it was usually / i / . For S2, when

epenthesis occurred, the vowel was almost always lal regardless of what C2 was.

Occasionally, when C2 was /w/, S2 inserted lul between the consonant elements. These

instances were found on the last probe when S2 was 4;6.

Coarticulatory Effects

In the discussion of timing dominant and articulation dominant systems, it was

stated that evidence of an articulation dominant system can be found in co-articulation

studies. It appears that for SI, there is a co-articulatory effect between the epenthetic

vowel and C2 which may suggest that the articulation system may be more highly

activated (i.e. there appears to be some linking of articulatory gestures as the epenthetic

vowel seems to function as a transition between CI and C2). It is interesting to note that

at the beginning of the original study, SJ. was observed to have mild oral motor difficulties

(a tongue thrust during speech).

For S2, his timing system may have more influence because the epenthetic vowel

appears to be functioning more as a neutral marker to help overcome timing demands.

Support for the hypothesis that S2 may be attending more to timing demands may also be

seen by examining some of his single element realizations of consonant clusters. It was

occasionally observed that S2 lengthened the duration of the singleton consonant to match

perceptually that of similar two-element consonant cluster realizations. It may be that S2

45

was attaching both tirning units to the single segment. In later consonant cluster

development where some co-articulation was noted between C2 and the epenthetic vowel,

the articulation system may be becoming more active.

Epenthetic Vowel Duration

Where an epenthetic vowel exists within a consonant cluster, is it dependent in terms of

length on the phrasal context or the duration of the syllable nucleus?

In considering epenthesis in consonant clusters, the duration of the epenthetic

vowel was also measured. In addition to the variations in the quality of the epenthetic

vowel that have been discussed, differences were seen in vowel length. For SI, the

duration of the epenthetic vowel ranged from 27ms to 264ms. For S2, the epenthetic

vowel duration ranged from 26ms to 237ms. In order to examine the possible influence of

the duration of the syllable nucleus on the duration of the epenthetic vowel, correlations

were calculated. Correlation coefficients were not significant. For SI, r = 0.293 and for

S2, r = 0.178. No correspondence was seen between the length (see Figure 6) or quality

of the underlying syllable nucleus and the length of the epenthetic vowel. It would seem

then that the epenthetic vowel is independent of the underlying word structure and is part

of the consonant cluster unit itself. Further, the consonant cluster appears to be governed

by its own timing system or constraints, separate from the timing of the word.

46

5 I

S1: Scatterplot of epenthetic vowel duration vs . syllable nucleus duration (r = 0.293)

ID

o

8

0.4 0.6

syllable nucleus duration (ms)

S2: Scatterplot of epenthetic vowel duration vs. syllable nucleus duration (r=0.178)

•E

0.2 0.4 0.6 0.8 1.0

syllable nucleus duration (ms)

Figure 6: Relationship between syllable nucleus duration and epenthetic vowel duration for Subjects 1 and 2.

47

IMPACT OF EPENTHESIS ON PROSODIC STRUCTURE

It is important to look farther at the change in the prosodic tier representation of a

word containing a consonant cluster when epenthesis occurs. As stated in the earlier

discussion of nonlinear phonology, information about prosodic structure is included in the

underlying representation. The prosodic tier is linked to the segmental tier according to

principles of association (Bernhardt, 1992). At the prosodic level, the word dorninates

feet which in turn dominate syllables. Different theories about the prosodic hierarchy exist

with regard to the status of the syllable as a constituent and its components.

Representation of Consonant Clusters

When looking at consonant clusters from the perspective of onset-rime theory,

both consonants branch from the onset and the vowel makes up the nucleus which

branches from the rime. In moraic representation, the two consonants are adjoined to the

mora (the actual point of joining is controversial) and the vowel is dominated by the mora.

If epenthesis occurs during consonant cluster production, the consonant cluster unit is

split, altering the syllable structure of the word (i.e. resyllabification occurs). In onset-

rime terms, the consonants are separated into two onset-rime units which are dominated

by a branching foot. In moraic terms, an extra weight unit (mora) is created. The two

moras create two syllables which are dominated by a branching foot (see Figure 7). In

both cases, the prosodic structure of the word is altered to allow the child to cope with the

articulatory and timing demands of consonant cluster production. However, as previously

discussed, the data suggest that the consonant cluster is governed by its own timing

constraints separate from the timing of the word. With these current theories, as shown in

Figure 7, there is no way representationalfy to attribute unitary status to C1 V eC2, unless

48

Moraic O-R

a

A A C M C M

A O R

A O R

C V e C V

Figure 7: Representation of epenthesis in moraic and onset-rime theories

Word

A, C,0 M,R C M,R

Figure 8: Representation of Cl V e C 2 as one unit

C 2 is represented as both a coda of the weak syllable and an onset to the strong (see

Figure 8). Although this is unusual, there are examples in English where intervocalic

consonants may be ambisyllabic (e.g. Dd in 'bucket' Ak^t/) (Bernhardt and Stemberge:

49

in preparation), although not usually in weak-strong syllable sequences (e.g. /behxa/, IM is

an onset). Furthermore, in English, weak syllables can have a coda. It is important to

note that even though it is unusual to think of glides as codas (i.e. /kaw/ = CVC), it is one

way to describe diphthongs ([aw], [ay] rather than [au ], [ai]). This issue needs to be

examined in much further detail before it can be resolved.

APPLICATION OF NONLINEAR PHONOLOGICAL FRAMEWORK

The impact of examining consonant cluster productions using a nonlinear

phonological framework is that it allows for reference to the underlying phonological

system. The main purpose of this paper was to look at epenthesis in consonant clusters

from phonological and phonetic perspectives. As stated, by applying the nonlinear

phonological theory to describe the underlying representation of consonant clusters, an

opportunity arises for describing how the underlying representation changes before

reaching the production stage.

In Chapter One, evidence was discussed to support the claim that both elements of

a consonant cluster are present in the underlying representation. It was shown earlier in

this chapter how this would be represented using a nonlinear phonological framework.

The next step then is to look at how the two-element consonant cluster in the underlying

representation is transformed into its phonetic form, especially when it includes an

epenthetic vowel. To look further at this, it is necessary to refer to the language

processing models described in Chapter One.

50

Epenthesis in a Serial Model of Language Processing

In Garrett's (1980, 1984) model of language processing, three distinct levels of

processing are described: the message level, the linguistic sentence level, and a motor

articulatory model. According to Garrett, information is passed linearly from one level to

the next. The first level where an epenthesis rule could be applied would be at the

sentence level where the general concept of the message undergoes the application of

logical and phonological rules. The other option would be for an epenthesis rule to be

applied at the final level, the articulatory level, where the signal undergoes phonetic and

prosodic encoding before being translated into instructions for the correct articulatory

sequencing.

If the epenthetic rule was applied at the sentence level, it means that it is a higher-

level process and thus associated closely with the underlying representation of the

consonant cluster. As stated in Chapter One, if this is the case, it would be expected that

phrasal context or the duration of the syllable nucleus would exert some influence on the

epenthetic vowel. As discussed, no evidence of this was seen in the consonant cluster

productions of the two subjects. Neither the length or quality of the syllable nucleus had

any effect on the duration or quality of the epenthetic vowel. Support for this is found in

the study by Peterson and Lebiste (1960) which looked at the effect of preceding and

following consonants on the duration of the syllable nucleus. They found that the syllable

nucleus was only influenced by the nature of the following consonants. This suggests that

if it is the following consonant that has the effect on duration of the syllable nucleus, the

consonant cluster may be separate from the word in its timing. Therefore, no anticipatory

lengthening would be expected in the epenthetic vowel.

It seems more likely that the epenthesis is occurring at the articulatory level when

the underlying signal undergoes phonetic encoding and articulatory sequencing. This is

51

supported by evidence from adults who pronounce foreign words with unfarniliar

consonant cluster combinations with an epenthetic vowel (Greenlee, 1974). However, as

discussed in Chapter One, it appears that the application of epenthesis cannot simply be

occurring at the time of the actual articulatory sequencing; otherwise it should be seen

more frequently in children's consonant cluster productions.

It is important to remember that according to Garrett's model, the occurrence of

the different processing levels and application of rules is a fixed, serial process. Therefore,

the epenthesis rule would always be applied at the same level during phonetic encoding

and should produce consistent results. However, an interesting difference was seen in the

use of epenthesis by the two subjects that cannot easily be account for by a linear model.

Epenthesis in a Parallel Interactive Model

In Chapter One, the possibility of a more interactive model of language processing

was introduced. Proponents of a parallel interactive model suggest that varying levels of

processing can be active at any given moment in time and that information interacts top

down and bottom up (Stemberger, 1985b). Initial processing begins when the speaker

forms an intent to talk thus activating pragmatic and semantic information. Each time a

new unit or level of processing is activated, a cascading effect of activation occurs.

Although phonemes are still considered lower-level units, as in Garrett's model, they are

not restricted to a final stage of processing. Once associated phonemes have been

activated, there can be a spreading of activation back up the processing levels (e.g. other

words containing the particular phonemes activated may then be partially activated).

Eventually the units that are most highly activated will be chosen and the correct

articulatory sequencing will occur.

52

When looking at how epenthesis in consonant cluster productions would be

accounted for in terms of a parallel interactive model, several factors need to be

considered. First, although an assumption of the model is that language is made up of

interconnected levels which are processed in parallel, this does not imply that processing

begins simultaneously at all levels. Second, within the parallel interactive model, rules can

be viewed as being units similar to any other unit. Stemberger states "structural

conditions on a rule are simply the set of input lines from other language units, as those

other units become activated they send activation to the phonological rule, and it also

begins to rise in activation level" (1985a, p.9). This means that the model assumes that

phonological information leads directly to the accessing of relevant rules. If the activation

level of a rule is high enough, the underlying information or representation is inhibited

before its activation level can become high enough for it to be accessed. Stemberger

(1985a) points out that this type of model was not developed to describe phonological rule

order. As phonological rules are equated with other units, they are treated as such and

therefore, said to be accessed and applied in the same manner as other units (e.g. words,

segments).

Since epenthesis is a phonological rule, it can be assumed that it occurs in the

mapping from segments to features (Stemberger, 1985a) which occurs at a lower

processing level. However, in contrast to a linear model, with a parallel interactive model,

the possibility exists that the application of a rule does not occur at the exact same point of

processing for different speakers. That is, the threshold point of activation which causes

the rule to be accessed may be reached more quickly in some speakers than in others.

Therefore, although the application of epenthesis, a phonological rule, will occur at a low-

level of processing, there may be varying points of application.

53

It was noted earlier that an interesting difference was seen in the surface realization

of the epenthetic vowel in the consonant cluster productions of the two subjects. Greater

coarticulatory effects were seen between the epenthetic vowel and C2 for Sl than for S2.

For S2, the epenthetic vowel appeared to function more as a neutral place marker.

In looking at Garrett's model of language processing, it was seen that due to its

linear nature, rule application occurs in a fixed serial process when specific requirement

conditions are met. Therefore, the application of epenthesis should be consistent between

speakers (i.e. the surface realizations should be phonetically similar) This does not mean

that a serial model would not be able to account for the differences between the two

subjects. However, it appears that detailed conditions would need to be outlined in order

to explain the production differences. In contrast, in a parallel interactive model of

language processing, rules function like any other unit and can, therefore, be activated at

various points in the language production process. The possibility exists that for Sl, the

application of epenthesis is activated just before articulatory sequencing but after feature

activation for C2 (hence the coarticulatory effects) whereas for S2, activation of

epenthesis may occur slightly earlier (for example, at the stage of mapping phonemes to

the CC prosodic structure).

STUDY LIMITATIONS

It is clear that this study has a number of limitations in terms of sample size, data

quality and measurement techniques. The small number of subjects limits generalizations

that can be made across children. Furthermore, only a relatively small sample of words

containing word-initial consonant clusters existed for both subjects. The sample size of

54

consonant clusters was particularly restricted for S2 since many of his consonant clusters

were realized as one element. Another limitation was that the data were not originally

collected for purposes of this study. This project arose from observations made during

Bernhardt's (1990) study. In terms of data quality, all measurements were made from

data which was transferred from original reel-to-reel tapes to audiocassette. Several

stimulus words needed to be discarded because of poor tape quality. Lastly, an essential

limitation in the measurement of segment durations was that of segmentation.

Historically, segmentation has been a major problem in speech analysis. Although there

are instances where cues signaling the onset and offset of a segment are relatively

unambiguous, there are many cases where it is difficult to determine accurately the

segmentation point. This was particularly true when determining the segmentation point

between an epenthetic vowel and C2 glide. It must be pointed out that although

segmentation criteria, as outlined in Chapter Two, were applied consistently, and reUabihty

calculated, measurement was still dependent on human judgment of speech characteristics.

FUTURE RESEARCH

The present study shows that epenthesis does occur in some children's productions

of consonant clusters and that these consonant cluster units differ in terms of length from

consonant clusters without an epenthetic vowel. Due to the small number of subjects

considered, it should serve as a preliminary effort in examining epenthesis in children's

consonant cluster productions. Further research is warranted to examine how widely

occurring epenthesis is in children's productions and what the acoustic and perceptual

characteristics of epenthesis are. Further acoustical analysis needs to be done to determine

55

if the frequency of epenthesis is being under-reported as a result of the epenthetic vowel

being considered as part of the following consonant.

Theory development is necessary in order to account better for the representation

of consonant cluster units containing an epenthetic vowel in terms of syllable structure.

Neither of the current theories, onset-rime or moraic, can satisfactorily account for a

consonant cluster with an epenthetic vowel as one unit, unless the second consonant is

considered ambisyllabic (not an optimum solution in a weak-strong syllable sequence).

CONCLUSION

This paper has investigated the occurrence of epenthesis in consonant cluster

productions of two subjects from both phonological and phonetic perspectives. In

Chapter One, previous research was examined and evidence supporting the hypothesis that

children have a two-element underlying representation for consonant clusters was

discussed. A current nonlinear phonological framework and two current models of

language processing were also reviewed.

Results of this study showed that consonant clusters produced with an epenthetic

vowel were significantly longer in duration than those produced without. This supports

the hypothesis presented by Gilbert and Purves (1970) that epenthesis is used as a strategy

for applying appropriate segmentation and overcoming timing demands. It was also found

that coarticulation occurred between the epenthetic vowel and C2 but not between the

epenthetic vowel and the syllable nucleus which suggests that the epenthetic vowel is part

of the consonant cluster unit and governed by its own timing system.

56

Another aspect of epenthesis in consonant clusters that was examined was the

effect of the epenthetic vowel on the prosodic structure of the word. It was shown, using

the nonlinear phonological framework discussed, how epenthesis in a consonant cluster

unit would be represented. Two current theories of the structure of the prosodic

hierarchy, onset-rime and moraic, were presented and it was shown that resyUabification

of the word occurs as a result of epenthesis in the consonant cluster. However, it is

important to note that the results of the study indicated that the consonant cluster unit is

governed by its own timing constraints, separate from the timing of the word. Taking this

into consideration, it was shown that in terms of both onset-rime theory and moraic

theory, there is no way to attribute unitary status to ClV e C2 unless C2 is ambisyllabic.

Further research will need to be conducted in this area to find a more optimum solution.

57

REFERENCES

Allen, George, D. (1973). Segmental tirning control in speech production. Journal of Phonetics. 1, 219-237.

Barton, D., Miller, R_, and Macken, M.A. (1980). Do children treat clusters as one unit or two? To appear in Papers and Reports on Child Language Development, 18, 105-137.

Bates,E., & MacWhinney, B. (1989). Functionalism and the competition model. InB. MacWmnney & E. Bates (Eds.). The Crosslinguistic Study of Sentence Processing. NY: Cambridge University Press, 3-73.

Bernhardt, B. (1990). Application of nonlinear phonological theory to intervention with six phonologicalry disordered children. (Unpublished PhD thesis, University of British Columbia).

Bernhardt, B. (1992). Developmental implications of nonlinear phonological theory. Clinical Linguistics & Phonetics. 6(4). 259-281.

Bernhardt, B. & Stemberger, J. (in progress). Nonlinear Phonology and Child Phonological Development.

Borden, G. I, & Harris, K S. (1984). Speech Science Primer: Physiology. Acoustics, and Perception of Speech. Baltimore, MD: Williams & Wilkins.

Chin, S.B., and Dinnsen, D.A. (1990). Consonant clusters in disordered speech: Correspondence strategies and constraints. Presented at the Fifteenth Boston Umversity Conference on Language Development.

Chin, S.B., and Dinnsen, D.A. (1992). Consonant clusters in disordered speech: Constraints and correspondence patterns. Journal of Child Language. 19, 259-285.

Daniloff, K G , and Hammarberg, RE. (1973). On defining coarticulation. Journal of Phonetics. 1, 239-248.

Dinnsen, D.A., Chin, S.B., Elbert, M., and PowelL T.W. (1990). Some constraints on functionally disordered phonologies: Phonetic inventories and phonotactics. Journal of Speech and Hearing Research. 33, 28-37.

58

Dunn, L.M., &Duiui, L.M. (1981). Peabody Picture Vocabulary Test. (Revised). Circle Pines, MN: American Guidance Service.

Edwards, M L . & Shriberg, L. (1983). Phonology: Applications in Communicative Disorders. San Diego, CA: College-Hill Press.

Fry, DB. (1979). The Physics of Speech. Cambridge: Cambridge University Press.

Garrett, M.F. (1980). Levels of processing in sentence production. In B. Butterworth (Ed.). Language Production. NY: Academic Press, 177-219.

Garrett, M. F. (1984). The organization of processing structure for language production: applications to aphasic speech. In D. Caplan, A.R Lecours, and A. Smith (Eds.). Biological Perspectives on Language. Cambridge, Mass: M.I. T., 172-193.

Gilbert, J.H.V., and Purves, B.A. (1977). Temporal constraints on consonant clusters in child speech production. Journal of Child Language, 4, 417-432.

Greenlee, M. (1974). Interacting processes in the child's acquisition of stop-liquid clusters. Papers and Reports on Child Language Development (Stanford University), 7, 85-100.

Grunwell P. (1985). Comment on the terms "phonetics" and "phonology" as applied in the investigation of speech disorders. British Journal of Disorders of Communication, 20, 165-170.

Haggard, M. P. (1973). Correlations between successive segment durations: values in clusters. Journal of Phonetics. 1. Ill- 116.

Hawkins, Sarah. (1973). Temporal coordination of consonants in the speech of children: preliminary data. Journal of Phonetics, 1(3), 181-218.

Hewlett, Nigel. (1985). Phonological versus phonetic disorders: some suggested modifications to the current use of the distinction. British Journal of Disorders of Communication. 20, 155-164.

Ingram, D. (1976). Current issues in child phonology. In D.M. Morehead and A.E. Morehead (Eds.). Normal and Deficient Child Language. Baltimore, Maryland: University Park Press.

Kornfeld, J.R (1976). Implications of studying reduced consonant clusters in normal and abnormal child speech. Presented at the Psychology of Language Conference, University of Stirling.

59

Kozhevnikov, V.A. & Chistovich, L.A. (1965). Speech: articulation and perception. Washington, D.C.: Joint Publications Research Service.

Laver, J. (1994). Principles of Phonetics. Cambridge: Cambridge University Press, 26-54.

Menyuk, P. (1972). Clusters as single underlying consonants: evidence from children's productions. In A. Rigault & K Charbonneau (Eds.). Proceedings of the Seventh International Congress of Phonetic Sciences. The Hague: Mouton, 1161-1165.

Menyuk, P., and Klatt, M. (1975). Voice onset time in consonant cluster production by children and adults. Journal of Child Language, 2, 223-231.

Ohala, J. (1970). Aspects of the control and productions of speech. U.C.L.A. Working Papers in Phonetics. 15.

Ohrnan, S.E.G. (1966). Coarticulation in VCV utterances: spectrographic measurements. Journal of the Acoustical Society of America. 39. 151-168.

Peterson, G.E. & Lehiste, I. (1969). Duration of syllable nuclei in English. Journal of the Acoustical Society of America. 32, 693-703.

Purves, B.A. (1976). Temporal aspects of children's production of consonant clusters. (Unpublished Masters thesis, University of British Columbia).

Roach, P. (1983). English Phonetics and Phonology: A Practical Course. Cambridge: Cambridge University Press.

Sloat, C , Taylor, S.H., & Hoard, J.E. (1978). Introduction to Phonology. Englewood Cliffs, NJ: Prentice-FIalL, Inc.

Smith, B.L., Sugarman, M.D., & Long, S.H. (1983). Experimental manipulation of speaking rate for studying temporal variability in children's speech. Journal of the Acoustical Society of America. 74, 744-749.

Smith, B.L. (1992). Relationships between duration and temporal varkbility in children's speech. Journal of the Acoustical Society of America. 91. 2165-2174.

Stemberger, J.P. (1985a). Phonological rule ordering in a model of language production. Proceedings of the IOP Regional Conference on Linguistics. Bloomington: Indiana University Linguistics Club.

60

Stemberger, J.P. (1985b). An interactive activation model of language production. In Andrew W. Ellis (Ed.). Progress in the Psychology of Language, vol. 1. Hillsdale, NJ: Lawrence Erlbaum Associates Ltd., 143-183.

Stemberger, J.P., and Treiman, R (1986). The internal structure of word-initial consonant clusters. Journal of Memory and Language, 25, 163-180.

Stoel-Gammon, C , & Cooper, J.A. (1984). Patterns of early lexical and phonological development. Journal of Child Language, 11, 247-271.

Stoel-Gammon, C , and Dunn, C. (1985). Normal and Disordered Phonology in Children. Baltimore: University Park Press.

Templin, M. C. (1957). Certain Language Skills in Children. Minneapolis: University of Minnesota Press.

Tingley, B.M., & Allen, G.D. (1975). Development of speech timing control in children. Child Development. 46, 186-194.

Zimmerman, I., Steiner, V., & Pond, R. (1979). Preschool Language Scale 2 -(Revised). Columbus, OH: Charles Merrill PubHshing Co.

61

APPENDIX ONE

Sl DATA

TestList ccv/cv duration cc duration v duration j c duration ep. v duration

sleeping f(w)ipin 0.254 0.155 0.1 oj 0.000 no epenthesis

bread bwAd(9}__ 0.335 0.205 0.13] OJpTjOj no epenthesis

snowman no(3)m£en(s) 0.306 0.000 0.31 0.112 no epenthesis

snowing nowen 0.447 0.000 0.45 0.202 no epenthesis

brush V W A 8 0.305 0.105 0.20 0.000 no epenthesis

airplane eupein 0.361 0.000 0.36 0.112 no epenthesis no epenthesis broken b(w oukm 0.274 0.118 0.16 0.000 no epenthesis no epenthesis

glove 0.207 0.115 0.09 0.000 no epenthesis

glovey g A v i ( ? ) 0.232 0.000 0.23 0.090 no epenthesis

grey gawei(j) 0.000 0.000 0.00 o.ooot plum p/bAm 0.166 0.000 0.17 0.085 no epenthesis

brushing bA(s)9m 0.000 0.000 0.00 0.000 snake (s)n:eik 0.783 0.452 0.33 0.000 no epenthesis

truck f W A k 0.777 0.394 0.38 0.000 no_epenthesis

trailer vweiio 0.274 0.129 0.14 0.000 no epenthesis

soring fvwin 0.386 0.249 0.14 o.ooo' no epenthesis

clown q(i)aun 0.531 0.133 0.40 0.000 no epenthesis

sleep _^ fwip 0.287 0.077 0.21 0.000 no epenthesis

three fwi(i) 0.470 0.086 0.38 0.000 noepenthesis

flower 0.386 0.000 0.39 0.016 no epenthesis

c r a y r a _ _ crayon

kuweijo£g)n 0.389 0.288 0.10 °5go 0.033

" 108 c r a y r a _ _ crayon keiwAn 0.250 0.000 0.25

°5go 0.033 no epenthesis

screwdriver f/f(w)uv(w)aiVA 0.279 0.049 0.23 0.000 no epenthesis

crayons fweipnQ 0.284 0.153 0.13 0.000 n^epjnthesis

black bask 0.452 0.000 0.45 0.027 no epenthesis

twenty fwenti(i) 0.193 0.150 0.04 0.000 no epanrthesis

cry fwai 0.516 0.060 0.46 0.000 no epenthesis

tried vwaU 0.425 0.148 0.28 0.000 Ino epenthesis

train dswem(9) 0617 0.270 0.35 0.000! 0.138 brothers b(w)ASa9 0.180 0.100 0.08 0.000 no epenthesis

twin dswm 0.415 0.241 0.17 0.000 0.120

blue b(i)iu 0.465 0.046 0.42 0.000 no epenthesis

blue fbwu I 0.334 0.0861 0.25 I 0.000 no epenthesis

62

twins t(a)wmz 0.556 0.229 0.33 O.OOOl 0.035

obstacle obsOika 0.407 0.217 0.19 O.OOOmo epenthesis

quiet k(a)waiAt 0.725 0.343 0.38 O.OOO) 0.060

glove q(i)JAf 0.476 0.153 0.32 O.OOOl 0.042

gloves g(j)Avz 0.355 0.088 0.27 O.OOOjno epenthesis

glove _ dress

gAy 0.207 0.000 0.21 0.021 \no epenthesis glove _ dress fweG 0.316 0.162 0.15 0.000|no epenthesis

string t(u)wm 0.449 0.193 0.26 0.000 0.052

spring pwerj 0.378 0.228 0.15 0.000 no epenthesis

play_ pwei 0.410 0.206 0.20 0.000 no epenthesis

break bwei 0.614 0.117 0.50 0.000 no epenthesis

screwdriver f(w)uvwaiv(w)a(a; 0.585 0.300 0.29 0.000 no epenthesis

green gwin 0.478 0.192 0.29 0.000 no epenthesis

bring vwirj 0.266 0.110 0.16 0.000 no epenthesis

plane pwecn 0.600 0.2071 0.39 0.000 no epenthesis no epenthesis airplane e(o)pwei:n 0.636 0.298 0.34 0.000

no epenthesis no epenthesis

truck dowAk 0.774 0.427 0.35 0.000 0.231

plum p/bwAm 0.579 0.217 0.36 0.000 no eg«nthesis

twenty fwenti 0.295 0.165 0.13 0.000 nojgpenthesis

black b(3)wak 0.441 0.224 0.22 0.000 0.042

queen k/g(a)w:in 0.946 0.392 0.55 0.000 0.047

broke bwok 0.304 0.110 0.19 O.OOOjno epenthesis

crown kwaun 0.469 0.29 O.OOOfno epenthesis

quick k/g(u)wik 0.372 0.157 0.22 0.000 0.050

flower fauwa 0.507 0.000 0.51 0.102 no epenthesis

grape 9 3 w e l L _ _ _ _ _ _ - 0.555 0.230 0.33 0.000 0.146

brush brush

bWAS 0.439 0.224 0.21 0.000 no erjenthesis brush brush DO:WA9 0.615 0.409 0.21 0.000 0.238

sleeping Gajipen 0.636 0.250 0.39 0.000 0.104

sleep (e)tjip 0.745 0.532 0.21 0.000 no epenthesis

snowman snoiwmsn 1.162 0.476 0.69 0.000 no epenthesis

snowing snoiwin 1.207 0.440 0.77 0.000 no epenthesis

squirrel f.oijau 0.333 0.000 0.33 0.060 no epenthesis

snake 0.703 0.256 0.45 0.000 no epenthesis

dropped fwopt 0.175 0.000 0.17 0.046 no epenthesis no epenthesis growl qwau:(w) 1.170 0.251 0.92 0.000 no epenthesis no epenthesis

cry k/gwec 1.334 0.249 1.09 0.000 no epenthesis

cry kwei 1.064 0.126 0.94 0.000 no epenthesis

throw fwo 0.558 0.285 0.27 0.000 no epenthesis

throwing fowin 0.268 0.000 0.27 0.081 no epenthesis

twins t(u)win5 0.489 0.321 0.17 0.000 0.061

gross guwos: 0.711 0.390 0.32 0.0001 0.168

elovev aiAvi 0.385 0.250 0.13 O.OOOlno enenthesis

63

brow b£u)wais 1.075 0.304 0.77 O.OOOl 0.091

sis. g(s)wm 0.306 0.189 0.12 O.OOOl _ _ P i > ± 3 0.065 sweater th(u)weio(?) 0.371 0.269 0.10 o.oool

_ _ P i > ± 3 0.065

sweater 0t(u)weJa 0.523 0.387 0.14 O.OOOl 0.045

sweater tuweJo(s) 0.445 0.341 0.10 O.OOOs 0.111

sweater 6t(u)web(3) 0.520 0.297 0.22 0.000 0.068

sweater GweJo/a 0.490 0.369 0.12 0.000 0.048

sweater 0uweJo(3) 0.486 0.364 (J 1 J 0.000 0.114

plane plein 0.287 0.087 0.20 0.000 no ep nti esis

airplane 0.824 0.401 0.42 0.000 0.037

crayon k(u)weijan 0.454 0.254 0.20 0.000 0.076

clock kh(u)bkh 0.645 0.337 0.31 0.000 0.027

brush bWA0 0.204 0.125 0.08 0.000 no epenthesis

brush brush

D ( U ) W A J 0.433 0.241 0.19 0.000 0.079 brush brush D ( O J W A8 0.345 0.160 0.19 0.000 0.048

brush M U ) W A | 0.269 0.146 0.12 0.000 0.043

clock k(u)bkh 0.345 0.180 0.16 0.000 0.034

three fuwi: 1.403 0.498 0.91 0.000 0.239

transformer twsenSfoms 0.172 0.074 0.10 0.000 no epenthesis

microscope mAkw30kouph 0.111 0.048 0.06 0.000 no epenthesis

bread buwed(s) 0.721 0.458 0.26 0.000 0.194

bread 0.488 0.223 0.27 0.000 0.075

quarter khozdo(a) 1.171 o.ow 1.17 0.076 n^pe^thesis 0.219 black balak 0.628 0.331 0.30 0.000

n^pe^thesis 0.219

throwing fuwo(w)m 0.779 0.601 0.18 0.000 0.264

throwing fuwowm 0.000 0.000 0.00 0.000

throw f(u)wo 0.603 0.438 0.16 0.000 0 054

play p(9)lei 0.524 0.281 0.24 0.000 0.045

dressed duwest 0.428 0.324 0.10 0.000 0.150

plum pgUm 0.526 0.294 0.23 0.000 O.044

grows gwouS 0.861 0.432 0.43 0.000 no epenthesis

quick khuwik 0.597 0.331 0.27' 0.000 0.175

broke bwok 0.624 0.255 0.37 0.000 no epjenthesis

crown k(u)waun 0.556 0.224 0.33 0.000 0.044

queen kwin 0.698 0.221 0.48 0.000 no epenthesis

great gweit 0.205 0.109 0.10 0.000 no epenthesis

trample th(3)waempy 0.478 0.252 0.23 0.0001 0.041

glove q(3)lAV 0.353 0.195 0.16 0.000 0.051

glove g(3)Uv 0.306 0.164 0.14 0.000 0.061

glove g(3)Uv 0.356 0.195 0.16 0.000 0.045

dress d(ojwe0 0.368 0.223 0.15 0.000 0.053

dress d(u)wa9 0.485 0.273 0.21 0.000 0.093

truck dru")WAkh 0.423 0.266 0.16 0.000 0.062

64

screwdriver |t/d(u)wudwaivo(a) drive |d(a)waiv

0.610 0.297 0.31 o.oogj 0.050

screwdriver |t/d(u)wudwaivo(a) drive |d(a)waiv 0.441 0.170 0.27 0.0001 0.050

screw t(s)wu 0.677 0.261 0.42 0.000 0.044

brush b(u)wAj 0.460 0.318 0.14 0.000 0.045

green guwi:n 1.211 0.553 0.66 0.000 0.211

green ffijwjn___ 0.835 0.629 0.21 0.000 0.179

flower f(9)lawo(s) 0.523 0.178 0.34 0.000 0.037

snowman t(s)nomaen 0.892 0.384 0.51 0.000 noepenthesis

snowing s:nowen 0.842 6.696I 0.1? 0.000 no epenthesis

broom b^wj^n^ 0.795 0.349 0.45. 0.000 0.037

sleeping 03liphen 0.807 0.479 0.33! 0.000 0.064

sleep Oslip 0.000 0.000 0.00* 0.000 0.000 snake sneik 0.715 0.408 0.31: 0.000 0.000 no epenthesis

black blaek 0.267 0.106 0.161 0.000 no epenthesis

black b(a/u)laek 0.248 0.061 0.1 gl o.ooo 0.019

stuck s(9)Ak 0.463 0.292 0.17f 0.000 no epenthesis

E ^ i £ ! L _ _ _ _ p(a)lei6 0.542 0.187 0.36 0.000 0.029

monster manGu 0.190 0.000 0.19 0.083 no epenthesis

crew k(u)wu(u) 0.266 0.114 0.15 O.OOOl 0.027

tricked thwikt 0.267 0.195 0.07 0.000 no epenthesis

dressing dwesGin 0.219 0.154 0.06 0.000 no epenthesis

monster manGu 0.287 0.000 0.29 0.127 no_epenthesis

stuck GAk 0.244 0.000, 0.24 0.138 no epenthesis

crying k(u}waKJm 1.223 0.603 0.62 0.000 i 0.061

broke b(u)wok 0.393 0.228 0.17 0.000 0.063

three a Guwija 1.041 0.644 0.40 0.000 0.190

three a Gwija 0.965 0.687 0.28 _ a o o o 0.000

nojggenthesis

string Guwirj 1.816 1.017 0.80 _ a o o o

0.000 0.176

black b(u)lsek 0.531 0.292 0.24 O.OOOl 0.034

gloves qalAVZ 0.714 0.395 0.32 O-OOOJ 0.167

throw fwou 0.232 0.149 0.08 O.OOOjno epenthesis

tree th(u)wi(i) 0.853 0.348 0.50 0.000 0.065

skate skeit 1.088 0.547 0.54 0.000 no egejn hesis

bridge b ]wic^ 0.599 0.319 0.28 0.000 0.049

thread Gswed 1.141 0.813 0.33 0.000 t""~ ai?2

treasure th(u)wed50 0.479 0.315 0.16 0.000 0.057

snowman snoumae(3)n 0.598 0.254 0.34 0.000 no epenthesis

plum pUm 0.236 0.110 0.13 0.000 no epenthesis

plum pUm 0.312 0.120 0.19 0.000 no epenthesis

clock k(Y)bkh 1.090 0.671 0.42 0.000

clock k(u)lak 0.394 0.191 0.20 0.000 0.049

strucked thWAkt 0.375 0.273 0.10 0.000 no epenthesis

airolane eod)l£in 0.404 0.244 0.16 0.000 no eoenthesis

65

sleep s(s)lip 0.884 0.461 0.42 0.000 0.037

sleeping 63lip(?)m 0.669 0.380 0.29 0.000 0.060

truck tWAk 0.381 0.195 0.19 0.000 noepenthesis

spring spuwirj 1.101 0.619 0.48 0.000 0.189

throw fwou 0.272 0.154 0.12 0.000 no_epenthesis

grows 0.567 0.318 0.25 0.000 0.037

squirrel skwab3 0.591 "O480 1 0.11 0.000 no epenthesis

cry kw:ai 1.200 0.321 0.88 0.000 no epenthesis

screwdriver tCu wudCuWaivu

66

APPENDIX TWO

S2 DATA

Test List | ccv/cv duration cc duration v duration c duration ep. v duration no epenthesis snowman noum(w)aen 0.606 0 0.61 0.09

ep. v duration no epenthesis

snowing snowin 0.359 0 0.36 0.089 no epenthesis

snake reei(?) 0.657 0 0.66 0.282 no epenthesis

snakey 1 green

n:ei?i 0.729 0 0.73 0.531 noj^enthesjs no epenthesis

snakey 1 green din 0.22 0 0.22 0.038

noj^enthesjs no epenthesis

black balas? 0.506 0.283 0.22 0 0.117

crayon theijan 0 0.00 0 0

glasses dslasiz 0 0.00 0 0 S3 .... , _

clothes taloQ 0.696 0.322 0.37 0 0.131

glove daUd 0.572 0.314 0.26 0 0.12

glovey dsLvblit?) 0.47 0.299 0.17 0 0.13

string dm 0 0.00 0 0

screwdriver tudaUa 0.322 0 0.32 0.033 jnojpeiujiesis

airplane e:p(3)lein 0.6 0.293 0.31 0 0.037

truck dA(3?) 0.149 0 0.15 0.015 jiojspenthesis

sgoon pun 0.777 i 0.78 0.092J no_egenthesis

sleeping lip?in 0.327 0 0.33 0.101 no epenthesis

black bala/as? 0.49 0.24 0.25 0 0.095

squirrel thewsl 0.313 0 0.31 0.029 no epenthesis

plum p/baUm 0.53? 0.285 0.25 0 0.107

srjring lphm 0.263 0 0.26 0.027 no epenthesis

clouds Jslaudz 0.831 0.318 0.51 0 0.117

airplane e(3)palein 0.858 0.403 0.46 0 0.132

glowing dress

t/dowin 0.287 0 0 2 9 0.02 no epenthesis glowing dress dswAG 0.699 0.48 0.22 0 0.181

screwdriver thu?h(3)od9waidA 0.418 0 0.42 0.042 no epenthesis

driver dawaidA 1.314 0.534 0.78 0 0.237

snake s::nei? 2.24 1.844 j 0.40 0 noepenthesis

snakey s:nei?i 1.066 0.809 0.26 0 no epenthesis

snakey s:nei?i 1.144 0.956 0.19 0 no epenthesis

snake n:ei? 0.686 0.69 0.537 no epenthesis

snake snrela 3.247 2.704 0.54 0 noepenthesis

snake rnei(?) 0.566 0 0.57 0.314 no epenthesis

sauirrel n ewaCu') 0.423 0 0.42 0.036 no eoenthesis

67

flower laA(u)we(a) 0.979 0 0.98 0.159 no epenthesis

sleeping tilip?in 0 0.00 0 0

snowman nomae(a)n 0.138 0 0.14 0.043 no epenthesis

snowing no(r)wjn. 0.747 0 0.75 0.327 no epenthesis

broom th-w(i)um 0.365 0 0.36 0.015 jic ejgerrthesis

broom b£(?)wum p haide(3)

0.785 0.584 0.20 0 no epenthesis

spider

b£(?)wum p haide(3) 0.551 0 0.55 0.096 no epenthesis

plum 0.515 0.279 0.24' _! 0.128

§ E 2 2 S _ try

pewirL 0.367 0.239 0.13 0 0.144 § E 2 2 S _ try thai 0 0.00 2

0 0

green dswjm 0 apoj 2 0 0

glove dslAb h 0.735 0.431 0.30 0 0.204

spoon pu(u)n 0.468 0 0.47 0.034 jiojerjentihiesis

black 0.64 0.237 0.40 0 0.102

string t h e W E n ^ _ _ _ _ 0.622 0.411 0.21 0 0.218

crayons theijan(t) 0.224 0 0.22 0.036 no epenthesis

stream din 0.411 0 0.41 0.018 no epenthesis

blue bslu 0.657 0.332 _ , 0 0.173

glovey dalAvi 0.669 0.259 0.41 0 0.112

sleeping slipin e(o)pslein

0.875 0.482 | 0.39 , 0 no epenthesis

airplane slipin e(o)pslein 0.664 0.409 0.26 0 0.228

broke biwoot 1.09 0.40? 0.69 0 0.219

spoon spi(u)n 0.92 0.691 0.23 0 no epenthesis

driving dAwaivin 0.896 0.357 0.54 0 0.182

snoring i[chcjc}nawjn_ 0.823 0.407 0.42 b no epenthesis

squirrel t3Wl(0)WA(u) 0.687 0.319 0.37 0 0.121

snake s:nei(?) 0.888 0.593 0.29 0 no epenthesis

Pjum_ pUm 0.632 0.358 0.27 0 0.144

fly f(s)lai 0.885 0.32 0.56 0 no epenthesis

spider spaide(a) 0.902 0.489 0.41 0 no epenthesis

crown t h 3 w a u n 0.68 0.274 0.41 0 0.12

queen _ tswin 0.785 0.38 0.41 0 0.217

spring spswin 0.838 0.609 0.23 0 0.195

try tawai 0 0.00 ^ , 0

0

blue b£(A)luju^ 0.727 0.33 rz_o.4o 0 0.139

black bolae(?) 0.551 0.274 0.28 I 0 0.079

string tuwm 0.69 0.471 0.22 0 0.217

pjay phslei 0.825 0.411 0.41 0 0.127

crayons daweijant 0.781 0.448 0.33 I 0 0.105

stream dawin 0.881 0.352 0.53 0 0.175

breaking buwei?in 0.727 0.439 0.29 1 o 0.183

break buwei(?) 0.679 0.47 0.21 0 0.186

brush bu\VA0 0.481 0.329 0.15 0 0.143

68

screwdriver i t3w(i)uduwaiv(j)5 0.739 0.374 0.37 0.237

green dawin 0.673 0.378 0.30 0.111

flower flaup(3) 0.738 0.3 0.44 Oj no e gnthesis

drawing dowawin 0.536 0.298 0.24 0.079

snowman snoumae(3)n 0.734 0.416 0.32 no epenthesis

snowing snowin 0.644 0.372 0.27 Olnoepenthesis

splashing pula;tin 1.088 0.402 0.69 Of 0.107

splashing pulaetin 0.477 0.221 0.26 0 0.102

glove dalAv(3) 0.807 0.474 0.33 0 0.212

spring sp(A)wm 0.779 0.587 0.19 0 0.158

plum paUm 0.644 0.338 0.31 0 0.176

spoon sph(i)u:n 0.771 0.571 0.20 0 no epenthesis

clock kala? 0.525 0.268 0.26 0 0.079

crown kuwaum 0.555 0.29 0.27 0.103

broom bsyy fjOom 0.514 0.293 0.22 0 0.114

snowing onowin 0.495 0 0.50 0.213 no epenthesis

glasses 0.461 0.279 0.18 0.096

draw d(u)wa(3^ 0.885 0.198 0.69 0 0.068

snake s(s)neik 0.707 0.513 0.19 0 0.026

snakey sneiki: 0.754 0.451 0.30 0 no epenthesis

squirrel guwia5 0.109 0 0.11 0.031 no epenthesis

brush bWA(3)0 0.684 0.496 0.19 0 noepenthesis

brushing buwAGin 0.45 0.306 0.14 0 0.162

broke buwout | __ 0.414 0.62 0 0.162

crying kuwaijin 0.658 0.257 0.40 0 0.095

cry k(3)wai 0.876 0.481 0.39 0 0.063

cjuack k9Wffi(3)kh 0.504 0.223 0.28 0 0096

flower 0.648 0.328 0.32 0 no epenthesis

string sPtawin 1.243 1.04 0.20 L 0 0.092

dress dswas 0.509 0.275 0.23 0 0.139

flipping f(9)lup?in 0.453 0.327 0.13 0 0.057

sleepy (t)6(a)lipi 0.364 0.207 0.16 0 O.041

screwdriver ski - tudswaive 0 0.00 L _ o 0

truck tawAk 0.479 0.248 0.23 0 0.097

sleep 9(3)lip 0.417 0.199 0.22 0 H.048

green dawin 0.44 0.263 0.18 0 _ _ J M 4 7

black balse: 1.259 0.215 1.04 0 0.108

black bslaek11 0.883 0.341 0.54 0 0.158

crayons 0.557 0.225 0.33 0 0.162

airpj ane e^pslein 0.388 0.25 0.14 0 0.102

glove aj_l_\y_ 0.624 0.338 0.29 0 0.138

dressing duwA0in 0.318 0.19 0.13 0 0.086

69