 

.1.
End.
y

. . — _L
2m... we.»

“I.

i

..i

 

 

‘
s
W... :

I

 

 

This is to certify that the
dissertation entitled

NATIVE AND NONNATIVE DIFFERENCES IN THE PERCEPTION
AND PRODUCTION OF VOWELS

presented by

DENNIE HOOPINGARNER

has been accepted towards fulﬁllment
of the requirements for the

 

 

 

 

Doctoral degree in Linguistics
Major Professor’s Signature
// Mg“ Mﬂ¢‘
Date

MSU is an Afﬁrmative Action/Equal Opportunity Institution

 

  
     

 
 

.UBRARY
Michigan State
University

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

DATE DUE DATE DUE DATE DUE

%
1'- Ii 20% ’:~

 

6/01 c:/CIRC/DateDue.p65-p.15

NATIVE AND NONNATIVE DIFFERENCES IN THE PERCEPTION AND
PRODUCTION OF VOWELS

By
Dennie Hoopingarner

A DISSERTATION
Submitted to
Michigan State University

in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Linguistics and Germanic, Slavic, Asian, and African Languages

2004

ABSTRACT

NATIVE AND NONNATIVE DIFFERENCES IN THE PERCEPTION AND
PRODUCTION OF VOWELS

By

Dennie Hoopingarner

That nonnative speakers typically speak with a foreign accent is
uncontroversial, but the extent to which production reflects the linguistic
systems of second language learners is less clear. The relationship between
perception and production of nonnative speakers has not been examined in a
thorough and objective way.

In a perception task, a computer program presented participants with a
continuum of synthesized speech samples that represented the vowel space.
For each of 11 monothongal English vowels, participants chose the sounds
that they judged was the best exemplar of the vowel. Native and nonnative
speaker groups participated in the study. In a production task, participants
read a word list that contained the same English vowels. Acoustic analysis
software was used to extract the formant values of the vowels.

Results indicate that nonnative speakers show more variation in their
language systems, and that there is less consistency between perception and
production among nonnative speakers than among native speakers. In addition,
there is evidence that the native language influences both the perception

and production of the second language.

To my family. Thank you for your support, love and understanding.

iii

ACKNOWLEDGEMENTS

I would like to thank the members of my committee, Alan Beretta, Yen—Hwei
Lin, Susan Gass and Grover Hudson, for their help as I worked on this dissertation. The
document is stronger as a result of their input. Grover was especially helpful as chair of
the dissertation committee, reading through many versions of the document, and
providing comments and suggestions. While working with me, he managed to address
issues with the organization and wording of my writing, as well as bi g-picture issues of
theory and analysis.

Thanks also to Dennis Preston, for our many informal talks, usually in his or my
ofﬁce, or in the hallway in between. Dennis shared his experience in synthesized speech
and acoustical analysis, and provided insight into variation within a speech community,
that helped in my analysis. The staff at the Language Learning Center, under the
management of Michael Kramizeh, configured the computer lab into a customized
environment for my data collection.

My friends and professional colleagues both at Michigan State University and at
other institutions were supportive of my efforts to complete this dissertation, and I thank
you all for your kind words of encouragement.

I would not have been able to get as far as I have without the support and love of
my family. They put up with me hiding in my study for many, many evenings while I
was writing. My wife Stacy, and my sons Ian and Evan, made it possible for me to
dedicate the time necessary to completing the dissertation, and gave me the reason to

persevere. Thank you for believing in me and for encouraging me to keep going.

TABLE OF CONTENTS

LIST OF TABLES ......................................................................... vii
LIST OF FIGURES ........................................................................ ix
CHAPTER 1
INTRODUCTION .......................................................................... l
1.1 Scope of the Dissertation ................................................. 1
1.2 Context of the study ......................................................... 2
1.2.1 Competence and performance in a speech community ...... 2
1.2.2 Acceptability tasks in ﬁrst and second language ............... 4
1.2.3 Limitations of previous studies ........................................ 5
1.3 Importance of this study ................................................... 6
1.3.1 Comparison between native and nonnative speakers ........ 6
1.3.2 Objective measure of perception and production .............. 6
1.4 Theoretical background .................................................... 7
1.4.1 Perception of speech ........................................................ 7
1.4.1.1 Problems of speech perception ......................................... 7
1.4.1.2 Categorical perception ..................................................... 9
1.4.1.3 Normalization .................................................................. 10
1.5 Research questions and hypotheses .................................. 15
1.6 Summary ......................................................................... 19
CHAPTER 2
REVIEW OF PREVIOUS STUDIES .................................. 20
2.0 Introduction ..................................................................... 20
2.1 First language transfer and

adult second language acquisition .................................... 21
2.2 Speech perception ............................................................ 25
2.2.1 Language and general cognition: Voice Onset Time ......... 25
2.2.2 Language-speciﬁc: Perceptual magnets ............................ 27
2.2.3 The influence of language experience .............................. 33
2.3 Perception and production of second-language vowels ..... 38
2.3.0 Introduction ..................................................................... 38
2.3.1 Perception of first and second language vowels in adults. 39
2.3.1.1 First language vowel perception ....................................... 39
2.3.1.2 Perception and production of L2 vowels .......................... 42
2.4 Acoustic comparison of Korean and English Vowels ....... 49
CHAPTER 3
METHOD ...................................................................................... 56
3.0 Introduction ..................................................................... 56
3. 1 Participants ...................................................................... 56
3.1.1 Native speakers ................................................................ 56
3.1.2 Nonnative speakers .......................................................... 57

3.2 Methodology .................................................................... 58
3.2.1 Rationale for the design of the instrument ........................ 58
3.2.2 Synthesized speech samples ............................................. 59
3.2.3 The vowel matrix ............................................................. 59
3.3 Procedure ......................................................................... 61
3.3.1 Introduction ..................................................................... 61
3.3.2 Task 1: Perception ........................................................... 61
3.3.3 Task 2: Production ........................................................... 63
3.4 Data ................................................................................. 64
CHAPTER 4

RESULTS ...................................................................................... 66
4.0 Introduction ..................................................................... 66
4.1 Native speakers ................................................................ 67
4.1.1 Task 1: Perception ........................................................... 67
4. 1.2 Production ....................................................................... 74
4.2 Nonnative speakers .......................................................... 82
4.2.1 Perception ........................................................................ 82
4.2.2 Production ....................................................................... 90
4.3 Perception-production differences .................................... 94
4.3.0 Introduction ..................................................................... 94
4.3.1 Intra—group perception versus production ......................... 96
4.3.2 Cross-group perception and production ............................ 98
CHAPTER 5

DISCUSSION AND CONCLUSIONS ........................................... 104
5.0 Introduction ..................................................................... 104
5.1 Review of research questions and hypotheses .................. 104
5.2 Evaluation of hypotheses ................................................. 106
5.2.1 Hypothesis 1 .................................................................... 106
5.2.2 Hypothesis 2 .................................................................... 109
5.2.3 Hypothesis 3 .................................................................... 110
5.2.4 Summary ......................................................................... 1 10
5.3 Additional findings of the study ....................................... 111
5.3.1 Production less variable than perception .......................... 111
5.3.2 Conflation of /a/ and /o/ in native speakers’ perception 111
5.4 Conclusions ..................................................................... 1 12
5.4.1 Research question 1 ......................................................... 113
5.4.2 Research question 2 ......................................................... 114
5.4.3 Research question 3 ......................................................... 115
5.4.4 Research question 4 ......................................................... 116
5.5 Areas for further research ................................................. 116
References ...................................................................................... 1 18

vi

LIST OF TABLES

Table 1: The average values of F1 and F2 for Korean vowels as produced by
the 20 native speaker participants .......................................................................... 49

Table 2: The average values of F1 and F2 for American English vowels as
produced by the 20 native speaker participants ...................................................... 50

Table 3: average F1 and F2 values for the vowels that are common for Korean
and English. .......................................................................................................... 54

Table 4: Age of native speaker participants ........................................................... 57

Table 5: Nonnative speaker participants’ age, years studying English and

length of stay in the US. ........................................................................................ 58
Table 6: Word lists from task 1 .............................................................................. 63
Table 7: Means for the native speaker perception task ........................................... 71

Table 8: The results of a pairwise comparison from an ANOVA test shows the
source of variation between pairs of vowels in native speakers’ responses to
the perception task in the F1 formant (top) and F2 formant (bottom). Only two
pairs of vowels (/e/-/e/ and /a/-/o/) showed no significant difference between

them. ..................................................................................................................... 73

Table 9: Main effects of the Northern Cities Shift. Column 1 shows the vowel
that is affected. Column 2 shows the vowel space that the shifting vowel
moves into (from Labov 1997). ............................................................................. 75

Table 10: Mean formant values for native speakers: production task ...................... 79
Table 11: The results of a pairwise comparison from an ANOVA test shows

the source of variation between pairs of vowels in native speakers’ responses
to the perception task in the F1 Formant (top) and F2 formant (bottom). Only

one pair of vowels (/o/-/U/) showed no significant difference between them. ......... 81
Table 12: Means for the nonnative speaker perception task .................................... 85
Table 13: Between-category t-tests: nonnative speaker perception ......................... 87

Table 14: The results of a pairwise comparison from an ANOVA test shows
the source of variation between pairs of vowels in nonnative speakers’
responses to the perception task for the F1 formant (Top) and F2 formant
(Bottom). Several pairs of vowels (/i/-/1/, /e/—/e/, /e/-/ze/, /e/-/'ce/, /a/-/o/, /a/-/o/,

/a/-/o/, /0/-/o/, and /U/-/u/) showed no significant difference between them. ........... 89

vii

Table 15: Means for the nonnative speaker production task ................................... 92

Table 16: The results of a pairwise comparison from an ANOVA test shows
the source of variation between pairs of vowels in nonnative speakers’
responses to the production task for the F1 formant (Top) and F2 formant
(Bottom). Several pairs of vowels (/i/—/I/, /e/-/$/, /8/—/ae/, /a/-/o/, /o/-/o/, /o/—/u/,

/o/-u/, and /U/—/u/) showed no significant difference between them. ....................... 93

Table 17: Average values for perception and production, native and nonnative
speakers ................................................................................................................ 95

Table 18: The results of ANOVA tests comparing perception and production

among native speaker subjects for F1 (Top) and F2 (Bottom). Only the vowel

pair /a/ - /0/ shows no signiﬁcant difference (p>.05) between pairs for both F 1

and F2. .................................................................................................................. 96

Table 19: The results of ANOVA tests comparing perception and production
among native speaker subjects for F1 (Top) and F2 (Bottom). The vowel pairs
/i/-/I/, /e/—/e/,/e/-/a=:/,/8/-/ae/,/o/-/a/,/o/-/o/,and /U/-/u/ show no significant

difference (p>.05) between pairs for both F 1 and F2. ............................................. 97

Table 20: Standard deviations for F1 and F2 values of vowels in the

perception and production tasks for native and nonnative speakers. For each

vowel, the ﬁrst line is the perception value, and the second line is the

production value. The column “PB” is the values from Peterson and Barney

(1952). The value for the group with the greatest variation is in boldface ............... 99

Table 21: The results of ANOVA tests comparing perception between native
and nonnative speaker subjects for F1 (Top) and F2 (Bottom). The vowel
pairs /e/-/e/, /a/-/o/, /a/-/o/ show no significant difference (p>.05) between

pairs for both F 1 and F2. ....................................................................................... 101

Table 22: The results of ANOVA tests comparing production between native

and normative speaker subjects for F1 (Top) and F2 (Bottom). The vowel

pairs /I/-/e/, /a/-/o/, /O/-/U/ show no significant difference (p>.05) between

pairs for both F1 and F2. ....................................................................................... 102

Table 23: Mean and standard variation values on the perception task for 11

simple vowel phonemes of English, native and nonnative speakers. The higher
value of native and nonnative speakers is shown in boldface. ................................ 107

viii

LIST OF FIGURES

Figure 1: The F1 and F2 values of vowel samples collected from adult male,

adult female, and child speakers in Peterson and Bamey’s (1952) study plotted

on a grid shows considerable variation within each vowel, and signiﬁcant

overlap between vowels ......................................................................................... 11

Figure 2: The formant frequency values F1 and F2 of vowels can be plotted
onto an XY graph. Values for F2 are plotted along the X axis, and values for
F1 are plotted on the Y axis. .................................................................................. 14

Figure 3: The X-Y grid maps formant frequency values of vowels in a way
corresponding to positions of the vowels in the traditional vowel quadrangle. ....... 15

Figure 4: The Ontogeny Model (Major 1987) ........................................................ 24

Figure 5: Visual representation of tokens from Kuhl’s 1991 study. Filled areas
on the matrix indicate the F1-F2 values of vowel sounds. The areas labeled
“P” and “NP” are the F1-F2 values of the token selections from Peterson and
Barney (1952). The label “P” indicates the prototype sound that was closer to
the sound produced by more subjects in Peterson and Bamey’s study. The
sound labeled “NP” is the nonprototype sound, a sound that was produced by
one subject, but deviates from the average of all subjects In the study to a
greater degree than the “P” sound. The ﬁlled areas surrounding the prototypes
are the F1-F2 values of vowel sounds that were synthesized for this study. The
F1-F2 values of these sounds differ incrementally from the sample sounds
from Peterson and Barney. .................................................................................... 29

Figure 6 The average F1 and F2 values of American English and Korean
vowels plotted on X-Y grids. The X axis is the F2 value, and the Y axis is the
F 1 value. ............................................................................................................... 51

Figure 7: The respective positions of the F1 and F2 values of the common
vowels between English and Korean, /i e e a e o u/, plotted on the same graph ...... 53

Figure 8: The vowel matrix ................................................................................... 60
Figure 9: Graphic representation of the vowel area . The gray shaded area

represents F1 and F2 values that were not represented by sounds in the
perception task, but that could possibly be produced by participants during the

production task. ..................................................................................................... 67
Figure 10: Average of Native Speaker Perception .................................................. 68
Figure 11: Native speaker choices for perception task for each vowel .................... 70

Figure 12: Average Native Speaker Perception and Production ............................. 76

Figure 13: Native speaker performance on production task .................................... 77
Figure 14: Average Normative Speaker Perception ................................................ 83
Figure 15: Nonnative speaker choices for perception task ...................................... 84
Figure 16: the Normative vowel areas: perception task .......................................... 88
Figure 17: Average Normative Speaker Perception and Production ........................ 90
Figure 18: Normative speaker performance on production task .............................. 91

Chapter 1: Introduction
1.1. Scope of the Dissertation

This study examines the acquisition of a second language phonology. The goal of
this study is to determine the extent to which nonnative speakers of English have
acquired the vowels of English. I compare the perception and production of eleven
monothongal English vowels by native and nonnative speakers of English. I then explore
the relationship between perception and production within a speech community, and
describe differences in perception and production between native and normative speakers
of English.

Many studies of second-language phonology concentrate on the spoken output of
nonnative speakers. This study adds the dimension of intuition data, acquired via a
judgment task, as a means of measuring the acquisition of the second language vowel
system. In a study of phonology, the use of intuition data, the type usually gathered in the
form of tasks in second language acquisition research, has the advantage over production
data in that foreign accent is eliminated as a factor in participants’ responses.

This study seeks to measure participants’ intuitions about eleven monothongal
vowels in English, and compare that intuitional data to their performance on a production
task. By comparing participants’ intuitions with their productions, foreign accent can be
isolated as an element of linguistic performance separate from linguistic competence. In
order to establish baseline data with which to compare nonnative speakers, data from
native speakers was also gathered. In this way, first—language competence and

performance can be compared to second-language competence and performance.

1.2. Context of the study

1.2.1. Competence and performance in a speech community

Linguistic theory distinguishes between competence and performance in linguistic
behavior among native speakers of a language. At least since Chomsky (196524)
explicitly distinguished linguistic “competence” from “performance,” the focus of
linguistics has been on competence. The topic of inquiry is thus the abstract, internal
system that a linguistic adult attains through the process of language acquisition.
Chomsky (1995: 14) characterized linguistic competence as “some array of cognitive
traits and capacities, a particular component of the human mind/brain.”

Although instantiated within the mind/brain of individuals, the grammars of
members of a speech community have been assumed to be homogenous (Saussure
1916:19, Chomsky 1965z3). While individuals may exhibit different language
performance, the underlying structure of the language system is uniform. It is commonly
accepted that linguistic competence is a more reliable indicator of underlying structure
than performance. Linguistic competence is unaffected by factors external to language,
such as slips of the tongue or psychological factors such as memory and attention. These
external factors can influence linguistic performance, however.

Phonological competence being more well-formed than performance is well
documented in the literature on child language phonology. The so-called “fis
phenomenon” (Berko and Brown 1960) is an example of how competence and

performance can differ in an individual’s grammar. It describes the condition in which a

child cannot produce a particular sound correctly (in this case, the sound /§/), but can

distinguish it when uttered by others. This condition is developmental in nature, and is
documented in the ﬁrst-language literature, but it may be applicable to second language
acquisition in that it indicates a system that is in transition.

Although it is a common assumption that competence should be the focus of
linguistic theory, it can be very difficult to accurately determine the underlying structure
based on the output. Production data is often riddled with errors. In second-language
phonology, the difference in performance between native and non-native speakers is
especially salient. Foreign accent is clearer marker of a speaker as non—native than syntax
or morphology errors. First-language inﬂuence on the second-language sound system is a
commonly-cited source of foreign accent. While the existence of foreign accent is not
controversial, it is an open question whether non—nativelike production is an indicator of
nonnative-like competence. In other words, the inability to produce the second language
in a native-like manner is not necessarily indicative of the nature of the speaker’s

underlying phonological competence in the second language.

1.2.2. Acceptability tasks in ﬁrst and second language

The predominant method used to study the grammar of a language is to call upon
the intuition of speakers of the language. A commonly-used instrument in investigations
into second language acquisition of syntax is the acceptability task (Schachter 1989,
White 1989, Broselow and Finer 1991, Al—Banyan 1996, Liu and Gleason 2002), in
which the participant indicates whether a sentence in the target language is grammatical
or not. There have been objections raised to this method. Sorace (1996) points out that
this kind of task forces the participant to make a categorical, binary decision about a form,
when the actual judgment of the form may be gradient. White (2003: 17) also
characterizes as a “myth” the assumption that acceptability tasks provide a direct
reﬂection on second-language linguistic competence. Nonetheless, the acceptability task
remains a commonly used tool for measuring linguistic competence in syntax. It is
commonly accepted that this method of examining linguistic structure is the least likely
to be influenced by outside influences.

In the area of theoretical phonology, the use of acceptability judgments is also the
means to discover underlying structure. In fact, it is argued that Optimality Theory
(Prince and Smolensky 1993), as a method of analysis, codifies the process by which a
native speaker arrives at a judgment about a phonological form by ranking constraints.
However, in the field of second language phonology, little word has been done to
measure participants’ intuitions linguistic forms. Rather, the focus has been on measuring
output and making inferences about the internal representation based on that output. In a
sense, research that uses second language speakers’ output as the only source of data may

amount to a study of foreign accent, which may not necessarily represent the speaker’s

internal system accurately. Studies in the role of the native language in speech perception,
which is discussed in Chapter 2, indicate that a given speech stream can be perceived
differently by speakers of different languages. Speech perception is thus part of an

individual’s grammar.

1.2.3. Limitations of previous studies

As mentioned in the section above, perception and production often operate
differently, and so focusing only on output may not give a complete picture of the
linguistic system. Consistently in studies of second language phonology, the topic of
study has largely been output. Nonnative-like production is often attributed to the
influence of the first language on the second. Understanding the differences between the
two languages was the goal of Lado’s (1957) Contrastive Analysis, based on contrast
between two languages. This method of analysis was applied as a method to predict
errors in pronunciation. Eckman’s (1977) Markedness Differential Hypothesis refined the
prediction mechanism to include universal tendencies of phonological complexity. He
used production data to evaluate this model.

Most studies of this nature drew their conclusions solely on the basis of an
analysis of the participants’ output. The measurement of the production data has typically
been done using subjective measures. Many studies rely on the judgments of human
raters to evaluate production (Eckman 1987, Broselow and Finer 1991, Eckman and

Iverson 1994, Major and Faudree 1996). Few studies use computer-based speech analysis

tools to extract the acoustic properties of the speech that participants produced, such as in

the procedure used by Flege (1987, 1991) and Flege et at (1997).

1.3. Importance of this study

1.3.1. Comparison between native and nonnative speakers

This study seeks to acoustically define specific vowel categories for both native
and normative speakers of English, both in identiﬁcation of vowels, and in production.
The phonological phoneme inventory is in some respects a closed system. A given
language has a ﬁnite number of consonant and vowel contrasts, and so this aspect of the
system is quantifiable. Although differences between vowel inventories and systems in
different languages has been documented for ﬁrst languages in studies such as Schwartz
et al (1997), there has been little empirical work to systematically map a normative
speaker’s entire vowel system, and compare it to that of a native speaker. The

contribution of this study is the addition of this dimension of phonological competence.

1.3.2. Objective measure of perception and production

Another potential contribution of this study is the objective nature of the analysis.
As was mentioned in section 1.1.2.3. above, in the bulk of second-language phonology
studies, the data was analyzed after it was transcribed. Doucherty and Foulkes (2000:

113) point out potential problems with transcribed data, chief among which, there is

always the danger that the data is not transcribed reliably. Even the best-trained
phoneticians can inadvertently filter speech sounds through their own phonological
systems. This can result in transcriptions that lack accuracy. The current study seeks to

avoid that problem by using fully objective measures.

1.4. Theoretical background

1.4.1. Perception of Speech

1.4.1.]. Problems of speech perception

A prerequisite for oral linguistic communication is that the listener understand
what the speaker says. That a native speaker of a given language can understand that
language when it is spoken by another speaker is an empirical fact, and yet an empirical
model for the process by which that occurs still eludes us. In a discussion of speech
perception, Chomsky and Miller (1963) discuss two problems that underlie the
phenomenon of speech perception: linearity and invariance. The problem of linearity is
that the speech stream is not a progression of clearly delineated and acoustically
recoverable phonemes, but is instead a fluid mixture of coarticulated sounds. Consonants
and vowels are combined together in the speech stream through coarticulation. Hockett
(1955: 210) made the analogy that recovering phonological elements from the speech
stream is similar to trying to reassemble colored Easter eggs that have been crushed as

they move along a conveyor belt. Liberman (1970) pointed out that it is impossible to

take a tape recording of a CVC syllable, and cut the tape to isolate the consonants and the
vowels. This is because consonants are represented in the speech stream as formant
contours around vowels, inseparable from the vowel sound. The seminal study by
Delattre et al (1955) showed that consonants can be synthesized by manipulating the ﬁrst
two formants in a vowel sound. By creating contours up or down from the value of the
formants that produce the vowel sound, a consonant-like sound was produced. Consonant
onset sounds corresponding to /b/, /d/ and /g/ were synthesized in this manner.
Participants in the study reported hearing a consonant onset. Delattre et al concluded that
the process of speech perception hinges on the formant frequencies of the utterance, and
furthermore, that the ﬁrst two formants, F1 and F2, carry enough information to
distinguish both consonants and vowels. Even though consonants and vowels may not be
objectively present in the speech stream, listeners are still able to perceive them.
Linearity is also instantiated in the phonological environment of the utterance.
The shape of the vowel tract for the production of a given phoneme will differ depending
on what sounds are to be articulated before and after it. Katamba (1989: 19) illustrates
how coarticulation influences speech production with the example of the English phrase
“car keys.” The /k/ sound is produced by the tongue interacting with different parts of the
oral cavity as the speaker anticipates the following vowel sound. The vowel /a/ is
produced by opening the oral cavity, and so the /k/ in the word “car” is produced by the
back part of the tongue touching the rear part of the soft palate. The vowel /i/ is produced
with a raised tongue, and so the /k/ is “keys” is produced with the tongue more forward.
The second problem of speech perception is invariance, or speaker variation. An

acoustic event in speech is not related in a one-to-one fashion to the phonemic element

that it invokes. One factor that causes variation in utterances is idiosyncratic
characteristics of the vocal tract, such as length and shape, that make each speaker’s
voice distinctive. The length of the vocal tract determines the fundamental frequency of
the speaker’s voice. Since not every speaker has a vocal tract of the exact same length,
different speakers uttering the same vowel sound will produce a sound with a different
pitch. Changing the shape of the oral cavity by positioning the jaw, tongue and lips,
produces different vowel sounds. Not every speaker will position their oral cavity exactly
the same, and so the vowel sounds of different speakers will be more or less different.
The problems of speech perception can be approached within the framework of
two theoretical concepts. First, using the concept of categorical perception, we can seek
to deﬁne the acoustic threshold of what we perceive as a phonological element. This
process seems to account for the linearity problem of speech perception. The “acoustic
space” can be divided up into categories. Any elements of the speech stream that “ﬁt”
into a particular acoustic section are interpreted as a single segment. Secondly, the
invariance problem can be addressed by models of speaker normalization. This theory
attempts to describe how listeners compensate for inter-speaker differences in voicing

and vowel quality.

1.4.1.2. Categorical perception

Categorical perception can be used to explain the ability to discriminate between

some sounds, and the inability to discriminate between others. The human brain has the

ability to gather sounds of language into perceptual categories. Within a category,

variants of a vowel may differ in a measurable, absolute way, and yet the human brain
can not distinguish the difference. Two variants of vowels that fall in different categories,
however, can differ by the same small amount as between sounds in the same category,
and participants can hear the difference between them.

The evidence of the use of categorical perception in language use has strong
empirical support (Liberman et al 1957, Fry et al 1962, Lisker and Abramson 1964,
Eimas et al 1971, Scoles 1968, Kuhl 1980, Gass 1984, Repp 1984, Werker and Tees 1984,
Kuhl 1987, Kuhl 1991, Werker 1994). However, as Kuhl (1987) points out, the effects of
categorical perception are not limited to language, and not limited to humans. It may be
worth questioning whether categorical perception is linguistic in the modular sense of
Fodor (1983), or whether, as part of general cognition, it is an ability that humans share
with other animals (Harnad 1987). Nevertheless, it does seem to play a part in language,
seems to be a part of the innate human cognitive system, and changes with language
acquisition as the infant begins to specialize its sensitivity to salient differences in the

native language (Kuhl et al 1992).

1.4.1.3. Normalization

Every speaker’s voice is different. Vowel productions vary across speakers. For
example, the voices of male speakers, as a general rule, have lower F0 frequencies than
those of female speakers. Differences in length of the vowel tract results in a distinct
audio quality for each speaker. The seminal study by Peterson and Barney (1952)

demonstrated that age and gender are the source of much variation in the production of

10

vowels, largely due to physiological differences between speakers. Figure 1 shows the
measurements of vowels collected from speech samples from speakers of different age
and gender. There is considerable variation among speakers, and considerable overlap

among vowels.

F2 0

 

F1

 

 

Figure 1: The F1 and F2 values of vowel samples collected from adult male, adult female,
and child speakers in Peterson and Barney’s (1952) study plotted on a grid shows
considerable variation within each vowel, and significant overlap between vowels.

In addition, the acoustic values of vowels produced by any given speaker can vary
as a result of context. However, the variation between speakers does not prevent the
listener from understanding. When hearing the same utterance by different speakers, the
listener perceives the utterance as the same in some sense. In order to understand spoken
language, listeners must perceptually remove this speaker-specific variation and perceive

the utterances using an objective norm. The process of removing the speaker-specific

differences is referred to as normalization. Native speakers seem to compensate

11

effortlessly and automatically for variations in the acoustic productions of different
speakers.

The problem of understanding such normalization, as pointed out by Hindle
(1978), comes to light when comparing two different speakers. The listener must
transform the formant values of the two speakers so that they coincide for comparison. A
common assumption is that vowel identiﬁcation can be done via an objective process,
using relative positions in the vowel space, similar to Joos’ (1948:68) concept of a
“template.” Joos suggested that, in speech perception, the listener divides the vowel space
into zones. Vowels that fall within a given zone are interpreted as that vowel. This
division of the vowel space is a listener’s template. The process of normalization,
according to Joos, consists of mapping the speaker’s utterances onto the vowel space
perceived in such zones.

Later studies of speech perception (Ladefoged and Broadbent, 1957, Gerstman
1968, Lehiste and Meltzer 1973, Assman et al 1982) support the position that listeners
use a subjective standard to perceive a given speaker’s vowels. Studies of normalization
techniques by Neary (1989) and Miller (1989) also support the position that listeners do
not use objective measures to perceive vowels in terms of absolute formant values.

Normalization is the process of identifying vowel sounds in the speech stream by
means of a norm. This norm is a system of vowels, which Kuhl (1991) calls “prototypes.”
An assumption of the process of normalization is that listeners have an internal system
that they are referring to and trying to map the speech stream to. Members of the same
speech community have the same internal system, and so normalization does not interfere

with communication. Nonnative speakers of the language may or may not have the same

12

internal system, and so normalization may or may not cause comprehension problems.
Part of the task of learning a second language is developing a system of vowels that is
used in comprehending spoken language. One of the aims of this study is to explore to
what extent second language learners have developed a system to use when normalizing

English speech.

1.4.2. Measuring vowels

The source-ﬁlter theory of vowel production (Fant 1960) sees vowel production
as a combination of two factors. The vocal folds restrict the flow of air through the
pharynx, producing vibration and sound. The sound passes through the oral cavity, where
it resonates. The shape of the oral cavity makes the sound resonate at speciﬁc frequencies.
The combination of the resonating frequencies produces a vowel sound. The speaker can
change the shape of the oral cavity by altering the position of the jaw, tongue and lips, to
produce different resonant frequencies that are perceived as different vowel sounds.

Analyzing vowels in terms of the source—ﬁlter theory allows us to isolate the
factors that make vowels distinctive. A given speaker will produce a source that is
consistent among utterances, because the quality of the source is determined by
physiological factors such as gender and size of the vocal tract. The speaker’s filter,
which changes with the shape of the oral cavity, determines which vowel is produced.
The wide variety of vowel sounds that can be produced is attributed to the ability to vary
the position of the tongue, lips and jaw. It is the filter that makes vowel sounds distinctive,

and so vowels are identiﬁed by analyzing the ﬁlter.

13

We can measure vowel sounds in terms of their component formant frequencies.
Peterson (1951) and Delattre et al (1955) showed that vowels can be identified solely on
the basis of the formant frequencies F1 and F2. Different values of the formant
frequencies produce different vowel sounds. The values of the formants F1 and F2 for

vowels can be plotted onto a graph to show their relative values. Figure 2 shows such a

graph.

2600 F2 0

 

F1

 

 

1000

 

Figure 2: The formant frequency values F1 and F2 of vowels can be plotted onto an XY
graph. Values for F2 are plotted along the X. axis, and values for F1 are plotted on the Y
axrs.

Values for F2 are plotted along the X axis, and values for F1 are plotted on the Y
axis. This grid differs from other grids in that the origin point in the vowel graph is in the
upper—right hand corner, rather than in the lower—left hand corner. By formatting the
graph in this manner, when vowels are plotted on the graph, their relative positions are
consistent with the traditional vowel quadrangle, as shown in Figure 3. High front vowels
such as /i/ have a relatively low F1 value and a relatively high F2 value. Plotting those

vowels on this grid would put them in the upper—left hand quadrant, corresponding to

their position on the vowel quadrangle. Similarly, high back vowels such as /u/, with their

14

low F1 and F2 values, would fall in the upper-right hand quadrant. Vowels like /a/ have a

high F1 value and a medium-range F2 value.

2600 F2 0

 

 

F1

 

 

 

1000

 

Figure 3: The X-Y grid maps formant frequency values of vowels in a way corresponding
to positions of the vowels in the traditional vowel quadrangle.

Studies by Peterson and Barkey (1952) and Hillenbrand et al (1995) showed that
the frequency range for adult male speakers is 300 —- 750 Hz for F 1, and 1000 — 2400 for
F2. These values would be higher for female voices, due to the shorter length of the

vowel tract.

1.5 Research questions and hypotheses

This study examines second language speech perception and production,

comparing the intuitions of native speakers and nonnative speakers intuitions about

vowel phonemes in English, and comparing perceptions to production of the vowels by

nonnatives. It seeks to map the English vowel system for both native and non—native

15

speakers for perception and production. The specific research questions to be addressed

by the dissertation are as follows:

1. To what extent can we determine the acoustic properties of the ideal vowel
phoneme of each monothongal vowel in English based on native speakers’
intuitions?

2. To what extent do nonnative speakers of English agree on the ideal vowel
phoneme of each monothongal vowel in English?

3. Are the intuitions of normative learners of English similar to the intuitions of
native speakers with regard to the vowel phonemes of English?

4. How similar is production of vowels in English by both native speakers of
English and normative learners of English to their respective identiﬁcation of

an ideal vowel?

The motivation behind research question (1) is the theory that native speakers
have established phonemic categories for the vowels of their native language, and that
they can use their intuitions to indicate whether-or not a given token of speech falls
within the category for a given vowel (Liberman 1970). This study asks whether the
intuitions of native speakers converge on an acoustic value for a given vowel. The degree
of consensus among the native speaker group can be compared to the degree of consensus
among nonnative speakers.

The basis of research question (2) is the supposition that native speaker members

of a speech community have a common linguistic system, including vowel categories.

16

Nonnative speakers, as learners of the language, are assumed to have a language system
that is different from native speakers’. It is unclear, however, to what extent nonnative
speakers have a speech community, with vowel categories that are homogenous to an
analogous degree that is found among native speakers. Research question 2 addresses this
issue. Using the same task for native and normative speakers, this study measures the
intuitions of normative speakers about the English monothongal vowels.

Research question (3) asks whether, if there is a speech community of nonnative
speakers, the intuitions of native speakers are similar to those of normative speakers.
This question is motivated by the assumption that an important part of acquiring a second
language is acquiring the internal system of the language. Intuitional data can provide
valuable insight into an individual’s grammar.

Research question (4) builds on two assumptions. The ﬁrst is that there is a
difference between linguistic perception and production. The vowel sound that a speaker
produces may be different from what the same speaker would identify as the correct
vowel sound. The second assumption is that nonnative speakers’ foreign accents are not
necessarily an accurate reflection of their second-language perceptual vowel systems.
The first language may inﬂuence a nonnative speaker’s production of a vowel sound
without inﬂuencing the perception.

This study seeks to address these questions by comparing the perception and
production of English vowels by two groups of participants: native speakers from
Michigan and normative speakers of English from Korea. With regard to research

questions 1 and 2, it is expected that native speakers, as members of the same speech

17

community, will agree on vowel categories. Normative speakers will show more variation
than the native speaker group.

The comparison of native and normative speaker participant groups, as indicated
in research question 3, will show the extent to which nonnative speakers share the same
intuition about vowel categories in English as native speakers do. This comparison will
eliminate foreign accent as a factor, and will thus be a better indicator of the degree to
which the participants have acquired the English vowel system.

Finally, a comparison of perception and production will reveal differences
between the two groups. Research question 4 examines the role of accent in linguistic
performance. Ideally, there would be no difference between perception and production. In
the real world, we expect to see other factors inﬂuencing linguistic performance.
Comparing native and normative speakers, we can expect to see a closer match among
native speakers than nonnative speakers. Foreign accent is expected to emerge as a factor
among nonnative speakers, resulting in a greater dissimilarity between perception and
production. Native speakers are expected to show more consistency between perception
and production.

The specific hypotheses of the study are as follows.

1. Given a perception task in which participants identify an English vowel from a
continuum of F1—F2 combinations (the “perception task”), there will be less
variation among native speakers than among nonnative speakers.

2. In the comparison of the performance of native speakers and normative speakers
on the perception task and a “production task” in which participants produce

words containing the English vowels, there will be less variation between the two

groups’ performance on the perception task than there will be on the production
task.
3. In a perception and a production task, there will be less variation between the two

tasks among native speakers than among nonnative speakers.

1.6 Summary

Native speakers of a language have a grammar that is embodied in the speaker’s
mind/brain in the form of a linguistic competence. The linguistic competence of members
of a speech community is relatively homogenous. An individual’s language system
inﬂuences both speech perception and production. Due to factors unrelated to
competence, linguistic performance can differ among speakers. Performance is visible
and measurable, but linguistic theory is more concerned with competence. Since it is a
part of the mind/brain, competence cannot be measured directly. The acceptability task is
one of the most popular ways to explore competence. It is commonly used in studies of
second language syntax, but not in second language phonology.

This study seeks to use both performance and competence data on both native and
normative speakers, to try to gain a more accurate picture of second language phonology.

Fully objective measures will increase the accuracy of reporting and analysis.

19

Chapter 2: Review of previous studies
2.0. Introduction

In this chapter I review studies concerning four areas. The first area is factors that
inﬂuence adult second language acquisition. A great deal of research points to qualitative
differences between language acquisition in adults and children. As the participants of
this study are adults, it is relevant to acknowledge these differences and how they can
affect the results of the study.

The second area is speech perception. Speech perception is an important part of
this study, and so it is relevant to define its domain. I review studies that examine the
phenomenon of speech perception as a linguistic phenomenon as opposed to a general
cognitive function. I then review studies that point to differences between perception of
language in general as compared to perception of language-speciﬁc structures. Finally, I
examine studies that explore the role of language experience on second language speech
perception.

The third area that I discuss is the relationship between perception and production
in second language acquisition. The potential for the two to differ in first language
performance is well known, but is an area that has received relatively little attention in
second language research. This study focuses on the difference between perception and
production in second language, and the review of the literature is intended to give some
background to the topic.

The fourth area is an acoustic comparison between the vowels of Korean and
English. The nonnative speaker participants of this study are Korean, and the native

speaker participants are American. This area uses data from a study that collected

20

production data from native Korean and American English participants producing vowels
in their native languages. The data from that study can provide reference material for this

study.

2.1. First language transfer and adult second language acquisition
As Gass (1996) points out, although the details of the role of the native language
in second language acquisition have been a continuous subject of debate, the inﬂuence of
the native language has not been in dispute. A common term for first-language inﬂuence
on the second language is “transfer.” White (1989) takes the position that UG is available
to learners, but acknowledges the role of the first language in the formation of the
second-language grammar. Leather and James (1996: 275-276) take as a base assumption
that listeners use the phonetic categories of their native language when labeling auditory
stimuli, a position taken earlier in the last century by Trubetzkoy (1969 [1939]: 52—53).
Trubetzkoy saw the ﬁrst language as a ﬁlter through which the second language was
perceived:
“The phonological system of a language is like a sieve through which everything
that is said passes. Only those phonic marks that are relevant for the identity of
the phoneme remain in it... when [a person] hears another language spoken he
intuitively uses the familiar ‘phonological sieve’ of his mother tongue to analyze
what is said. However, since this sieve is not suited for the foreign language,
numerous mistakes and misinterpretations are the result. The sounds of the
foreign language receive an incorrect phonological interpretation since they are
strained through the ‘phonological sieve’ of one’s own mother tongue.”
Work on loanwords also suggests that the native language is operant in the

process by which a foreign word is altered to fit the structure and constraints of the

adopting language. As foreign words become part of the native language, something of

21

their structure often changes as the words are reinterpreted by the adopting language’s
phonology. Silvermann (1992) proposed a two-stage process by which a word is ﬁrst
perceived by native speakers of the adopting language, and then a structure for the word
is abstracted out inductively. In Kenstowicz’ (2003) Optimality Theory analysis of Fijian
loanwords, the process of nativization consists of trying to maintain faithfulness to salient
aspects of the foreign word by making use of repair strategies in the native language
grammar.

Odlin (1989), in his discussion of language transfer, distinguishes transfer effects
in various aspects of phonology. The characteristics of the native language phonological
categories, phonemes, rules and syllable structure can all inﬂuence the interlanguage
phonology.

Even phonological categories that are similar in both the ﬁrst and second
languages often differ at the subphonemic level. Flege (1987) describes phonetic
difference between the Dutch and English phonemes /i/ and /u/. Dutch /i/ has lower F2
and higher F lvalues than English /i/, placing it lower in the acoustic space than English
/i/. Dutch /u/ has a higher F2 value than English /u/, making it closer phonetically to

English /u/ than to English /u/. Flege found that Dutch learners of English produced /u/

with less accuracy than native speakers, due to the inﬂuence of the Dutch system. Riney
and Takagi (1999) investigated the inﬂuence of the differences in the voice onset time
(VOT) in /p t k/ in Japanese and English had on the production of the English consonants
by native speakers of Japanese. They found that although native—like VOT correlated to
overall proficiency in English production, VOT values did not become more English-like

over time. That is, participants who had native-like VOT did not seem to develop the

22

native-like pronunciation over time. Riney and Takagi attributed the lack of change over
time to the close similarity between Japanese and English VOT values. Flege (1987)
claims that second-language sounds that are perceived by the learner as equivalent to a
native-language sound will be produced as the native—language sound. Riney and Takagi
claim that this “equivalence classiﬁcation” is operational in their study, and that it tended
to prevent native-like acquisition by their participants. The authors suggest that the
inﬂuence of the L1 phonology was so strong that it impeded the L2 pronunciation of
many participants.

Phonemic differences between languages can also inﬂuence interlanguage.
Studies by Sheldon and Strange (1982), Yamada et al (1997), Takagi (2002) and many
others have documented the chronic inability of Japanese learners to correctly distinguish
and produce the English segments /r/ and /l/. Even with intensive discrimination training,
learners still do not uniformly develop a native-like system for distinguishing between the
two phonemes. The difficulty of the learners seems to stem from the fact that in the
acoustic space where Japanese has only one phoneme, English has two. The inﬂuence of
the ﬁrst language seems to interfere with the establishment of new phonemic categories
in the second language. Wayland (1997) studied the acquisition of Thai consonants,
vowels, and tones by native English speakers, and found significant native-language
effects for all categories.

Major’s (1987) Ontogeny Model of second language phonology development is
an effort to account for developmental patterns in the interlanguage, while explaining the
role of the first language in the second language system. According to the model, the

nature of errors in the second language will change over time. In the early stages of

23

acquisition, the inﬂuence of the first language will dominate the phonological system,
causing interference-like errors. These errors will decline over time, as the learners
acquire the second language phonology. Developmental errors will be nonexistent in the
beginning, will increase in frequency, and then decrease. Figure 4 below illustrates the

model.

 

 

 

 

m U)
H H
o O
t: t:
0) 0)
9-4 9-;
o O
a a
0" 6*
Q) Q)
d: «1::
time time
Interference Developmental

Figure 4: The Ontogeny Model (Major 1987)

Major tested the Ontogeny Model with Brazilian learners of English and their
acquisition of English /r/, ﬁnal consonant clusters, and voiced and unvoiced word-ﬁnal
obstruents, and tentatively concluded that the learners followed the sequence outlined in
his model. This model has been criticized for being vague in some aspects (James 1988),
but Major does make one strong claim: that there is no fundamental difference between
first and second language acquisition with regards to the progress of developmental errors.
The difference between adults and children, Major claims, is the initial state. While
children start with no established language system, adults start with an intact language
system, and it is this native language system that interferes with the development of the
second language phonology. A revision of the model proposed by Major (2001) is the

Ontogeny Phylogeny Model, which adds the role of the L2 as a factor to the other two

24

factors of L1 and developmental inﬂuences, and makes a clearer statement that

“development” indicates universal patterns of language acquisition.

2.2. Speech Perception
2.2.1. Language and general cognition: Voice Onset Time

A common topic for second-language speech research is the acquisition of voice
onset time (VOT). Adults can discriminate between voiced and voiceless segments in
their native language in a categorical manner, as demonstrated by Lisker and Abramson
(1964). Presented with a collection of synthetic speech segments whose only difference
was a gradual variation in the voice onset time, participants could identify each segment
as voiced or voiceless with great consistency and uniformity of responses. The ability to
discriminate differences in it is evident in very young infants. Eimas et al (1971) tested
infants as young as 1 month old, and found that they could distinguish VOT differences
in a categorical manner. The ability to detect differences in VOT seems to be innate.

The language-specific values for VOT must be learned, however. Lasky et al
(1975) examined the ability of infants to detect cross-language VOT values. Infants from
Spanish and English-speaking environments were tested. The two languages have
different VOT values to distinguish voiced and voiceless onsets. English onset VOT is
more consistent with other languages. Voiced onsets have a typical VOT of 20
milliseconds, and voiceless onsets have an average VOT of 40 milliseconds. Spanish, on
the other hand, has a more marked system of dividing the VOT space. Voiced onsets in
Spanish have a pre-voiced VOT of —20 milliseconds, and voiceless onsets have a VOT of

20 milliseconds.

25

The infants listened to a series of sounds that varied from [ba] to [pa] in gradient
steps. Using a head—tum method, the infants indicated when they heard a change in the
category from /b/ to /p/. The results from testing the responses of infants showed that
infants 4-6 months old from both language backgrounds responded in a uniform manner.
Even infants in a Spanish environment indicated the change at the VOT values of 20 and
40 milliseconds, even though in the language of their environment, VOT times ranged
from ——20 to 20 milliseconds. The result of this study indicates that there are innate
phonetic values that may provide a basis from which language-speciﬁc values can be
built, and at the age of 4—6 months, the language—specific values have not yet been set.

The language-specific value of VOT is one aspect of the second language that
must be learned. Gass (1984) examined second language learners’ perceptions of VOT by
using a forced-judgment task with synthesized stimuli. Participants indicated whether the
sound that they heard began with a voiced or a voiceless bilabial segment. Compared
with native speakers of English, whose judgments typically show a clear categorical
distinction consistent with studies done with native speakers (Lisker and Abramson 1964,
Liberman 1970), the normative speakers showed a fuzzy, continuous distinction, with no
clear categorical boundaries between segments. Gass suggests that the participants are
inﬂuenced by the categorical boundaries in their native language. Their judgments seem
to reﬂect a system in ﬂux. The participants could have been in the process of developing
an interlanguage system, and so their responses showed that they had not yet developed
clear distinctions of the VOT boundaries. Another factor that may have inﬂuenced the
participants is their relative short time in an English—speaking environment (half of the

participants had been in the US between 2 days and 1 month).

26

Work has also been done in measuring the ability to detect VOT differences in
non-human animals. Kuhl and Miller (1975) found that chinchillas could detect VOT in a
manner almost identical to that of humans. Other evidence that animals can perceive
speech with patterns similar to humans was found in studies by Kluender et al (1987) and
Kluender and Lotto (1994). These studies suggest that the speech perception mechanisms
may not be modular and specific to language, but are part of the general cognitive

processing capabilities of the nervous system.

2.2.2. Language-speciﬁc: Perceptual magnets

Kuhl’s (1991) seminal study introduced what she termed “perceptual magnets”
for speech perception. Kuhl proposed a model of native language perception that
differentiates categorical perception from the linguistic process of phonemic perception.
Although categorical perception has been shown in nonlinguistic modules of perception
in humans and in nonhuman animals, Kuhl attempted to determine whether phonemic
perception was only found in humans. While not refuting categorical perception as a real
phenomenon, Kuhl’s study was an attempt to isolate the linguistic component of vowel
perception.

For her study, Kuhl chose two tokens of the vowel /i/ from Peterson and Barney’s
(1952) study. She termed one of the tokens from the Peterson and Barney study a
“prototype,” because it was close to the average value for /i/ in Peterson and Bamey’s
study. The other token vowel from the Peterson and Barney study was termed a “non-
prototype” of the vowel /i/, because although a token of /i/ with its values was found in

the study, it was an outlier token in the study. Kuhl then synthesized two series of 32

27

vowels for each token, each of which differed from the token in measured distances
based on values of the ﬁrst and second formants. The relative values of the stimuli are
shown in Figure 5 below. The “P” and “NP” indicate the relative positions of the
prototype and nonprototype sounds, and the variant sounds, as they would appear on an
F2-F1 X-Y grid. As mentioned in section 1.4.2, the F1 and F2 formant values are

indicators of vowel height and frontness, respectively.

28

500
i 200

 

F1

I"
I
._I
Y
.

600
Figure 5: Visual representation of tokens from Kuhl’s 1991 study. Filled areas on the
matrix indicate the F1-F2 values of vowel sounds. The areas labeled “P” and “NP” are
the F1-F2 values of the token selections from Peterson and Barney (1952). The label “P”
indicates the prototype sound that was closer to the sound produced by more subjects in
Peterson and Barney’s study. The sound labeled “NP” is the nonprototype sound, a sound
that was produced by one subject, but deviates from the average of all subjects In the
study to a greater degree than the “P” sound. The filled areas surrounding the prototypes
are the F1-F2 values of vowel sounds that were synthesized for this study. The F1-F2
values of these sounds differ incrementally from the sample sounds from Peterson and
Barney.

One group of adult participants listened to the group of sounds that were based on
the prototype vowel. Using a goodness rating task, participants evaluated each sound for
the goodness to ﬁt for the vowel /i/ on a scale of 1 to 7. Kuhl found that participants
showed a clear preference for the prototype over the other sounds that were variants of
the prototype. This indicated that participants agreed on a “best ﬁt” paradigm of the
vowel /i/. In the task, the more a sound stimulus varied from the prototype sound, the
worse it was ranked by participants.

A second group of adult participants performed the same judgment task, but the
sounds that they listened to were based on the nonprototype sound. Although the

nonprototype sound was an instantiation of a sound that was actually uttered by a subject

in the Peterson and Barney study, participants showed a stronger preference for sounds

29

that were closer in value to the prototype sound, the sound that was closer to the average
value of utterances from the Peterson and Barney study. This ﬁnding supports the
suggestion of the first task, that participants had an intuition about the best fit for /i/. Even
though the prototype sound was not available to the second group of participants, they
nevertheless showed preference for those sounds that were closer to the prototype sound
in the first task.

Kuhl’s conclusion from the study was that adults have established a clear
phonological category for the vowel /i/. This vowel category was part of the participants’
linguistic system. Furthermore, they could call on their linguistic competence to evaluate
other sounds and indicate how similar those sounds were to their linguistic category for
the vowel sound. Kuhl also noted that the response patterns of all the subjects were
consistent within the group. This suggested to Kuhl that the “internal standard for the
vowel /i/” is quite similar among speakers of the same speech community.

Another task in Kuhl’s study used the same set of stimuli. This task measured
whether participants could detect differences between sounds. A tenet of categorical
perception is that within categories, differences between tokens are difﬁcult to detect,
while across category boundaries, differences are easier to detect. Participants listened to
a pair of sounds, and indicated whether the sounds were the same or different. It was in
this task that the linguistic factor seemed to influence participants’ judgments.
Participants had more difficulty detecting differences between sounds that were both
close to the prototype. They had less difﬁculty detecting differences between sounds that
were farther away from the prototype. Kuhl dubbed this the “perceptual magnet effect.”

Kuhl found that participants tended to identify sounds that were close to the prototype as

30

being the same, but were able to distinguish between sounds that were not close to the
prototype. Kuhl proposed that the prototype acted as a magnet, drawing to it sounds that
were close to it in the acoustic space, and consequently making them harder to
distinguish.

Human infants aged 6-7 months and monkeys were tested on the same stimuli, to
see if they could detect the gradual differences in the sounds. Participants listened to two
sounds, and indicated whether the sounds were the same or different. The responses of
human infants were gathered using a head-turn technique (Kuhl 1981). Monkeys were
trained to press a button for a food reward when they heard a different sound. The results
for human infants mirrored those for adults. Kuhl concluded from this that human infants
also have a mental prototype for the vowel /i/. Rhesus monkeys could detect differences
between stimuli, but their response patterns were quite different from those of human
participants. Kuhl concluded that since humans and monkeys used different bases to
judge the phonetic material, that only humans showed evidence for mental prototypes for
linguistic segments.

Kuhl’s ﬁnding has important theoretical implications for speech perception. A
nagging problem with theories of speech perception is that non—human animals, which
are not supposed to be capable of linguistic behavior as humans are, exhibit the
categorical perception effects that supposedly indicated linguistic behavior. The concept
of a perceptual magnet captures the distinction between discrimination between
categorical perception, which may be a general cognitive function, from phonemic

identification, which should be linguistic and restricted to humans. There is no doubt that

31

categorical perception underlies the perceptual magnet effect, but Kuhl’s study indicates
that there is more to perceptual magnets than categorical perception.

Kuhl concluded from this study that phonetic categories are unique to humans,
since they were not evident in the response patterns of monkeys. Furthermore, phonetic
categories inﬂuence categorical perception in humans. The presence of phonetic
categories in infants could be due to an innate endowment, but as Kuhl points out, six-
month-old infants have already been exposed to a considerable amount of language, and
it is possible that categories have been formed by that age. The effect of language
experience on category formation is discussed in section 2.2.3 below.

Kuhl and Iverson (1995) suggest that the acoustic space is divided up into
“natural magnets,” which they indicate can be seen as innate phonemic categories.
During language acquisition, as infants gain more exposure to the language of their
environment, the acoustic space is re-divided, and new phonemes (“magnets”) are formed.
Some magnets may even disappear. The well-documented difficulty that Japanese
speakers have in discriminating the English /r/ and /l/ distinction (e. g., Sheldon and
Strange 1982, Yamada, Tohkura, and Kobayashi 1997, Gordon, Keyes and Yung 2001)
could be explained as the development of a perceptual magnet in Japanese speakers that
blurs the phonetic distinction between the two sounds.

A study by Iverson et al (2003) explored the inﬂuence of first language on speech
perception. Their hypothesis was that participants would be less able to detect contrasts
that do not exist in their L1 than contrasts that are found in their L1. The target of their
investigation was the distinction of /r/ and /l/ in English, and its acquisition by native

speakers of Japanese. They employed as stimuli 18 synthesized tokens of a CV syllable.

32

The stimuli were synthesized to vary the F2 and F3 values in equal steps, forming a series
of sounds that varied between /ra/ and /la/. Participants, native speakers of English,
German and Japanese, listened to the stimuli in pairs, and indicated whether the sounds
were the same or different.

Results showed that Japanese participants are less sensitive to the distinction
between /ra/ and /la/. A particular difference between native speakers and Japanese
speakers was sensitivity to differences in the F3 formant. English—speaking participants,
but not Japanese speakers, indicated that stimuli with different F3 values were different
sounds. Iverson et al concluded that native and normative speakers attended to different
aspects of the sound. Changes in the F2 formant seemed to be the determining factor for
Japanese participants to make their choice in labeling the sound, in contrast to the F3,
which native speakers seemed to use. They explain their results by suggesting that the
participants’ Japanese phonological systems were “mis—tuned” to the English contrast.
They had developed perceptual categories for Japanese, and used these categories to
interpret the English sounds. Because their systems did not attend to the indicative factor,

F3 values, they did not accurately perceive the normative contrast.

2.2.3. The influence of language experience

The generalization of some aspects of speech perception to general auditory
processing strategies has forced a reﬁnement of theories of speech perception. Kuhl’s
research suggests that phonetic categories (Kuhl’s “prototypes”) are formed early in life.
Her study suggested that categories were formed at least by 6 months of age. The

formation of language-speciﬁc prototypes was the focus of Werker and Tees (1984).

33

They studied infants’ abilities to distinguish native and normative phonetic contrasts. The

difference between [t] and the retroﬂex Lt] that is found in Hindi could be detected by

infants younger than 10 months regardless of whether the distinction is present in the
language of the environment. Infants aged 10-12 months of age were less able than 6
month-old infants from the same language environment to distinguish between [t] and the

retroﬂex Lt], while 10-12 month old infants from Hindi—speaking environments could

detect the difference. Werker and Tees conclude that the loss in ability to detect
nonnative contrasts is a result of language experience.

Werker and Lalonde (1988) replicated the approach used by Werker and Tees
(1984). They asked whether adult speakers of English and Hindi would categorize the
same continuum of sound differently. Their stimuli consisted of eight synthesized sounds
that varied the onset formants to form a /ba/ to /da/ continuum. Using an ABX matching
task, Werker and Lalonde were able to determine where participants’ categorical
boundaries fell. The English-speaking participants divided the continuum into two
categories, labial (/ba/) and alveolar (/da/), but the Hindi speakers divided the continuum

into three categories: labial (/ba/), dental (/da/) and retroﬂex (/da/). They then tested two

groups of infants using the same stimuli. Infants aged 6—8 months, regardless of the
language of their environments, could discriminate the contrasts that Hindi speaking
adults indicated. Infants ages 1 l—l3 months from an English-speaking environment could
not detect the Hindi contrasts. Werker and Lalonde conclude that language experience in
the ﬁrst year of life causes a developmental change the way that spoken language is
perceived. Infants lose sensitivity to some contrasts as they develop perceptual

categories.

34

Kuhl et al (1992) tested the ability of 6-month old infants in English and Swedish
environments to detect native and normative vowels. A background assumption of this
study was that by the age of 6 months, perceptual categories have already been formed,
and these categories will affect the perception of speech. Kuhl’s perceptual magnet effect
predicts that the categories that are formed as a result of linguistic experience serve as
prototypes. The effect of the perceptual magnet effect is a reduction in the ability to
discriminate between small variations from the prototype. The vowels that are near the
prototypes will not be distinguished as different from the prototype. There were two
assumptions of this study. One was that the prototypes for /i/ are different for English and
Swedish speakers (a comparison of American and Swedish adults’ prototypes of /i/ was
measured in Willerman and Kuhl 1996, reﬂecting a measurable difference). The other
assumption was that the Swedish participants had in their vowel systems the high front
vowel /y/, and the English speakers did not.

Kuhl et al chose good examples (“prototypes”) of English /i/ and Swedish /y/, and
synthesized variants of them by altering the F1 and F2 values. The F1 and F2 values of
the stimuli varied in measured steps from the prototypes. Participants were American and
Swedish infants. Both groups of participants listened to the stimuli based on both
languages’ prototypes. The prediction was that the /i/ prototype would be a magnet for
the American but not the Swedish infants, and the /y/ prototype would be a magnet for
the Swedish but not the American infants.

Using a head-tum technique (Kuhl 1981), infants listened to stimuli in 2 second
intervals, and, during a learning phase, learned that when the stimulus changed, a toy bear

pounded a drum. Infants turned their heads to anticipate the performance of the toy bear.

35

The American infants did not show signs of detecting variation from the English
/i/ prototype to the degree that the Swedish infants did. This ﬁnding was predicted by the
perceptual magnet theory. A prototype sound acts as a magnet, reducing discriminability
of similar sounds. If Swedish infants did not have the same prototype, then they would
not show reduced ability to detect differences. Similarly, Swedish infants could not detect
variation from the /y/ prototype to the degree that the American infants could. The
Swedish infants showed signs that they had developed a prototype for /y/, and thus
displayed perceptual magnet effects. The results from this study give evidence that by age
6 months, infants had been affected by the language of their environments, and they had
developed linguistic categories for those vowels.

An implication of Kuhl’s (1981) study is that language-specific categories seem
to be formed earlier for vowels than for consonants. Polka and Werker (1994) had similar
findings. They examined the development of perception of L1 vowels and the
accompanying loss of ability to detect nonnative vowel contrasts in Canadian infants.
Variants of the German high front vowel /y/ were presented in a CV C syllable /dyt/.
Variants of the high back rounded vowel /u/, found both in Canadian English and
German, were also prepared in the same CVC environment (/dut/). Two groups of infants,
aged 6-8 months and 10—12 months, were presented with the stimuli and tested for
discrimination using the head-turn method described in Kuhl (1981). Comparing the
results of this vowel-discrimination study with those of infant consonant-discrimination
studies, the authors found that the 6—8 month old infants responded with more language-
speciﬁc responses than was found in that age group in discrimination tasks involving

consonants. Running the same experiment with 4-month old infants showed more

36

language-neutral results, the kind shown in 6-8 month old infants for consonants. The
results of this study suggest that the shift in discrimination ability from language-neutral
to language—specific happens at an earlier age for vov‘vels than for consonants. Later
research by Bosch and Sebastian—Gallés (1997) with 4—month old infants supports the
finding that at that age, infants are already able to discriminate between their native
language and other, even closely related, languages.

In summary, research on the effect of language experience on speech perception
shows that from an early age, the native language changes the way that speech is
perceived. The process of language acquisition seems to include the development of
perceptual categories. These categories affect the way that the native language is
perceived, and also reduces sensitivity to nonnative contrasts. This process of
specialization to a language system results in an internal language system that is
specialized for one language. This system may assist members of the speech community
to disregard the individual differences of speakers, which will aid speech perception.
However, specialization to the acoustic properties of one language may also impede the
acquisition of a second language. The process of second language acquisition may
require the learner to alter the existing categories, and may require the development of
new categories. Measuring whether a learner of a second language has categories similar
to those of native speakers is one way to measure second language acquisition.

The research also illustrates methods by which speech perception can be tested. In
many studies of speech perception, the use of synthesized speech is a way to control the

stimuli that is presented to participants. A goodness of ﬁt task in which adult participants

37

compare the stimuli with their internal systems is a common method to investigate

phonological competence.

2.3. Perception and production of second-language vowels
2.3.0. Introduction

The relationship between vowel perception and production is central to this study.
The relationship is analogous to the competence-performance relationship in linguistic
theory. Part of an individual’s linguistic competence is the vowel categories in the
individual’s phonological system. As the discussion of the research in section 2.2 above
suggests, the native language of the perceiver inﬂuences how vowels are perceived in
speech. Part of the task of the adult language learner is to adjust the perceptual categories
of the native language to accommodate the categories of the second language. In other
words, the language learner has to change or develop a new phonological competence.

A salient part of an individual’s linguistic performance is speech production. But
just like other instances of linguistic performance, factors other than competence can
inﬂuence speech performance. This is especially evident in second-language speech.
Foreign accent is an indicator that the learner has not mastered the phonology of the
second language. However, just as definitive conclusions about linguistic competence
should not be made based on performance, so also must we exert caution about reaching
conclusions about second-language phonology based on second-language speech. This
section discusses research that explores the relationship between vowel perception and

production. First, I brieﬂy review research into the nature of vowel perception in general.

38

 

Following that is a discussion of research in second-language vowel perception. Finally, I

review studies of the relationship of second-language vowel perception and production.

2.3.1 Perception of first and second-language vowels in adults
2.3.1. 1. First language vowel perception

Ladefoged and Broadbent (1957) synthesized vowels in a /th/ context, and
preceded the word with the phrase “Please say what this word is...” Participants were
asked to identify the vowel in the last word, choosing between bit, bat, bet or but. In
tokens in which the ﬁrst formant of the “carrier phrase” that preceded the /th/ word was
shifted up or down, but the /th/ word was the same, participants identiﬁed the vowel in
the /th/ word differently. The quality of the vowels in the preceding phrase seemed to
inﬂuence the perception of the following word. Their study suggests that perception of
vowels is greatly dependent on context. Ladefoged and Broadbent suggest that the
preceding acoustic signal, which they term a “carrier signal,” gives a crucial reference of
vowel quality to the listener. This study illustrates the phenomenon of speech
normalization, as discussed in section 1.4. 1.3 above. Listeners need to adjust their
perception based on the qualities of the speaker’s voice. The carrier phrase in this study
inﬂuenced listeners to normalize the speech stream up or down, and the /th/ word was
normalized with the carrier phrase.

Fry et al (1962) explored participants’ ability to recognize phoneme boundaries
and intra-category discrimination of vowels. They produced 13 synthetic vowels that

varied on a continuum that covered the vowels /I/, /e/ and /ae/. Participants were

presented with the stimuli using a “forced-choice ABX” method. Two sounds were

39

 

 

presented (“A” then “B”), followed by a third sound (“X”). Participants were asked
whether the third sound (“X”) was the same as the first (“A”) or the second (“B”). The
stimuli were presented in isolation (different from Ladefoged and Broadbent (1957),
which included a carrier signal) with no context. There was significant overlap in
participants’ responses. Unlike studies of categorical perception, in which there is
considerable agreement among participants for consonant boundaries, no such general
agreement was found for the vowel boundaries. The results of this study indicated that
boundaries between vowels not clear-cut. Fry et al concluded that the perception of
vowels is continuous rather than categorical. In the discrimination task, there was no
evidence of categorical perception effects.

The study by Fry et al complements Ladefoged and Broadbent (1957) in showing
that vowels are relative in nature. When presented in a context, in relation to other vowels
in a phrase, listeners are able to make consistent judgments about vowels, as in the
Ladefoged and Broadbent study. However, if there is no point of reference, as in the Fry
et al study, listeners do not make categorical judgments about the vowel sound. In
explaining their results, Fry et al point to Joos’ (1948:68) “template” model of speech
perception that suggests the listener mentally maps the speech signal into vowels within
the acoustic space (see the discussion of Joos’ model in the discussion of normalization in
section 1.4.1.3 above). In the absence of context, listeners are unable to normalize a
vowel sound.

Scholes (1967a) asked whether participants would associate synthetic vowels with
phonemes in their native languages. Participants indicated whether each of 69

synthesized vowels was a representation of a sound in their language. The vowel sounds

4O

 

were synthesized from a continuum of F1 and F2 values over a normal frequency range
for human speech. Participants chose from a list of words in their native language as a
match for each of the synthesized vowels. The stimuli were presented in a scrambled
order. With almost no exception, participants were able to match each synthesized vowel
with a vowel sound in their language. The methodology allowed the mapping of
boundaries for each vowel for each participant. There was signiﬁcant agreement among
speakers of the various languages as to the category boundaries for the vowels. There was
some overlap, however. Some stimuli were identified as different vowel phonemes by
some speakers. This study showed that there is some speaker—speciﬁc variation in the
perception of vowels, similar to that found by Fry et al (1962). While there was not
complete agreement among all participants with regard to vowel category boundaries, the
overall level of agreement among participants indicates that the identification technique
reﬂects the homogeneity of a language community’s vowel phoneme categories.

In a followup study, Scholes (1968) had nonnative English speaker participants
give words in their native language that represented the vowel sound that they heard in
the synthesized sounds. The synthesized vowel sounds provided an objective
environment in which to compare vowel phoneme patterns of various languages.
Participants from the same language background gave words with largely similar
phonemes in their native languages. Scholes claimed that this task supported the results
of the 1967 study, that participants can associate native-language vowel phonemes with
the synthetic stimuli. Participants’ second task was to listen to the vowel sounds again,

and give an English word that used the vowel sound that they heard.

41

 

Scholes then mapped the F1 and F2 values of the words collected from the two
tasks onto an X—Y grid. The areas covered by the vowel sounds of the two groups of
words overlapped signiﬁcantly. Scholes claimed that this showed similar categorization
between vowels in their native language and in English. He concluded that the non—native
speakers of English seemed to perceive English according to their native language
phonemes, and that their native language categories inﬂuenced their performance in a
second-language task. Scholes found that the categorization of synthetic vowels by
nonnative speakers is the same for the native language and nonnative language, as long as
the L1 and the L2 have counterparts. Nonnative speakers will hear the normative
language through their native language system.

Scholes’ ﬁndings are consistent with those of the studies in section 2.2.3 that
show evidence for a specialization of the perceptual system for the phonetic categories of
a particular language. Evidence for the inﬂuence of the first language on the second can
be seen through performances on tasks that make use of perceptual categories.
Participants in Scholes’ studies tended to identify second language vowels in terms of

vowel categories in their first languages.

2.3.1.2. The interrelationship of perception and production of L2 vowels

Flege’s (1995) Speech Learning Model represents an attempt to account for the
differential performance in perception and production in second language phonology.
Specifically, it addresses a phenomenon in second language phonology that on the
surface may seem puzzling: that sounds in the second language that are quite different

from sounds in the native language are acquired with more accuracy than sounds that

42

have close correlates in the native language. Contrastive Analysis (Lado 1957259)
predicts that areas of the native and target languages that are similar would be acquired
with greater case than areas in which the two languages are not similar. Stockwell and
Bowen (1965:9-18) present a systematic method of predicting difficulty based on a
comparison of the phonological systems of two languages. Following that claim, then, we
would predict that phonological segments in the target language that correlate to those in
the native language should then be easier for learners to acquire. However, empirical
evidence shows just the opposite.

Flege and Hillenbrand (1984) compared production of the English onset /t/ by
French-speaking learners of English. Acoustical comparison of the onset in the two
languages showed that VOT values for the segment /t/ differ phonetically in French and
English. Native speakers of French, when speaking English, tend to use French VOT
values to produce an English /t/. The interpretation of this phenomenon using the Speech
Learning Model is that the French learners of English perceive the English /t/ as similar
enough to the French /t/, so no adjustment of the phoneme is necessary.

Flege (1987) had measured the performance of native speakers of English in
producing the French segments /t/, /u/ and /y/. The first two segments Flege classified
“similar,” because although not phonetically the same, English and French both have
those segments. The segment /y/ is a “new” sound for Anglophone learners of French.
Flege found that their production of this new sound was more nativelike than their
production of the French categories that had similar counterparts in English.

What is interesting in these examples is that the Contrastive Analysis Hypothesis

would have predicted that the new sound /y/ would have given the American learners of

43

French more difficulty than the segments that were similar to segments in the leamers’
native language. The similar segments should have been easier to learn, and so the
participants’ performance should have been better. The answer that Flege gives to the
contradiction to the Contrastive Analysis Hypothesis is that learners will establish new
categories for the segments in the second language that are perceived by the learner as
new, and for sounds that are classified as similar to categories in the native language, the
learner will simply substitute those categories.

This model of second-language speech is an interesting corollary to Eckman’s
(1977) Markedness Differential Hypothesis, which was a refinement of Lado’s (1957)
Contrastive Analysis Hypothesis. Lado predicted that in comparing the L1 and L2, those
elements that are similar to the learner’s native language will be easy to learn, and those
elements that are different will be difficult. Flege’s SLM predicts that those categories in
the L2 that are similar to the L1 will be more accented, and those elements that are new
in the L2 have the potential to be acquired with native-like accuracy.

Bohn and Flege (1997) tested this hypothesis in a study involving the acquisition
by native speakers of German of the English vowel /a:/. German has the vowels /a/, /e/,

and /e/, but not /ae/. The authors asked whether adult learners of a second language could

acquire a new category, and how their perception and production would compare.

Two groups of normative speakers were compared to a group of native speakers.
One group, which they labeled “experienced” learners, has a mean length of stay of 7.5
years. The other group of normative speakers, the “inexperienced,” had a mean length of
stay of 0.5 years. For the production task, the two groups of German learners of English

read short sentences in English, ending in words that contained the /8/ or lae/ sound.

Acoustic analysis of the production showed that the speech of the more experienced
group of normative speakers was more similar to the native speakers, and the speech of
the group of less experienced participants was less native-like. There was little overlap
between the two vowels in the production of the native speakers, but no distinction
between the two vowels by the inexperienced nonnative speakers. Bohn and Flege
conclude from this that the inexperienced speakers do not distinguish a distinct category
for the /2e/ sound. They suggest that length of exposure to the second language could have
a greater effect on the production than on the perception of the second language.

For the perception test, the authors synthesized a gradient of 33 vowels that varied

between /8/ and /aa/ on a continuum, varying F1, F2 and F3, plus the duration of the

segments. Using a forced-response method, participants indicated whether each sound
corresponded to the word bet or bar. Items were presented in random order. Native

speakers responded with a categorical pattern, indicating a clear separation between /8/

and /ae/ in their vowel systems. Overall, the normative group that had had more exposure
to English responded in a more native-like manner than the nonnative group with
relatively less exposure on the production test. The responses of the experienced group
showed a more distinct differentiation between the two vowels than those of the less
experienced group. However, the pattern of perception by both groups of normative
speakers differed from that of the native speakers in that it reﬂected a more continuous
pattern, with no clear separation between the segments, than was found in the native
speakers.

A closer analysis of the data by the authors indicated that nonnative speakers

referred to different acoustic cues than native speakers. The pattern of responses among

45

native speakers showed that this groups detected a category change along the continuum
of F1, indicating vowel height. Vowel height was the salient factor that separated vowels
for native speakers. The pattern of responses for nonnative speakers did not show such a
trend. There was no clear categorical distinction among nonnative speakers that
correlated with F 1. Instead, the response pattern of normative speakers indicated that they
associated vowel duration with category change. For native speakers of English, duration
of the segment was not a factor in making a judgment, but “inexperienced” German
learners of English seemed to rely on duration as the prime factor in making their
judgments. This ﬁnding indicates that the participants had developed categories for the
second language vowels that use different criteria for distinguishing the vowels than
those that are used by native speakers.

An interesting aspect of this study is the apparent disconnect between perception
and production. The experienced nonnative group produced the vowels to a more native—
like degree of accuracy than the inexperienced group, yet their performance on the
perception task showed differences suggestive of how the vowel is represented in the
participants’ grammars. This result was similar to that of the VOT study by Gass (1984),
which indicated that perception and production could be disconnected among nonnative
speakers.

The indication that the normative speakers in Bohn and Flege’s study relied on a
different acoustic aspect of the vowel than native speakers for identifying vowels
parallels the findings of the study of Japanese learners of English by Iverson et al (2003),
in which relying on changes in F3 instead of F2 led nonnative speakers to inaccurate

judgments.

 

 

Ingram and Park (1996) investigated the perception and production of Australian
English vowels by Korean and Japanese learners of English. In the first task, participants
listened to recordings of speakers reading /th/ words containing the Australian English

vowels /i/, /I/, /e/, /a=./ and /a/. Participants performed a forced—choice identiﬁcation task to

identify the vowels. Japanese participants could correctly identify all of the vowels with a
high degree of accuracy (92% - 100%), but Korean participants responded with a much
more mixed pattern of judgment for the vowels /e/ and /a3/, with accuracy measures
between 46% and 54% for those vowels. Korean participants with more exposure to
English did better on the task of identifying /e/, but misidentiﬁed several tokens of /a:/ as
/e/, and vice versa.

The authors surmised that the two groups of participants were using different
strategies for arriving at their judgments. Their hypothesis was that Japanese participants
were using duration as a factor in perception, possibly because length is contrastive in
Japanese, but not in Korean. The inﬂuence of the L1, the authors suggest, helped the
Japanese arrive at the correct judgment. Since long and short vowels are not distinctive in
Korean, the authors continue, length did not inﬂuence Korean participants’ judgments.
Korean participants relied on other cues to distinguish vowels. The study did not include
a discussion of the formant values of the vowels, focusing instead on vowel duration as a
factor in perception and production.

The second task was reading aloud the words that were in the first task.
Participants’ voices were recorded and analyzed for vowel duration. The results showed a
clear distinction between the two groups. Japanese participants produced vowels with

internally-consistent values, in lengths that differed across segments. The authors assume

47

that the participants were transferring the moraic vowel length from Japanese to the
Australian English target vowels. The Korean participants, on the other hand, produced
vowels in a pattern of lengths that was closer to that of the target language.

The third task was a “native rating” of the L2 vowels. Participants listened to
tokens spoken by two native speakers of Australian English, and were told to write the
vowel in their native language that best represented the vowel sound that they heard.
Additionally, they were told to indicate whether the vowel that they heard was long or
short. The Japanese participants categorized the vowels consistently for each of the two
speakers. Their responses did not indicate sensitivity to inter-speaker differences. The
Korean participants’ judgments seemed to have been more speaker-dependent, however.
They rated the same vowel from the two speakers differently. While Japanese
participants responded to the stimuli in a categorical way, Korean participants responded
to absolute differences in token vowel duration.

The authors assumed that the Korean participants were using phonetic cues to
make their judgments. Japanese participants, however, seemed to have normalized the
speakers’ vowels for length, compensated for speaker differences, and responded on the
basis of phonological cues. Since vowel length is distinctive in Japanese, but not Korean,
the authors conclude that the inﬂuence of the participants’ native languages was at work
in their perception of the second language. Japanese subjects used processing strategies
for speaker normalization that are operational in their native language. The responses
were very consistent among that group, a pattern that would be expected among members

of the same speech community using their native language competence.

48

2.4. Acoustic Comparison of Korean and English vowels: Yang 1996

Yang (1996) compared the production of English and Korean vowels, with the
purpose to compare the acoustic properties of the vowel phonemes of the two languages.
The data for the study was samples of vowel utterances from native speakers of English
and Korean who read word lists containing the vowel phonemes of the languages.
Participants’ voices were recorded for later acoustic analysis. The 20 English speaker
participants were from the South or Southwest of the United States. The 20 Korean
participants were from Seoul and spoke standard Korean.

Table 1 and Table 2 show the average F1 and F2 values for the Korean vowels /i
y i e s2) 8 a A o u/ and English vowels /i I e 8 ac a a o 0 U u/ from the study as produced

by the participants from the two language backgrounds.

Korean F 1
i 343
y 3
i 4
57

12

0 47
u 3 1001

 

Table 1: The average values of F1 and F2 for Korean vowels as produced by the 20
native speaker participants in Yang (1996).

49

 

 

 

 

 

 

 

 

 

 

 

 

 

English F 1 F2
1 338 2572
I 438 2193
C 495 2309
8 581 2072
a: 756 1901
a 710 1 169
9 647 1486
3 720 1083
0 513 1 167
U 469 1409
u 375 1452

 

 

 

 

 

Table 2: The average values of F1 and F2 for American English vowels as produced by
the 20 native speaker participants in Yang (1996).

Figure 6 shows the same values plotted on an X-Y grid, showing their respective

positions within the vowel space.

50

Yang 1996 Korean Vowels

 

2600 F2 475 0
+11 200
+l_+_y

+i

+0 F1
+¢
+e
+ E
+ A
+21
1000

 

 

 

Yang 1996 English Vowels

 

2600 F2 475 0
+u 200
+i
+1
F1
+0 +0
+c
+9
+30
+3-
1000

 

 

 

Figure 6 The average F1 and F2 values of American English and Korean vowels plotted
on X—Y grids. The X axis is the F2 value, and the Y axis is the F1 value.

51

 

For purposes of comparison, Yang assumed that the two languages had vowels in
common. In the literature, vowels that are written with the same phonetic symbols across
languages are often acoustically similar, and should cluster when graphed according to

their F1 and F2 values. Yang compared the vowels that are written /i e e a o u/. We can
also include in the comparison the Korean vowel /’\/ and the English /a/, because the two

vowels have similar acoustic properties.
Figure 7 below shows where the values of each of these vowels lie within the

vowel space. The Korean vowel /’\/ is labeled /a/ for purposes of comparison.

52

Yang 1996 Common Vowels

 

 

2600 F2 475 0
200
-E.
+41“ E
11
+ +uK
K
O
+eE +05 + F1
+93:- {E
+£K +3E
K
a
+ +aE
+aK
1 000

 

 

Figure 7: The respective positions of the F1 and F2 values of the common vowels
between English and Korean, /i e e a a o u/, plotted on the same graph

For many of the vowels, the relative positions of the corresponding vowels are
quite similar. The vowel /i/ for example is very similar between languages. Larger
differences are found in the vowels /a/, /9/ and /u/.

An interesting aspect of the differences between the vowel systems is the relative

values for /e/ and /e/. The Korean vowels /e/ and /e/ are lower and more back than their

53

English counterparts. There is greater similarity between the Korean /e/ and the English

/8/ than there is between the English and Korean /e/ or /£/.

The average F1 and F2 values of the common vowels is given in the Table 3

 

 

 

 

 

 

 

 

 

below.
F1 F2
Vowel English Korean English Korean

i 338 343 2572 2517
e 495 570 2193 2173
e 581 634 2072 2067
a 710 862 1 169 1583
a 647 687 1486 1246
o 513 476 1 167 987

u 375 396 1452 1001

 

 

 

 

 

 

 

Table 3: average F1 and F2 values for the vowels that are common for Korean and
English in Yang (1996).

There are two aspects of Yang’s study that are relevant to this study. The first is
that the vowels that are indicated as corresponding vowels do indeed seem generally to
correspond. When acoustic measurements of tokens of the vowels that use the same
notation in the literature are mapped on a grid using the F1 and F2 values as x and y
Coordinates, the respective positions of the vowels in the acoustic are similar across the

two languages.

54

 

The second significant aspect of this study is that it shows that while the vowel
systems of English and Korean have some similarities, they are still different. Through an
examination of the formant values of corresponding vowels across languages, Yang
found statistically signiﬁcant differences between the two language systems. A series of

t-tests revealed cross—language differences in nearly every pair.

55

 

Chapter 3: Method

3.0 Introduction

As mentioned in Chapter 1, this study is an investigation into the relationship
between perception and production of American English vowels by native and normative
speakers. This study investigates perception in a way that differs from other perception
studies. This chapter describes the participants, research instrument, and data collection

procedure.

3.1. Participants

3.1.1. Native Speakers

N inety-ei ght male and female native speakers of English participated in the study.
They were recruited from an undergraduate sociolinguistics course at Michigan State
University, and received partial course credit in return for their participation. The
majority of the participants ranged in age from 18 to 22 years old. Of those participants,
six were not from Michigan, and the data from five others was not collectable because of
technical difficulties. The data from eighty—seven participants was usable, and is reported

in this study.

56

 

 

 

 

 

 

 

 

 

 

Age Number of
participants
18 1
19 15
20 4O
21 22
22 7
over 22 2
Total 87

 

 

 

Table 4: Age of native speaker participants

3.1.2. Nonnative Speakers:

Twenty-seven Korean nonnative speakers of English participated in the study.
They were recruited from English as a Second Language classes that were offered by
Michigan State University for members of the community. The participants ranged in age
from 18 to 42 years old. One participant did not report an age. Their length of residence
in the US ranged from a few months to 12 years, with the majority less than one year.

One participant did not report the length of time studying English or residing in the US.

57

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Age Years studied English Months in the US. Gender
18 6 just F
18 4 10 months F
19 10 2 months F
19 5 2 years M
19 6 4 months F
20 4 4 months F
20 8 12 months M
21 1 7 months F
21 9 4 months F
22 not reported not reported N/A
22 10 8 months F
22 10 9 months F
22 3 3 months F
22 3 4 months F
22 6 24 months M
23 1 12 months M
23 5 7 months F
24 6 3 months M
24 8 12 months N/A
25 7 4 months M
27 3 8 months F
36 14 12 years F
36 22 7 months M
37 10 9 months M
38 5 12 months M
40 3 8 months F
42 6 4 months M

 

 

 

 

 

 

Table 5: Nonnative speaker participants’ age, years studying English and length of stay in
the US.

3.2 Methodology
3.2.1. Rationale for the design of the instrument

The instrument used in this study was designed to gather data in a non-subjective
way. The hypotheses being tested required the availability of a continuum of vowel
sounds that varied in controlled, incremental steps. Synthesized speech samples were the

most effective way to present participants with such a continuum.

58

Using a computer-based instrument facilitated objectivity of data collection. The
data was stored in a central database, and so administering the instrument on a networked
computer system was the most practical option. The instrument that was developed was

able to accommodate a large number of participants working independently

3.2.2. Synthesized speech samples

The synthesized speech sounds used in this study consisted of 306 artificial
vowels each of 500 milliseconds duration, synthesized with the software system Praat,
version 4.0 (Boersma and Weenick 1999-2000), and presented in a web-based software
program written in Macromedia Flash MX.

The acoustic properties of the speech sounds were based on those used in a study
by Frieda et al (1999). The speech sounds approximated a male speaker with a voice
pitch that began at 300 Hz, and fell to 130 Hz over the duration of each sound. The audio
files were created at a 22,050 Hz sample rate, and compressed using the Speech codec of

Flash MX at 22 kHz.

3.2.3. The vowel matrix

A 17x18 cell matrix represented the range of F1 and F2 value combinations. F1
values ranged in 50 Hz increments from 200 to 1000 Hz in 17 steps with a constant
bandwidth of 50. F2 ranged in 125-Hz increments from 475 Hz to 2600 Hz in 18 steps at
a bandwidth interval of 100 Hz. For all samples, F3 was held constant at 3000 Hz,
B3=150. The 306 vowels were presented to participants in a matrix as shown below.

Clicking on any of the squares in the matrix played the vowel with the F1 and F2 values

59

that corresponded to that square’s relative position in the matrix. By clicking on adjacent
squares, participants could hear the continuum of vowel sounds that was represented in

the matrix.

 

Figure 8: The vowel matrix

The matrix represents a 2-dimensi0nal continuum of vowel sounds. Vowel height
is represented in the vertical axis, and vowel frontness is represented by the horizontal
axis. The sounds were arranged so that the F1 and F2 values of sounds were lowest at the
origin point. In the section of chapter 1 on measuring vowels (section 1.4.2), I discussed
how arranging the F1 and F2 values in this way aligns the values with the positions of
vowels in the traditional vowel quadrangle. Cells further to the left had a higher F2 value.

Cells further below the origin had a higher F1 value.

60

3.3. Procedure

3.3.1. Introduction:

The data collection took place via the Internet. The only requirements were that
the participants’ computers have a web browser with the Flash 6 plugin, be connected to
the Internet, and have the capability to play and record sound. The advantage of
developing the instrument in Flash was that the program was playable on any platform or
operating system that supported the Flash player, and so there were no cross-platform
differences in the presentation or functionality of the instrument. This lent more
ﬂexibility in finding a location to administer the instrument.

Computer laboratories at various locations at Michigan State University were
used to collect data. The researcher was present at all data collection sessions to answer
questions that participants had about the study or about their tasks, and to assist with
computer problems.

All participants performed the same tasks in the same order. All items in each task

were presented in the same order.

3.3.2. Task 1: Perception

The first task was to identify a specific vowel sound. The software ﬁrst collected
some demographic data, including place of birth, age, and native language. This was to
identify the two groups to be compared: natives of Michigan and Korean learners of

English. Next, it presented the participants with a brief introduction to the vowel

61

quadrangle, and to computer-generated speech. Some brief tasks in vowel identification
were presented in the form of a game. The purpose of this was to assist them in
navigating the software and familiarize them with synthesized speech sounds. The few
participants that had problems with the software were able to overcome them before
beginning the actual task.

This orientation phase of the program gradually introduced the participants to the
concept of the vowel space, and how this was modeled in the vowel grid that was the
medium of the first task. Just before beginning the task, participants were able to explore
the vowel grid at their leisure, familiarizing themselves with the way the vowel quality
gradually changed with their movement of the cursor through the gradient of F1 and F2
values presented in the matrix.

The first task was to identify the vowel sound that best matched the sound found

in each of 11 sets of words. The vowel sounds to identify were /i I e e m a a o o u u/.

These represent the typical analysis of Standard American English monothon gal vowel
phonemes. For each vowel sound, participants were presented with a list of 5
monosyllabic words with the vowel sound. The five words represented a variety of
environments and spellings. The words lists are presented in Table 6. Participants were
presented with a word list in writing only. There was no audio recording of words.
Participants were instructed to explore the vowel grid by clicking on squares in the matrix
to locate the sound that best matched the vowel sound in those words. They identified the
sound by double-clicking the square. After the participant identiﬁed a vowel, the program
presented the next word list. The computer program tracked the participants’ mouse

clicks, and recorded to a server the list of clicks and the double-clicked square. Each

62

participant had a unique record in the database, which was tagged with an identification

number.

 

Vowel Word list
i eat, he, bead, sleep, peak

 

 

H

hit, bid, lip, pick, slip

 

hate, late, paid, race, lake

 

pet, head, tell, red, wet

 

hat, dad, sap, tap, rat

rod, hot, dock, sod, nob

but, duck, cut, hush, rub
caught, law, draw, paw, pause

 

 

 

 

blow, so, rose, toe, post

 

book, hood, look, push, would

 

 

 

 

Cicooomeamo

who, loose, you, blue, spoon

 

Table 6: Word lists from task 1.

3.3.3. Task 2: Production

The software program prompted the participants to read a short word list,
presented in writing, into their computer’s microphone. The list of words was heat, hit,
hate, pet, hat, hot, hut, pause, toe, book, who. The two considerations in choosing the
words on the list were familiarity to the participants, and the acoustic properties of the
vowels produced when speaking them. These words were chosen because they were
thought to be more familiar to nonnative speakers. The word list was in an order that
circumscribed the vowel quadrangle. This was done in an effort to maximize contrast
between the vowel sounds in the words by the participants when they read the list. The
word list was presented in the same order to each participant, on the assumption that

attempting to maximize dispersion was preferable to counter-balancing.

63

Because certain onset consonants, such as voiced bilabials, produce a formant
transition in the following vowel, the word list avoided the use of voiced onsets. This
decreased the likelihood that the vowel sound produced by the participants was
inﬂuenced by its phonetic environment.

The words were presented in an order that followed the outline of the vowel
quadrangle. The vowel sound of each word on the list was adjacent to a word whose
vowel sound was adjacent to it in the vowel space. This was done with the intention of
maximizing contrast between vowels.

The audio was captured using the Flash MX microphone control, and was
streamed to the server in real-time using the Flash MX Communication Server.
Participants used Telex headsets with a built-in boom microphone. The audio was
sampled digitally at 22,050 Hz with a 16-bit sample rate. The audio ﬁle for each
participant was tagged with the participant’s unique identification number, which linked
the participants’ responses on both tasks to the demographic data.

After the participants clicked a button indicating that they had finished recording,
the program informed them that they had completed the task, and thanked them for their
participation. Native speaker participants completed both tasks in 10 to 20 minutes.
Normative speakers finished both tasks in 10 to 45 minutes, with most completing within

25 minutes.

3.4 Data
Since participants’ responses to the identiﬁcation task were recorded to a database,

they were readily available for recovery and analysis. Each square on the vowel grid was

internally tagged with a unique code, which would be used to recover the participants’
choices for each vowel.

The software program Praat was used to extract the F1 and F2 values for each
vowel from the production data of each participant. The values were taken from the
earliest point in each syllable where the vowel became stable, typically between 75 and
125 milliseconds after the beginning of the onset. These values were entered into the
database as well. Several native speaker participants did not speak clearly or loudly
enough, and so their audio ﬁles could not be used. Some other participants chose not to
record their voices. Audio data from 46 of the 87 native speaker participants could be

used, and 24 of the 27 nonnative speaker participants’ audio ﬁles could be used.

65

 

Chapter 4: Results

4.0 Introduction

In this chapter, I first show the results of both the perception and production tasks
in graphic and numeric format. I examine the data in terms of within-group homogeneity
and perception—production differences, and then look at between—group differences. Next,
I show a statistical comparison of within-group and within-task results.

Using custom software that accesses the data from the server and plots it onto a
grid, the average F1 and F2 values of participants’ responses were plotted in the X-Y
grid of Figure 9. Values for F2 make up the X axis values, and F1 values are arranged on
the Y axis. By positioning the 0 point of both axes in the upper-right hand corner, instead
of the lower-left hand corner, the vowels can be plotted into positions analogous to the
traditional vowel diagram. The F2 parameter reﬂects frontness of the vowel, and the F1
represents height. Fronter vowels fall to the left of the origin point, and higher vowels
towards the top.

The data were stored automatically into a database on the server during data
collection. Scripts written in the PHP programming language ran on the server to retrieve
the data, and to calculate averages. The results were collected by software written in

Macromedia Flash, which displayed the data in graphical form.

66

2600 F2 475 o

 

200

F1

 

 

1000

 

Figure 9: Graphic representation of the vowel area . The gray shaded area represents F1
and F2 values that were not represented by sounds in the perception task, but that could
possibly be produced by participants during the production task.

The statistical procedures used for analysis largely consist of heteroscodastic t-
tests. According to Howell (1995:246—247), the t-test is the best choice for comparing the

means of two independent groups. ANOVA was also used to identify the component

factors of variation.

4.1 Native Speakers

4.1.1 Task 1: Perception

Figure 10 shows the average judgments of the 87 participants who identified

themselves as native speakers from Michigan. The positions of the vowels relative to

each other are in a pattern acoustically consistent with the English vowel system, as in

Ladefoged (1993:212).

67

 

2600 F2 475 O

 

 

 

1000

 

Figure 10: Average of Native Speaker Perception

This positioning of the vowels indicates that in this study, as in the studies by
Scholes (1967b, 1968), the participants were able to associate the synthesized sounds
with vowel sounds in their own language systems. The data from this task was consistent
with other acoustic measurements of English vowels.

The average numbers shown in Figure 10 belie the variation in responses. Figure
11 shows the choices of all native speaker participants for each vowel sound. Trends are
clearer for some vowel sounds than for others. The responses for /i/, /ze/ and /u/ appear to

cluster together more than those for /e/ and /e/, for example. This variation may be a

68

reﬂection of natural variation within the speech community. This kind of wide variation
was also found in the study by Frieda et al (1999). In that study, participants chose the
synthesized v0wel sound that best matched a particular vowel sound. There was
considerable variation among the responses. It could be claimed that synthesized vowels
are not natural-sounding enough to make a clear distinction, or that confusion or fatigue
on the part of the participants inﬂuenced their responses. However, in light of the fact
that this is not the only study that found this kind of variation, it seems more likely that

this phenomenon is a reﬂection of inter—speaker differences.

69

 

 

 

 

 

 

 

 

 

 

 

 

Juan-.4
«Add .1 a- .r .1 .I
dIrad a] I u 4
a..a. III .xauu .- 4 l
a“ an Ida 4.1 .i .1
.4 argue: d d .‘u «an
a .. .4 .J .1 die-1.1.5.101 41 A .11.» u .t‘
.’ dida‘dudd .— o‘oluéuu .4 a |>~~N6 4.1 .1
.1 “Isaac; ~ .IJdé» .5; Add-'9
do .i dfu‘éd'noe dot-'9
u .1 .14 4.6» «K...» .. I~"o‘o‘.-’
. .H u-a .— «1.3.1.90»;
’ .1 . Jun
.1 j a at .6 «- 4a.: .1
.4 ‘ I4: I aw
.i .- '..~
'7 A C r >_— — T— _— _—‘—u——.J
H /I/ /e/ /e/
l I
Av i I .i
.:i ~.
.‘le .3.- ,I. K“ .2
o" n .; .:._ .0 .3 . . .i
.5 .2. 1
ol 0.1 I ul‘s‘ «a .J
I .0 ..:..> .
I *l - .J u «a
l w—nvdv-‘A I g..‘..)..‘ .. ~
.A .103 I ".36).! 03¢») . . s‘
.‘d 63.).-.) ..‘ . e. ’0 o‘.
.1 .J‘I‘l ..| 4...; 050-304 o'- 039’”
dd‘u..l€o.‘-J cod-.3093“; . . . .2.).
.1 adage: .4 I a ._‘..>.. ..‘ . .:.: . up»
I ﬂue-J o w I -— -J t o o’ e3~~0 >3
‘ - 9.)-.- do" .1 w -1 i Q a. . .
_.,, have? .1 “5‘“‘“_‘ “1r' ’ ’ gr “ '— .‘ r 0:*~*“

 

 

 

 

 

 

 

 

 

 

 

 

-— - ~ e u... . - ... .a. - . .
l
, -L ~31 '- «14.4
. u‘ m .2. «an Al .0 J: Jr.“
1 .c- .L M w 9'. w l a. «1.2.9:: .0
I W- .1 w .rou .u I .11 .1: wind»: .u
o»; m - w .u discus-Jul! I .94! urcll‘uc‘dw
(Jaws ococ o-er.) .L‘ .uwuvl «UV-H ! .1 ell-5| .um
i g.) .~ .qu'w E .‘l .ri'm'.
I oﬁvoﬁ'ﬁw .o ‘ .11.le ‘0 i oil .1'
a mew-t lava -. wow ,1- | .0
I ,{3 .r .0 .: r3 2.) .1 I
' 1.!)9‘ u .3 .41.: .1 .i
I .o .o
0’." u'
I .z: .0
l N c' - u
i .n #3“ .a .l _
. _ 7 22. .. __ _ W——"" l . ... w-.__-_ i. ._. ‘3 ._.. .. .. r__l

/0/ /U/ /u/

 

Figure 11: Native speaker choices for perception task for each vowel

70

 

 

 

Table 7 shows the means of F 1 and F2 for the vowels that were tested for in task 1.

Standard deviations are given below the means, in parentheses.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Vowel F1 F2
1 346 2320
(176) (459)
I 529 2123
(137) (450)
e 632 2220
(135) (480)
e 652 2156
(172) (405)
a: 803 2138
(153) (422)
a 815 1283
(163) (391)
o 652 1331
(171) (390)
o 779 1291
(150) (429)
o 613 909
(178) (356)
u 508 1067
(175) (346)
u 409 953
(184) (365)

 

 

 

Table 7: Means for the native speaker perception task. Standard deviations for each mean
are under the mean, in parentheses.

Two pairs of vowels (/e/-/e/, /a/-/o/) had F1 and F2 values that are very close

together, within 40 Hertz for the F1 values and within 60 Hertz for the F2 values. When
plotted onto the vowel space, these pairs of vowels were very close together. This
similarity of position suggests the possibility that the participants did not distinguish
between the vowel categories. Two-tailed t-tests were performed to determine if the

differences between the means were significant or not. Heteroscedastic t-tests were

71

 

performed to test two samples with unequal sample sizes. The t-test for the /e/-/8/

distinction yielded t(175) = 0.96, p = 0.34 for F2, and t(171) = -0.89, p = 0.38 for F1. The

t—test for the /a/—/o/ distinction yielded t(179) = —0. 14, p = 0.89 for F2, and t(179) = 1.56,

p = 0.12 for F1. In the t-test to see if there was a difference between the means, the null
hypothesis was that there would be no difference. In all four t-tests, the observed value
was less than the critical value of t, 1.97, and so we cannot reject the null hypothesis. We
cannot assume that there is any significant difference between the F1 or F2 values for the
pairs of vowels in question.

To examine the system of native speakers as a whole, an ANOVA test was
performed that included a post-hoc pairwise comparison test using the Tukey method.
This test revealed the level of variation between vowels in the perception test. Table 8

shows the results of the pairwise comparison test for F1 and F2.

72

o
N:
is

cooomamo
009.099.6993

0.
0.
0.
0.
0.
0.
0.
0.
0.

t—i r-I-

cooomamo
OOOOOOOO
Ov—‘pppppppp

.735

 

Table 8: The results of a pairwise comparison from an ANOVA test shows the source of
variation between pairs of vowels in native speakers’ responses to the perception task in
the F1 formant (top) and F2 formant (bottom). Only two pairs of vowels (/e/-/e/ and /a/—

/o/) showed no significant difference between them.

Vowel pairs that showed no significant difference were those with a p value less
than 0.05. In evaluating this data, it is necessary to look at both the results for F1 and F2
for each vowel pair. One formant will only show one dimension of the vowel, and it
would be inaccurate to say that vowels are not distinct on the basis of only one formant.

For example, the F1 values of /1/ and /u/ are very close, showing a p value of 0.999.

73

 

However, we cannot conclude from this that the vowels are not distinct. In fact, we would
expect the F1 value for those two vowels to be very close, because they are both high
vowels, and the F1 formant indicates vowel height. The difference between these two
vowels is revealed by the F2 formant. The comparison of the F2 values for those vowels

shows a p value of 0.000. Again, this is to be expected, since /I/ is a front vowel, and /U/

is a back vowel, and F2 indicates vowel frontness. For purposes of this comparison, then
we assume that vowels are not distinct only if the p value for both F1 and F2 are over

0.05. Under that assumption, only the vowel pairs /e/-/e/ and /a/-/o/ showed no significant

difference between both the F1 and F2 values for native speakers’ perception. This
ANOVA test indicates that all the other vowels that were identified by the native

speakers in the perception task are distinct.

4.1.2 Production

Figure 12 shows the average values for perception, from Figure 10 above, along
with the average values for production. As with the perception task, the averages of the
F1 and F2 values for the 87 participants are consistent with previous theoretical and
empirical work. For most vowels, there seems to be a close match between native i
speakers’ perception and production. The two exceptions are /e/ and /a:/. Those two
segments are higher and more forward in production than they are in perception.
This phenomenon is consistent with the Northern Cities vowel Shift (Labov et al 1973,
Labov 1994, Labov et a1 1997). The shift involves the raising of low front vowels into the
range occupied by mid and high front vowels. The acoustic properties of vowels on .
American English are changing as the vowel system undergoes a vowel chain shift. A

chain shift is a repositioning of vowels within a system. This chain shift involves a

74

shifting of the positions of six low and mid vowels. As one vowel moves into the acoustic
space occupied by another vowel, the other vowel moves as well. The cascade of
movement results in a reorganization of the vowel system. Labov (1994: 178)
characterizes the Northern Cities Shift as “one of the most vigorous sound changes now
in progress in the United States.”

This shift has resulted in a measurable change in the pronunciation of vowels in
the affected areas of the United States. According to Labov et al (1997), Michigan is in

the area affected by the vowel shift. The major trends in the shift are summarized in

 

 

 

 

 

 

 

Table 9 below.
Original Post-shift
/a/ /2e/
/0/ /a/
/a/ /e/
/I/ /e/
/e/ /a/
/o/ /o/

 

 

 

 

Table 9: Main effects of the Northern Cities Shift. Column 1 shows the vowel that is
affected. Column 2 shows the vowel space that the shifting vowel moves into (from
Labov et a1 1997).

The research program examining this vowel shift centers on measuring production
data. While the shift in pronunciation is well established, the relationship between the
pronunciation of the vowels and the perception, if any, has been unclear. This vowel shift
is changing the quality of vowels that native speakers of English in Michigan produce.
The result of this shift will be a difference between native speakers’ intuitions about the
acoustic values of vowels, and those that they actually produce. There will be, in other

words, a widening competence-performance distinction among speakers in areas affected

by this vowel shift.

75

 

In that light, the results of the production task are not surprising. What is

interesting is that the shift in the low front vowels is only evident in production, not in

 

 

 

 

perception.
2600 F2 475 0
200
4i *
11
I ’0 F1
’q, 9 '0
'a e ’3
1000
Perception

”Fl- 741577
’11
'I
c U
ac}. '9’
a
'a
Production

Figure 12: Average Native Speaker Perception and Production

76

200

F1

 

1000

 

Figure 13 shows the aggregate production data for each vowel. The production of

native speaker participants is much more uniform than their perception data, with very

few within-group outliers.

 

 

 

 

 

 

 

 

 

0|
J 'f' 4.- u-
r, w n 40"“; .. '
1‘} ‘ “'1'...“ ‘ .4» .4
{I .4 11 :7.” 4 I.
.1 a .,
| :L" *1 d h “u. ’i
‘ I
I 4.“ h
I "7 J: Jr —: u‘
i I ‘3
l _ ___-_v- E__ _ ...... _ _E c E W . ,_._ 7'
I
I
I l I T T l 7 MW H
l l
‘ I l I'
I ~ ‘ . l
I Q I I l
I n I I r . i
' .Ze ~_ .."“_ 3. I I
‘ ' ‘1 d .A ~ “ :9 _.
ti.‘ 3”.» ‘91: :‘ | T
at «i- “has . .c. - “ -‘ .. ..
me: "’ ‘ N- ._ .. ,f‘ I *3, ,:
I «6:119 if - A ‘.‘._ 7 ~‘ kt I '3 -'
“as? . ”~ is i 22:?"
‘ ~._ 9 .1
I q: ‘ -.- I ‘ 5 1‘1
. «was; I s: I ‘ . ‘3 4
I _ __> do. i Q ., E ,l ’
I TI ‘ V F“ A“ 7_ — -7
: I i ‘
' ‘ l
i ; I l
| I .1 "it;
i 2. ‘ ..~"-.:
I . ..... u~$émrﬂ
, v. .u 'I - ‘ .I
””43" '3‘" -
I ‘ e u, I“ .' ,u
:1 ~ 4» -'
'5 ._ ,n
I I
I I

 

Figure 13: Native speaker performance on production task

Several tokens of /i/ and /a/ fell outside the limits of the graph. Two factors may

have inﬂuenced the values found for /i/. The gender of participants may have been a

contributing factor. There were several female participants, while the synthesized vowels

were in the range of a male speaker. Since the typical female vocal tract is shorter than

77

 

 

that of the typical male, the formant frequencies of utterances by females are generally
higher.

The other factor could have been the co-called “hyperspace effect” hypothesis of
Johnson (2000), according to which participants choose hyperarticulated vowels in an
apparent effort to maximize vowel contrast. Although there seemed to be a preference for
/i/ sounds with very high F2 values, this effect seems to be limited to front vowels. If
there was preference to hyperarticulate vowels to maximize vowel contrast, then we
would expect to see a preference for back vowels to have lower F2 values. In fact,
however, there was no evidence of preference for the other high vowel in this study, /u/,
to have very low F2 values. In fact, only one participant chose the 450 Hz F2 value for /u/,
which was the back-most value available to participants, and only four made an F2
selection that was less than 1000 Hz. In production, the minimum F2 value was 727 Hz,
and only 5 participants produced an F2 less than 1000 Hz.

The other unexpected ﬁnding was the range of F1 values of the vowel /a/. This
vowel had higher F1 values than was predicted in the literature. Ladefoged (19931193)
indicates a Fl value for /a/ at 710 Hz. In this study, however, the F1 values for /a/ ranged
from 459 Hz to 1279 Hz for production. That F1 values would be as high as they were in
the production task was not anticipated in the creation of the perception task of the
instrument. The vowel matrix provided vowel sounds with a maximum F1 value of 1000
Hz. In the perception task, the range for /a/ was 250 to 1000 Hz. The high F1 values in
production suggest that the perception task should have included a greater range for F1.
Ten participants chose the maximum F1 value in the perception task (100 Hz), and 9

participants produced F1 values in excess of 1000 Hz.

78

The means for the vowels are given in Table 10. Standard deviations for each

mean are shown in parentheses below the mean.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Vowel F1 F2

i 305 2673
(100) (263)

I 505 2191
(95) (255)

e 461 2480
(88) (254)

e 738 1912
(140) (240)

a 699 2249
(183) (261)

a 879 1489
(160) (193)

a 706 1396
(139) (154)

o 779 1219
(96) (223)

o 607 1377
(102) (202)

U 576 1226
(90) (3 18)

u 407 1352
(57) (353)

 

 

Table 10: Mean formant values for native speakers: production task. The standard
deviation for each mean is shown in parentheses below the mean.

F1 and F2 means for the vowels /e/ and /a3/ appeared to be close together. The F1

values differed by only 39 Hz, and the F2 values differed by 237 Hz. T-tests were
performed to determine the significance of the differences. The null hypothesis was that
there would be no difference. The t—test for the means for F2 yielded t(91) = -6.51,

p = 0.00, and the t—test for F1 yielded t(86) = 1.19, p = 0.24. The observed value for F2

exceeded the critical value of 1.99, but the observed value for F1 did not. We can thus

79

 

reject the null hypothesis that there is no difference between the respective F2 values for

/e/ and /ae/. There is a statistically significant difference in frontness between these two

vowels. However, we cannot reject the null hypothesis that there is no difference between
the F1 values for that pair of vowels.

An ANOVA test was performed to show the level of variation among vowels in
the production system. The p values for each vowel pair in the post-hoc pairwise

comparison is given in Table 11.

80

 

C

)—l

99.09.0999
99.09.09.099

cooompamo
\o
\l

0.
0.
0.
0.
0.
0.
0.
0.
0.
0.

H v—I-

sax—9.0.0.099

9&9999999
9.09.09.09.09

cooomajmo

 

.0
D.)
00

Table 11: The results of a pairwise comparison from an ANOVA test shows the source of
variation between pairs of vowels in native speakers’ responses to the perception task in
the F1 Formant (top) and F2 formant (bottom). Only one pair of vowels (/o/-/u/) showed

no signiﬁcant difference between them.

As outlined in the section on native speaker perception in 4.1. 1, we assume that

vowels are not distinct only if the p value for both F1 and F2 are over 0.05. Only one pair

of vowels, /o/ and /U/, showed no significant difference between both the F1 and F2

81

 

 

values for native speakers’ production. This ANOVA test indicates that all the other

vowels that were identified by the native speakers in the production task are distinct.

4.2 Normative speakers

4.2.1 Perception

Figure 14 shows the average values of the normative speakers’ choices in the
perception task. While the vowels are arranged in the general pattern that indicates the
participants identiﬁed the synthesized vowels as speech sounds analogous to natural
speech, the pattern has noticeable differences from that of native speaker participants.

First is the close convergence of /i/ and /I/. The average of F2 for /i/ was 2382 Hz, and
2419 Hz for /I/. The average F1 was the same for both vowels: 335 Hz.
The very low placement of /e/ and /e/ clusters these very closely with /2e/. It

seems that nonnative speakers had difficulty discerning these three vowels in the

perception task. The average F1 for /e 8 ae/ were 676, 713, 759, respectively, and the F2

values were 2160, 2174, 2128, respectively.

Normative participants also placed /u/ close to /u/, which may mean that they had
difﬁculty differentiating those two vowels as well. The placement of /0/ was higher than

/a/, near /o/.

82

2600 F2 475 o

 

 

 

200
H” +11
TU
F1
1 3 I 0
+9
+8
*36‘ "- 9+3

1000

 

Figure 14: Average Normative Speaker Perception

Individual participants’ choices are shown in the scatterplot diagrams in Figure 15.
As with native speakers, there is considerable variation in the positions of the tokens for
each vowel, although for each vowel, a general trend can be seen. The grouping of /i/ and

/I/ along the left edge of the grid indicates that the participants could have preferred

tokens with F2 values that were even higher than those that were available for this task.
Fifteen participants, more than half, chose the maximum F2 value for /i/, and fourteen

chose the maximum F2 value for /r/. For all the front vowels /i, I, e, 8, $/, there is a

preference trend toward more fronting, represented by higher F2 values. This could be a

result of ﬁrst language inﬂuence.

83

 

 

 

 

o'e-
.4
u.- d
d:
.Eu
4..

 

-10
J"
tut-- us.
.P
I
.m
. "‘
.k'l‘ -1“
.3th
.9 new ' "'nr ‘
d
.4
“a
v:
__ _ __ _ ._ P- -J'

"/8/

 

 

 

' —_"_ _ -—"‘ ' 1
l
I
0' l
a u |
... a .. I
djot .I
g.
" l
l
l
I
,ﬁw _
I
s)
l
.I .1
— .l I c-J
u) .3.
as :3 3‘.
_ __ ._ _ ..---..,I
I
p) I
U
,I
l
I
___ _ _ j..—

 

.4
I 7'

'4

.5
r10

‘- ..
.‘u 4

v

4
to .r
I. _ _
\
I
I
I
.J
6.2
.‘ x.”
a;
l

'1
i
I
.
|
.
i

 

‘. 1.1. L

/o/"

 

Figure 15: Normative speaker choices for perception task

84

 

 

 

 

Table 12 shows the means of F l and F2 for the vowels that were tested for in the

perception task for nonnative speakers.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Vowel F1 F2
i 335 2382
(127) (467)
I 335 2419
(103) (284)
e 676 2160
(182) (516)
8 713 2174
(203) (484)
ae 759 2128
(212) (515)
a 774 1244
(232) (446)
a 743 1378
(156) (393)
o 606 1220
(185) (539)
0 615 989
(193) (523)
U 400 1072
(164) (405)
u 361 1142
(151) (527)

 

 

Table 12: Means for the normative speaker perception task

T-tests were performed to test the difference between the means for several pairs
of vowels that varied by less than 50 Hz for F1, and 100 Hz for F2. The results are
summarized in Table 13 below. In every t—test, we could not reject the null hypothesis

that the means were statistically the same. The vowels /i I/ were virtually the same. Three
2-way t-tests were performed on the combinations of /e 8 ae/. There was no statistically

significant difference found in any of the tests. It may be accurate to say that the

85

participants’ responses to the perception task revealed a 5-vowel interlanguage system, as

shown in Figure 16.

86

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Distinction Formant Observed Signiﬁcant
value

i-I F1 0.00 NS

F2 -0.35 NS

e-e F1 -0.71 NS

F2 -0. 10 NS

e-a F1 -1.55 NS

F2 0.23 NS

e-a: F1 082 NS

F2 0.34 NS

a-a F1 —0.58 NS

F2 1.17 NS

u-u Fl 0.91 NS

F2 0.54 NS

o-o Fl -0. l 8 NS

F2 1.60 NS

 

 

 

 

 

 

Table 13: Between-category t—tests: nonnative speaker perception

87

 

 

 

 

2000 F2 475
200
/’/A’\
«an I -
\ v/ ’11] ‘I
. "u ,
F1
(’1‘ 5 ﬂ VT6\
/" 7' \ \c, .
Kt? .2
i a
. we " 1" .
.\ :/ \er I fl/
1000
Figure 16: the Normative vowel areas: perception task

An ANOVA test was performed to show the level of variation among vowels in

nonnative speakers’ perception. The p values for each vowel pair in the post-hoc pairwise

comparison is given in Table 14.

88

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I C 8 & a 9 O O U 11
i 1.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.960 1.000
I 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.960 1.000
c 1.000 0.820 0.625 0.952 0.932 0.974 0.000 0.000
8 0.997 0.974 1.000 0.488 0.625 0.000 0.000
as 1.000 1.000 0.059 0.101 0.000 0.000
a 1.000 0.023 0.042 0.000 0.000
9 0.148 0.229 0.000 0.000
9 1.000 0.001 0.000
0 0.001 0.000
0 0.999
I C 8 a a 9 O O U u
i 1.000 0.814 0.868 0.654 0.000 0.000 0.000 0.000 0.000 0.000
I 0.628 0.703 0.448 0.000 0.000 0.000 0.000 0.000 0.000
9 1.000 1.000 0.000 0.000 0.000 0.000 0.000 0.000
8 1.000 0.000 0.000 0.000 0.000 0.000 0.000
as 0.000 0.000 0.000 0.000 0.000 0.000
a 0.994 1.000 0.654 0.961 0.999
9 0.978 0.088 0.375 0.750
9 0.772 0.986 1.000
0 1.000 0.983
0 1.000

 

 

 

 

 

 

 

 

 

 

 

 

Table 14: The results of a pairwise comparison from an ANOVA test shows the source of
variation between pairs of vowels in nonnative speakers’ responses to the perception task
for the F1 formant (Top) and F2 formant (Bottom). Several pairs of vowels (/i/—/I/, /e/-/€/,

/e/-/ze/, /8/-/a3/, /a/-/9/, /9/-/o/, /9/-/o/, /o/-/o/, and /U/-/u/) showed no significant difference
between them.

The results of the ANOVA test were consistent with the t-tests. There was no

difference between several vowel pairs: /i/—/I/, /e/-/e/, /e/-/&/, /e/-/2e/, /a/-/9/, /9/-/o/, /9/-/o/,

/o/-/o/, and /U/—/u/.

89

4.2.2. Production

The difference between perception and production of normative speakers is shown
in Figure 17. In the production task, there is greater separation of /i/ and /1/, however, /I/
is higher than /i/ and is almost as fronted. The positions of /8/ and /ae/ are even closer in
the production task than in the perception task, almost completely overlapping. Because

of the great variation of /0/ among the participants, its average position is between /e/ and

 

 

 

 

 

 

 

/£/.
2600 F2 475 0 l2600 F2 475 0
200 1 200
11 Rb ‘ iI 11

F1 ‘ +6 '0 F1

* ‘3 '0 i 901 e

€16 ‘ a l '3
1000 ‘ a 1000

Perception Production

Figure 17: Average Nonnative Speaker Perception and Production

The plots of all tokens for the production task in Figure 18 show the variation
among tokens, but similar to the production data in native speakers, there was greater
uniformity among tokens in the production task than for the perception task. As with the
native speakers, there were several instances of high F2 values exceeding the limits of the
grid. This again could be because of the gender of the participants. Female voices tend to
have higher F2 values than males because of their shorter vocal tract. Also similar to

native speakers, there were many instances of a high value of F1 for the vowel /a/.

90

As with the native speakers, the nonnative speaker participants exhibited a

preference for a greater range in F1 than anticipated. There were 5 instances of F1 values

in excess of 1000 Hz for /a/, and two cases of F1 values over 1000 Hz for /o/. The greater

range of F1 in the production task suggests that the perception task would have been able

to more accurately reflect the intuitions of the participants if the synthesized vowels in

the vowel matrix had included a greater range for F1.

 

 

 

 

 

 

 

/1/

 

Lat

.".
‘ -
1 ‘1 i
~u.___ ‘vww .__ p_~_—

>1

Au

 

 

/e/

 

 

 

cl 1
‘4
. I“ w ~~

«131:9

 

 

/u/

 

/o/

 

Figure 18: N onnative speaker performance on production task

91

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Vowel F1 F2

i 356 2584
(51) (369)

I 399 2537
(58) (389)

e 572 2349
(97) (409)

8 705 2106
(135) (339)

a: 674 2058
(121) (422)

a 879 1368
(202) (284)

9 656 1096
(132) ( 153)

o 628 1275
(21 1) (536)

o 547 1 195
(1 18) (47 l)

U 437 l 175
(144) (413)

u 429 1062
( 139) (264)

 

 

 

 

Table 15: Means for the normative speaker production task. The standard deviation for
each mean is shown in parentheses below the mean.

The averages for the vowels /i I/ and /8 218/ were very close to each other, all within 50 Hz,
so t—tests were performed to test for signiﬁcance of difference. The t-test for the pair /i-I/,

the t-test yielded t(48) = 0.44, p = 0.66 for F2, and t(47) = -2.77, p = 0.01 for F 1. For the

pair /e-&/, the t-test yielded t(46) = 0.45, p = 0.66 for F2, and t(47) = 0.85, p = 0.40 for
F1. The only case in which the null hypothesis could be rejected was the F1 values of /i-I/,

which had an observed value of —2.66, greater than the critical value of 2.01. Those

vowels could be distinguished only on the basis of height.

92

 

nonnative speakers’ production. The p values for each vowel pair in the post-hoc pairwise

An ANOVA test was performed to show the level of variation among vowels in

comparison is given in Table 16.

 

 

 

 

 

 

 

 

 

 

 

 

I e 8 a a
1. 0.5 0.001 0. 0. 0. 0. 0.
0. 0. 0.001 0. 0. 0. 0.

e 0.53 0. 0. 0. 0. 0.
8 1. O. 0. 0. 0.
33 0. 0. 0. 0.
a 0. 0. 0.
9 0. 1. 1.
0 1. 0. 0.
0 1. 0.
U 0.

1 e e m a a o 0 U u
i 0.993 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.644 0.773
I 0.001 0.000 0.000 0.000 0.000 0.000 0.013 0.997 1.000
6 0.043 0.292 0.000 0.587 0.954 1.000 0.035 0.019
8 1.000 0.001 0.981 0.698 0.005 0.000 0.000
a 0.000 1.000 0.986 0.064 0.000 0.000
a 0.000 0.000 0.000 0.000 0.000
9 1.000 0.198 0.000 0.000
0 0.649 0.000 0.000
0 0.189 0.118
0 1.000

 

 

 

 

 

 

 

 

 

 

 

 

Table 16: The results of a pairwise comparison from an ANOVA test shows the source of
variation between pairs of vowels in nonnative speakers’ responses to the production task
for the F1 formant (Top) and F2 formant (Bottom). Several pairs of vowels (/i/-/I/, /e/-/EB/,

fe/-/2e/, /9/-/o/, /9/—/o/, /o/-/o/, /O/-/U/, /o/—u/, and /U/—/u/) showed no significant difference
between them.

93

The results of this ANOVA test show that there was no difference between

several vowel pairs: /i/—/I/, /e/-/ae/, /8/—/ae/, /0/—/o/, /9/-/o/, /o/-/o/, /O/-/U/, /o/-u/, and /u/-

/u/.

4.3 Perception—production differences
4.3.0. Introduction

Table 17 shows the mean F1 and F2 values for perception and production for each
vowel, separated by native and normative speaker group. The standard deviation for each

mean is shown in parentheses below the mean.

94

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Native Speakers Nonnative Speakers
Vowel Perception Production Perception Production

i F1 346 290 335 390
(176) (101) (127) (52)

F2 2320 2659 23 82 2640

(459) (266) (467 ) (377)

I F1 529 501 335 350
(137) (96) (103) (59)

F2 2123 2185 2419 2617

(450) (258) (284) (398)

e F1 632 460 676 689
(135) (89) (182) (99)

F2 2220 2477 2160 2475

(480) (257) (516) (418)

8 F1 652 728 7 l 3 689
(172) (141) (203) (138)

F2 2156 1912 2174 2074

(405) (243) (484) (346)

2e F1 803 700 759 692
(153) (185) (212) (123)

F2 2138 2240 2128 2079

(422) (264) (515) (431)

a F1 815 875 774 992
(163) (161) (232) (206)

F2 1283 1493 1244 1385

(391) (191) (446) (290)

9 F1 652 703 743 763
(171) (141) (156) (135)

F2 1331 1398 1378 1041

(390) (156) (393) (156)

0 F1 779 777 606 624
(150) (97) (185) (216)

F2 1291 1222 1220 2188

(429) (225) (539) (547 )

0 F1 613 61 1 615 696
(178) (104) (193) (120)

F2 909 1385 989 2123

(356) (204) (523) (481)

U F1 508 578 400 495
(175) (91) (164) (147)

F2 1067 1183 1072 1401

(346) (322) (405) (422)

u F1 409 407 361 399
(184) (57) (151) (142)

F2 953 1333 1142 1239

(365) (357) (527) (270)

 

 

Table 17: Mean and standard deviation values for perception and production, native and
nonnative speakers

95

4.3.1 Intra—group perception versus production

Tables 18 and 19 show the results of ANOVA tests measuring the difference
between the perception and production tasks. One prediction of the study was that both
perception and production would be more closely aligned among native speakers than

nonnative speakers.

0
C

99.099999
O\
U]
p—L
b»)

oocsbpppz—mo

C
8
E
a
9
O
O
U

92.099.099.09

6 8

0.1 0.
0.013 0.871
0.

H v—--

OOOOOOOO
99.09.099.09

 

cooowﬁmo
t—‘r-‘QSDSDSDPPPP

Table 18: The results of ANOVA tests comparing perception and production among
native speaker subjects for F1 (Top) and F2 (Bottom). Only the vowel pair /a/ - /3/ shows

no significant difference (p>.05) between pairs for both F1 and F2.

96

 

 

 

 

 

 

 

 

 

 

 

 

I e 8 a3 a 8 3 O U u
i 1.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.454 0.916
I 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.864 0.999
c 0.262 0.128 0.000 0.400 1.000 0.950 0.000 0.000
8 1.000 0.016 1.000 0.115 0.004 0.000 0.000
33 0.044 1.000 0.048 0.001 0.000 0.000
a 0.007 0.000 0.000 0.000 0.000
9 0.199 0.009 0.000 0.000
9 0.994 0.000 0.000
0 0.000 0.000
U 1.000

 

 

 

 

 

 

 

 

 

 

 

 

H.

I

0.43
0.93 0.
0.931 0.85
1. 1.

1.
Table 19: The results of ANOVA tests comparing perception and production among
nonnative speaker subjects for F1 (Top) and F2 (Bottom). The vowel pairs /i/-/1/, /e/-

/8/,/e/-/ze/,/8/-/ae/,/o/-/9/,/o/—/o/,and /U/-/u/ show no significant difference (p>.05) between
pairs for both F1 and F2.

0. 0.
0.74
0.7

 

cooomama

The pairwise comparison of perception and production among native speakers
that is shown in Table 18 shows that the contrast between only one pair of vowels, /a/ and

/0/, is not distinct between the two tasks. The data for Normative speakers shows a lack of
contrast among several pairs of vowels: /i/-/I/, /e/-/e/,/e/-/2e/,/e/-/2£/,/o/-/9/,/o/-/o/,and /U/—

/u/. This data strongly supports the hypothesis that native speakers would be more

97

consistent between perception and production than nonnative speakers would.

4.3.2. Cross-group perception and production

Table 20 gives the standard deviations of F1 and F2 for each vowel for native
speakers (NS) and normative speakers (NNS). For comparison purposes, and where
comparable vowels are available, the standard deviations from the study by Peterson and
Barney (1952) are also given, labeled “PB.” The category that has the greatest standard
deviation, and thus the greatest amount of within-group variation, is marked in bold.
There is a striking difference between F2 and F1. Nonnative speakers have the greater
variation in F2, which measures frontness, than native speakers for almost all vowels.
The two groups are almost evenly split in number of categories with the more variation
for F1, which measures vowel height.

One hypothesis of the study was that the native speaker group in this study, which
was composed of members from the same speech community, would exhibit less
variation than nonnative speakers. While the data in Table 20 seems to support that
hypothesis for vowel frontness, as measured by F2, there seems to be no trend either way

in vowel height, measured by F1.

98

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

F2 F1
Vowel PB NS NNS PB NS NNS
i Perception 459 467 176 127
Production 374 266 377 60 101 52
I Perception 450 284 137 103
Production 337 258 398 75 96 59
e Perception 480 516 135 182
Production 336 257 418 97 89 99
e Perception 405 484 172 203
Production 243 346 141 13 8
a: Perception 422 515 153 212
Production 288 264 431 172 185 123
a Perception 391 446 163 232
Production 157 195 290 146 161 206
a Perception 390 393 171 156
Production 190 156 156 1 13 141 135
o Perception 429 539 150 185
Production 225 547 97 216
o Perception 356 523 178 193
Production 144 204 481 96 104 120
U Perception 346 405 175 164
Production 194 322 422 71 91 147
u Perception 365 527 184 15 1
Production 220 357 270 76 57 142

 

 

 

 

 

 

 

 

 

 

 

Table 20: Standard deviations for F1 and F2 values of vowels in the perception and
production tasks for native and nonnative speakers. For each vowel, the first line is the
perception value, and the second line is the production value. The column “PB” is the

values from Peterson and Barney (1952). The value for the group with the greatest

variation is in boldface.

99

Another hypothesis of the study was that nonnative speaker perception would be
more native-like than their production. To test this hypothesis, an ANOVA test was
performed, comparing native and normative speakers’ performance on both the
perception and the production tasks. Table 21 summarizes the results of the pairwise
comparison for perception, and Table 22 summarizes the results of the pairwise

comparison for production.

100

 

 

 

 

 

 

 

 

 

 

 

 

I C 8 ae a 9 0 O U u
i 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.297
I 0.000 0.000 0.000 0.000 0.000 0.000 0.000 1.000 0.003
6 0.990 0.000 0.000 0.943 0.000 0.964 0.000 0.000
3 0.000 0.000 1.000 0.034 0.344 0.000 0.000
a 1.000 0.000 0.320 0.000 0.000 0.000
a 0.000 0.081 0.000 0.000 0.000
9 0.085 0.180 0.000 0.000
9 0.000 0.000 0.000
0 0.000 0.000
0 0.004

 

 

 

 

 

 

 

 

 

 

 

H H-

cooowamo

e

0.421
1.

8

0.
1.
0.

28

0.01
0.
0.97
1.

0.
0.
0.
0.
0.
0.
0.
0.

999999999

 

9999999999

 

Table 21: The results of ANOVA tests comparing perception between native and
normative speaker subjects for F1 (Top) and F2 (Bottom). The vowel pairs /e/-/8/, /a/-/o/,

/9/-/:)/ show no significant difference (p>.05) between pairs for both F1 and F2.

101

C

00000000
999999999

H

p—n
O\
’—

999999999
9
y—a

9999999999

6
8
2B
a
9
O
O
U
u

99999999

?

(D
s:

9

t—t
9
U3
U]

91‘999999
92‘9999999

. 0.93
0. . 0.
0. 1. 1. 0. 0.

?9."‘!"‘??????

scooomﬁamo

 

9999999999
99999999

Table 22: The results of ANOVA tests comparing production between native and
normative speaker subjects for F1 (Top) and F2 (Bottom). The vowel pairs /I/—/e/, /9/—/o/,

/O/—/U/ show no significant difference (p>.05) between pairs for both F 1 and F2.

The data in Tables 21 and 22 show the source of the variation between subject
groups in the two tasks. In the perception task, there was no significant difference
between 24 out of 110 pairs of vowels in the F1 formant, and 30 out of 110 for the F2.
For the production task, there were no significant differences between 22 out of 110 pairs

in the F1 formant, and 28 out of 110 for the F2 formant. Three vowel pairs (/e/-/8/, /a/—/o/,

102

/9/-/o/) showed no difference between both groups in both F1 and F2 on the perception
task, and three different pairs of vowels (/e/-/e/, /a/-/o/, /9/—/o/) showed no difference in

both formants between the two groups on the production task. The data thus show no

clear trend favoring either perception or production between groups.

103

Chapter 5: Discussion and Conclusions

5.0. Introduction

In this section, I review the research questions and hypotheses of the study, and
evaluate them in light of the results. Following that, I discuss some of the new findings of
the study, and the implications of the ﬁndings for the fields of linguistics and second
language acquisition. Finally, I discuss the directions that future studies could take to

further the exploration along the lines of this study.

5.1. Review of research questions and hypotheses

The research questions of this study were first given in Chapter 1, section 1.5, and are

repeated here:

1. To what extent can we determine the acoustic properties of the ideal vowel
phoneme of each monothongal vowel in English based on native speakers’
intuitions?

2. To what extent do nonnative speakers of English agree on the ideal vowel
phoneme of each monothongal vowel in English?

3. Are the intuitions of nonnative learners of English similar to the intuitions of

native speakers with regard to the vowel phonemes of English?

104

 

4. How similar is production of vowels in English by both native speakers of
English and normative learners of English to their respective identification of an

ideal vowel?

This study sought to address the research questions. The hypotheses were that
nonnative speakers’ first language would influence their performance on the production
task, and their perception would be more native-like than their production. In Chapter 1,

the following hypotheses were made:

1. Given a perception task in which participants identify an English vowel from a
continuum of F1-F2 combinations (the “perception task”), there will be less
variation among native speakers than among nonnative speakers.

2. In the comparison of the performance of native speakers and normative speakers
on the perception task and a “production task” in which participants produce
words containing the English vowels, there will be less variation between the two
groups’ performance on the perception task than there will be on the production
task.

3. In a perception and a production task, there will be less variation between the two

tasks among native speakers than among nonnative speakers.

105

 

5.2. Evaluation of hypotheses

5.2.1. Hypothesis 1: Native speakers will show less variation than nonnative speakers in

perception

This hypothesis was motivated by the assumption that members of a language
community have a common linguistic competence. If all the native speaker participants
of the study were from the same language community, then they should show similar
responses on the perception task. The data largely supports this hypothesis. There is more
variation among nonnative speakers than among native speakers.

The range in values of responses by native speakers was unexpected, given the
assumption of uniformity that underpinned Hypothesis 1. Other studies have shown a
similar phenomenon, however, suggesting that the variation shown by native speakers in
this study may actually be the norm for native speakers. A wide range in judgments on a
perception task was noted in Frieda et al (1999). The Peterson and Barney (1952) study
showed a wide range of responses in production as well. In light of the results of previous
studies, the wide range of values found in the perception task in this study is not unusual.

The variation of responses within each group of participants of this study was i
measured by calculating the standard deviation for each vowel. The means and standard
deviation values for the F1 and F2 value for each of the eleven monothongal English
vowels that were used in this study are shown in Table 23 below. The table shows the

data by native speaker and nonnative speaker participant groups. The hypothesis was that

106

higher standard deviation from the mean.

the normative speaker group would show more variation, which could be evidenced by a

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Vowel Native Speakers N onnati ve Speakers
Mean Standard Mean Standard
Deviation Deviation
i F 1 346 176 335 127
F2 2320 459 2382 467
I F1 529 137 335 103
F2 2123 450 2419 282
6 F1 632 135 676 182
F2 2220 480 2160 516
8 F1 652 172 7 13 203
F2 2156 405 2174 484
a: F l 803 153 759 212
F2 2138 422 2128 515
a F1 815 163 774 232
F2 1283 391 1244 446
9 F1 652 171 743 156
F2 1331 390 1378 393
9 F1 77 9 150 606 185
F2 1291 429 1220 539
0 F1 613 178 615 193
F2 909 356 989 523
U F1 508 175 400 164
F2 1067 346 1072 405
u F1 409 184 361 151
F2 953 365 1 142 527

 

 

 

 

 

 

Table 23: Mean and standard variation values on the perception task for 11 simple vowel
phonemes of English, native and nonnative speakers. The higher value of native and

normative speakers is shown in boldface.

107

 

 

 

 

 

Table 23 shows in boldface type the higher standard deviation value, whether of
native or nonnative speaker participant group. In only 6 out of 22 of the formant values
there was greater variation among native speaker participants than nonnative speakers.
This data supports Hypothesis 1. As reﬂected in standard deviation, nonnative speaker
participants showed more variation than native speakers in the majority of vowels.

Although greater variation was seen among nonnative speakers in 16 out of 22 of
the formant values, there were some items in which the data from native speakers had
greater within-group variation. Native speakers had greater variability in the F1 formant
for /i/. The tendency among some native speakers of English to prefer a hyperarticulated
/i/ was documented by Johnson et al (1993) and Johnson (2000). It is possible that the
greater variation of /i/ exhibited by native speakers does not reﬂect their actual
competence, but is influenced by what Johnson et al termed the “hyperspace effect,” in
which there is a preference for the /i/ phoneme that is beyond the scope of the speaker’s
normal range for the vowel. However, the greater variation was limited to one formant,
Fl, which reflects the height of the vowel. Native speakers show more consensus for the
dimension of frontness, which is measured by the F2 formant value.

Native speakers also showed more variety for /1/ than nonnatives both in F1 and

F2. This may be a result of the different category distinction between groups. As noted in
Chapter 4, a statistical analysis shows that nonnative speakers seem not to have separate

phonemes for /i/ and /I/. The nonnative group’s /I/ is statistically indistinguishable from

/i/. It seems that the nonnative speaker participants have not differentiated the two vowels

in English, and are quite uniform in their judgments that the two English vowels are the

108

 

same. If that is the case, then a comparison between nonnative /1/ and native /I/ may not

be meaningful.
The other items in which native speakers exhibited more variety were the F 1

values in the vowels /9 U u/.

5.2.2. Hypothesis 2: There will be a less variation between the two groups’ performance

on the perception task than there will be on the production task.

Hypothesis 2 was that there would be less variation between native and nonnative
speakers in perception than in production. This hypothesis goes beyond the observation
that nonnative speakers have foreign accents, and posited that in the course of second
language acquisition, perception could be more native-like than production. This
hypothesis was that foreign accents would affect the nonnative speakers’ production, but
would not necessarily reflect their perceptual competence. The hypothesis was tested by
comparing the two groups’ performance on the two tasks.

The ANOVA test that analyzed the differences between native and nonnative
speakers was discussed in section 4.3.2 of Chapter 4. The results of that test showed the
sources of differences between subject groups. As was mentioned in section 4.3.2, there
was no clear pattern of differences between the comparisons of perception and production

between native and nonnative speakers.

109

5.2.3 Hypothesis 3: There will be a less variation between the performance on the

perception task and the production task by native speakers than by nonnative speakers.

As discussed in section 4.3.1 of Chapter 4, Tables 18 and 19 showed ANOVA
comparisons of perception and production within participant groups. The native speakers
showed much more agreement between perception and production than the nonnative

speakers did.

5.2.4 Summary

Hypothesis 1 was that on the perception task, there would be less variation among
native speakers than among nonnative speakers. Although there was more variation
among the native speakers than was expected on the perception task, there was more
variation among nonnative speakers than native speakers. Hypothesis 1 was confirmed.

Hypothesis 2 was that there would be a closer correlation between the two
groups’ performance on the perception task than on the production task. The data that
was gathered could neither support nor refute Hypothesis 2.

Hypothesis 3 was that there would be a closer correlation between the
performance on the perception task and the production task by native speakers than by

nonnative Speakers. Hypothesis 3 was confirmed.

110

5.3 Additional findings of the study

5.3.1 Production less variable than perception

An unexpected ﬁnding of the study was that there was greater variation in the
perception task than in the production task. Table 20 above showed the standard
deviation value for each vowel, separated by task and participant group. On 18 of the 22
items, participants of both groups showed more internal consistency in the production
task than in the perception task.

While the general clustering of each token was in the correct general area for each
vowel on the perception task, there was much greater variation among participants than
was anticipated. The data for the perception task showed a larger number of outliers for
each vowel than was found in the production task. It could be that participants could not
make an accurate identification using the synthesized vowels. It could also be the case
that perception is more variable among individuals within a speech community than this

study assumed. More research that focuses on variation in perception is needed.

5.3.2 Conflation of /a—9/ in native speakers’ perception

A dialectical distinction in areas of the United States is the conflation of the

vowels /a/ and /o/. Research done by Labov et al (1997) indicates that Michigan is an area

in which the vowels typically show a distinction. The native speaker participants were all
from Michigan, and so would be predicted to maintain the distinction between the vowels.

Indeed, in the production task, native speakers produced statistically distinct vowels for

111

 

 

each vowel. Data from the production task is consistent with Labov et al’s study. Section
4.1 of Chapter 4 gave the results of statistical analyses of differences.

Labov (1994: 363) refers to inconsistency within a participant on vowel merging
as the “Bill Peters effect,” named for a participant who merged vowels on a formal task,
but distinguished them in casual speech. Labov attributes the difference in performance
to the level of formality while speaking. Labov’s study involved reading word lists.
However, a similarity can be drawn to the perception task of this study. Another of his
participants (“Mrs. V,” p. 363) hesitated before reading the cot/caught pair, stating
uncertainty as to the distinction between the two words before she read them aloud. On
the Don/Dawn pair, the participant evaluated the vowel sounds as “slightly different,” but
produced an almost indistinguishable vowel sound when producing them. Although not
the focus of the study, Labov’s work documents a perception-production distinction on

the /a-0/ vowels that this study also found.

5.4 Conclusions

This study examined the phonological systems of Korean learners of English as a
second language, and compared their systems to those of native speakers. While the data
supported some of the hypotheses, some unexpected results give reason to reconsider
some assumptions about language learners and phonological competence. I conclude with

addressing the research questions of Chapter 1.

112

5.4.1. Research question 1: To what extent can we determine the acoustic properties of
the ideal vowel phoneme of each monothongal vowel in English based on native

speakers’ intuitions?

Hypothesis 1 and Hypothesis 2 were predicated on the assumption of the uniform
judgments of an idealized speech community composed of homogenous hearer-speakers.
Although each group was made up of speakers from similar geographic regions and age
groups, the data from the perception task in this study did not reﬂect such uniformity.
The high level of variation suggests that with respect to vowel categories, even within a
speech community it is difficult to define something even as basic as the phonological
categories of the language. The scope of this study did not explore variation in depth,
partly because it was assumed that the amount of variation would be negligible, and so
this ﬁnding should be regarded as tentative.

In spite of the unexpected level of within-group variation, on average, native
speakers of English displayed less variation than the normative speakers did. When asked
to identify a given English vowel sound, native speakers were more uniform in their
choice than nonnative speakers were. It seems that it is more accurate to view the degree
of variation, not its absence or presence, in examining a vowel system. An answer to
Research question 1 could be that we can give a probabilistic prediction of the acoustic

properties of ideal vowels in English, and presumably, other languages as well.

113

 

5.4.2 Research question 2. To what extent do nonnative speakers of English agree on the

ideal vowel phoneme of each monothongal vowel in English?

In one aspect, the data from nonnative speakers was closer to the assumptions of
the study than the data from native speakers. The responses of the perception and
production tasks for nonnative speakers showed more consistency. There was far less
variance between nonnative speakers’ choices of a vowel on the perception task, and their
production of the same vowel.

Another aspect of the normative speaker group, however, did not display native-
like tendencies. The greater within-groupovariation on the perception task among
nonnative speakers did not indicate agreement within the group about the English vowel
categories. On this basis, it is difﬁcult to determine the value of an ideal vowel for
nonnative learners of English. With regards to a match between perception and
production, the nonnative speakers showed less evidence of being a “speech community”
as described in Chomsky (196523) and Saussure (1916: 19) than native speakers did.

Based on the performance of the nonnative speaker group in this study, the
answer to Research question 2 is that nonnative speakers show even less agreement for an

ideal vowel segment than the loose agreement that native speakers show.

114

 

5.4.3 Research question 3: Are the intuitions of normative learners of English similar to
the intuitions of native speakers with regard to the vowel phonemes of English?

The perception task was designed to measure participants’ intuitions about part of
the phonological inventory of English. What they perceived as the correct vowel among
the choices available to them is assumed to be their intuition about that English vowel.
The results of the perception task show that there is a significant correlation between the
intuitions of native and normative speakers. This study indicates that the interlanguage
phonological inventories of the participants are native-like in many respects.

The role of the native language could be a factor in this ﬁnding. Although the
acoustic properties of vowel categories vary across languages, there is overlap in many
vowels. An individual’s intuition about a normative vowel category could be inﬂuenced
by the intuition about the corresponding vowel category in the native language.

N onnative speakers must develop categories for vowels in the second language that do

not exist in the first language. For Korean learners of English, these new vowels are /1, a3,
9, U/. The results of this study show that nonnative speakers do not show evidence of

having formed native-like categories for these new vowels. The data in Table 13 in
section 4.2.1 of Chapter 4 indicates that the intuitions of nonnative speakers about these
new vowels of English are not distinct from vowels that are similar to Korean vowels.
Comparisons of intuitions about new vowel categories in the second language need to

take into account the role of the ﬁrst language.

115

5.4.4 Research question 4: How similar is production of vowels in English by both native
speakers of English and normative learners of English to their respective identification of
an ideal vowel?

Although native and normative speaker groups agreed on F1 and F2 values of
most of the vowels on the perception task, their performance varied on the production
task. A surprising result of this study was the extent to which the normative speakers were
internally consistent in the two tasks, and the extent to which native speakers were not
internally consistent.

As mentioned in the discussion of Hypothesis 3 in section 5.2.3 above, the only
vowels for which native speakers’ pronunciation did not differ significantly from their

selection of ideal vowels were the vowels /a/ and /o/. There were statistically significant

differences between their perception and production in at least one of the F l or F2
formants in each of the other vowels in this study. The relationship between the
perception and pronunciation of English vowels among native speakers in this study was

closer than the same relationship among nonnative speakers.

5.5 Areas for future research

This study was a preliminary exploration of the relationship between perception
and production in native and normative speakers. The study used synthesized speech
samples for stimuli, which allowed a great deal of control over the acoustic properties of
the stimuli. However, we can’t necessarily assume that the participants responded to the

synthetic stimuli in the same way that they would. respond to natural speech. One way to

116

verify that would be to compare participants’ responses to natural speech to their
responses to synthetic speech with the same acoustic values.

This study forced participants to decide on a single, best exemplar of a vowel.
However, the variation found in this study suggests that categories are gradient rather
than categorical. A comparison of the vowel category boundaries in native and nonnative
speakers’ vowel systems, similar to the studies of Scholes (1967, 1968) could shed some

light on the degrees of variation found in both groups.

117

 

References:

Al-Banyan, Ahmed Abdullah M. 1996. The accessibility of universal grammar in
language acquisition: a cross-linguistic perspective. Unpublished doctoral dissertation:
Michigan State University.

Assman P.F. et al. 1982. Vowel identification: orthographic, perceptual, and acoustic
aspects Journal of the Acoustical Society of America 71, 975-982.

Birdsong, David. 1992. Ultimate attainment in second language acquisition. Language 68,
4, 706-755.

Birdsong, David. 1999. Introduction: Whys and why nots of the critical period hypothesis
for second language acquisition. In Birdsong (ed) Second language acquisition and the
critical period hypothesis (pp 1-22). Mahwah, NJ: Lawrence Erlbaum.

Bley-Vroman, Robert. 1989. What is the logical problem of foreign language learning?
in Gass, Susan and Jacqulyn Schacter (eds). Linguistic perspectives on second language
acquisition (pp 41-68). Cambridge: Cambridge University Press.

Boersma, Paul & David Weenink (1999—2000). Praat, a system for doing phonetics by
computer [Computer program]. Web site: www.praat.org.

Berko, J. and R. Brown. 1960. “Psycholinguistic Research Methods.” in Paul M. Mussen,
(ed). Handbook of Research methods in Child Development (pp 517-557). New York:
John Wiley.

Bohn, Ocke-Schwen and James Emil Flege. 1997. Perception and production of a new
vowel category by adults second language learners. In Leather, Jonathan and Allan James
(ed). Second-Language Speech: Structure and Process (pp. 53-74).Berlin: Mouton de
Gruyter.

Bosch, Laura and Nﬁria Sebastian-Gallés. Native-language recognition abilities in 4-
month-old infants from monolingual and bilingual environments. Cognition 65, 33-69.

Bradlow, Ann R. 1995. A Comparative Acoustic Study of English and Spanish Vowels.
Journal of the Acoustical Society of America 97,3, 1916- 1924.

Broselow, Ellen. 1984. An investigation of transfer in second language phonology.
International Review of Applied Linguistics 22, 253-269.

Broselow, Ellen. 1988. An investigation of transfer in second language phonology.
Studies in Descriptive Linguistics 17, 77-93.

118

Broselow, Ellen. 1992. Transfer and universals in second language acquisition. In Gass,
Susan and Larry Selinker (eds) Language transfer in language learning (pp 71-
86).Amsterdam: John Benjamins.

Broselow, Ellen and Daniel Finer. 1991. Parameter setting in second language phonology
and syntax. Second Language Research 7, 35—59.

Chomsky, Noam. 1964. Current issues in linguistic theory. The Hague: Mouton.
Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, M.I.T. Press.

Chomsky, Noam. 1986. Knowledge of language: its nature, origin, and use. Westport,
CT: Praeger.

Chomsky, Noam. 1995. The minimalist program. Cambridge, MA: The MIT Press.

Chomsky, Noam and Miller, 1963. Introduction to the formal analysis of natural
languages in Luce R. Duncan, Robert R. Bush and Eugene Galanter (eds). Handbook of
mathematical psychology (pp 269—321).New York: Wiley.

Clahsen, Harold. & P. Muysken (1986), The availability of universal grammar to adult

and child learners: A study of the acquisition of German word order, Second Language
Research 2, 93-119.

Clahsen, Harold. & P. Muysken (1989), The UG paradox in L2 acquisition. Second
Language Research 5, 1-29.

Coppieters, Ren_. 1987. Competence differences between native and normative speakers.
Language 63, 544—573.

Delattre, P.C., A.M Liberman, ES. Cooper. 1955. Acoustic loci and transitional cues for
consonants. Journal of the Acoustical Society of America 27, 769-773.

Diller, Karl Conrad. 1978. The language teaching controversy. Rowley, MA: Newbury
House.

Doucherty, Gerard and Paul Foulkes. 2000. Speaker, speech, and knowledge of sounds in
Burton-Roberts, Noel, Philip Carr, and Gerard Docherty (eds) Phonological knowledge:
conceptual and empirical issues (pp 105-130). Oxford: Oxford University Press.

Eckman, Fred. 1977. Markedness and the contrastive analysis hypothesis. Language
Learning 27, 315-330.

Eckman, Fred R. 1987. The reduction of word—final consonant clusters in interlanguage.
In James, Allen and Jonathan Leather (eds). Sound patterns in second language
acquisition (pp 143-162). Dordrecht: Foris.

119

Eckman, Fred R. and Gregory K. Iverson. 1994. Pronunciation difﬁculties in ESL: Coda
consonants in English interlanguage. In M. Yavas (ed). First and second language
phonology (pp 251—265).San Diego: Singular Press.

Eimas, P. D., Siqueland, E. R., Jusczyk, P., and Vigorito, J. (1971). Speech perception in
infants. Science 171, 303-306.

Ellis, Rod. 1994. The study of second language acquisition. Oxford: Oxford University
Press.

Fant, Gunthar. 1960. Acoustic theory of speech production. The Hague: Mouton.

Flege, James Emil. 1987. The Production of "New" and "Similar" Phones in a Foreign
Language: Evidence for the Effect of Equivalence Classification. Journal of Phonetics 15
(1), 47-65.

Flege, James Emil. 1991. Perception and production: the relevance of phonetic input to
L2 phonological learning. In T. Huebner and CA. Ferguson (eds.) Crosscurrents in
Second Language Acquisition (pp 249-289). Amsterdam: Benjamins.

Flege, James Emil. 1995 Second language speech learning: Theory, findings and
problems. in Strange, Winifred. (ed). Speech perception and linguistic experience:
Theoretical and methodological issues (pp 233-272). Timonium, MD: York Press.

Flege, James Emil. 1997. English vowel production by Dutch talkers: more evidence for
the similar vs new distinction. In Leather, Jonathan and Allan James (ed). Second-
Language Speech: Structure and Process (pp 11-52). Berlin: Mouton de Gruyter.

Flege, James Emil. 1999. Age of learning and second-language speech. In Birdsong,
David P. (ed), Second language acquisition and the critical period hypothesis (101-132).
Mahwah, NJ: Erlbaum.

Flege, James Emil and Robert Hillenbrand. 1984. Limits on pronunciation accuracy in
adult foreign language speech production. Journal of the acoustical society of America 76,
708-721.

Flege, James Emil; Bohn, Ocke-Schwen; Jang, Sunyoung. 1997. Effects of Experience on
Non-Native Speakers' Production and Perception of English Vowels. Journal of

Phonetics 25, 437-470.

Fodor, Jerry A. 1983. The Modularity of Mind. Bradford Books. MIT Press, Cambridge,
MA.

Foster-Cohen, Susan. 1999. An introduction to child language development. London:
Longman.

120

Frieda, Elaina M, Amanda C. Walley, James E. Flege, Michael E. Sloane. 1999. Adults’
Perception of Native and N onnative Vowels: Implications for the Perceptual Magnet
Effect. Perception and Psychophysics 61 (3), 561-577.

Fromkin, Victoria and Robert Rodman. 1993. An introduction to language. Fort Worth:
Harcourt Brace College Publishers.

Fry, D.B., Arthur S. Abramson, Peter D. Eimas, and Alvin M. Liberman. 1962. The
identification and discrimination of synthetic sounds. Language and Speech 5, 171-189.

Gass, Susan. 1984. Development of speech perception and speech production abilities in
adult second language learners. Applied Psycholinguistics 5, 51—74.

Gass, Susan. 1996. Second language acquisition and linguistic theory: the role of
language transfer. In Ritchie, William C. and Tej K. Bhatia (eds). Handbook of second
language acquisition (pp 317-345). San Diego: Academic Press.

Gass, Susan and Larry Selinker. 1994. Second language acquisition: an introductory
course. Hillsdale, NJ: Lawrence Erlbaum.

Gerstman, L.H. 1968 Classification of self-normalized vowels. Institute of Electrical and
Electronic Engineers, Transactions on Audio Electroacoustics 16, 78-80.

Gordon, Peter C., Lisa Keyes and Yiu-fai Yung. 2001. Ability in perceiving nonnative
contrasts: Performance on natural and synthetic speech stimuli. Perception and
Psychophysics 63, 746-758.

Gregg, Kevin 1989. Second language acquisition theory: the case for a generative
perspective. In Gass, Susan and Jacqulyn Schacter (eds). Linguistic perspectives on
second language acquisition (pp 1540). Cambridge: Cambridge University Press.

Harnad, Stevan. 1987. Psychophysical and cognitive aspects of categorical perception: A
critical overview. In Harnad, Stevan (ed.) Categorical Perception: The Groundwork of
Cognition (pp 1-52). New York: Cambridge University Press.

Herschensohn, Julia Rogers. 2000. The second time around: minimalism and L2
acquisition. Philadelphia, PA: J. Benjamins.

Hillenbrand, James M., Getty, L.A., Clark, M.J., and Wheeler, K. (1995). Acoustic
characteristics of American English vowels. Journal of the Acoustical Society of America,
97, 3099-3111.

Hindle, D. 1978. Approaches to Vowel Normalization in the Study of Natural
Speech. In: D. Sankoff (ed). Linguistic Variation: Models and Methods (pp 161-172).
New York. Academic Press.

121

Hockett, 1955. A manual of phonology. Baltimore, Waverly Press.

Howell, David C. 1995. Fundamental statistics for the behavioral sciences, 3rd edition.
Belmont, CA: Duxbury Press.

Ingram, David. 1989. First language acquisition: Method, description, and explanation.
Cambridge: Cambridge University Press.

Ingram, David. 1999. Phonological acquisition. In Barrett, Martyn (ed). The development
of language. Hove, East Sussex, UK: Psychology Press.

Ingram, John CL. and See-Gyoon Park. 1996. Inter-language vowel perception and
production by Korean and Japanese listeners. Proceedings of the Fourth International
Conference on Spoken Language Processing, Philadelphia, PA.

Iverson, Peter, Patricial K. Kuhl, Reiko Akahane-Yamada, Eugen Diesch, Yoh-ichi
Tohkura, Andreas Kettermann, and Claudio Siebert. 2003. A perceptual interference
account of acquisition difﬁculties for non-native phonemes. Cognition 87, B47—B57.

Jackendoff, Ray. 1994. Patterns in the mind: Language and human nature. New York:
BasicBooks.

Jackobson, Roman. 1968 [1941]. Child Language, Aphasia, and Phonological Universals.
The Hague, Paris: Mouton.

James, Allan R. 1988. The acquisition of a second language phonology: a linguistic
theory of developing sound structures. Tubingen: G. Narr.

Johnson, J. & Newport, Elizabeth. 1989. Critical period effects in second language
learning: the inﬂuence of maturational state on the acquisition of ESL. Cognitive
Psychology 12, 60-99.

Johnson, J. & Newport, Elizabeth. 1991. Critical period effects on universal properties of

language: the status of subjacency on in the acquisition of a second language. Cognition
39, 215-258.

Johnson, Keith. 2000. Adaptive dispersion in vowel perception. Phonetica 57, 181-188.

Johnson, Keith, E. Flemming, and R. Wright. 1993. The hyperspace effect: Phonetic
targets are hyperarticulated. Language 69: 505-528.

Joos, Martin. 1948. Acoustic Phonetics. Language Monographs 23. Baltimore: Linguistic
Society of America.

122

 

Jusczyk, Peter. 1986. Speech perception. In Handbook of perception and human
performance. New York: John Wiley and Sons.

 

Katamba, Francis. 1989. An introduction to phonology. London: Longman.

Kenstowicz, Michael. 1994. Phonology in generative grammar. Cambridge, MA:
Blackwell.

Kenstowicz, Michael. 2003. Salience and Similarity in Loanword Adaptation: a Case
Study from Fijian. ROA-609-0803, Rutgers Optimality Archive, http://roa.rutgers.edu.

Kluender, K. R., Diehl, R. L., & Killeen, P. R. 1987. Japanese quail can learn phonetic
categories. Science 237, 1195-1197.

Kluender, K. R., Lotto, A. J. 1994. Effects of first formant onset frequency on

[- voice ] judgments result from auditory processes not specific to humans. Journal of the
Acoustical Society of America 95, 1044—1052.

 

Kuhl, Patricia K. 1980. Perceptual constancy for speech sound categories in early infancy
in Yeni-Komshian, Grace H., James F. Kavanagh, Charles A. Ferguson (eds) Child
Phonology New York: Academic Press.

Kuhl, Patricia K. 1981. Auditory category formation and development of speech
perception. In Stark, Rachel (ed), Language behavior in infancy and early childhood.
New York: Elsevier North—Holland.

Kuhl, Patricia K. 1987. The special-mechanisms debate in speech research:
Categorization tests on animals and infants. in Stevan Harnad, (ed). Categorical
perception: The groundwork of cognition (pp 355-386). New York: Cambridge
University Press.

Kuhl, Patricia K. 1991. Human adults and human infants show a Perceptual magnet effect
for the prototypes of speech categories, monkeys do not. Perception and Psychophysics
50, 93-107.

Kuhl, Patricia K. and James D. Miller. 1975. Speech perception by the chinchilla: voiced-
voiceless distinction in alveolar plosive consonants. Science 190, 4209, 69-72.

Kuhl, Patricia K. Karen A. Williams, Francisco Lacerda, Kenneth N. Stevens, and Bjorn
Lindblom. 1992. Linguistic Experience Alters Phonetic Perception in Infants by 6
Months of Age. Science 255, 606-608.

Kuhl, Patricia K. and Peter Iverson. 1995. Linguistic experience and the Perceptual
magnet effect. in Strange, Winnifred (ed). Speech perception and linguistic experience:
issues in cross-language research (pp 121-154). Baltimore: York Press.

123

Labov, William. 1994. Principals of linguistic change: Internal factors. Cambridge, MA:
Blackwell.

Labov, William, Malcah Yaeger, and Richard Steiner. 1973. The quantitative study of
sound change in progress. Philadelphia: US. Regional Survey.

Labov, William, Sharon Ash and Charles Boberg. 1997. A National Map of the Regional
Dialects of American English. The Linguistics Laboratory, Department of Linguistics,
University of Pennsylvania. Retreived November 5, 2003 from
http://www.ling.upenn.edu/phono_atlas/NationalMap/NationalMap.html

Ladefoged, Peter. 1993. A course in phonetics. Orlando: Harcourt Brace & Company.

Ladefoged, Peter. 2001. Vowels and consonants: an introduction to the sounds of
languages. Malden, Mass: Blackwell.

Ladefoged, Peter and Broadbent, Donald. 1957. Information conveyed by vowels.
Journal of the Acoustical Society of America 29, 98-104.

Lado, Robert. 1957. Linguistics across cultures. Ann Arbor: University of Michigan
Press.

Lasky, R. E., Syrdal-Lasky, A., & Klein, R. E. (1975). VOT discrimination by four to six
and a half month old infants from Spanish environments. Journal of Experimental Child
Psychology, 20, 215-225.

Leather, Jonathan and Allan James. 1996. Second language speech. In Ritchie, William C.
and Tej K. Bhatia (eds). Handbook of second language acquisition (pp 269-316). San
Diego: Academic Press.

Leather, Jonathan. 1997. Interrelation of perceptual and productive learning in the initial
acquisition of second-language tone. In Leather, Jonathan and Allan James (ed). Second-

Language Speech: Structure and Process (pp 75-101). Berlin: Mouton de Gruyter.

Lehiste, I. And D. Meltzer. 1973. Vowel and speaker identiﬁcation in natural and
synthetic speech. Language and Speech 16, 356-364.

Lenneberg, E. 1967. Biological foundations of language. New York: Wiley.

Liberman, Alvin. 1970. Some characteristics of perception in the speech mode.
Perception and its Disorders 48, 23 8—254.

Liberman, Alvin.M., K. Harris, H. Hoffman, and B. Grifﬁth. 1957. The discrimination of

speech sounds within and across phoneme boundaries. Journal of Experimental
Psychology 54, 358-368.

124

 

Lin, Yuh-Huey. 2001. Syllable Simpliﬁcation Strategies: A Stylistic Perspective.
Language Learning 51, 681-718.

Lisker, L. and A. Abramson. 1964. A cross language study of voicing in initial stops:
acoustical measurements. Word 20, 384-422.

Lisker, L. and A. Abramson. 1967. Some effects of context on voice onset time in
English stops. Language and Speech 10, 1-28.

Liu, Dilin and Johanna L. Gleason. 2002. Acquisition of the article the by nonnative
speakers of English. Studies in Second Language Acquisition 24, 1-33.

Long, Michael. 1990: Maturational constraints on language development. Studies in
Second Language Acquisition 12, 251-285.

Major, Roy. 1997 . A model for interlanguage phonology. In Ioup, Georgette and Steven
H. Weinberger (eds) Interlanguage phonology (pp 101-124). Cambridge, MA: Newbury.

Major, Roy. 2001. Foreign accent: the ontogeny and phylogeny of second language
phonology. Mahwah, NJ: Laurence Erlbaum associates.

Major, Roy and Faudree, Michael C. 1996. Markedness Universals and the Acquisition of
Voicing Contrasts by Korean Speakers of English. Studies in Second Language
Acquisition, 1996, 18, 1, Mar, 69-90

Miller, J .D. 1989. Auditory perceptual interpretation of the vowel. Journal of the
Acoustical Society of America 85, 2114-2134.

Nearey, TM. 1989. Static, dynamic, and relational properties in vowel perception.
Journal of the Acoustical Society of America 85, 2088-2113.

Odlin, Terence. 1989. Language transfer: Cross-linguistics inﬂuence on language
learning. Cambridge: Cambridge University Press.

Patkowski, Mark. 1990. Age and Accent in a Second Language: A Reply to James Emil
Flege. Applied Linguistics 11, 1 73-89.

Peterson, GE. 1951. The phonetic value of vowels. Language 27, 541

Peterson, GE. and H.L. Barney. 1952. Control methods used in a study of the vowels
Journal of the Acoustical Society of America 24, 175-184.

Pinker, Steven. 1994. The language instinct: How the mind creates language. New York:
W. Morrow and Co.

125

Polka, Linda and Janet F. Werker. 1994. Developmental changes in perception of

normative vowel contrasts. Journal of Experimental Psychology: Human Perception and
Performance 20, 2, 421-435.

Prince, Alan, and Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in
Generative Grammar. RuCCS Technical Report 2. Piscataway, NJ: Rutgers Center for
Cognitive Science, Rutgers University, and Boulder, CO: Department of Computer
Science, University of Colorado.

Repp, EH. 1984. Categorical perception: Issues, methods, ﬁndings. In Lass, N.J., ed.
Speech and language: Advances in basic research and practice vol 10. New York:
Academic Press.

Riney, Timothy and Naoyuki Takagi. 1999. Global foreign accent and Voice onset time
among Japanese EFL speakers. Language Learning 42, 275-302.

Saussure, Ferdinand de. 1916. A course in general linguistics. Edited by Charles Bally
and Albert Reidlinger. Translated from the French by Wade Baskin. New York,
Philosophical Library.

Schachter, Jacquelin. 1989. Testing a proposed universal. In Gass, Susan and Jacqueline
Schachter (eds). Linguistic perspectives on second language acquisition (pp 73-88).
Cambridge: Cambridge University Press.

Schachter, Jacquelin. 1996. Maturation and the issue of universal grammar in second
language acquisition. In Ritchie, William C. and Tej K. Bhatia (eds). Handbook of second
language acquisition (pp 159-193). San Diego: Academic Press.

Schwartz, Jean-Luc, Louis-Jean Boe, Nathalie Vallde and Christian Abry. 1997 . Major
trends in vowel system inventories. Journal of Phonetics 25, 233-253.

Scholes, Robert J. 1967a. Phoneme categorization of synthetic vocalic stimuli by
speakers of Japanese, Spanish, Persian and American English. Language and Speech 10,
1, 46-68.

Scholes, Robert J. 1967b. Categorical responses to synthetic vocalic stimuli by speakers
of various languages. Language and Speech 10, 252-282.

Scoles, Robert J. 1968. Phonemic interference as a perceptual phenomenon. Language
and Speech 11, 86-103.

Selinker, Larry. 1972. Interlanguage. International Review of Applied Linguistics 10,
209-231.

126

Sheldon, Amy and Winifred Strange. 1982. The acquisition of /r/ and /l/ by Japanese
learners of English: Evidence that speech production can precede perception. Applied
‘ Psycholinguistics 3, 243-261.

Silverman, Daniel. 1992. Multiple scansions in loanword phonology: evidence from
Cantonese. Phonology 92:289-328.

Sorace, Antonella. 1996. The use of acceptability judgments in second language
acquisition research in Ritchie, William C. and Tej K. Bhatia (eds). Handbook of second
language acquisition (pp 375-409). San Diego: Academic Press.

Stevens, Kenneth N. 1998. Acoustic phonetics. Cambridge, The MIT Press.

Stockwell, Robert P. and J. Donald Bowen. 1965. The Sounds of English and Spanish.
Chicago: Chicago University Press.

Strozer, J .R. 1994. Language acquisition after puberty. Washington, DC: Georgetown
University Press.

Takagi, Naoyuki. 2002. The Limits of Training Japanese Listeners to Identify English /r/
and /l/: Eight Case Studies. The Journal of the Acoustical Society of America 111, 2887-
2896.

Trubetzkoy, Nikoli. S. 1969. Grundziige der Phonologie. Translated by Christiane A. M.
Baltaxe. Berkeley: University of California Press.

Vihman, Mary May. 1996. Phonological development: the origins of language in the
child. Cambridge, MA: Blackwell.

Wayland, Ratree. 1997. Non-native production of Thai: acoustic measurements and
accentedness ratings. Applied Linguistics, 18, 3, 345-373.

Weinberger, Steven. 2003. Foreign accent archive. Retrieved June 26, 2003 from http://
classweb.gmu.edu/accent/

Werker, Janet F. 1994. Cross-language speech perception: Developmental change does
not involve loss. In Goodman, Judith C. and Howard C. Nusbaum (eds). The development
of speech perception: The transition from speech sounds to spoken words (93-120).
Cambridge, MA: MIT Press.

Werker, Janet F. and RC. Tees. 1984. Cross-language speech perception: Evidence for a

perceptual reorganization during the ﬁrst year of life. Infant behavior and development 7,
49-64.

Werker, Janet E. and Chris B. Lalonde. 1988. Cross-Language Speech Perception: Initial
Capabilities and Developmental Change. Developmental Psychology 24—5, 672—683.

127

 

Wexler, Ken. & Manzini, M. 1987 . “Parameters and Learnability in Binding Theory”, in
‘ T. Roeper & E. Williams (eds.), Parameters and Linguistic Theory (pp 41-76). Dordrecht,
The Netherlands: Reidel.

White, Lydia. 1989. Universal grammar and second language acquisition. Amsterdam:
John Benjamins.

White, Lydia. (2000). Second Language Acquisition: From Initial to Final State. in
Archibald, J. (ed.). Second Language Acquisition and Linguistic Theory (pp 130-155).
Blackwell Publishers: Malden, Massachusetts.

White, Lydia. 2003. Second language acquisition and universal grammar. Cambridge:
Cambridge University Press.

White, Lydia and Fred Genesee. 1996. How native is near-native? The issue of ultimate
attainment in adult second language acquisition. Second Language Research 12, 233-265.

Willerman, Raquel and Patricia K. Kuhl. 1996. Cross-language speech perception:
Swedish, English, and Spanish Speakers’ perception of front rounded vowels. In H.
Timothy Bunnell & William Idsardi (eds) Proceedings of ICSLP 96 1: 442-445.

Yamada, Reiko, Yoh-ichi Tohkura and Noriko Kobayashi. 1997. Effect on word
familiarity on non-native phoneme perception: identiﬁcation of English /r/, /l/ and /w/ by
native speakers of Japanese. In Leather, Jonathan and Allan James (ed). Second-
Language Speech: Structure and Process (pp 103-117). Berlin: Mouton de Gruyter.

Yang, Byonggon. 1996. A comparative study of American English and Korean vowels
produced by male and female speakers. Journal of Phonetics 24, 245-261.

128

 

 

 

.. nYTr. $7.vi . .

5.1:.» X.

 

4.." p...) with?

 

 

 

5.511;.
0..

3:.
.. .. :1?
x; 1.5:: . . .
1:. .